2010 Mathematics Subject Classification. 35Q35, 76B15.
Key words and phrases. hydroelastic waves; fluid–structure interaction; asymptotic models; well-posedness.
Introduction
The interaction between incompressible fluids and deformable elastic structures stands as one of the most challenging areas in modern mathematical analysis of partial differential equations. The past two decades have seen a substantial expansion of the field, driven by applications ranging from naval hydrodynamics [3 ] and aeroelasticity to the mathematics of oceanic waves interacting with sea ice or floating plates, [28 ] . At a mathematical level, these problems combine free-boundary fluid dynamics with the nonlinear geometry and high-order mechanics of thin structures, leading to systems in which hyperbolic (inertia), dispersive (bending), and sometimes dissipative (viscoelastic) effects interact in a genuinely nonlocal way.
In this context, one of the most prominent and physically relevant settings is hydroelasticity, where surface waves propagate beneath a deformable medium such as sea ice or an elastic plate, and where geometric shell models provide a natural framework to describe the elastic forces. In contrast with classical water-wave theory, in which the interface is massless and obeys the Bernoulli dynamic condition, the hydroelastic interface carries mass, elasticity, and, in many situations, damping. A fundamental reference in this direction is the work of Plotnikov and Toland [27 ] , who derived a general geometrically nonlinear model for an elastic shell of Cosserat type interacting with an irrotational Euler fluid. Existing studies of hydroelastic waves are primarily devoted to traveling-wave solution, either establishing their existence or computing them, most notably in the massless case [29 ] and, when the interface mass is included, in [9 , 30 ] . In two spatial dimensions the situation has also been extensively studied, most notably in the vortex-sheet formulation introduced by Ambrose and Liu [7 ] . There, the interface is assumed to be a one-dimensional curve with elastic properties, and the fluid velocity is described by the Birkhoff–Rott integral. Despite the apparent simplicity of the geometry, the coupling between elasticity and the Kelvin–Helmholtz instability of the vortex sheet leads to a highly delicate system. The authors obtained local well-posedness in Sobolev spaces and studied regularity mechanisms arising from the elastic terms.
Beyond hydroelasticity, several lines of work have addressed fluid–structure interaction problems involving three-dimensional incompressible fluids and deformable solids. One classical approach, initiated by Coutand and Shkoller [15 ] , employs a fully Lagrangian framework for the elastodynamics of the solid and transforms the fluid equations to the material coordinates of the structure. This method has proven particularly effective for fluid–solid systems in which the structure retains a volumetric geometry. More recently, Kukavica, Tuffaha, and Ziane [22 , 23 ] developed a sharp regularity theory for incompressible fluids interacting with elastic solids, establishing local well-posedness under minimal Sobolev assumptions. Their work highlights the need for precise control of the deformation of the elastic region, the mapping of the fluid equations to a moving coordinate system, and the geometric terms appearing in the transformed system.
In the context of thin structures, models based on the Koiter shell energy have emerged as a natural limit of three-dimensional elasticity. The work of Muha and Canić [26 ] represents a landmark result in this direction: they establish the existence of weak solutions for the coupling between a viscous incompressible fluid and a Koiter shell, using an operator-splitting approach together with Aleksandrov transformations of the moving domain. Their setting assumes a cylindrical structure and exploits symmetries to obtain compactness and energy bounds that are otherwise difficult to establish. See also [24 ] where the existence theory has also been extended to thermally coupled fluid–structure interaction models.
A further development is due to Balakrishna, Kukavica, Muha, and Tuffaha [10 ] , who studied the interaction between the three-dimensional Euler equations and a nonlinear two-dimensional Koiter plate represented as a graph. Their model incorporates the full geometric nonlinearity of the Koiter energy, the inertia of the plate, and Kelvin–Voigt damping, and it is posed on a time-dependent fluid domain whose upper boundary is the unknown graph. Related inviscid free-boundary fluid–structure models, including variants with alternative structural laws, different solution concepts, and global-in-time results in weak frameworks have also been investigated in [1 , 2 , 8 , 20 , 21 , 31 ] .
Contributions and main results
This work addresses a three-dimensional inviscid fluid–structure interaction problem in which an incompressible Euler flow evolves beneath a nonlinear viscoelastic plate represented as a graph. Starting from the geometric Euler–plate system (2.14), we pursue a twofold program. On the one hand, we derive reduced interface equations in a weakly nonlinear, small-steepness regime, obtaining asymptotic models that capture the leading hydroelastic dynamics. On the other hand, we provide a well-posedness theory for these reduced equations. A similar program has been implemented for fluid-solid interaction problems in the case of Stokes law [11 , 12 ] and Darcy law [6 ] .
Assuming irrotationality, we reduce the Euler–plate system to a boundary formulation and perform a weakly nonlinear small-steepness expansion in ε = H / L \varepsilon= H/L ε = H / L . Truncating at quadratic accuracy yields two closed bidirectional interface equation for the renormalized displacement f = η ( 0 ) + ε η ( 1 ) f = \eta^{(0)} + \varepsilon\eta^{(1)} f = η ( 0 ) + ε η ( 1 ) ,
( I + T Λ ) f t t + δ Λ 3 f t + ( Λ + β 4 Λ 5 ) f = ε M [ f ] , (1) \left(I+\mathcal{T}\Lambda\right)f_{tt}+\delta\Lambda^{3}f_t+\left(\Lambda+\frac{\beta}{4}\Lambda^{5}\right)f=\varepsilon\mathcal{M}[f], \tag*{(1)} ( I + T Λ ) f tt + δ Λ 3 f t + ( Λ + 4 β Λ 5 ) f = ε M [ f ] , (1)
where the quadratic forcing M [ f ] \mathcal{M}[f] M [ f ] depends on the bidirectional model under consideration. On the one hand, the first bidirectional model has a nonlinearity given by (73) . This first bidirectional model is doubly nonlinear in the sense that there is a nonlinear elliptic operator acting on the acceleration f t t f_{tt} f tt of the membrane. On the other hand, a second bidirectional model (with the same accuracy) has a nonlinearity given by explicitly in (80) . Moreover, by restricting to one horizontal dimension and introducing the characteristic variables
ξ = x − t , τ = ε t , f ( x , t ) = F ( ξ , τ ) , \xi=x-t,\qquad\tau=\varepsilon t,\qquad f(x,t)=F(\xi,\tau), ξ = x − t , τ = εt , f ( x , t ) = F ( ξ , τ ) ,
we obtain two unidirectional slow-modulation models. The first one, associated with the quadratic-precision bidirectional equation (73) , can be written as
F τ = 1 ϵ ( a + b H ) ( ( I + γ Λ ) F ξ + β 4 Λ F ξ ξ ξ + δ Λ F ξ ξ + Λ F ) − ( a + b H ) ( 2 F ξ Λ F − [ Λ , F ] F ξ ) , (2) F_{\tau}=\frac{1}{\epsilon}(a+b\mathcal{H})\left((I+\gamma\Lambda)F_{\xi}+\frac{\beta}{4}\Lambda F_{\xi\xi\xi}+\delta\Lambda F_{\xi\xi}+\Lambda F\right)-(a+b\mathcal{H})(2F_{\xi}\Lambda F-[\Lambda,F]F_{\xi}), \tag*{(2)} F τ = ϵ 1 ( a + b H ) ( ( I + γ Λ ) F ξ + 4 β Λ F ξ ξ ξ + δ Λ F ξ ξ + Λ F ) − ( a + b H ) ( 2 F ξ Λ F − [ Λ , F ] F ξ ) , (2)
where a = a ( Λ , ∂ ξ ) a=a(\Lambda,\partial_{\xi}) a = a ( Λ , ∂ ξ ) and b = b ( Λ , ∂ ξ ) b=b(\Lambda,\partial_{\xi}) b = b ( Λ , ∂ ξ ) are the real Fourier multipliers defined in (91) (all multipliers understood with respect to ξ \xi ξ ). The second one, associated with the alternative quadratic closure (79) , takes the form
F τ = 1 ϵ ( α + γ H ) [ ( I + γ Λ ) F ξ ξ + β 4 Λ F ξ ξ ξ ξ + δ Λ F ξ ξ + Λ F ] − ( α + γ H ) [ 1 2 Λ ( F ξ 2 − ( Λ F ) 2 ) − [ Λ , F ξ ] F ξ + F ξ ξ Λ F + [ Λ , F ] T F − F ξ H T F + β 4 ( [ Λ , F ] T F ξ ξ ξ − F ξ Λ T F ξ ξ ) + δ ( [ Λ , F ] T F ξ ξ − F ξ Λ T F ξ ) ] , (3) \begin{aligned}
F_{\tau}={}&\frac{1}{\epsilon}(\alpha+\gamma\mathcal{H})\left[(I+\gamma\Lambda)F_{\xi\xi}+\frac{\beta}{4}\Lambda F_{\xi\xi\xi\xi}+\delta\Lambda F_{\xi\xi}+\Lambda F\right]\\
&-(\alpha+\gamma\mathcal{H})\left[\frac{1}{2}\Lambda(F_{\xi}^{2}-(\Lambda F)^{2})-[\Lambda,F_{\xi}]F_{\xi} + F_{\xi\xi}\Lambda F+[\Lambda,F]\mathcal{T}F-F_{\xi}\mathcal{H}\mathcal{T}F\right.\\
&\left.\qquad+\frac{\beta}{4}\left([\Lambda,F]\mathcal{T}F_{\xi\xi\xi}-F_{\xi}\Lambda\mathcal{T}F_{\xi\xi}\right)+\delta\left([\Lambda,F]\mathcal{T}F_{\xi\xi}-F_{\xi}\Lambda\mathcal{T}F_{\xi}\right)\right], \tag*{(3)}
\end{aligned} F τ = ϵ 1 ( α + γ H ) [ ( I + γ Λ ) F ξ ξ + 4 β Λ F ξ ξ ξ ξ + δ Λ F ξ ξ + Λ F ] − ( α + γ H ) [ 2 1 Λ ( F ξ 2 − ( Λ F ) 2 ) − [ Λ , F ξ ] F ξ + F ξ ξ Λ F + [ Λ , F ] T F − F ξ H T F + 4 β ( [ Λ , F ] T F ξ ξ ξ − F ξ Λ T F ξ ξ ) + δ ( [ Λ , F ] T F ξ ξ − F ξ Λ T F ξ ) ] , (3)
where α = α ( Λ , ∂ ξ ) \alpha=\alpha(\Lambda,\partial_{\xi}) α = α ( Λ , ∂ ξ ) and γ = γ ( Λ , ∂ ξ ) \gamma=\gamma(\Lambda,\partial_{\xi}) γ = γ ( Λ , ∂ ξ ) are the real Fourier multipliers introduced in (96) , and T = ( I + γ Λ ) − 1 Λ \mathcal{T}=(I+\gamma\Lambda)^{-1}\Lambda T = ( I + γ Λ ) − 1 Λ is defined in (76) .
The second purpose of this paper is to establish a well-posedness theory for the reduced models derived above. More precisely:
f 0 ∈ H 0 3 ( T 2 ) , f t ( ⋅ , 0 ) ∈ H 0 1 ( T 2 ) , ∥ f 0 ∥ H 3 sufficiently small , f_{0}\in H_{0}^{3}(\mathbb{T}^{2}),\qquad f_{t}(\cdot,0)\in H_{0}^{1}(\mathbb{T}^{2}),\qquad\|f_{0}\|_{H^{3}}\ \text{sufficiently small}, f 0 ∈ H 0 3 ( T 2 ) , f t ( ⋅ , 0 ) ∈ H 0 1 ( T 2 ) , ∥ f 0 ∥ H 3 sufficiently small ,
see Theorem 4.1 . A key difficulty in the bidirectional model (73) is that, once written as the first-order system (98) , the coupling involves the unknown time derivative through N ( f , ∂ t v ) \mathcal{N}(f,\partial_{t}v) N ( f , ∂ t v ) , and the problem is not a standard semilinear evolution and in fact it has a doubly nonlinear structure akin to
v t + N 1 ( v t ) = N 2 ( v ) , v_{t}+N^{1}(v_{t})=N^{2}(v), v t + N 1 ( v t ) = N 2 ( v ) ,
with N 1 N^{1} N 1 being a nonlinear elliptic operator. Our approach is therefore based on a two-parameter regularization and a nested fixed-point scheme; see Steps 1–5 in the proof of Theorem 4.1 .
Unidirectional local well-posedness and small-data global decay (first reduction). For the unidirectional slow-modulation model (92) (equivalently (2) ), we establish local well-posedness for arbitrary mean-zero data F 0 ∈ H 2 ( T ) F_{0}\in H^{2}(\mathbb{T}) F 0 ∈ H 2 ( T ) ; see Theorem 5.1 . The argument relies on a priori energy estimates in the symmetrized formulation together with a classical Friedrichs mollification scheme. Moreover, for sufficiently small H 2 H^{2} H 2 data we prove global-in-time existence and exponential decay of the natural energy; see Theorem 5.2 .
Global small-data well-posedness (second reduction). For the more singular unidirectional model (97) (equivalently (3) ), we obtain global well-posedness for sufficiently small mean-zero data F 0 ∈ H 3 ( T ) F_{0}\in H^{3}(\mathbb{T}) F 0 ∈ H 3 ( T ) ; see Theorem 5.3 . The proof is based on an energy method with a bootstrap/absorption
step which allows the highest-order nonlinear terms to be controlled by the linear dissipation as long as the solution remains small.
Notation and preliminary estimates
In what follows we collect the notation and the analytic tools that will be used throughout the paper, in particular basic properties of the Fourier multipliers involved and the Sobolev/product estimates needed in the well-posedness analysis.
Notation and functional setting. We work with horizontal variables x = ( x 1 , x 2 ) ∈ T 2 x=(x_1,x_2)\in\mathbb{T}^2 x = ( x 1 , x 2 ) ∈ T 2 , vertical variable x 3 ∈ ( − ∞ , 0 ) x_3\in(-\infty,0) x 3 ∈ ( − ∞ , 0 ) , and time t ≥ 0 t\ge0 t ≥ 0 . We write
∇ = ( ∂ x 1 , ∂ x 2 , ∂ x 3 ) , Δ = ∂ x 1 2 + ∂ x 2 2 + ∂ x 3 2 , ∇ x = ( ∂ x 1 , ∂ x 2 ) , Δ x = ∂ x 1 2 + ∂ x 2 2 , \nabla=(\partial_{x_1},\partial_{x_2},\partial_{x_3}),\qquad\Delta=\partial_{x_1}^2+\partial_{x_2}^2+\partial_{x_3}^2,\qquad\nabla_x=(\partial_{x_1},\partial_{x_2}),\qquad\Delta_x=\partial_{x_1}^2+\partial_{x_2}^2, ∇ = ( ∂ x 1 , ∂ x 2 , ∂ x 3 ) , Δ = ∂ x 1 2 + ∂ x 2 2 + ∂ x 3 2 , ∇ x = ( ∂ x 1 , ∂ x 2 ) , Δ x = ∂ x 1 2 + ∂ x 2 2 ,
and ∂ t \partial_t ∂ t for the time derivative. Throughout the paper the notation A ≲ B A\lesssim B A ≲ B means A ≤ C B A\le CB A ≤ C B for a constant C > 0 C>0 C > 0 which may depend on fixed parameters (e.g. Υ , δ , β \Upsilon,\delta,\beta Υ , δ , β ) but not on the evolving functions.
For d ∈ { 1 , 2 } d\in\{1,2\} d ∈ { 1 , 2 } and f : T d → C f:\mathbb{T}^d\to\mathbb{C} f : T d → C we write the Fourier series
f ( x ) = ∑ k ∈ Z d f ^ ( k ) e i k ⋅ x , f ^ ( k ) = 1 ( 2 π ) d ∫ T d f ( x ) e − i k ⋅ x d x , ∣ k ∣ = ( k 1 2 + ⋯ + k d 2 ) 1 / 2 . f(x)=\sum_{k\in\mathbb{Z}^d}\widehat{f}(k)e^{ik\cdot x},\qquad\widehat{f}(k)=\frac{1}{(2\pi)^d}\int_{\mathbb{T}^d}f(x)e^{-ik\cdot x}\,dx,\qquad|k|=(k_1^2+\cdots+k_d^2)^{1/2}. f ( x ) = k ∈ Z d ∑ f ( k ) e ik ⋅ x , f ( k ) = ( 2 π ) d 1 ∫ T d f ( x ) e − ik ⋅ x d x , ∣ k ∣ = ( k 1 2 + ⋯ + k d 2 ) 1/2 .
For s ∈ R s\in\mathbb{R} s ∈ R we define
∥ f ∥ H s ( T d ) 2 : = ∑ k ∈ Z d ( 1 + ∣ k ∣ 2 ) s ∣ f ^ ( k ) ∣ 2 , ⟨ f ⟩ : = 1 ( 2 π ) d ∫ T d f , H 0 s ( T d ) : = { f ∈ H s ( T d ) : ⟨ f ⟩ = 0 } . \|f\|_{H^s(\mathbb{T}^d)}^2:=\sum_{k\in\mathbb{Z}^d}(1+|k|^2)^s|\widehat{f}(k)|^2,\qquad\langle f\rangle:=\frac{1}{(2\pi)^d}\int_{\mathbb{T}^d}f,\qquad H_0^s(\mathbb{T}^d):=\{f\in H^s(\mathbb{T}^d):\langle f\rangle=0\}. ∥ f ∥ H s ( T d ) 2 := k ∈ Z d ∑ ( 1 + ∣ k ∣ 2 ) s ∣ f ( k ) ∣ 2 , ⟨ f ⟩ := ( 2 π ) d 1 ∫ T d f , H 0 s ( T d ) := { f ∈ H s ( T d ) : ⟨ f ⟩ = 0 } .
On mean-zero functions we also use the homogeneous seminorm
∥ f ∥ H ˙ s ( T d ) 2 : = ∑ k ≠ 0 ∣ k ∣ 2 s ∣ f ^ ( k ) ∣ 2 , \|f\|_{\dot H^s(\mathbb{T}^d)}^2:=\sum_{k\ne0}|k|^{2s}|\widehat{f}(k)|^2, ∥ f ∥ H ˙ s ( T d ) 2 := k = 0 ∑ ∣ k ∣ 2 s ∣ f ( k ) ∣ 2 ,
and interpret Λ − 1 \Lambda^{-1} Λ − 1 mode-by-mode on k ≠ 0 k\ne0 k = 0 . By Poincaré, for s ≥ 0 s\ge0 s ≥ 0 and f ∈ H 0 s ( T d ) f\in H_0^s(\mathbb{T}^d) f ∈ H 0 s ( T d ) ,
∥ f ∥ H s ( T d ) ∼ ∥ f ∥ H ˙ s ( T d ) . (4) \|f\|_{H^s(\mathbb{T}^d)}\sim\|f\|_{\dot H^s(\mathbb{T}^d)}. \tag*{(4)} ∥ f ∥ H s ( T d ) ∼ ∥ f ∥ H ˙ s ( T d ) . (4)
Basic estimates and inequalities. On T 2 \mathbb{T}^2 T 2 we denote Λ : = ( − Δ x ) 1 / 2 \Lambda:=(-\Delta_x)^{1/2} Λ := ( − Δ x ) 1/2 , that is Λ f ^ ( k ) = ∣ k ∣ f ^ ( k ) \widehat{\Lambda f}(k)=|k|\widehat{f}(k) Λ f ( k ) = ∣ k ∣ f ( k ) , and we use the Riesz transforms
R j f ^ ( k ) = − i k j ∣ k ∣ f ^ ( k ) , k ≠ 0 , j = 1 , 2 , R f = ( R 1 f , R 2 f ) . \widehat{\mathcal{R}_j f}(k)=-i\frac{k_j}{|k|}\widehat{f}(k),\qquad k\ne0,\quad j=1,2,\qquad\mathcal{R}f=(\mathcal{R}_1f,\mathcal{R}_2f). R j f ( k ) = − i ∣ k ∣ k j f ( k ) , k = 0 , j = 1 , 2 , R f = ( R 1 f , R 2 f ) .
They are bounded on Sobolev spaces: for all s ∈ R s\in\mathbb{R} s ∈ R ,
∥ R j f ∥ H s ( T 2 ) ≲ ∥ f ∥ H s ( T 2 ) , ∥ R j f ∥ H ˙ s ( T 2 ) ≲ ∥ f ∥ H ˙ s ( T 2 ) . (5) \|\mathcal{R}_j f\|_{H^s(\mathbb{T}^2)}\lesssim\|f\|_{H^s(\mathbb{T}^2)},\qquad\|\mathcal{R}_j f\|_{\dot H^s(\mathbb{T}^2)}\lesssim\|f\|_{\dot H^s(\mathbb{T}^2)}. \tag*{(5)} ∥ R j f ∥ H s ( T 2 ) ≲ ∥ f ∥ H s ( T 2 ) , ∥ R j f ∥ H ˙ s ( T 2 ) ≲ ∥ f ∥ H ˙ s ( T 2 ) . (5)
We also use the Sobolev embedding
H 2 ( T 2 ) ↪ W 1 , ∞ ( T 2 ) . (6) H^2(\mathbb{T}^2)\hookrightarrow W^{1,\infty}(\mathbb{T}^2). \tag*{(6)} H 2 ( T 2 ) ↪ W 1 , ∞ ( T 2 ) . (6)
On H 0 s ( T 2 ) H_0^s(\mathbb{T}^2) H 0 s ( T 2 ) the resolvent ( I + Υ Λ ) (I+\Upsilon\Lambda) ( I + ΥΛ ) is invertible mode-by-mode with symbol ( 1 + Υ ∣ k ∣ ) − 1 (1+\Upsilon|k|)^{-1} ( 1 + Υ∣ k ∣ ) − 1 for k ≠ 0 k\ne0 k = 0 , and in particular
∥ ( I + Υ Λ ) − 1 G ∥ H s ( T 2 ) ≲ ∥ G ∥ H s ( T 2 ) , G ∈ H 0 s ( T 2 ) . (7) \|(I+\Upsilon\Lambda)^{-1}G\|_{H^s(\mathbb{T}^2)}\lesssim\|G\|_{H^s(\mathbb{T}^2)},\qquad G\in H_0^s(\mathbb{T}^2). \tag*{(7)} ∥ ( I + ΥΛ ) − 1 G ∥ H s ( T 2 ) ≲ ∥ G ∥ H s ( T 2 ) , G ∈ H 0 s ( T 2 ) . (7)
When restricting to one horizontal dimension we work on T \mathbb{T} T with variable ξ \xi ξ . Fourier multipliers Λ = ∣ D ξ ∣ \Lambda=|D_\xi| Λ = ∣ D ξ ∣ and Λ − 1 \Lambda^{-1} Λ − 1 are understood with respect to ξ \xi ξ . We use the Hilbert transform H \mathcal{H} H ,
H g ^ ( k ) = − i sgn ( k ) g ^ ( k ) , k ∈ Z , \widehat{\mathcal{H}g}(k)=-i\operatorname{sgn}(k)\widehat{g}(k),\qquad k\in\mathbb{Z}, H g ( k ) = − i sgn ( k ) g ( k ) , k ∈ Z ,
so that H ∗ = − H \mathcal{H}^*=-\mathcal{H} H ∗ = − H on L 2 ( T ) L^2(\mathbb{T}) L 2 ( T ) and H 2 = − I \mathcal{H}^2=-I H 2 = − I on mean-zero functions. On nonzero modes we have the identities
Λ = H ∂ ξ , ∂ ξ Λ − 1 = − H , ∂ ξ H = H ∂ ξ , (8) \Lambda=\mathcal{H}\partial_\xi,\qquad\partial_\xi\Lambda^{-1}=-\mathcal{H},\qquad\partial_\xi\mathcal{H}=\mathcal{H}\partial_\xi, \tag*{(8)} Λ = H ∂ ξ , ∂ ξ Λ − 1 = − H , ∂ ξ H = H ∂ ξ , (8)
and, for mean-zero f f f and any s ∈ R s\in\mathbb{R} s ∈ R ,
∥ H f ∥ H ˙ s ( T ) = ∥ f ∥ H ˙ s ( T ) . (9) \|\mathcal{H}f\|_{\dot{H}^s(\mathbb{T})}=\|f\|_{\dot{H}^s(\mathbb{T})}. \tag*{(9)} ∥ H f ∥ H ˙ s ( T ) = ∥ f ∥ H ˙ s ( T ) . (9)
We also use Tricomi’s identities: for sufficiently regular real-valued f , g f,g f , g ,
H ( f H g + g H f ) = ( H f ) ( H g ) − f g , H ( f H f ) = 1 2 ( ( H f ) 2 − f 2 ) . (10) \mathcal{H}(f\mathcal{H}g+g\mathcal{H}f)=(\mathcal{H}f)(\mathcal{H}g)-fg,\qquad\mathcal{H}(f\mathcal{H}f)=\frac{1}{2}\left((\mathcal{H}f)^2-f^2\right). \tag*{(10)} H ( f H g + g H f ) = ( H f ) ( H g ) − f g , H ( f H f ) = 2 1 ( ( H f ) 2 − f 2 ) . (10)
Next, we gather products, commutators and auxiliary multipliers estimates in one dimension. For a linear operator A A A and a function f f f we write [ A , f ] g : = A ( f g ) − f A g [A,f]g:=A(fg)-fAg [ A , f ] g := A ( f g ) − f A g . We will repeatedly use that for s > 1 2 s>\frac{1}{2} s > 2 1 and f ∈ H 0 s ( T ) f\in H_0^s(\mathbb{T}) f ∈ H 0 s ( T ) ,
∥ f ∥ L ∞ ( T ) ≲ s ∥ f ∥ H ˙ s ( T ) . (11) \|f\|_{L^\infty(\mathbb{T})}\lesssim_s\|f\|_{\dot{H}^s(\mathbb{T})}. \tag*{(11)} ∥ f ∥ L ∞ ( T ) ≲ s ∥ f ∥ H ˙ s ( T ) . (11)
so in particular H s ( T ) H^s(\mathbb{T}) H s ( T ) is a Banach algebra for s > 1 2 s>\frac{1}{2} s > 2 1 , with
∥ f g ∥ H s ( T ) ≲ s ∥ f ∥ H s ( T ) ∥ g ∥ H s ( T ) , ∥ f g ∥ H s ( T ) ≲ s ∥ f ∥ L ∞ ∥ g ∥ H s + ∥ g ∥ L ∞ ∥ f ∥ H s . (12) \|fg\|_{H^s(\mathbb{T})}\lesssim_s\|f\|_{H^s(\mathbb{T})}\|g\|_{H^s(\mathbb{T})},\qquad\|fg\|_{H^s(\mathbb{T})}\lesssim_s\|f\|_{L^\infty}\|g\|_{H^s}+\|g\|_{L^\infty}\|f\|_{H^s}. \tag*{(12)} ∥ f g ∥ H s ( T ) ≲ s ∥ f ∥ H s ( T ) ∥ g ∥ H s ( T ) , ∥ f g ∥ H s ( T ) ≲ s ∥ f ∥ L ∞ ∥ g ∥ H s + ∥ g ∥ L ∞ ∥ f ∥ H s . (12)
We also use the Gagliardo–Nirenberg inequality
∥ f ∥ W ˙ 1 , 4 ( T ) 2 = ∥ f ξ ∥ L 4 ( T ) 2 ≲ ∥ f ∥ L ∞ ( T ) ∥ f ∥ H ˙ 2 ( T ) . (13) \|f\|_{\dot{W}^{1,4}(\mathbb{T})}^2=\|f_\xi\|_{L^4(\mathbb{T})}^2\lesssim\|f\|_{L^\infty(\mathbb{T})}\|f\|_{\dot{H}^2(\mathbb{T})}. \tag*{(13)} ∥ f ∥ W ˙ 1 , 4 ( T ) 2 = ∥ f ξ ∥ L 4 ( T ) 2 ≲ ∥ f ∥ L ∞ ( T ) ∥ f ∥ H ˙ 2 ( T ) . (13)
Given γ > 0 \gamma>0 γ > 0 , we define the Fourier multiplier
T : = ( I + γ Λ ) − 1 Λ , T f ^ ( k ) = ∣ k ∣ 1 + γ ∣ k ∣ f ^ ( k ) , k ≠ 0 , (14) \mathcal{T}:=(I+\gamma\Lambda)^{-1}\Lambda,\qquad\widehat{\mathcal{T}f}(k)=\frac{|k|}{1+\gamma|k|}\widehat{f}(k),\qquad k\ne0, \tag*{(14)} T := ( I + γ Λ ) − 1 Λ , T f ( k ) = 1 + γ ∣ k ∣ ∣ k ∣ f ( k ) , k = 0 , (14)
so that for all s ∈ R s\in\mathbb{R} s ∈ R and mean-zero f f f ,
∥ T f ∥ H ˙ s ( T ) ≤ ∥ f ∥ H ˙ s ( T ) , ∥ Λ T f ∥ H ˙ s ( T ) ≲ γ ∥ f ∥ H ˙ s + 1 ( T ) . (15) \|\mathcal{T}f\|_{\dot{H}^s(\mathbb{T})}\leq\|f\|_{\dot{H}^s(\mathbb{T})},\qquad\|\Lambda\mathcal{T}f\|_{\dot{H}^s(\mathbb{T})}\lesssim_\gamma\|f\|_{\dot{H}^{s+1}(\mathbb{T})}. \tag*{(15)} ∥ T f ∥ H ˙ s ( T ) ≤ ∥ f ∥ H ˙ s ( T ) , ∥Λ T f ∥ H ˙ s ( T ) ≲ γ ∥ f ∥ H ˙ s + 1 ( T ) . (15)
Finally, for ν ∈ ( 0 , 1 ] \nu\in(0,1] ν ∈ ( 0 , 1 ] consider J ν \mathcal{J}_\nu J ν the heat kernel on T d \mathbb{T}^d T d (d = 1 , 2 d=1,2 d = 1 , 2 ) defined by
J ν f ^ ( k ) : = e − ν ∣ k ∣ 2 f ^ ( k ) , k ∈ Z d . \widehat{\mathcal{J}_\nu f}(k):=e^{-\nu|k|^2}\widehat{f}(k),\qquad k\in\mathbb{Z}^d. J ν f ( k ) := e − ν ∣ k ∣ 2 f ( k ) , k ∈ Z d .
Then J ν \mathcal{J}_\nu J ν is self-adjoint on L 2 L^2 L 2 , it commutes with ∂ ξ \partial_\xi ∂ ξ and Λ \Lambda Λ (and with H \mathcal{H} H when d = 1 d=1 d = 1 ), and J ν f → f \mathcal{J}_\nu f\to f J ν f → f in H s H^s H s as ν → 0 \nu\to0 ν → 0 for all s ∈ R s\in\mathbb{R} s ∈ R . Moreover, for every s ∈ R s\in\mathbb{R} s ∈ R and m ≥ 0 m\geq0 m ≥ 0 ,
∥ J ν f ∥ H s ≲ ∥ f ∥ H s , ∥ J ν f ∥ H s + m ≲ ν − m ∥ f ∥ H s . (16) \|\mathcal{J}_\nu f\|_{H^s}\lesssim\|f\|_{H^s},\qquad\|\mathcal{J}_\nu f\|_{H^{s+m}}\lesssim\nu^{-m}\|f\|_{H^s}. \tag*{(16)} ∥ J ν f ∥ H s ≲ ∥ f ∥ H s , ∥ J ν f ∥ H s + m ≲ ν − m ∥ f ∥ H s . (16)
To close this preliminary section, we record a standard representation formula for the Poisson problem:
Lemma 1.1 (Poisson problem with Dirichlet data). Let b = b ( x , x 3 ) b=b(x,x_3) b = b ( x , x 3 ) and g = g ( x ) g=g(x) g = g ( x ) be smooth, 2 π 2\pi 2 π –periodic in x ∈ T 2 x\in\mathbb{T}^2 x ∈ T 2 , and assume that b ( ⋅ , x 3 ) b(\cdot,x_3) b ( ⋅ , x 3 ) decays sufficiently fast as x 3 → − ∞ x_3\to-\infty x 3 → − ∞ . Consider the problem
{ Δ u = b in Ω : = T 2 × ( − ∞ , 0 ) , u = g on Γ : = T 2 × { 0 } , u ( ⋅ , x 3 ) → 0 as x 3 → − ∞ . (17) \begin{cases}
\Delta u=b & \text{in }\Omega:=\mathbb{T}^2\times(-\infty,0),\\
u=g & \text{on }\Gamma:=\mathbb{T}^2\times\{0\},\\
u(\cdot,x_3)\to0 & \text{as }x_3\to-\infty. \tag*{(17)}
\end{cases} ⎩ ⎨ ⎧ Δ u = b u = g u ( ⋅ , x 3 ) → 0 in Ω := T 2 × ( − ∞ , 0 ) , on Γ := T 2 × { 0 } , as x 3 → − ∞. (17) Then (17) admits a unique solution u u u , and for every nonzero Fourier mode k ∈ Z 2 ∖ { 0 } k\in\mathbb{Z}^2\setminus\{0\} k ∈ Z 2 ∖ { 0 } one has
∂ x 3 u ^ ( k , 0 ) = ∣ k ∣ g ^ ( k ) + ∫ − ∞ 0 e ∣ k ∣ y 3 b ^ ( k , y 3 ) d y 3 . (18) \partial_{x_3}\widehat{u}(k,0)=|k|\widehat{g}(k)+\int_{-\infty}^{0}e^{|k|y_3}\widehat{b}(k,y_3)\,\mathrm{d}y_3. \tag*{(18)} ∂ x 3 u ( k , 0 ) = ∣ k ∣ g ( k ) + ∫ − ∞ 0 e ∣ k ∣ y 3 b ( k , y 3 ) d y 3 . (18) Equivalently, in operator form on mean–zero functions,
∂ x 3 u ( ⋅ , 0 ) = Λ g + ∫ − ∞ 0 e y 3 Λ b ( ⋅ , y 3 ) d y 3 . (19) \partial_{x_3}u(\cdot,0)=\Lambda g+\int_{-\infty}^{0}e^{y_3\Lambda}b(\cdot,y_3)\,\mathrm{d}y_3. \tag*{(19)} ∂ x 3 u ( ⋅ , 0 ) = Λ g + ∫ − ∞ 0 e y 3 Λ b ( ⋅ , y 3 ) d y 3 . (19) Proof. Taking the Fourier series in x x x reduces (17) to
( ∂ x 3 2 − ∣ k ∣ 2 ) u ^ ( k , x 3 ) = b ^ ( k , x 3 ) , u ^ ( k , 0 ) = g ^ ( k ) , u ^ ( k , x 3 ) → 0 as x 3 → − ∞ , (\partial_{x_3}^2-|k|^2)\widehat{u}(k,x_3)=\widehat{b}(k,x_3),\qquad\widehat{u}(k,0)=\widehat{g}(k),\qquad\widehat{u}(k,x_3)\to0\text{ as }x_3\to-\infty, ( ∂ x 3 2 − ∣ k ∣ 2 ) u ( k , x 3 ) = b ( k , x 3 ) , u ( k , 0 ) = g ( k ) , u ( k , x 3 ) → 0 as x 3 → − ∞ , which is solved explicitly by variation of constants. Differentiating at x 3 = 0 x_3=0 x 3 = 0 gives (18) . A detailed derivation can be found for instance in [16 , 17 Lemma A.1] .
Organization of the manuscript
The manuscript is organized as follows. In Section 2 we introduce the full three-dimensional Euler–plate system, discuss the geometric bending energy and the associated structural forces, and rewrite the problem in potential variables. We then nondimensionalize the equations and derive an Arbitrary Lagrangian–Eulerian (ALE) formulation on a fixed reference domain. In Section 3 we perform a small-steepness asymptotic expansion of the ALE system and obtain, up to quadratic order, two bidirectional nonlocal interface models for the surface displacement. In Section 4 we derive two one-dimensional unidirectional reductions capturing one-way propagation while retaining the leading dispersive and dissipative effects. In Section 4 we prove local well-posedness for the first of the bidirectional model for small H 3 H^3 H 3 initial data. In Section 5 we establish well-posedness results for the unidirectional models, including local well-posedness for arbitrary data and global well-posedness for small data. Finally, Appendix A collects geometric identities for graph surfaces, and Appendix B contains the details of the non-dimensionalisation.
Setting of the problem
The fluid structure interaction system
In this section we present a complete formulation of the fluid–structure interaction system under consideration. The model consists of a three-dimensional incompressible, inviscid fluid evolving below a deformable elastic surface. The fluid occupies the time-dependent domain
Ω ( t ) = { ( x 1 , x 2 , x 3 ) ∈ R 3 : − L π < x 1 , x 2 < L π , x 3 < η ( t , x 1 , x 2 ) , t ∈ [ 0 , T ] } . (20) \Omega(t)=\{(x_1,x_2,x_3)\in\mathbb{R}^3:-L\pi<x_1,x_2<L\pi,\quad x_3<\eta(t,x_1,x_2),\quad t\in[0,T]\}. \tag*{(20)} Ω ( t ) = {( x 1 , x 2 , x 3 ) ∈ R 3 : − L π < x 1 , x 2 < L π , x 3 < η ( t , x 1 , x 2 ) , t ∈ [ 0 , T ]} . (20)
where L > 0 L > 0 L > 0 and the elastic free–boundary surface occupied by the plate is represented as the graph
Γ ( t ) = { ( x 1 , x 2 , η ( x 1 , x 2 , t ) ) : ( x 1 , x 2 ) ∈ L S 1 × L S 1 , t ∈ [ 0 , T ] } , (21) \Gamma(t)=\{(x_1,x_2,\eta(x_1,x_2,t)):(x_1,x_2)\in LS^1\times LS^1,\quad t\in[0,T]\}, \tag*{(21)} Γ ( t ) = {( x 1 , x 2 , η ( x 1 , x 2 , t )) : ( x 1 , x 2 ) ∈ L S 1 × L S 1 , t ∈ [ 0 , T ]} , (21)
The fluid is assumed to be incompressible, inviscid and of constant density ρ f > 0 \rho_f>0 ρ f > 0 . Therefore, velocity u = ( u 1 , u 2 , u 3 ) u=(u_1,u_2,u_3) u = ( u 1 , u 2 , u 3 ) and pressure p p p in the fluid region satisfy the incompressible Euler equations
ρ f ( ∂ t u + ( u ⋅ ∇ ) u ) = − ∇ p − ρ f g e 3 in Ω ( t ) , (22) \rho_f\left(\partial_tu+(u\cdot\nabla)u\right)=-\nabla p-\rho_f g e_3 \qquad\text{in }\Omega(t), \tag*{(22)} ρ f ( ∂ t u + ( u ⋅ ∇ ) u ) = − ∇ p − ρ f g e 3 in Ω ( t ) , (22)
∇ ⋅ u = 0 in Ω ( t ) . (23) \nabla\cdot u=0 \qquad\text{in }\Omega(t). \tag*{(23)} ∇ ⋅ u = 0 in Ω ( t ) . (23)
where g > 0 g>0 g > 0 denotes gravity and e 3 = ( 0 , 0 , 1 ) e_3=(0,0,1) e 3 = ( 0 , 0 , 1 ) . Since the domain extends infinitely deep, we impose the decay condition
u ( t , x 1 , x 2 , x 3 ) → 0 , p ( t , x 1 , x 2 , x 3 ) → p ∞ ( t ) , x 3 → − ∞ . (24) u(t,x_1,x_2,x_3)\to0,\qquad p(t,x_1,x_2,x_3)\to p_\infty(t),\qquad x_3\to-\infty. \tag*{(24)} u ( t , x 1 , x 2 , x 3 ) → 0 , p ( t , x 1 , x 2 , x 3 ) → p ∞ ( t ) , x 3 → − ∞. (24)
Moreover, the fluid is assumed to be periodic in the horizontal directions, i.e., for k = ( k 1 , k 2 ) ∈ Z 2 k=(k_1,k_2)\in\mathbb{Z}^2 k = ( k 1 , k 2 ) ∈ Z 2 we impose
u ( t , x 1 + 2 π L k 1 , x 2 + 2 π L k 2 , x 3 ) = u ( t , x 1 , x 2 , x 3 ) , t ∈ [ 0 , T ] , (25) u(t,x_1+2\pi Lk_1,x_2+2\pi Lk_2,x_3)=u(t,x_1,x_2,x_3),\qquad t\in[0,T], \tag*{(25)} u ( t , x 1 + 2 π L k 1 , x 2 + 2 π L k 2 , x 3 ) = u ( t , x 1 , x 2 , x 3 ) , t ∈ [ 0 , T ] , (25)
and analogously for the pressure
p ( t , x 1 + 2 π L k 1 , x 2 + 2 π L k 2 , x 3 ) = p ( t , x 1 , x 2 , x 3 ) . (26) p(t,x_1+2\pi Lk_1,x_2+2\pi Lk_2,x_3)=p(t,x_1,x_2,x_3). \tag*{(26)} p ( t , x 1 + 2 π L k 1 , x 2 + 2 π L k 2 , x 3 ) = p ( t , x 1 , x 2 , x 3 ) . (26)
2.2. Boundary conditions, geometric bending energy, and structural forces. The motion of the free boundary Γ ( t ) \Gamma(t) Γ ( t ) is governed by two conditions: a kinematic condition expressing that the interface is transported by the fluid, and a dynamic condition expressing balance of normal stresses.
The kinematic boundary condition. Since the free surface consists of material particles of the fluid, its normal velocity must coincide with that of the fluid. Noticing that the upward unit normal to Γ ( t ) \Gamma(t) Γ ( t ) is given by
n ( x , t ) = ( − ∇ x η ( x , t ) , 1 ) T 1 + ∣ ∇ x η ( x , t ) ∣ 2 , x = ( x 1 , x 2 ) , n(x,t)=\frac{(-\nabla_x\eta(x,t),1)^T}{\sqrt{1+|\nabla_x\eta(x,t)|^2}},\qquad x=(x_1,x_2), n ( x , t ) = 1 + ∣ ∇ x η ( x , t ) ∣ 2 ( − ∇ x η ( x , t ) , 1 ) T , x = ( x 1 , x 2 ) ,
we have that
∂ t η = u ⋅ 1 + ∣ ∇ x η ∣ 2 n , on Γ ( t ) (27) \partial_t\eta=u\cdot\sqrt{1+|\nabla_x\eta|^2}\,n,\quad\text{on }\Gamma(t) \tag*{(27)} ∂ t η = u ⋅ 1 + ∣ ∇ x η ∣ 2 n , on Γ ( t ) (27)
Geometric bending energy, inertia, and damping. Before formulating the dynamic boundary condition, we describe the forces acting on the elastic plate. These consist of (i) a nonlinear bending force obtained as the first variation of a curvature-dependent energy, (ii) an inertial response arising from the mass density of the plate, and (iii) a dissipative contribution modelling structural damping. Together these form the structural side of the evolution equation on the free boundary. Following the hydroelastic shell model of Plotnikov and Toland [27 ] , we assume that the restoring elastic force derives from a curvature-based bending energy of the form
E b [ η ] = ∫ L S 1 × L S 1 W b ( H ( η ) ) 1 + ∣ ∇ x η ∣ 2 d x , (28) E_b[\eta]=\int_{LS^1\times LS^1}W_b(H(\eta))\sqrt{1+|\nabla_x\eta|^2}\,dx, \tag*{(28)} E b [ η ] = ∫ L S 1 × L S 1 W b ( H ( η )) 1 + ∣ ∇ x η ∣ 2 d x , (28)
where H ( η ) H(\eta) H ( η ) denotes the mean curvature of the graph Γ ( t ) \Gamma(t) Γ ( t ) and W b W_b W b is a prescribed bending density. In the context of hydroelasticity, membrane strains are typically negligible compared to bending, and the energy depends solely on curvature. A widely used and physically relevant choice is
W b ( H ) = B 2 H 2 , W_b(H) = \frac{B}{2}H^2, W b ( H ) = 2 B H 2 ,
which leads to the Willmore-type energy
E b [ η ] = ∫ L S 1 × L S 1 B 2 H ( η ) 2 1 + ∣ ∇ x η ∣ 2 d x . (29) E_b[\eta] = \int_{\mathbb{L}S^1 \times\mathbb{L}S^1} \frac{B}{2}H(\eta)^2\sqrt{1+\lvert\nabla_x\eta\rvert^2}\,dx. \tag*{(29)} E b [ η ] = ∫ L S 1 × L S 1 2 B H ( η ) 2 1 + ∣ ∇ x η ∣ 2 d x . (29)
The elastic force associated with this energy is obtained from its first variation. As shown in [27 ] , one obtains the fourth-order nonlinear geometric operator
E ( η ) = B 2 Δ Γ H ( η ) + B H 3 ( η ) − B H ( η ) K ( η ) , (30) \mathcal{E}(\eta) = \frac{B}{2}\Delta_\Gamma H(\eta) + BH^3(\eta) - BH(\eta)K(\eta), \tag*{(30)} E ( η ) = 2 B Δ Γ H ( η ) + B H 3 ( η ) − B H ( η ) K ( η ) , (30)
where Δ Γ \Delta_\Gamma Δ Γ is the Laplace–Beltrami operator and K K K is the Gauss curvature of the evolving surface. The operator retains the full geometric nonlinearity of the plate and depends explicitly on η \eta η , ∇ x η \nabla_x\eta ∇ x η , and ∇ x 2 η \nabla_x^2\eta ∇ x 2 η . For convenience, explicit expressions for H H H , K K K , Δ Γ \Delta_\Gamma Δ Γ , and the metric coefficients in terms of η \eta η are collected in Appendix § 6 .
In addition to curvature-driven elasticity, the plate possesses inertia: with mass density ρ s h \rho_s h ρ s h per unit area, vertical acceleration produces the inertial force
ρ s h ∂ t t η . (31) \rho_s h\,\partial_{tt}\eta. \tag*{(31)} ρ s h ∂ tt η . (31)
Here ρ s \rho_s ρ s denotes the volumetric density of the solid material and h > 0 h>0 h > 0 is the (constant) thickness of the plate, so that ρ s h \rho_s h ρ s h is the mass per unit area of the structure. This hyperbolic term reflects the kinetic response of the elastic sheet and plays a central role in the time-dependent hydroelastic dynamics. Finally, structural or viscoelastic damping may be present. A standard and analytically convenient choice is Kelvin–Voigt damping, modelled by
D ( η t ) = − γ ∂ t Δ x η , γ ≥ 0 , (32) \mathcal{D}(\eta_t) = -\gamma\,\partial_t\Delta_x\eta,\qquad\gamma\ge0, \tag*{(32)} D ( η t ) = − γ ∂ t Δ x η , γ ≥ 0 , (32)
corresponding to a Rayleigh dissipation potential R ( η ) = γ 2 ∫ ∣ ∂ t ∇ x η ∣ 2 d x \mathcal{R}(\eta) = \frac{\gamma}{2}\int\lvert\partial_t\nabla_x\eta\rvert^2\,dx R ( η ) = 2 γ ∫ ∣ ∂ t ∇ x η ∣ 2 d x . This term introduces high-order dissipation compatible with the geometric structure of the bending operator and appears naturally in Koiter-type plate models [10 ] .
These three contributions combine to give the complete structural response of the plate: the nonlinear bending operator E ( η ) \mathcal{E}(\eta) E ( η ) in (30) , the inertial force ρ s h η t t \rho_s h\eta_{tt} ρ s h η tt in (31) , and the damping term D ( η t ) \mathcal{D}(\eta_t) D ( η t ) in (32) . Together they form the left-hand side of the evolution equation on the free boundary, and their balance with the normal fluid traction yields the dynamic boundary condition on Γ ( t ) \Gamma(t) Γ ( t ) considered in the next subsection.
Dynamic boundary condition: balance of normal forces. With the elastic, inertial, and damping forces identified, we now state the dynamic boundary condition on the free surface. For an inviscid, incompressible fluid, the Cauchy stress tensor is σ f = − p I \boldsymbol{\sigma}_f=-pI σ f = − p I . The traction exerted by the fluid across the interface with outward unit normal n n n is therefore σ f n = − p n \boldsymbol{\sigma}_f n=-pn σ f n = − p n .
In the plate equation, however, the relevant quantity is the normal force exerted by the fluid on the structure. By Newton’s third law, this force is the opposite of the traction exerted by the structure on the fluid. Consequently, the normal fluid force acting on the plate is
T f l u i d = − ( σ f n ) ⋅ n = p , on Γ ( t ) . T_{\mathrm{fluid}}=-(\boldsymbol{\sigma}_f n)\cdot n=p,\qquad\text{on }\Gamma(t). T fluid = − ( σ f n ) ⋅ n = p , on Γ ( t ) .
Balancing this fluid force with the inertial, elastic, and damping contributions of the plate yields Newton’s second law on the moving boundary:
ρ s h ∂ t t η + E ( η ) − γ ∂ t Δ x η = p on Γ ( t ) . (33) \rho_s h\,\partial_{tt}\eta+\mathcal{E}(\eta)-\gamma\partial_t\Delta_x\eta=p\qquad\text{on }\Gamma(t). \tag*{(33)} ρ s h ∂ tt η + E ( η ) − γ ∂ t Δ x η = p on Γ ( t ) . (33)
Equation (33) constitutes the fully nonlinear dynamic boundary condition governing the evolution of the elastic surface.
The full Euler–plate system and dimensionless system
Equation (33) , together with the kinematic condition (27) and the Euler equations (2.3), completes the formulation of the fluid–structure interaction problem. For convenience, we collect the full system below:
ρ f ( ∂ t u + ( u ⋅ ∇ ) u ) = − ∇ p − ρ f g e 3 in Ω ( t ) , (34) \rho_f\left(\partial_tu+(u\cdot\nabla)u\right)=-\nabla p-\rho_f g e_3\qquad\text{in }\Omega(t), \tag*{(34)} ρ f ( ∂ t u + ( u ⋅ ∇ ) u ) = − ∇ p − ρ f g e 3 in Ω ( t ) , (34)
∇ ⋅ u = 0 in Ω ( t ) , (35) \nabla\cdot u=0\qquad\text{in }\Omega(t), \tag*{(35)} ∇ ⋅ u = 0 in Ω ( t ) , (35)
∂ t η = u ⋅ 1 + ∣ ∇ x η ∣ 2 n on Γ ( t ) , (36) \partial_t\eta=u\cdot\sqrt{1+|\nabla_x\eta|^2}\,n\qquad\text{on }\Gamma(t), \tag*{(36)} ∂ t η = u ⋅ 1 + ∣ ∇ x η ∣ 2 n on Γ ( t ) , (36)
ρ s h ∂ t t η + E ( η ) − γ ∂ t Δ x η = p on Γ ( t ) . (37) \rho_s h\,\partial_{tt}\eta+\mathcal{E}(\eta)-\gamma\partial_t\Delta_x\eta=p\qquad\text{on }\Gamma(t). \tag*{(37)} ρ s h ∂ tt η + E ( η ) − γ ∂ t Δ x η = p on Γ ( t ) . (37)
where E ( η ) \mathcal{E}(\eta) E ( η ) is given in (30) . Assuming that the flow is irrotational, we introduce a potential ϕ \phi ϕ such that u = ∇ ϕ u=\nabla\phi u = ∇ ϕ and denote by ψ ( x 1 , x 2 , t ) = ϕ ( t , x 1 , x 2 , η ( x 1 , x 2 , t ) ) \psi(x_1,x_2,t)=\phi(t,x_1,x_2,\eta(x_1,x_2,t)) ψ ( x 1 , x 2 , t ) = ϕ ( t , x 1 , x 2 , η ( x 1 , x 2 , t )) the surface potential. Taking the trace of the Bernoulli relation
p = − ρ f ( ∂ t ϕ + 1 2 ∣ ∇ ϕ ∣ 2 + g x 3 ) in Ω ( t ) , p=-\rho_f\left(\partial_t\phi+\frac{1}{2}|\nabla\phi|^2+gx_3\right)\qquad\text{in }\Omega(t), p = − ρ f ( ∂ t ϕ + 2 1 ∣∇ ϕ ∣ 2 + g x 3 ) in Ω ( t ) ,
yields
p = − ρ f ( ∂ t ϕ + 1 2 ∣ ∇ ϕ ∣ 2 + g η ) on Γ ( t ) . p=-\rho_f\left(\partial_t\phi+\frac{1}{2}|\nabla\phi|^2+g\eta\right)\qquad\text{on }\Gamma(t). p = − ρ f ( ∂ t ϕ + 2 1 ∣∇ ϕ ∣ 2 + g η ) on Γ ( t ) .
Moreover, using the chain rule, we can write
∂ t ψ = ∂ t ϕ + ∂ x 3 ϕ ∂ t η , on Γ ( t ) . \partial_t\psi=\partial_t\phi+\partial_{x_3}\phi\,\partial_t\eta,\qquad\text{on }\Gamma(t). ∂ t ψ = ∂ t ϕ + ∂ x 3 ϕ ∂ t η , on Γ ( t ) .
and using the kinematic boundary condition in (36) ,
∂ t ψ = ∂ t ϕ + ∂ x 3 ϕ ( ∇ ϕ ⋅ 1 + ∣ ∇ x η ∣ 2 n ) , on Γ ( t ) . \partial_t\psi=\partial_t\phi+\partial_{x_3}\phi\left(\nabla\phi\cdot\sqrt{1+|\nabla_x\eta|^2}\,n\right),\qquad\text{on }\Gamma(t). ∂ t ψ = ∂ t ϕ + ∂ x 3 ϕ ( ∇ ϕ ⋅ 1 + ∣ ∇ x η ∣ 2 n ) , on Γ ( t ) .
Thus, the fluid–structure interaction system admits the following potential formulation
Δ ϕ = 0 in Ω ( t ) , (38) \Delta\phi= 0 \qquad\text{in } \Omega(t), \tag*{(38)} Δ ϕ = 0 in Ω ( t ) , (38)
ϕ = ψ on Γ ( t ) , (39) \phi= \psi\qquad\text{on } \Gamma(t), \tag*{(39)} ϕ = ψ on Γ ( t ) , (39)
∂ t η = ∇ ϕ ⋅ 1 + ∣ ∇ x η ∣ 2 n on Γ ( t ) , (40) \partial_t\eta= \nabla\phi\cdot\sqrt{1+|\nabla_x\eta|^2}\,n \qquad\text{on } \Gamma(t), \tag*{(40)} ∂ t η = ∇ ϕ ⋅ 1 + ∣ ∇ x η ∣ 2 n on Γ ( t ) , (40)
∂ t ψ = − ρ s h ρ f ∂ t 2 η − 1 ρ f E ( η ) + γ ρ f ∂ t Δ x η + ∂ x 3 ϕ ( ∇ ϕ ⋅ 1 + ∣ ∇ x η ∣ 2 n ) − 1 2 ∣ ∇ ϕ ∣ 2 − g η on Γ ( t ) . (41) \begin{aligned}
\partial_t\psi&= -\frac{\rho_s h}{\rho_f}\partial_t^2\eta-\frac{1}{\rho_f}\mathcal{E}(\eta)+\frac{\gamma}{\rho_f}\partial_t\Delta_x\eta\\
&\quad+\partial_{x_3}\phi(\nabla\phi\cdot\sqrt{1+|\nabla_x\eta|^2}\,n)-\frac{1}{2}|\nabla\phi|^2-g\eta
\qquad\text{on } \Gamma(t). \tag*{(41)}
\end{aligned} ∂ t ψ = − ρ f ρ s h ∂ t 2 η − ρ f 1 E ( η ) + ρ f γ ∂ t Δ x η + ∂ x 3 ϕ ( ∇ ϕ ⋅ 1 + ∣ ∇ x η ∣ 2 n ) − 2 1 ∣∇ ϕ ∣ 2 − g η on Γ ( t ) . (41)
where we recall that E ( η ) \mathcal{E}(\eta) E ( η ) is given in (30) .
Remark 2.1 (General bending energies). The structure of the model remains unchanged for general bending energies of the form
E b [ η ] = ∫ L S 1 × L S 1 W b ( H ( η ) ) 1 + ∣ ∇ x η ∣ 2 d x , E_b[\eta]=\int_{L^{S^1}\times L^{S^1}}W_b(H(\eta))\sqrt{1+|\nabla_x\eta|^2}\,dx, E b [ η ] = ∫ L S 1 × L S 1 W b ( H ( η )) 1 + ∣ ∇ x η ∣ 2 d x ,
with W b W_b W b smooth. The corresponding elastic operator is
E ( η ) = 1 2 Δ Γ W b ′ ( H ) + 2 H ( H W b ′ ( H ) − W b ( H ) ) − K W b ′ ( H ) . \mathcal{E}(\eta)=\frac{1}{2}\Delta_{\Gamma}W_b'(H)+2H(HW_b'(H)-W_b(H))-KW_b'(H). E ( η ) = 2 1 Δ Γ W b ′ ( H ) + 2 H ( H W b ′ ( H ) − W b ( H )) − K W b ′ ( H ) .
The choice W b ( H ) = B 2 H 2 W_b(H)=\frac{B}{2}H^2 W b ( H ) = 2 B H 2 is made for physical and analytic clarity, but more general models arise in Cosserat shell theory.
Remark 2.2 (Dependence on the Gauss curvature). Bending energies depending on both H H H and K K K can also be considered:
E b [ η ] = ∫ L S 1 × L S 1 W ( H , K ) 1 + ∣ ∇ x η ∣ 2 d x . E_b[\eta]=\int_{L^{S^1}\times L^{S^1}}W(H,K)\sqrt{1+|\nabla_x\eta|^2}\,dx. E b [ η ] = ∫ L S 1 × L S 1 W ( H , K ) 1 + ∣ ∇ x η ∣ 2 d x .
Such energies lead to additional terms involving the variation of K K K and may require higher regularity. These appear in the literature on elastic shells and biomembranes, though we do not pursue them here.
Dimensionless quantities and system. In order to derive the asymptotic models, we provide the non-dimensionalization of the system (38) –(41) . Let L L L denote the typical horizontal wavelength and H H H the typical vertical amplitude of the elastic sheet. We introduce the dimensionless spatial and temporal variables
x = L x ~ , x 3 = L x ~ 3 , t = L g t ~ , x=L\tilde{x},\qquad x_3=L\tilde{x}_3,\qquad t=\sqrt{\frac{L}{g}}\,\tilde{t}, x = L x ~ , x 3 = L x ~ 3 , t = g L t ~ ,
and the non-dimensional surface displacement, potential and surface potential
η ( x , t ) = H η ~ ( x ~ , t ~ ) , ϕ ( x , x 3 , t ) = H g L ϕ ~ ( x ~ , x ~ 3 , t ~ ) , ψ ( x , t ) = H g L ψ ~ ( x ~ , t ~ ) . \eta(x,t)=H\tilde{\eta}(\tilde{x},\tilde{t}),\qquad\phi(x,x_3,t)=H\sqrt{gL}\,\tilde{\phi}(\tilde{x},\tilde{x}_3,\tilde{t}),\qquad\psi(x,t)=H\sqrt{gL}\,\tilde{\psi}(\tilde{x},\tilde{t}). η ( x , t ) = H η ~ ( x ~ , t ~ ) , ϕ ( x , x 3 , t ) = H g L ϕ ~ ( x ~ , x ~ 3 , t ~ ) , ψ ( x , t ) = H g L ψ ~ ( x ~ , t ~ ) .
Accordingly, we define the non-dimensionalized fluid domain
Ω ~ ( t ) = { ( x ~ 1 , x ~ 2 , x ~ 3 ) ∈ R 3 : − π < x ~ 1 , x ~ 2 < π , x ~ 3 < ϵ η ~ ( x ~ , t ~ ) , t ∈ [ 0 , T ] } . \widetilde{\Omega}(t)=\{(\tilde{x}_1,\tilde{x}_2,\tilde{x}_3)\in\mathbb{R}^3:-\pi<\tilde{x}_1,\tilde{x}_2<\pi,\ \tilde{x}_3<\epsilon\tilde{\eta}(\tilde{x},\tilde{t}),\ t\in[0,T]\}. Ω ( t ) = {( x ~ 1 , x ~ 2 , x ~ 3 ) ∈ R 3 : − π < x ~ 1 , x ~ 2 < π , x ~ 3 < ϵ η ~ ( x ~ , t ~ ) , t ∈ [ 0 , T ]} .
and free surface
Γ ~ ( t ) = { ( x ~ 1 , x ~ 2 , ε η ( x ~ , t ) ) : − π < x ~ 1 , x ~ 2 < π , t ∈ [ 0 , T ] } . \widetilde{\Gamma}(t)=\{(\widetilde{x}_1,\widetilde{x}_2,\varepsilon\eta(\widetilde{x},t)):-\pi<\widetilde{x}_1,\widetilde{x}_2<\pi,\quad t\in[0,T]\}. Γ ( t ) = {( x 1 , x 2 , ε η ( x , t )) : − π < x 1 , x 2 < π , t ∈ [ 0 , T ]} .
Here we introduce the nondimensional parameters
γ : = ρ s h ρ f L , β : = H B ρ f g L 4 , δ : = γ ρ f g L 3 , ε : = H L . \gamma:=\frac{\rho_s h}{\rho_f L},\qquad\beta:=\frac{HB}{\rho_f gL^4},\qquad\delta:=\frac{\gamma}{\rho_f\sqrt{gL^3}},\qquad\varepsilon:=\frac{H}{L}. γ := ρ f L ρ s h , β := ρ f g L 4 H B , δ := ρ f g L 3 γ , ε := L H .
The non-dimensionalisation of the potential formulation (2.15a)–(2.15d) is carried out in detail in Appendix B. In particular, after introducing the dimensionless variables and parameters and dropping tildes for notational simplicity, the fluid–structure system takes the form
Δ ϕ = 0 in Ω ( t ) , (42) \Delta\phi=0 \qquad\text{in }\Omega(t), \tag*{(42)} Δ ϕ = 0 in Ω ( t ) , (42)
ϕ = ψ on Γ ( t ) , (43) \phi=\psi\qquad\text{on }\Gamma(t), \tag*{(43)} ϕ = ψ on Γ ( t ) , (43)
∂ t η = ∇ ϕ ⋅ ( − ε ∇ x η , 1 ) on Γ ( t ) , (44) \partial_t\eta=\nabla\phi\cdot(-\varepsilon\nabla_x\eta,1) \qquad\text{on }\Gamma(t), \tag*{(44)} ∂ t η = ∇ ϕ ⋅ ( − ε ∇ x η , 1 ) on Γ ( t ) , (44)
∂ t ψ = − γ ∂ t t η − β E ( η ; ε ) + δ ∂ t Δ x η + ε ∂ x 3 ϕ ( ∇ ϕ ⋅ ( − ε ∇ x η , 1 ) ) − ε 2 ∣ ∇ ϕ ∣ 2 − η on Γ ( t ) . (45) \partial_t\psi=-\gamma\partial_{tt}\eta-\beta\mathcal{E}(\eta;\varepsilon)+\delta\partial_t\Delta_x\eta+\varepsilon\partial_{x_3}\phi\left(\nabla\phi\cdot(-\varepsilon\nabla_x\eta,1)\right)-\frac{\varepsilon}{2}|\nabla\phi|^2-\eta\qquad\text{on }\Gamma(t). \tag*{(45)} ∂ t ψ = − γ ∂ tt η − β E ( η ; ε ) + δ ∂ t Δ x η + ε ∂ x 3 ϕ ( ∇ ϕ ⋅ ( − ε ∇ x η , 1 ) ) − 2 ε ∣∇ ϕ ∣ 2 − η on Γ ( t ) . (45)
The dimensionless operator E ( η ; ε ) \mathcal{E}(\eta;\varepsilon) E ( η ; ε ) is given explicitly by
E ( η ; ε ) = 1 2 L Γ ( η ; ε ) + ε 2 ( H ( η ; ε ) 3 − H ( η ; ε ) K ( η ; ε ) ) , \mathcal{E}(\eta;\varepsilon)=\frac{1}{2}\mathcal{L}_{\Gamma}(\eta;\varepsilon)+\varepsilon^2\left(\mathcal{H}(\eta;\varepsilon)^3-\mathcal{H}(\eta;\varepsilon)\mathcal{K}(\eta;\varepsilon)\right), E ( η ; ε ) = 2 1 L Γ ( η ; ε ) + ε 2 ( H ( η ; ε ) 3 − H ( η ; ε ) K ( η ; ε ) ) ,
where
L Γ ( η ; ε ) = 1 α ∂ x i ( α α i j ∂ x j H ( η ; ε ) ) , \mathcal{L}_{\Gamma}(\eta;\varepsilon)=\frac{1}{\sqrt{\alpha}}\partial_{x_i}\left(\sqrt{\alpha}\alpha^{ij}\partial_{x_j}\mathcal{H}(\eta;\varepsilon)\right), L Γ ( η ; ε ) = α 1 ∂ x i ( α α ij ∂ x j H ( η ; ε ) ) ,
and
H ( η ; ε ) = 1 2 α ( α 11 η x 1 x 1 + 2 α 12 η x 1 x 2 + α 22 η x 2 x 2 ) , K ( η ; ε ) = η x 1 x 1 η x 2 x 2 − η x 1 x 2 2 α 2 . \mathcal{H}(\eta;\varepsilon)=\frac{1}{2\sqrt{\alpha}}\left(\alpha^{11}\eta_{x_1x_1}+2\alpha^{12}\eta_{x_1x_2}+\alpha^{22}\eta_{x_2x_2}\right),\qquad\mathcal{K}(\eta;\varepsilon)=\frac{\eta_{x_1x_1}\eta_{x_2x_2}-\eta_{x_1x_2}^2}{\alpha^2}. H ( η ; ε ) = 2 α 1 ( α 11 η x 1 x 1 + 2 α 12 η x 1 x 2 + α 22 η x 2 x 2 ) , K ( η ; ε ) = α 2 η x 1 x 1 η x 2 x 2 − η x 1 x 2 2 .
α = 1 + ε 2 ∣ ∇ x η ∣ 2 , α 11 = 1 + ε 2 η x 2 2 α , α 22 = 1 + ε 2 η x 1 2 α , α 12 = − ε 2 η x 1 η x 2 α . \alpha=1+\varepsilon^2|\nabla_x\eta|^2,\qquad\alpha^{11}=\frac{1+\varepsilon^2\eta_{x_2}^2}{\alpha},\qquad\alpha^{22}=\frac{1+\varepsilon^2\eta_{x_1}^2}{\alpha},\qquad\alpha^{12}=-\frac{\varepsilon^2\eta_{x_1}\eta_{x_2}}{\alpha}. α = 1 + ε 2 ∣ ∇ x η ∣ 2 , α 11 = α 1 + ε 2 η x 2 2 , α 22 = α 1 + ε 2 η x 1 2 , α 12 = − α ε 2 η x 1 η x 2 .
In this section, our goal is to transform the free-boundary problem into a fixed reference domain
Ω = T 2 × ( − ∞ , 0 ) , Γ = T 2 × { 0 } . \Omega=\mathbb{T}^2\times(-\infty,0),\qquad\Gamma=\mathbb{T}^2\times\{0\}. Ω = T 2 × ( − ∞ , 0 ) , Γ = T 2 × { 0 } .
To this end, we introduce a time-dependent diffeomorphism
Ψ ( ⋅ , t ) : Ω ⟶ Ω ( t ) , \Psi(\cdot,t):\Omega\longrightarrow\Omega(t), Ψ ( ⋅ , t ) : Ω ⟶ Ω ( t ) ,
defined by
Ψ ( x 1 , x 2 , x 3 , t ) = ( x 1 , x 2 , x 3 + ε η ( x 1 , x 2 , t ) ) . \Psi(x_1,x_2,x_3,t)=(x_1,x_2,x_3+\varepsilon\eta(x_1,x_2,t)). Ψ ( x 1 , x 2 , x 3 , t ) = ( x 1 , x 2 , x 3 + ε η ( x 1 , x 2 , t )) .
This mapping straightens the moving free surface x 3 = ε η ( x 1 , x 2 , t ) x_3=\varepsilon\eta(x_1,x_2,t) x 3 = ε η ( x 1 , x 2 , t ) into the fixed boundary x 3 = 0 x_3=0 x 3 = 0 . Given a scalar function f f f defined on Ω ( t ) \Omega(t) Ω ( t ) , we define its ALE counterpart on the reference domain Ω \Omega Ω by composition with Ψ \Psi Ψ , that is,
F = f ∘ Ψ . F=f\circ\Psi. F = f ∘ Ψ.
In particular, we introduce the ALE potential
Φ = ϕ ∘ Ψ \Phi=\phi\circ\Psi Φ = ϕ ∘ Ψ
We now compute the gradient of the diffeomorphism Ψ \Psi Ψ . A direct calculation yields
∇ Ψ = ( 1 0 0 0 1 0 ε ∂ x 1 η ( x , t ) ε ∂ x 2 η ( x , t ) 1 ) , ( ∇ Ψ ) − 1 = ( 1 0 0 0 1 0 − ε ∂ x 1 η ( x , t ) − ε ∂ x 2 η ( x , t ) 1 ) . \nabla\Psi=
\begin{pmatrix}
1 & 0 & 0\\
0 & 1 & 0\\
\varepsilon\partial_{x_1}\eta(x,t) & \varepsilon\partial_{x_2}\eta(x,t) & 1
\end{pmatrix},
\qquad
(\nabla\Psi)^{-1}=
\begin{pmatrix}
1 & 0 & 0\\
0 & 1 & 0\\
-\varepsilon\partial_{x_1}\eta(x,t) & -\varepsilon\partial_{x_2}\eta(x,t) & 1
\end{pmatrix}. ∇Ψ = 1 0 ε ∂ x 1 η ( x , t ) 0 1 ε ∂ x 2 η ( x , t ) 0 0 1 , ( ∇Ψ ) − 1 = 1 0 − ε ∂ x 1 η ( x , t ) 0 1 − ε ∂ x 2 η ( x , t ) 0 0 1 .
Adopting the Einstein summation convention and defining the ALE pullback F : = f ∘ Ψ F:=f\circ\Psi F := f ∘ Ψ , the chain rule yields
( ∂ x j f ) ∘ Ψ = A j k ∂ x k F , A = ( ∇ Ψ ) − 1 . (\partial_{x_j}f)\circ\Psi=A_j^k\partial_{x_k}F,
\qquad
A=(\nabla\Psi)^{-1}. ( ∂ x j f ) ∘ Ψ = A j k ∂ x k F , A = ( ∇Ψ ) − 1 .
Consequently,
( Δ f ) ∘ Ψ = A j i ∂ x i ( A j k ∂ x k F ) . (\Delta f)\circ\Psi=A_j^i\partial_{x_i}\left(A_j^k\partial_{x_k}F\right). ( Δ f ) ∘ Ψ = A j i ∂ x i ( A j k ∂ x k F ) .
Using the explicit form of A A A and expanding up to first order in ε \varepsilon ε , we obtain
( Δ f ) ∘ Ψ = Δ F − ε [ ( η x 1 x 1 + η x 2 x 2 ) ∂ x 3 F + η x 1 ( ∂ x 3 x 1 F + ∂ x 1 x 3 F ) + η x 2 ( ∂ x 3 x 2 F + ∂ x 2 x 3 F ) ] + O ( ε 2 ) . (\Delta f)\circ\Psi=\Delta F-\varepsilon\left[(\eta_{x_1x_1}+\eta_{x_2x_2})\partial_{x_3}F+\eta_{x_1}(\partial_{x_3x_1}F+\partial_{x_1x_3}F)+\eta_{x_2}(\partial_{x_3x_2}F+\partial_{x_2x_3}F)\right]+\mathcal{O}(\varepsilon^2). ( Δ f ) ∘ Ψ = Δ F − ε [ ( η x 1 x 1 + η x 2 x 2 ) ∂ x 3 F + η x 1 ( ∂ x 3 x 1 F + ∂ x 1 x 3 F ) + η x 2 ( ∂ x 3 x 2 F + ∂ x 2 x 3 F ) ] + O ( ε 2 ) .
Finally, the composition of equations (42) –(45) the system with the diffeomorphism reads to
Δ Φ − ε [ ( η x 1 x 1 + η x 2 x 2 ) Φ x 3 + η x 1 ( Φ x 3 x 1 + Φ x 1 x 3 ) + η x 2 ( Φ x 3 x 2 + Φ x 2 x 3 ) ] = O ( ε 2 ) in Ω , Φ = ψ on Γ , ∂ t η = Φ x 3 − ε ( η x 1 Φ x 1 + η x 2 Φ x 2 ) + O ( ε 2 ) on Γ , ∂ t ψ = − Y ∂ t η − β 4 Δ x 2 η + δ ∂ t Δ x η − η + ε 2 ( Φ x 3 2 − ∣ ∇ Φ ∣ 2 ) + O ( ε 2 ) on Γ . (46) \begin{aligned}
\Delta\Phi-\varepsilon\left[(\eta_{x_1x_1}+\eta_{x_2x_2})\Phi_{x_3}+\eta_{x_1}(\Phi_{x_3x_1}+\Phi_{x_1x_3})+\eta_{x_2}(\Phi_{x_3x_2}+\Phi_{x_2x_3})\right]&=\mathcal{O}(\varepsilon^2) &&\text{in }\Omega,\\
\Phi=\psi&&\text{on }\Gamma,\\
\partial_t\eta=\Phi_{x_3}-\varepsilon(\eta_{x_1}\Phi_{x_1}+\eta_{x_2}\Phi_{x_2})+\mathcal{O}(\varepsilon^2) &&\text{on }\Gamma,\\
\partial_t\psi=-\mathcal{Y}\,\partial_t\eta-\frac{\beta}{4}\Delta_x^2\eta+\delta\,\partial_t\Delta_x\eta-\eta+\frac{\varepsilon}{2}\left(\Phi_{x_3}^2-\lvert\nabla\Phi\rvert^2\right)+\mathcal{O}(\varepsilon^2) &&\text{on }\Gamma. \tag*{(46)}
\end{aligned} ΔΦ − ε [ ( η x 1 x 1 + η x 2 x 2 ) Φ x 3 + η x 1 ( Φ x 3 x 1 + Φ x 1 x 3 ) + η x 2 ( Φ x 3 x 2 + Φ x 2 x 3 ) ] Φ = ψ ∂ t η = Φ x 3 − ε ( η x 1 Φ x 1 + η x 2 Φ x 2 ) + O ( ε 2 ) ∂ t ψ = − Y ∂ t η − 4 β Δ x 2 η + δ ∂ t Δ x η − η + 2 ε ( Φ x 3 2 − ∣ ∇Φ ∣ 2 ) + O ( ε 2 ) = O ( ε 2 ) on Γ , on Γ , on Γ. in Ω , (46)
Φ = ψ (47) \Phi=\psi \tag*{(47)} Φ = ψ (47)
∂ t η = Φ x 3 − ε ( η x 1 Φ x 1 + η x 2 Φ x 2 ) + O ( ε 2 ) (48) \partial_t\eta=\Phi_{x_3}-\varepsilon(\eta_{x_1}\Phi_{x_1}+\eta_{x_2}\Phi_{x_2})+\mathcal{O}(\varepsilon^2) \tag*{(48)} ∂ t η = Φ x 3 − ε ( η x 1 Φ x 1 + η x 2 Φ x 2 ) + O ( ε 2 ) (48)
∂ t ψ = − Y ∂ t η − β 4 Δ x 2 η + δ ∂ t Δ x η − η + ε 2 ( Φ x 3 2 − ∣ ∇ Φ ∣ 2 ) + O ( ε 2 ) (49) \partial_t\psi=-\mathcal{Y}\,\partial_t\eta-\frac{\beta}{4}\Delta_x^2\eta+\delta\,\partial_t\Delta_x\eta-\eta+\frac{\varepsilon}{2}\left(\Phi_{x_3}^2-\lvert\nabla\Phi\rvert^2\right)+\mathcal{O}(\varepsilon^2) \tag*{(49)} ∂ t ψ = − Y ∂ t η − 4 β Δ x 2 η + δ ∂ t Δ x η − η + 2 ε ( Φ x 3 2 − ∣ ∇Φ ∣ 2 ) + O ( ε 2 ) (49)
Remark 2.3 (Leading-order reduction of the bending operator). At leading order in the small-steepness parameter ε \varepsilon ε , the geometric bending operator E ( η ; ε ) \mathcal{E}(\eta;\varepsilon) E ( η ; ε ) reduces to a linear biharmonic operator. Indeed,
α = 1 + O ( ε 2 ) , α i j = δ i j + O ( ε 2 ) , \alpha=1+\mathcal{O}(\varepsilon^2),
\qquad
\alpha^{ij}=\delta_{ij}+\mathcal{O}(\varepsilon^2), α = 1 + O ( ε 2 ) , α ij = δ ij + O ( ε 2 ) ,
so that
H ( η ; ε ) = 1 2 Δ x η + O ( ε 2 ) , E ( η ; ε ) = 1 4 Δ x 2 η + O ( ε 2 ) . \mathcal{H}(\eta;\varepsilon)=\frac{1}{2}\Delta_x\eta+\mathcal{O}(\varepsilon^2),
\qquad
\mathcal{E}(\eta;\varepsilon)=\frac{1}{4}\Delta_x^2\eta+\mathcal{O}(\varepsilon^2). H ( η ; ε ) = 2 1 Δ x η + O ( ε 2 ) , E ( η ; ε ) = 4 1 Δ x 2 η + O ( ε 2 ) .
Accordingly, in equation (63) the elastic response of the plate is governed by a biharmonic operator, while nonlinear geometric effects are encapsulated in the O ( ε 2 ) \mathcal{O}(\varepsilon^2) O ( ε 2 ) terms.
We now derive a hierarchy of approximate models by performing a formal small–steepness expansion in the parameter ε \varepsilon ε in the ALE system (60) –(3.14). Specifically, we seek Φ \Phi Φ , ψ \psi ψ and η \eta η as power series in ε \varepsilon ε of the form
Φ ( x 1 , x 2 , x 3 , t ) = ∑ n = 0 ∞ ε n Φ ( n ) ( x 1 , x 2 , x 3 , t ) , (50) \Phi(x_1,x_2,x_3,t)=\sum_{n=0}^{\infty}\varepsilon^n\Phi^{(n)}(x_1,x_2,x_3,t), \tag*{(50)} Φ ( x 1 , x 2 , x 3 , t ) = n = 0 ∑ ∞ ε n Φ ( n ) ( x 1 , x 2 , x 3 , t ) , (50)
ψ ( x 1 , x 2 , t ) = ∑ n = 0 ∞ ε n ψ ( n ) ( x 1 , x 2 , t ) , η ( x 1 , x 2 , t ) = ∑ n = 0 ∞ ε n η ( n ) ( x 1 , x 2 , t ) , (51) \psi(x_1,x_2,t)=\sum_{n=0}^{\infty}\varepsilon^n\psi^{(n)}(x_1,x_2,t), \qquad\eta(x_1,x_2,t)=\sum_{n=0}^{\infty}\varepsilon^n\eta^{(n)}(x_1,x_2,t), \tag*{(51)} ψ ( x 1 , x 2 , t ) = n = 0 ∑ ∞ ε n ψ ( n ) ( x 1 , x 2 , t ) , η ( x 1 , x 2 , t ) = n = 0 ∑ ∞ ε n η ( n ) ( x 1 , x 2 , t ) , (51)
Substituting (50) –(51) into (60) –(63) and collecting terms of equal powers of ε \varepsilon ε yields, at each order n ≥ 0 n\geq0 n ≥ 0 , a linear boundary value problem for Φ ( n ) \Phi^{(n)} Φ ( n ) in Ω \Omega Ω coupled to evolution equations for ψ ( n ) \psi^{(n)} ψ ( n ) and η ( n ) \eta^{(n)} η ( n ) on Γ = { x 3 = 0 } \Gamma=\{x_3=0\} Γ = { x 3 = 0 } . In particular, the leading order n = 0 n=0 n = 0 provides the principal (linear) hydroelastic model, while n = 1 n=1 n = 1 captures the first nonlinear correction.
The initial data are expanded consistently as
Φ ( x , x 3 , 0 ) = ∑ n = 0 ∞ ε n Φ ( n ) ( x , x 3 , 0 ) , ψ ( x , 0 ) = ∑ n = 0 ∞ ε n ψ ( n ) ( x , 0 ) , η ( x , 0 ) = ∑ n = 0 ∞ ε n η ( n ) ( x , 0 ) , \Phi(x,x_3,0)=\sum_{n=0}^{\infty}\varepsilon^n\Phi^{(n)}(x,x_3,0), \qquad\psi(x,0)=\sum_{n=0}^{\infty}\varepsilon^n\psi^{(n)}(x,0), \qquad\eta(x,0)=\sum_{n=0}^{\infty}\varepsilon^n\eta^{(n)}(x,0), Φ ( x , x 3 , 0 ) = n = 0 ∑ ∞ ε n Φ ( n ) ( x , x 3 , 0 ) , ψ ( x , 0 ) = n = 0 ∑ ∞ ε n ψ ( n ) ( x , 0 ) , η ( x , 0 ) = n = 0 ∑ ∞ ε n η ( n ) ( x , 0 ) ,
so that Φ ( n ) ( ⋅ , ⋅ , 0 ) \Phi^{(n)}(\cdot,\cdot,0) Φ ( n ) ( ⋅ , ⋅ , 0 ) , ψ ( n ) ( ⋅ , 0 ) \psi^{(n)}(\cdot,0) ψ ( n ) ( ⋅ , 0 ) and η ( n ) ( ⋅ , 0 ) \eta^{(n)}(\cdot,0) η ( n ) ( ⋅ , 0 ) represent the n n n -th order initial profiles in the expansion.
The case n = 0 n=0 n = 0 . We substitute the formal expansions (50) –(51) into the ALE system (60) –(63) and retaining only the leading-order contributions yields the following system for ( Φ ( 0 ) , ψ ( 0 ) , η ( 0 ) ) (\Phi^{(0)},\psi^{(0)},\eta^{(0)}) ( Φ ( 0 ) , ψ ( 0 ) , η ( 0 ) ) :
Δ Φ ( 0 ) = 0 in Ω , (52) \Delta\Phi^{(0)}=0 \qquad\text{in }\Omega, \tag*{(52)} Δ Φ ( 0 ) = 0 in Ω , (52)
Φ ( 0 ) = ψ ( 0 ) on Γ , (53) \Phi^{(0)}=\psi^{(0)} \qquad\text{on }\Gamma, \tag*{(53)} Φ ( 0 ) = ψ ( 0 ) on Γ , (53)
∂ t η ( 0 ) = ∂ x 3 Φ ( 0 ) on Γ , (54) \partial_t\eta^{(0)}=\partial_{x_3}\Phi^{(0)} \qquad\text{on }\Gamma, \tag*{(54)} ∂ t η ( 0 ) = ∂ x 3 Φ ( 0 ) on Γ , (54)
∂ t ψ ( 0 ) = − Γ ∂ t η ( 0 ) − β 4 Δ x 2 η ( 0 ) + δ ∂ t Δ x η ( 0 ) − η ( 0 ) on Γ . (55) \partial_t\psi^{(0)}=-\Gamma\partial_t\eta^{(0)}-\frac{\beta}{4}\Delta_x^2\eta^{(0)}+\delta\partial_t\Delta_x\eta^{(0)}-\eta^{(0)} \qquad\text{on }\Gamma. \tag*{(55)} ∂ t ψ ( 0 ) = − Γ ∂ t η ( 0 ) − 4 β Δ x 2 η ( 0 ) + δ ∂ t Δ x η ( 0 ) − η ( 0 ) on Γ. (55)
Taking the Fourier series in x ∈ T 2 x\in\mathbb{T}^2 x ∈ T 2 , the unique decaying solution of (52) –(53) is
Φ ^ ( 0 ) ( k , x 3 , t ) = ψ ^ ( 0 ) ( k , t ) e ∣ k ∣ x 3 , k ∈ Z 2 , x 3 ≤ 0 , \widehat{\Phi}^{(0)}(k,x_3,t)=\widehat{\psi}^{(0)}(k,t)e^{|k|x_3}, \qquad k\in\mathbb{Z}^2,\quad x_3\leq0, Φ ( 0 ) ( k , x 3 , t ) = ψ ( 0 ) ( k , t ) e ∣ k ∣ x 3 , k ∈ Z 2 , x 3 ≤ 0 ,
and in particular
∂ x 3 Φ ( 0 ) ∣ Γ = Λ ψ ( 0 ) . \left.\partial_{x_3}\Phi^{(0)}\right|_{\Gamma}=\Lambda\psi^{(0)}. ∂ x 3 Φ ( 0 ) Γ = Λ ψ ( 0 ) .
Therefore, the kinematic condition (54) becomes
∂ t η ( 0 ) = Λ ψ ( 0 ) on Γ . (56) \partial_t\eta^{(0)}=\Lambda\psi^{(0)} \qquad\text{on }\Gamma. \tag*{(56)} ∂ t η ( 0 ) = Λ ψ ( 0 ) on Γ. (56)
Differentiating (56) in time and using (55) yields a closed equation for η ( 0 ) \eta^{(0)} η ( 0 ) :
∂ t t η ( 0 ) = Λ ( − Γ ∂ t t η ( 0 ) − β 4 Δ x 2 η ( 0 ) + δ ∂ t Δ x η ( 0 ) − η ( 0 ) ) , on Γ . (57) \partial_{tt}\eta^{(0)}=\Lambda\left(-\Gamma\partial_{tt}\eta^{(0)}-\frac{\beta}{4}\Delta_x^2\eta^{(0)}+\delta\partial_t\Delta_x\eta^{(0)}-\eta^{(0)}\right), \qquad\text{on }\Gamma. \tag*{(57)} ∂ tt η ( 0 ) = Λ ( − Γ ∂ tt η ( 0 ) − 4 β Δ x 2 η ( 0 ) + δ ∂ t Δ x η ( 0 ) − η ( 0 ) ) , on Γ. (57)
Equivalently,
( I + Υ Λ ) ∂ t η ( 0 ) + β 4 Λ Δ x 2 η ( 0 ) − δ Λ ∂ t Δ x η ( 0 ) + Λ η ( 0 ) = 0. (58) (I+\Upsilon\Lambda)\partial_t\eta^{(0)}+\frac{\beta}{4}\Lambda\Delta_x^2\eta^{(0)}-\delta\Lambda\partial_t\Delta_x\eta^{(0)}+\Lambda\eta^{(0)}=0. \tag*{(58)} ( I + ΥΛ ) ∂ t η ( 0 ) + 4 β Λ Δ x 2 η ( 0 ) − δ Λ ∂ t Δ x η ( 0 ) + Λ η ( 0 ) = 0. (58)
Let η ^ ( 0 ) ( k , t ) \widehat{\eta}^{(0)}(k,t) η ( 0 ) ( k , t ) denote the Fourier coefficient of η ( 0 ) \eta^{(0)} η ( 0 ) . For k ≠ 0 k\ne0 k = 0 , using Λ f ^ ( k ) = ∣ k ∣ f ^ ( k ) \widehat{\Lambda f}(k)=|k|\widehat{f}(k) Λ f ( k ) = ∣ k ∣ f ( k ) and Δ x f ^ ( k ) = − ∣ k ∣ 2 f ^ ( k ) \widehat{\Delta_x f}(k)=-|k|^2\widehat{f}(k) Δ x f ( k ) = − ∣ k ∣ 2 f ( k ) , equation (57) becomes
( 1 + Υ ∣ k ∣ ) ∂ t 2 η ^ ( 0 ) ( k , t ) + δ ∣ k ∣ 3 ∂ t η ^ ( 0 ) ( k , t ) + ( ∣ k ∣ + β 4 ∣ k ∣ 5 ) η ^ ( 0 ) ( k , t ) = 0. (59) (1+\Upsilon|k|)\partial_t^2\widehat{\eta}^{(0)}(k,t)+\delta|k|^3\partial_t\widehat{\eta}^{(0)}(k,t)+\left(|k|+\frac{\beta}{4}|k|^5\right)\widehat{\eta}^{(0)}(k,t)=0. \tag*{(59)} ( 1 + Υ∣ k ∣ ) ∂ t 2 η ( 0 ) ( k , t ) + δ ∣ k ∣ 3 ∂ t η ( 0 ) ( k , t ) + ( ∣ k ∣ + 4 β ∣ k ∣ 5 ) η ( 0 ) ( k , t ) = 0. (59)
The case n = 1 n=1 n = 1 . Similarly, in the case n = 1 n=1 n = 1 the leading-order contributions yields the following system for ( Φ ( 1 ) , ψ ( 1 ) , η ( 1 ) ) (\Phi^{(1)},\psi^{(1)},\eta^{(1)}) ( Φ ( 1 ) , ψ ( 1 ) , η ( 1 ) ) :
Δ Φ ( 1 ) = [ ( η x 1 ( 0 ) + η x 2 ( 0 ) ) Φ x 3 ( 0 ) + η x 1 ( 0 ) ( Φ x 3 x 1 ( 0 ) + Φ x 1 x 3 ( 0 ) ) + η x 2 ( 0 ) ( Φ x 3 x 2 ( 0 ) + Φ x 2 x 3 ( 0 ) ) ] in Ω , (60) \Delta\Phi^{(1)}=\left[(\eta_{x_1}^{(0)}+\eta_{x_2}^{(0)})\Phi_{x_3}^{(0)}+\eta_{x_1}^{(0)}\left(\Phi_{x_3x_1}^{(0)}+\Phi_{x_1x_3}^{(0)}\right)+\eta_{x_2}^{(0)}\left(\Phi_{x_3x_2}^{(0)}+\Phi_{x_2x_3}^{(0)}\right)\right]\quad\text{in }\Omega, \tag*{(60)} Δ Φ ( 1 ) = [ ( η x 1 ( 0 ) + η x 2 ( 0 ) ) Φ x 3 ( 0 ) + η x 1 ( 0 ) ( Φ x 3 x 1 ( 0 ) + Φ x 1 x 3 ( 0 ) ) + η x 2 ( 0 ) ( Φ x 3 x 2 ( 0 ) + Φ x 2 x 3 ( 0 ) ) ] in Ω , (60)
Φ ( 1 ) = ψ ( 1 ) on Γ , (61) \Phi^{(1)}=\psi^{(1)}\quad\text{on }\Gamma, \tag*{(61)} Φ ( 1 ) = ψ ( 1 ) on Γ , (61)
∂ t η ( 1 ) = Φ x 3 ( 1 ) − ( η x 1 ( 0 ) Φ x 1 ( 0 ) + η x 2 ( 0 ) Φ x 2 ( 0 ) ) on Γ , (62) \partial_t\eta^{(1)}=\Phi_{x_3}^{(1)}-\left(\eta_{x_1}^{(0)}\Phi_{x_1}^{(0)}+\eta_{x_2}^{(0)}\Phi_{x_2}^{(0)}\right)\quad\text{on }\Gamma, \tag*{(62)} ∂ t η ( 1 ) = Φ x 3 ( 1 ) − ( η x 1 ( 0 ) Φ x 1 ( 0 ) + η x 2 ( 0 ) Φ x 2 ( 0 ) ) on Γ , (62)
∂ t ψ ( 1 ) = − Υ ∂ t η ( 1 ) − β 4 Δ x 2 η ( 1 ) + δ ∂ t Δ x η ( 1 ) − η ( 1 ) + 1 2 ( ( Φ x 3 ( 0 ) ) 2 − ∣ ∇ x Φ ( 0 ) ∣ 2 ) on Γ . (63) \partial_t\psi^{(1)}=-\Upsilon\partial_t\eta^{(1)}-\frac{\beta}{4}\Delta_x^2\eta^{(1)}+\delta\partial_t\Delta_x\eta^{(1)}-\eta^{(1)}+\frac{1}{2}\left((\Phi_{x_3}^{(0)})^2-|\nabla_x\Phi^{(0)}|^2\right)\quad\text{on }\Gamma. \tag*{(63)} ∂ t ψ ( 1 ) = − Υ ∂ t η ( 1 ) − 4 β Δ x 2 η ( 1 ) + δ ∂ t Δ x η ( 1 ) − η ( 1 ) + 2 1 ( ( Φ x 3 ( 0 ) ) 2 − ∣ ∇ x Φ ( 0 ) ∣ 2 ) on Γ. (63)
A convenient form for the bulk source term is to write
b ( x 1 , x 2 , x 3 , t ) = Δ x η ( 0 ) ∂ x 3 Φ ( 0 ) + 2 ∇ x η ( 0 ) ⋅ ∇ x ∂ x 3 Φ ( 0 ) . b(x_1,x_2,x_3,t)=\Delta_x\eta^{(0)}\partial_{x_3}\Phi^{(0)}+2\nabla_x\eta^{(0)}\cdot\nabla_x\partial_{x_3}\Phi^{(0)}. b ( x 1 , x 2 , x 3 , t ) = Δ x η ( 0 ) ∂ x 3 Φ ( 0 ) + 2 ∇ x η ( 0 ) ⋅ ∇ x ∂ x 3 Φ ( 0 ) .
Therefore, by using Lemma 1.1 we can compute
∂ x 3 Φ ^ ( 1 ) ( k , 0 , t ) = ∣ k ∣ ψ ^ ( 1 ) ( k , t ) + ∫ − ∞ 0 b ^ ( k , y 3 , t ) e ∣ k ∣ y 3 d y 3 , k = ( k 1 , k 2 ) ∈ Z 2 . (64) \partial_{x_3}\widehat{\Phi}^{(1)}(k,0,t)=|k|\widehat{\psi}^{(1)}(k,t)+\int_{-\infty}^{0}\widehat{b}(k,y_3,t)e^{|k|y_3}\,dy_3,\qquad k=(k_1,k_2)\in\mathbb{Z}^2. \tag*{(64)} ∂ x 3 Φ ( 1 ) ( k , 0 , t ) = ∣ k ∣ ψ ( 1 ) ( k , t ) + ∫ − ∞ 0 b ( k , y 3 , t ) e ∣ k ∣ y 3 d y 3 , k = ( k 1 , k 2 ) ∈ Z 2 . (64)
In order to provide an explicit expression for the previous integral term, we first notice that
b ^ ( k , x 3 ) = ∑ m ∈ Z 2 Δ x η ^ ( 0 ) ( k − m ) ∂ 3 Φ ^ ( 0 ) ( m , x 3 ) + 2 ∑ m ∈ Z 2 ∇ η ^ ( 0 ) ( k − m ) ⋅ ∇ ∂ 3 Φ ^ ( 0 ) ( m , x 3 ) = ∑ m ∈ Z 2 ( − ∣ k − m ∣ 2 ) ( ∣ m ∣ ψ ^ ( 0 ) ( m ) e ∣ m ∣ x 3 ) η ^ ( 0 ) ( k − m ) + 2 ∑ m ∈ Z 2 ( i ( k − m ) η ^ ( 0 ) ( k − m ) ) ⋅ ( i m ∣ m ∣ ψ ^ ( 0 ) ( m ) e ∣ m ∣ x 3 ) = − ∑ m ∈ Z 2 ∣ m ∣ ( ∣ k ∣ 2 − ∣ m ∣ 2 ) η ^ ( 0 ) ( k − m ) ψ ^ ( 0 ) ( m ) e ∣ m ∣ x 3 . (65) \begin{aligned}
\widehat{b}(k,x_3)={}&\sum_{m\in\mathbb{Z}^2}\widehat{\Delta_x\eta}^{(0)}(k-m)\,\partial_3\widehat{\Phi}^{(0)}(m,x_3)+2\sum_{m\in\mathbb{Z}^2}\widehat{\nabla\eta}^{(0)}(k-m)\cdot\widehat{\nabla\partial_3\Phi}^{(0)}(m,x_3)\\
={}&\sum_{m\in\mathbb{Z}^2}\left(-|k-m|^2\right)\left(|m|\widehat{\psi}^{(0)}(m)e^{|m|x_3}\right)\widehat{\eta}^{(0)}(k-m)\\
&+2\sum_{m\in\mathbb{Z}^2}\left(i(k-m)\widehat{\eta}^{(0)}(k-m)\right)\cdot\left(im|m|\widehat{\psi}^{(0)}(m)e^{|m|x_3}\right)\\
={}&-\sum_{m\in\mathbb{Z}^2}|m|\left(|k|^2-|m|^2\right)\widehat{\eta}^{(0)}(k-m)\widehat{\psi}^{(0)}(m)e^{|m|x_3}. \tag*{(65)}
\end{aligned} b ( k , x 3 ) = = = m ∈ Z 2 ∑ Δ x η ( 0 ) ( k − m ) ∂ 3 Φ ( 0 ) ( m , x 3 ) + 2 m ∈ Z 2 ∑ ∇ η ( 0 ) ( k − m ) ⋅ ∇ ∂ 3 Φ ( 0 ) ( m , x 3 ) m ∈ Z 2 ∑ ( − ∣ k − m ∣ 2 ) ( ∣ m ∣ ψ ( 0 ) ( m ) e ∣ m ∣ x 3 ) η ( 0 ) ( k − m ) + 2 m ∈ Z 2 ∑ ( i ( k − m ) η ( 0 ) ( k − m ) ) ⋅ ( im ∣ m ∣ ψ ( 0 ) ( m ) e ∣ m ∣ x 3 ) − m ∈ Z 2 ∑ ∣ m ∣ ( ∣ k ∣ 2 − ∣ m ∣ 2 ) η ( 0 ) ( k − m ) ψ ( 0 ) ( m ) e ∣ m ∣ x 3 . (65)
Thus, using (64) and (65) , we obtain
∂ x 3 Φ ( 1 ) = Λ ψ ( 1 ) − [ Λ , η ( 0 ) ] Λ ψ ( 0 ) , on Γ . (66) \partial_{x_3}\Phi^{(1)}=\Lambda\psi^{(1)}-\left[\Lambda,\eta^{(0)}\right]\Lambda\psi^{(0)},\quad\text{on }\Gamma. \tag*{(66)} ∂ x 3 Φ ( 1 ) = Λ ψ ( 1 ) − [ Λ , η ( 0 ) ] Λ ψ ( 0 ) , on Γ. (66)
Combining (66) and (62) , the evolution equation for η ( 1 ) \eta^{(1)} η ( 1 ) becomes
∂ t η ( 1 ) = Λ ψ ( 1 ) − [ Λ , η ( 0 ) ] Λ ψ ( 0 ) − ( η x 1 ( 0 ) Φ x 1 ( 0 ) + η x 2 ( 0 ) Φ x 2 ( 0 ) ) , on Γ . (67) \partial_t\eta^{(1)}=\Lambda\psi^{(1)}-\left[\Lambda,\eta^{(0)}\right]\Lambda\psi^{(0)}-\left(\eta_{x_1}^{(0)}\Phi_{x_1}^{(0)}+\eta_{x_2}^{(0)}\Phi_{x_2}^{(0)}\right),\quad\text{on }\Gamma. \tag*{(67)} ∂ t η ( 1 ) = Λ ψ ( 1 ) − [ Λ , η ( 0 ) ] Λ ψ ( 0 ) − ( η x 1 ( 0 ) Φ x 1 ( 0 ) + η x 2 ( 0 ) Φ x 2 ( 0 ) ) , on Γ. (67)
Differentiating (67) in time and using (63) we deduce
∂ t t η ( 1 ) = Λ ( − Υ ∂ t t η ( 1 ) − β 4 Δ x 2 η ( 1 ) + δ ∂ t Δ x η ( 1 ) − η ( 1 ) + 1 2 ( ( Λ ψ ( 0 ) ) 2 − ∣ ∇ x ψ ( 0 ) ∣ 2 ) ) − ∂ t ( [ Λ , η ( 0 ) ] Λ ψ ( 0 ) − ∂ t ( η x 1 ( 0 ) Φ x 1 ( 0 ) + η x 2 ( 0 ) Φ x 2 ( 0 ) ) ) . (68) \partial_{tt}\eta^{(1)}=\Lambda\left(-\Upsilon\partial_{tt}\eta^{(1)}-\frac{\beta}{4}\Delta_x^2\eta^{(1)}+\delta\partial_t\Delta_x\eta^{(1)}-\eta^{(1)}+\frac{1}{2}\left((\Lambda\psi^{(0)})^2-\lvert\nabla_x\psi^{(0)}\rvert^2\right)\right)-\partial_t\left([\Lambda,\eta^{(0)}]\Lambda\psi^{(0)}-\partial_t\left(\eta_{x_1}^{(0)}\Phi_{x_1}^{(0)}+\eta_{x_2}^{(0)}\Phi_{x_2}^{(0)}\right)\right). \tag*{(68)} ∂ tt η ( 1 ) = Λ ( − Υ ∂ tt η ( 1 ) − 4 β Δ x 2 η ( 1 ) + δ ∂ t Δ x η ( 1 ) − η ( 1 ) + 2 1 ( ( Λ ψ ( 0 ) ) 2 − ∣ ∇ x ψ ( 0 ) ∣ 2 ) ) − ∂ t ( [ Λ , η ( 0 ) ] Λ ψ ( 0 ) − ∂ t ( η x 1 ( 0 ) Φ x 1 ( 0 ) + η x 2 ( 0 ) Φ x 2 ( 0 ) ) ) . (68)
Rearranging the terms and collecting all contributions involving η ( 1 ) \eta^{(1)} η ( 1 ) on the left-hand side, we obtain
( I + Υ Λ ) ∂ t t η ( 1 ) + β 4 Λ Δ x 2 η ( 1 ) − δ Λ ∂ t Δ x η ( 1 ) + Λ η ( 1 ) = 1 2 Λ ( ( Λ ψ ( 0 ) ) 2 − ∣ ∇ x ψ ( 0 ) ∣ 2 ) − ∂ t ( [ Λ , η ( 0 ) ] Λ ψ ( 0 ) ) − ∂ t ( η x 1 ( 0 ) Φ x 1 ( 0 ) + η x 2 ( 0 ) Φ x 2 ( 0 ) ) . (69) \begin{aligned}
(I+\Upsilon\Lambda)\partial_{tt}\eta^{(1)}+\frac{\beta}{4}\Lambda\Delta_x^2\eta^{(1)}-\delta\Lambda\partial_t\Delta_x\eta^{(1)}+\Lambda\eta^{(1)} \\
&=\frac{1}{2}\Lambda\left((\Lambda\psi^{(0)})^2-\lvert\nabla_x\psi^{(0)}\rvert^2\right)-\partial_t\left([\Lambda,\eta^{(0)}]\Lambda\psi^{(0)}\right)-\partial_t\left(\eta_{x_1}^{(0)}\Phi_{x_1}^{(0)}+\eta_{x_2}^{(0)}\Phi_{x_2}^{(0)}\right). \tag*{(69)}
\end{aligned} ( I + ΥΛ ) ∂ tt η ( 1 ) + 4 β Λ Δ x 2 η ( 1 ) − δ Λ ∂ t Δ x η ( 1 ) + Λ η ( 1 ) = 2 1 Λ ( ( Λ ψ ( 0 ) ) 2 − ∣ ∇ x ψ ( 0 ) ∣ 2 ) − ∂ t ( [ Λ , η ( 0 ) ] Λ ψ ( 0 ) ) − ∂ t ( η x 1 ( 0 ) Φ x 1 ( 0 ) + η x 2 ( 0 ) Φ x 2 ( 0 ) ) . (69)
Next, we eliminate ψ ( 0 ) \psi^{(0)} ψ ( 0 ) in favour of η ( 0 ) \eta^{(0)} η ( 0 ) using the relation (56) On nonzero Fourier modes, this yields
ψ ( 0 ) = Λ − 1 ∂ t η ( 0 ) , Λ ψ ( 0 ) = ∂ t η ( 0 ) , ∇ x ψ ( 0 ) = ∇ x Λ − 1 ∂ t η ( 0 ) . (70) \psi^{(0)}=\Lambda^{-1}\partial_t\eta^{(0)},\qquad\Lambda\psi^{(0)}=\partial_t\eta^{(0)},\qquad\nabla_x\psi^{(0)}=\nabla_x\Lambda^{-1}\partial_t\eta^{(0)}. \tag*{(70)} ψ ( 0 ) = Λ − 1 ∂ t η ( 0 ) , Λ ψ ( 0 ) = ∂ t η ( 0 ) , ∇ x ψ ( 0 ) = ∇ x Λ − 1 ∂ t η ( 0 ) . (70)
Therefore, (69) can be rewritten as
( I + Υ Λ ) ∂ t t η ( 1 ) + β 4 Λ Δ x 2 η ( 1 ) − δ Λ ∂ t Δ x η ( 1 ) + Λ η ( 1 ) = 1 2 Λ ( ( ∂ t η ( 0 ) ) 2 − ∣ ∇ x Λ − 1 ∂ t η ( 0 ) ∣ 2 ) − ∂ t ( [ Λ , η ( 0 ) ] ∂ t η ( 0 ) ) \begin{aligned}
(I+\Upsilon\Lambda)\partial_{tt}\eta^{(1)}+\frac{\beta}{4}\Lambda\Delta_x^2\eta^{(1)}-\delta\Lambda\partial_t\Delta_x\eta^{(1)}+\Lambda\eta^{(1)} \\
&=\frac{1}{2}\Lambda\left((\partial_t\eta^{(0)})^2-\lvert\nabla_x\Lambda^{-1}\partial_t\eta^{(0)}\rvert^2\right)-\partial_t\left([\Lambda,\eta^{(0)}]\partial_t\eta^{(0)}\right)
\end{aligned} ( I + ΥΛ ) ∂ tt η ( 1 ) + 4 β Λ Δ x 2 η ( 1 ) − δ Λ ∂ t Δ x η ( 1 ) + Λ η ( 1 ) = 2 1 Λ ( ( ∂ t η ( 0 ) ) 2 − ∣ ∇ x Λ − 1 ∂ t η ( 0 ) ∣ 2 ) − ∂ t ( [ Λ , η ( 0 ) ] ∂ t η ( 0 ) )
− ∂ t ( ∇ x η ( 0 ) ⋅ ∇ x Λ − 1 ∂ t η ( 0 ) ) . (71) -\partial_t\left(\nabla_x\eta^{(0)}\cdot\nabla_x\Lambda^{-1}\partial_t\eta^{(0)}\right). \tag*{(71)} − ∂ t ( ∇ x η ( 0 ) ⋅ ∇ x Λ − 1 ∂ t η ( 0 ) ) . (71)
Combining the leading-order equation for η ( 0 ) \eta^{(0)} η ( 0 ) with ε \varepsilon ε times (71) , and recalling that
f = η ( 0 ) + ε η ( 1 ) , f=\eta^{(0)}+\varepsilon\eta^{(1)}, f = η ( 0 ) + ε η ( 1 ) ,
we obtain, up to terms of order O ( ε 2 ) \mathcal{O}(\varepsilon^2) O ( ε 2 ) ,
( I + Υ Λ ) ∂ t t f + β 4 Λ Δ x 2 f − δ Λ ∂ t Δ x f + Λ f = ε [ 1 2 Λ ( ( ∂ t η ( 0 ) ) 2 − ∣ ∇ x Λ − 1 ∂ t η ( 0 ) ∣ 2 ) − ∂ t ( [ Λ , η ( 0 ) ] ∂ t η ( 0 ) ) − ∂ t ( ∇ x η ( 0 ) ⋅ ∇ x Λ − 1 ∂ t η ( 0 ) ) ] . (72) \begin{aligned}
(I+\Upsilon\Lambda)\partial_{tt}f+\frac{\beta}{4}\Lambda\Delta_x^2f-\delta\Lambda\partial_t\Delta_xf+\Lambda f \\
&=\varepsilon\left[\frac{1}{2}\Lambda\left((\partial_t\eta^{(0)})^2-\lvert\nabla_x\Lambda^{-1}\partial_t\eta^{(0)}\rvert^2\right)-\partial_t\left([\Lambda,\eta^{(0)}]\partial_t\eta^{(0)}\right)-\partial_t\left(\nabla_x\eta^{(0)}\cdot\nabla_x\Lambda^{-1}\partial_t\eta^{(0)}\right)\right]. \tag*{(72)}
\end{aligned} ( I + ΥΛ ) ∂ tt f + 4 β Λ Δ x 2 f − δ Λ ∂ t Δ x f + Λ f = ε [ 2 1 Λ ( ( ∂ t η ( 0 ) ) 2 − ∣ ∇ x Λ − 1 ∂ t η ( 0 ) ∣ 2 ) − ∂ t ( [ Λ , η ( 0 ) ] ∂ t η ( 0 ) ) − ∂ t ( ∇ x η ( 0 ) ⋅ ∇ x Λ − 1 ∂ t η ( 0 ) ) ] . (72)
Since ε f = ε η ( 0 ) + O ( ε 2 ) \varepsilon f=\varepsilon\eta^{(0)}+\mathcal{O}(\varepsilon^2) ε f = ε η ( 0 ) + O ( ε 2 ) , we may replace η ( 0 ) \eta^{(0)} η ( 0 ) by f f f in the right-hand side at the cost of an O ( ε 2 ) \mathcal{O}(\varepsilon^2) O ( ε 2 ) error. Hence, neglecting O ( ε 2 ) \mathcal{O}(\varepsilon^2) O ( ε 2 ) terms and recalling that Δ x 2 = Λ 4 \Delta_x^2=\Lambda^4 Δ x 2 = Λ 4 and R = ∇ x Λ − 1 \mathcal{R}=\nabla_x\Lambda^{-1} R = ∇ x Λ − 1 the equation closes as
( I + Υ Λ ) f t t + δ Λ 3 f t + ( Λ + β 4 Λ 5 ) f = ε [ 1 2 Λ ( ( ∂ t f ) 2 − ∣ R ∂ t f ∣ 2 ) − ∂ t ( [ Λ , f ] ∂ t f ) − ∂ t ( ∇ x f ⋅ R ∂ t f ) ] . (73) \begin{aligned}
(I+\Upsilon\Lambda)f_{tt}+\delta\Lambda^3f_t+\left(\Lambda+\frac{\beta}{4}\Lambda^5\right)f \\
&=\varepsilon\left[\frac{1}{2}\Lambda\left((\partial_tf)^2-\lvert\mathcal{R}\partial_tf\rvert^2\right)-\partial_t\left([\Lambda,f]\partial_tf\right)-\partial_t\left(\nabla_xf\cdot\mathcal{R}\partial_tf\right)\right]. \tag*{(73)}
\end{aligned} ( I + ΥΛ ) f tt + δ Λ 3 f t + ( Λ + 4 β Λ 5 ) f = ε [ 2 1 Λ ( ( ∂ t f ) 2 − ∣ R ∂ t f ∣ 2 ) − ∂ t ( [ Λ , f ] ∂ t f ) − ∂ t ( ∇ x f ⋅ R ∂ t f ) ] . (73)
In the sequel we derive from (73) a different equation of the same order of precision, i.e., neglecting terms of order O ( ε 2 ) \mathcal{O}(\varepsilon^2) O ( ε 2 ) . To that purpose, we start from (71) and expand the time derivatives in the forcing. Using the product rule,
∂ t ( [ Λ , η ( 0 ) ] ∂ t η ( 0 ) ) = [ Λ , ∂ t η ( 0 ) ] ∂ t η ( 0 ) + [ Λ , η ( 0 ) ] ∂ t t η ( 0 ) , \partial_t\left(\left[\Lambda,\eta^{(0)}\right]\partial_t\eta^{(0)}\right)=\left[\Lambda,\partial_t\eta^{(0)}\right]\partial_t\eta^{(0)}+\left[\Lambda,\eta^{(0)}\right]\partial_{tt}\eta^{(0)}, ∂ t ( [ Λ , η ( 0 ) ] ∂ t η ( 0 ) ) = [ Λ , ∂ t η ( 0 ) ] ∂ t η ( 0 ) + [ Λ , η ( 0 ) ] ∂ tt η ( 0 ) ,
and
∂ t ( ∇ x η ( 0 ) ⋅ R ∂ t η ( 0 ) ) = ∇ x ∂ t η ( 0 ) ⋅ R ∂ t η ( 0 ) + ∇ x η ( 0 ) ⋅ R ∂ t t η ( 0 ) . \partial_t\left(\nabla_x\eta^{(0)}\cdot\mathcal{R}\partial_t\eta^{(0)}\right)=\nabla_x\partial_t\eta^{(0)}\cdot\mathcal{R}\partial_t\eta^{(0)}+\nabla_x\eta^{(0)}\cdot\mathcal{R}\partial_{tt}\eta^{(0)}. ∂ t ( ∇ x η ( 0 ) ⋅ R ∂ t η ( 0 ) ) = ∇ x ∂ t η ( 0 ) ⋅ R ∂ t η ( 0 ) + ∇ x η ( 0 ) ⋅ R ∂ tt η ( 0 ) .
Hence (71) may be rewritten as
( I + γ Λ ) ∂ t t η ( 1 ) + β 4 Λ Δ x 2 η ( 1 ) − δ Λ ∂ t Δ x η ( 1 ) + Λ η ( 1 ) = 1 2 Λ ( ( ∂ t η ( 0 ) ) 2 − ∣ R ∂ t η ( 0 ) ∣ 2 ) − ⟦ Λ , ∂ t η ( 0 ) ⟧ ∂ t η ( 0 ) − [ Λ , η ( 0 ) ] ∂ t t η ( 0 ) − ( ∇ x ∂ t η ( 0 ) ) ⋅ R ∂ t η ( 0 ) − ( ∇ x η ( 0 ) ) ⋅ R ∂ t t η ( 0 ) . (74) \begin{aligned}
(I+\gamma\Lambda)\partial_{tt}\eta^{(1)}+\frac{\beta}{4}\Lambda\Delta_x^2\eta^{(1)}-\delta\Lambda\partial_t\Delta_x\eta^{(1)}+\Lambda\eta^{(1)}
={}&\frac{1}{2}\Lambda\left((\partial_t\eta^{(0)})^2-\left\lvert\mathcal{R}\partial_t\eta^{(0)}\right\rvert^2\right)-\llbracket\Lambda,\partial_t\eta^{(0)}\rrbracket\partial_t\eta^{(0)}\\
&-\left[\Lambda,\eta^{(0)}\right]\partial_{tt}\eta^{(0)}-\left(\nabla_x\partial_t\eta^{(0)}\right)\cdot\mathcal{R}\partial_t\eta^{(0)}-\left(\nabla_x\eta^{(0)}\right)\cdot\mathcal{R}\partial_{tt}\eta^{(0)}. \tag*{(74)}
\end{aligned} ( I + γ Λ ) ∂ tt η ( 1 ) + 4 β Λ Δ x 2 η ( 1 ) − δ Λ ∂ t Δ x η ( 1 ) + Λ η ( 1 ) = 2 1 Λ ( ( ∂ t η ( 0 ) ) 2 − R ∂ t η ( 0 ) 2 ) − [ [ Λ , ∂ t η ( 0 ) ] ] ∂ t η ( 0 ) − [ Λ , η ( 0 ) ] ∂ tt η ( 0 ) − ( ∇ x ∂ t η ( 0 ) ) ⋅ R ∂ t η ( 0 ) − ( ∇ x η ( 0 ) ) ⋅ R ∂ tt η ( 0 ) . (74)
We now eliminate ∂ t t η ( 0 ) \partial_{tt}\eta^{(0)} ∂ tt η ( 0 ) by means of the leading-order closure (58) . Solving (58) for ∂ t t η ( 0 ) \partial_{tt}\eta^{(0)} ∂ tt η ( 0 ) gives
∂ t t η ( 0 ) = − ( I + γ Λ ) − 1 ( β 4 Λ Δ x 2 η ( 0 ) − δ Λ ∂ t Δ x η ( 0 ) + Λ η ( 0 ) ) . (75) \partial_{tt}\eta^{(0)}=-(I+\gamma\Lambda)^{-1}\left(\frac{\beta}{4}\Lambda\Delta_x^2\eta^{(0)}-\delta\Lambda\partial_t\Delta_x\eta^{(0)}+\Lambda\eta^{(0)}\right). \tag*{(75)} ∂ tt η ( 0 ) = − ( I + γ Λ ) − 1 ( 4 β Λ Δ x 2 η ( 0 ) − δ Λ ∂ t Δ x η ( 0 ) + Λ η ( 0 ) ) . (75)
Moreover, introducing the Fourier multiplier
T : = ( I + γ Λ ) − 1 Λ . (76) \mathcal{T}:=(I+\gamma\Lambda)^{-1}\Lambda. \tag*{(76)} T := ( I + γ Λ ) − 1 Λ. (76)
equation (75) becomes
∂ t t η ( 0 ) = − β 4 T Δ x 2 η ( 0 ) + δ T ∂ t Δ x η ( 0 ) − T η ( 0 ) . (77) \partial_{tt}\eta^{(0)}=-\frac{\beta}{4}\mathcal{T}\Delta_x^2\eta^{(0)}+\delta\mathcal{T}\partial_t\Delta_x\eta^{(0)}-\mathcal{T}\eta^{(0)}. \tag*{(77)} ∂ tt η ( 0 ) = − 4 β T Δ x 2 η ( 0 ) + δ T ∂ t Δ x η ( 0 ) − T η ( 0 ) . (77)
Substituting (77) into (74) yields a forcing depending only on η ( 0 ) \eta^{(0)} η ( 0 ) and ∂ t η ( 0 ) \partial_t\eta^{(0)} ∂ t η ( 0 ) :
( I + γ Λ ) ∂ t t η ( 1 ) + β 4 Λ Δ x 2 η ( 1 ) − δ Λ ∂ t Δ x η ( 1 ) + Λ η ( 1 ) = 1 2 Λ ( ( ∂ t η ( 0 ) ) 2 − ∣ R ∂ t η ( 0 ) ∣ 2 ) − ⟦ Λ , ∂ t η ( 0 ) ⟧ ∂ t η ( 0 ) − ( ∇ x ∂ t η ( 0 ) ) ⋅ R ∂ t η ( 0 ) + [ Λ , η ( 0 ) ] T η ( 0 ) + ( ∇ x η ( 0 ) ) ⋅ R ( T η ( 0 ) ) + β 4 ( [ Λ , η ( 0 ) ] T Δ x 2 η ( 0 ) + ( ∇ x η ( 0 ) ) ⋅ R ( T Δ x 2 η ( 0 ) ) ) − δ ( [ Λ , η ( 0 ) ] T ∂ t Δ x η ( 0 ) + ( ∇ x η ( 0 ) ) ⋅ R ( T ∂ t Δ x η ( 0 ) ) ) . (78) \begin{aligned}
(I+\gamma\Lambda)\partial_{tt}\eta^{(1)}+\frac{\beta}{4}\Lambda\Delta_x^2\eta^{(1)}-\delta\Lambda\partial_t\Delta_x\eta^{(1)}+\Lambda\eta^{(1)}
={}&\frac{1}{2}\Lambda\left((\partial_t\eta^{(0)})^2-\left\lvert\mathcal{R}\partial_t\eta^{(0)}\right\rvert^2\right)-\llbracket\Lambda,\partial_t\eta^{(0)}\rrbracket\partial_t\eta^{(0)}\\
&-\left(\nabla_x\partial_t\eta^{(0)}\right)\cdot\mathcal{R}\partial_t\eta^{(0)}+\left[\Lambda,\eta^{(0)}\right]\mathcal{T}\eta^{(0)}+\left(\nabla_x\eta^{(0)}\right)\cdot\mathcal{R}\left(\mathcal{T}\eta^{(0)}\right)\\
&+\frac{\beta}{4}\left(\left[\Lambda,\eta^{(0)}\right]\mathcal{T}\Delta_x^2\eta^{(0)}+\left(\nabla_x\eta^{(0)}\right)\cdot\mathcal{R}\left(\mathcal{T}\Delta_x^2\eta^{(0)}\right)\right)\\
&-\delta\left(\left[\Lambda,\eta^{(0)}\right]\mathcal{T}\partial_t\Delta_x\eta^{(0)}+\left(\nabla_x\eta^{(0)}\right)\cdot\mathcal{R}\left(\mathcal{T}\partial_t\Delta_x\eta^{(0)}\right)\right). \tag*{(78)}
\end{aligned} ( I + γ Λ ) ∂ tt η ( 1 ) + 4 β Λ Δ x 2 η ( 1 ) − δ Λ ∂ t Δ x η ( 1 ) + Λ η ( 1 ) = 2 1 Λ ( ( ∂ t η ( 0 ) ) 2 − R ∂ t η ( 0 ) 2 ) − [ [ Λ , ∂ t η ( 0 ) ] ] ∂ t η ( 0 ) − ( ∇ x ∂ t η ( 0 ) ) ⋅ R ∂ t η ( 0 ) + [ Λ , η ( 0 ) ] T η ( 0 ) + ( ∇ x η ( 0 ) ) ⋅ R ( T η ( 0 ) ) + 4 β ( [ Λ , η ( 0 ) ] T Δ x 2 η ( 0 ) + ( ∇ x η ( 0 ) ) ⋅ R ( T Δ x 2 η ( 0 ) ) ) − δ ( [ Λ , η ( 0 ) ] T ∂ t Δ x η ( 0 ) + ( ∇ x η ( 0 ) ) ⋅ R ( T ∂ t Δ x η ( 0 ) ) ) . (78)
Finally, defining the renormalized variable
f = η ( 0 ) + ε η ( 1 ) , f=\eta^{(0)}+\varepsilon\eta^{(1)}, f = η ( 0 ) + ε η ( 1 ) ,
adding the n = 0 n=0 n = 0 equation (58) to ε \varepsilon ε times (3.29), and neglecting O ( ε 2 ) O(\varepsilon^2) O ( ε 2 ) terms, we obtain a closed O ( ε ) \mathcal{O}(\varepsilon) O ( ε ) model for f f f of the form
( I + Υ Λ ) f t t + δ Λ 3 f t + ( Λ + β 4 Λ 5 ) f = ε N [ f ] , (79) (I+\Upsilon\Lambda)f_{tt}+\delta\Lambda^3f_t+\left(\Lambda+\frac{\beta}{4}\Lambda^5\right)f=\varepsilon\mathcal{N}[f], \tag*{(79)} ( I + ΥΛ ) f tt + δ Λ 3 f t + ( Λ + 4 β Λ 5 ) f = ε N [ f ] , (79)
where the nonlinear forcing is
N [ f ] = 1 2 Λ ( ( f t ) 2 − ∣ R f t ∣ 2 ) − [ Λ , f ] f t − ( ∇ x f ) ⋅ R f t + [ Λ , f ] T f + ( ∇ x f ) ⋅ R ( T f ) + β 4 ( [ Λ , f ] T Δ x 2 f + ( ∇ x f ) ⋅ R ( T Δ x 2 f ) ) − δ ( [ Λ , f ] T ∂ t Δ x f + ( ∇ x f ) ⋅ R ( T ∂ t Δ x f ) ) . (80) \begin{aligned}
\mathcal{N}[f]={}&\frac{1}{2}\Lambda\left((f_t)^2-\lvert\mathcal{R}f_t\rvert^2\right)-\left[\Lambda,f\right]f_t-(\nabla_xf)\cdot\mathcal{R}f_t\\
&+\left[\Lambda,f\right]\mathcal{T}f+(\nabla_xf)\cdot\mathcal{R}(\mathcal{T}f)\\
&+\frac{\beta}{4}\left(\left[\Lambda,f\right]\mathcal{T}\Delta_x^2f+(\nabla_xf)\cdot\mathcal{R}(\mathcal{T}\Delta_x^2f)\right)\\
&-\delta\left(\left[\Lambda,f\right]\mathcal{T}\partial_t\Delta_xf+(\nabla_xf)\cdot\mathcal{R}(\mathcal{T}\partial_t\Delta_xf)\right). \tag*{(80)}
\end{aligned} N [ f ] = 2 1 Λ ( ( f t ) 2 − ∣ R f t ∣ 2 ) − [ Λ , f ] f t − ( ∇ x f ) ⋅ R f t + [ Λ , f ] T f + ( ∇ x f ) ⋅ R ( T f ) + 4 β ( [ Λ , f ] T Δ x 2 f + ( ∇ x f ) ⋅ R ( T Δ x 2 f ) ) − δ ( [ Λ , f ] T ∂ t Δ x f + ( ∇ x f ) ⋅ R ( T ∂ t Δ x f ) ) . (80)
The unidirectional models
In this section we derive unidirectional asymptotic models associated with the bidirectional systems obtained above. To that purpose, as in [4 , 5 , 14 , 16 , 18 ] , we restrict to one horizontal dimension x ∈ T x\in\mathbb{T} x ∈ T and introduce the unidirectional variables
ξ = x − t , τ = ε t , f ( x , t ) = F ( ξ , τ ) . \xi=x-t,\qquad\tau=\varepsilon t,\qquad f(x,t)=F(\xi,\tau). ξ = x − t , τ = εt , f ( x , t ) = F ( ξ , τ ) .
The unidirectional model for system (73) . By the chain rule,
∂ t = − ∂ ξ + ε ∂ τ , ∂ x = ∂ ξ , (81) \partial_t=-\partial_\xi+\varepsilon\partial_\tau,\qquad\partial_x=\partial_\xi, \tag*{(81)} ∂ t = − ∂ ξ + ε ∂ τ , ∂ x = ∂ ξ , (81)
and consequently
f t = − F ξ + ε F τ , f t t = F ξ ξ − 2 ε F ξ τ + O ( ε 2 ) , f x x = F ξ ξ , f x x x x = F ξ ξ ξ ξ . (82) f_t=-F_\xi+\varepsilon F_\tau,\qquad f_{tt}=F_{\xi\xi}-2\varepsilon F_{\xi\tau}+O(\varepsilon^2),\qquad f_{xx}=F_{\xi\xi},\qquad f_{xxxx}=F_{\xi\xi\xi\xi}. \tag*{(82)} f t = − F ξ + ε F τ , f tt = F ξ ξ − 2 ε F ξ τ + O ( ε 2 ) , f xx = F ξ ξ , f xxxx = F ξ ξ ξ ξ . (82)
Substituting (82) into (73) and discarding O ( ε 2 ) O(\varepsilon^2) O ( ε 2 ) terms yields the ( ξ , τ ) (\xi,\tau) ( ξ , τ ) -equation
( I + Υ Λ ) ( F ξ ξ − 2 ε F ξ τ ) + β 4 Λ F ξ ξ ξ ξ + δ Λ F ξ ξ ξ ξ − ε δ Λ F τ ξ ξ + Λ F = ε [ 1 2 Λ ( F ξ 2 − ( Λ F ) 2 ) − ∂ ξ ( [ Λ , F ] F ξ ) + ∂ ξ ( F ξ Λ F ) ] , (83) \begin{aligned}
(I+\Upsilon\Lambda)(F_{\xi\xi}-2\varepsilon F_{\xi\tau})+\frac{\beta}{4}\Lambda F_{\xi\xi\xi\xi}+\delta\Lambda F_{\xi\xi\xi\xi}-\varepsilon\delta\Lambda F_{\tau\xi\xi}+\Lambda F\\
=\varepsilon\left[\frac{1}{2}\Lambda\left(F_\xi^2-(\Lambda F)^2\right)-\partial_\xi\left([\Lambda,F]F_\xi\right)+\partial_\xi\left(F_\xi\Lambda F\right)\right], \tag*{(83)}
\end{aligned} ( I + ΥΛ ) ( F ξ ξ − 2 ε F ξ τ ) + 4 β Λ F ξ ξ ξ ξ + δ Λ F ξ ξ ξ ξ − ε δ Λ F τ ξ ξ + Λ F = ε [ 2 1 Λ ( F ξ 2 − ( Λ F ) 2 ) − ∂ ξ ( [ Λ , F ] F ξ ) + ∂ ξ ( F ξ Λ F ) ] , (83)
where all Fourier multipliers are understood with respect to ξ \xi ξ , and we used ∂ t ∂ x f = − F ξ ξ + ε F τ ξ ξ \partial_t\partial_xf=-F_{\xi\xi}+\varepsilon F_{\tau\xi\xi} ∂ t ∂ x f = − F ξ ξ + ε F τ ξ ξ . Integrating (83) once in ξ \xi ξ and using that Λ \Lambda Λ commutes with ∂ ξ \partial_\xi ∂ ξ , we obtain (up to an additive function of τ \tau τ which vanishes under the mean-zero convention)
( I + Υ Λ ) ( F ξ − 2 ε F τ ) + β 4 Λ F ξ ξ ξ + δ Λ F ξ ξ − ε δ Λ F τ ξ + H F = ε [ 1 2 H ( F ξ 2 − ( Λ F ) 2 ) − [ Λ , F ] F ξ + F ξ Λ F ] . (84) \begin{aligned}
(I+\Upsilon\Lambda)(F_\xi-2\varepsilon F_\tau)+\frac{\beta}{4}\Lambda F_{\xi\xi\xi}+\delta\Lambda F_{\xi\xi}-\varepsilon\delta\Lambda F_{\tau\xi}+\mathcal{H}F\\
=\varepsilon\left[\frac{1}{2}\mathcal{H}\left(F_\xi^2-(\Lambda F)^2\right)-[\Lambda,F]F_\xi+F_\xi\Lambda F\right]. \tag*{(84)}
\end{aligned} ( I + ΥΛ ) ( F ξ − 2 ε F τ ) + 4 β Λ F ξ ξ ξ + δ Λ F ξ ξ − ε δ Λ F τ ξ + H F = ε [ 2 1 H ( F ξ 2 − ( Λ F ) 2 ) − [ Λ , F ] F ξ + F ξ Λ F ] . (84)
Moreover, by means of the Tricomi identity for the Hilbert transform (10) , we can write
1 2 H ( F ξ 2 − ( Λ F ) 2 ) = 1 2 H ( F ξ 2 − ( H F ξ ) 2 ) = F ξ H F ξ = F ξ Λ F . (85) \frac{1}{2}\mathcal{H}\left(F_\xi^2-(\Lambda F)^2\right)=\frac{1}{2}\mathcal{H}\left(F_\xi^2-(\mathcal{H}F_\xi)^2\right)=F_\xi\mathcal{H}F_\xi=F_\xi\Lambda F. \tag*{(85)} 2 1 H ( F ξ 2 − ( Λ F ) 2 ) = 2 1 H ( F ξ 2 − ( H F ξ ) 2 ) = F ξ H F ξ = F ξ Λ F . (85)
Thus, (84) simplifies to
( I + Υ Λ ) ( F ξ − 2 ε F τ ) + β 4 Λ F ξ ξ ξ + δ Λ F ξ ξ − ε δ Λ F τ ξ + H F = ε ( − [ Λ , F ] F ξ + 2 F ξ Λ F ) . (86) (I+\Upsilon\Lambda)(F_\xi-2\varepsilon F_\tau)+\frac{\beta}{4}\Lambda F_{\xi\xi\xi}+\delta\Lambda F_{\xi\xi}-\varepsilon\delta\Lambda F_{\tau\xi}+\mathcal{H}F=\varepsilon\left(-[\Lambda,F]F_\xi+2F_\xi\Lambda F\right). \tag*{(86)} ( I + ΥΛ ) ( F ξ − 2 ε F τ ) + 4 β Λ F ξ ξ ξ + δ Λ F ξ ξ − ε δ Λ F τ ξ + H F = ε ( − [ Λ , F ] F ξ + 2 F ξ Λ F ) . (86)
Rewriting (84) , moving the τ \tau τ -terms to the left-hand side, we obtain the evolution equation
ε ( 2 ( I + Υ Λ ) + δ Λ ∂ ξ ) F τ = ( I + Υ Λ ) F ξ + β 4 Λ F ξ ξ ξ + δ Λ F ξ ξ + H F − ε ( 2 F ξ Λ F − [ Λ , F ] F ξ ) . (87) \varepsilon\left(2(I+\Upsilon\Lambda)+\delta\Lambda\partial_\xi\right)F_\tau=(I+\Upsilon\Lambda)F_\xi+\frac{\beta}{4}\Lambda F_{\xi\xi\xi}+\delta\Lambda F_{\xi\xi}+\mathcal{H}F-\varepsilon\left(2F_\xi\Lambda F-[\Lambda,F]F_\xi\right). \tag*{(87)} ε ( 2 ( I + ΥΛ ) + δ Λ ∂ ξ ) F τ = ( I + ΥΛ ) F ξ + 4 β Λ F ξ ξ ξ + δ Λ F ξ ξ + H F − ε ( 2 F ξ Λ F − [ Λ , F ] F ξ ) . (87)
Starting from (87) , we write
ε M F τ = ( I + Υ Λ ) F ξ + β 4 Λ F ξ ξ ξ + δ Λ F ξ ξ + H F − ε [ 2 F ξ Λ F − [ Λ , F ] F ξ ] , (88) \varepsilon\mathcal{M}F_\tau=(I+\Upsilon\Lambda)F_\xi+\frac{\beta}{4}\Lambda F_{\xi\xi\xi}+\delta\Lambda F_{\xi\xi}+\mathcal{H}F-\varepsilon\left[2F_\xi\Lambda F-[\Lambda,F]F_\xi\right], \tag*{(88)} ε M F τ = ( I + ΥΛ ) F ξ + 4 β Λ F ξ ξ ξ + δ Λ F ξ ξ + H F − ε [ 2 F ξ Λ F − [ Λ , F ] F ξ ] , (88)
where
M : = 2 ( I + Υ Λ ) + δ Λ ∂ ξ , M ∗ : = 2 ( I + Υ Λ ) − δ Λ ∂ ξ . \mathcal{M}:=2(I+\Upsilon\Lambda)+\delta\Lambda\partial_\xi,\qquad\mathcal{M}^{*}:=2(I+\Upsilon\Lambda)-\delta\Lambda\partial_\xi. M := 2 ( I + ΥΛ ) + δ Λ ∂ ξ , M ∗ := 2 ( I + ΥΛ ) − δ Λ ∂ ξ .
Multiplying (88) on the left by M ∗ \mathcal{M}^{*} M ∗ and using
M ∗ M = 4 ( I + Υ Λ ) 2 − δ 2 Λ 2 ∂ ξ 2 , \mathcal{M}^{*}\mathcal{M}=4(I+\Upsilon\Lambda)^2-\delta^2\Lambda^2\partial_\xi^2, M ∗ M = 4 ( I + ΥΛ ) 2 − δ 2 Λ 2 ∂ ξ 2 ,
we obtain
ε ( M ∗ M ) F τ = M ∗ ( ( I + Υ Λ ) F ξ + β 4 Λ F ξ ξ ξ + δ Λ F ξ ξ + H F ) − ε M ∗ ( 2 F ξ Λ F − [ Λ , F ] F ξ ) . (89) \varepsilon(\mathcal{M}^{*}\mathcal{M})F_\tau=\mathcal{M}^{*}\left((I+\Upsilon\Lambda)F_\xi+\frac{\beta}{4}\Lambda F_{\xi\xi\xi}+\delta\Lambda F_{\xi\xi}+\mathcal{H}F\right)-\varepsilon\mathcal{M}^{*}\left(2F_\xi\Lambda F-[\Lambda,F]F_\xi\right). \tag*{(89)} ε ( M ∗ M ) F τ = M ∗ ( ( I + ΥΛ ) F ξ + 4 β Λ F ξ ξ ξ + δ Λ F ξ ξ + H F ) − ε M ∗ ( 2 F ξ Λ F − [ Λ , F ] F ξ ) . (89)
Applying ( M ∗ M ) − 1 (\mathcal{M}^{*}\mathcal{M})^{-1} ( M ∗ M ) − 1 (understood mode-by-mode on k ≠ 0 k\ne0 k = 0 ) yields the first inversion step
F τ = 1 ε ( M ∗ M ) − 1 M ∗ [ ( I + Υ Λ ) F ξ + β 4 Λ F ξ ξ ξ + δ Λ F ξ ξ + H F ] (90) F_\tau=\frac{1}{\varepsilon}(\mathcal{M}^{*}\mathcal{M})^{-1}\mathcal{M}^{*}\left[(I+\Upsilon\Lambda)F_\xi+\frac{\beta}{4}\Lambda F_{\xi\xi\xi}+\delta\Lambda F_{\xi\xi}+\mathcal{H}F\right] \tag*{(90)} F τ = ε 1 ( M ∗ M ) − 1 M ∗ [ ( I + ΥΛ ) F ξ + 4 β Λ F ξ ξ ξ + δ Λ F ξ ξ + H F ] (90)
− ( M ∗ M ) − 1 M ∗ [ 2 F ξ Λ F − [ Λ , F ] F ξ ] . -(\mathcal{M}^{*}\mathcal{M})^{-1}\mathcal{M}^{*}\left[2F_\xi\Lambda F-[\Lambda,F]F_\xi\right]. − ( M ∗ M ) − 1 M ∗ [ 2 F ξ Λ F − [ Λ , F ] F ξ ] .
On the Fourier side (k ≠ 0 k\ne0 k = 0 ), one checks that
( M ∗ M ) − 1 M ∗ = a ( Λ , ∂ ξ ) + b ( Λ , ∂ ξ ) H , (\mathcal{M}^{*}\mathcal{M})^{-1}\mathcal{M}^{*}=a(\Lambda,\partial_\xi)+b(\Lambda,\partial_\xi)\mathcal{H}, ( M ∗ M ) − 1 M ∗ = a ( Λ , ∂ ξ ) + b ( Λ , ∂ ξ ) H ,
where a ( Λ , ∂ ξ ) a(\Lambda,\partial_\xi) a ( Λ , ∂ ξ ) and b ( Λ , ∂ ξ ) b(\Lambda,\partial_\xi) b ( Λ , ∂ ξ ) are real Fourier multipliers with symbols
a k : = 2 ( 1 + Υ ∣ k ∣ ) 4 ( 1 + Υ ∣ k ∣ ) 2 + δ 2 ∣ k ∣ 2 k 2 , b k : = δ ∣ k ∣ 2 4 ( 1 + Υ ∣ k ∣ ) 2 + δ 2 ∣ k ∣ 2 k 2 , k ≠ 0 , a_k:=\frac{2(1+\Upsilon|k|)}{4(1+\Upsilon|k|)^2+\delta^2|k|^2k^2},\qquad b_k:=\frac{\delta|k|^2}{4(1+\Upsilon|k|)^2+\delta^2|k|^2k^2},\qquad k\ne0, a k := 4 ( 1 + Υ∣ k ∣ ) 2 + δ 2 ∣ k ∣ 2 k 2 2 ( 1 + Υ∣ k ∣ ) , b k := 4 ( 1 + Υ∣ k ∣ ) 2 + δ 2 ∣ k ∣ 2 k 2 δ ∣ k ∣ 2 , k = 0 ,
or equivalently,
a ( Λ , ∂ ξ ) = 2 ( I + Υ Λ ) 4 ( I + Υ Λ ) 2 − δ 2 Λ 2 ∂ ξ 2 , b ( Λ , ∂ ξ ) = δ Λ 2 4 ( I + Υ Λ ) 2 − δ 2 Λ 2 ∂ ξ 2 . (91) a(\Lambda,\partial_\xi)=\frac{2(I+\Upsilon\Lambda)}{4(I+\Upsilon\Lambda)^2-\delta^2\Lambda^2\partial_\xi^2},\qquad b(\Lambda,\partial_\xi)=\frac{\delta\Lambda^2}{4(I+\Upsilon\Lambda)^2-\delta^2\Lambda^2\partial_\xi^2}. \tag*{(91)} a ( Λ , ∂ ξ ) = 4 ( I + ΥΛ ) 2 − δ 2 Λ 2 ∂ ξ 2 2 ( I + ΥΛ ) , b ( Λ , ∂ ξ ) = 4 ( I + ΥΛ ) 2 − δ 2 Λ 2 ∂ ξ 2 δ Λ 2 . (91)
Substituting this identity into (90) yields
F τ = 1 ε ( a + b H ) ( ( I + Υ Λ ) F ξ + β 4 Λ F ξ ξ ξ + δ Λ F ξ ξ + H F ) − ( a + b H ) ( 2 F ξ Λ F − [ Λ , F ] F ξ ) , (92) F_\tau=\frac{1}{\varepsilon}(a+b\mathcal{H})\left((I+\Upsilon\Lambda)F_\xi+\frac{\beta}{4}\Lambda F_{\xi\xi\xi}+\delta\Lambda F_{\xi\xi}+\mathcal{H}F\right)-(a+b\mathcal{H})\left(2F_\xi\Lambda F-[\Lambda,F]F_\xi\right), \tag*{(92)} F τ = ε 1 ( a + b H ) ( ( I + ΥΛ ) F ξ + 4 β Λ F ξ ξ ξ + δ Λ F ξ ξ + H F ) − ( a + b H ) ( 2 F ξ Λ F − [ Λ , F ] F ξ ) , (92)
with a a a and b b b given by (91) .
The unidirectional model for (3.30)
We now derive the unidirectional reduction of (79) in the same manner as for the first model. More precisely, we restrict to one horizontal space dimension and introduce the slow variables
ξ = x − t , τ = ε t , f ( x , t ) = F ( ξ , τ ) . \xi= x-t,\qquad\tau= \varepsilon t,\qquad f(x,t)=F(\xi,\tau). ξ = x − t , τ = εt , f ( x , t ) = F ( ξ , τ ) .
With these conventions, up to O ( ε 2 ) O(\varepsilon^2) O ( ε 2 ) we have
f t = − F ξ + ε F τ , f t t = F ξ ξ − 2 ε F ξ τ + O ( ε 2 ) , f x x = F ξ ξ , f x x x x = F ξ ξ ξ ξ , f_t=-F_\xi+\varepsilon F_\tau,\qquad f_{tt}=F_{\xi\xi}-2\varepsilon F_{\xi\tau}+O(\varepsilon^2),\qquad f_{xx}=F_{\xi\xi},\qquad f_{xxxx}=F_{\xi\xi\xi\xi}, f t = − F ξ + ε F τ , f tt = F ξ ξ − 2 ε F ξ τ + O ( ε 2 ) , f xx = F ξ ξ , f xxxx = F ξ ξ ξ ξ ,
∂ t f x x = ( − ∂ ξ + ε ∂ τ ) F ξ ξ = − F ξ ξ ξ + ε F τ ξ ξ . \partial_t f_{xx}=(-\partial_\xi+\varepsilon\partial_\tau)F_{\xi\xi}=-F_{\xi\xi\xi}+\varepsilon F_{\tau\xi\xi}. ∂ t f xx = ( − ∂ ξ + ε ∂ τ ) F ξ ξ = − F ξ ξ ξ + ε F τ ξ ξ .
Substituting into equation (79) , doing some straightforward computations and neglecting O ( ε 2 ) O(\varepsilon^2) O ( ε 2 ) we find the equation
( I + Υ Λ ) ( F ξ ξ − 2 ε F ξ τ ) + β 4 Λ F ξ ξ ξ ξ + δ Λ F ξ ξ ξ − ε δ Λ F τ ξ ξ + Λ F = ε [ 1 2 Λ ( F ξ 2 − ( Λ F ) 2 ) − [ Λ , F ξ ] F ξ + F ξ ξ Λ F + [ Λ , F ] T F − F ξ H T F + β 4 ( [ Λ , F ] T F ξ ξ ξ − F ξ Λ T F ξ ξ ξ ) + δ ( [ Λ , F ] T F ξ ξ − F ξ Λ T F ξ ξ ) ] . (93) \begin{aligned}
(I+\Upsilon\Lambda)(F_{\xi\xi}-2\varepsilon F_{\xi\tau})+\frac{\beta}{4}\Lambda F_{\xi\xi\xi\xi}+\delta\Lambda F_{\xi\xi\xi}-\varepsilon\delta\Lambda F_{\tau\xi\xi}+\Lambda F
={}&\varepsilon\Bigl[\frac{1}{2}\Lambda\left(F_\xi^2-(\Lambda F)^2\right)-[\Lambda,F_\xi]F_\xi+F_{\xi\xi}\Lambda F\\
&+[\Lambda,F]\mathcal{T}F-F_\xi\mathcal{H}\mathcal{T}F\\
&+\frac{\beta}{4}\left([\Lambda,F]\mathcal{T}F_{\xi\xi\xi}-F_\xi\Lambda\mathcal{T}F_{\xi\xi\xi}\right)\\
&+\delta\left([\Lambda,F]\mathcal{T}F_{\xi\xi}-F_\xi\Lambda\mathcal{T}F_{\xi\xi}\right)\Bigr]. \tag*{(93)}
\end{aligned} ( I + ΥΛ ) ( F ξ ξ − 2 ε F ξ τ ) + 4 β Λ F ξ ξ ξ ξ + δ Λ F ξ ξ ξ − ε δ Λ F τ ξ ξ + Λ F = ε [ 2 1 Λ ( F ξ 2 − ( Λ F ) 2 ) − [ Λ , F ξ ] F ξ + F ξ ξ Λ F + [ Λ , F ] T F − F ξ H T F + 4 β ( [ Λ , F ] T F ξ ξ ξ − F ξ Λ T F ξ ξ ξ ) + δ ( [ Λ , F ] T F ξ ξ − F ξ Λ T F ξ ξ ) ] . (93)
We first move the τ \tau τ –dependent linear terms to the left-hand side and factor out the slow time derivative. Namely, (93) can be rewritten as
ε ( 2 ( I + Υ Λ ) ∂ ξ + δ Λ ∂ ξ 2 ) F τ = ( I + Υ Λ ) F ξ ξ + β 4 Λ F ξ ξ ξ ξ + δ Λ F ξ ξ ξ + Λ F − ε [ 1 2 Λ ( F ξ 2 − ( Λ F ) 2 ) − [ Λ , F ξ ] F ξ + F ξ ξ Λ F + [ Λ , F ] T F − F ξ H T F + β 4 ( [ Λ , F ] T F ξ ξ ξ − F ξ Λ T F ξ ξ ξ ) + δ ( [ Λ , F ] T F ξ ξ − F ξ Λ T F ξ ξ ) ] . (94) \begin{aligned}
\varepsilon\left(2(I+\Upsilon\Lambda)\partial_\xi+\delta\Lambda\partial_\xi^2\right)F_\tau
={}&(I+\Upsilon\Lambda)F_{\xi\xi}+\frac{\beta}{4}\Lambda F_{\xi\xi\xi\xi}+\delta\Lambda F_{\xi\xi\xi}+\Lambda F\\
&-\varepsilon\Bigl[\frac{1}{2}\Lambda\left(F_\xi^2-(\Lambda F)^2\right)-[\Lambda,F_\xi]F_\xi+F_{\xi\xi}\Lambda F\\
&+[\Lambda,F]\mathcal{T}F-F_\xi\mathcal{H}\mathcal{T}F\\
&+\frac{\beta}{4}\left([\Lambda,F]\mathcal{T}F_{\xi\xi\xi}-F_\xi\Lambda\mathcal{T}F_{\xi\xi\xi}\right)\\
&+\delta\left([\Lambda,F]\mathcal{T}F_{\xi\xi}-F_\xi\Lambda\mathcal{T}F_{\xi\xi}\right)\Bigr]. \tag*{(94)}
\end{aligned} ε ( 2 ( I + ΥΛ ) ∂ ξ + δ Λ ∂ ξ 2 ) F τ = ( I + ΥΛ ) F ξ ξ + 4 β Λ F ξ ξ ξ ξ + δ Λ F ξ ξ ξ + Λ F − ε [ 2 1 Λ ( F ξ 2 − ( Λ F ) 2 ) − [ Λ , F ξ ] F ξ + F ξ ξ Λ F + [ Λ , F ] T F − F ξ H T F + 4 β ( [ Λ , F ] T F ξ ξ ξ − F ξ Λ T F ξ ξ ξ ) + δ ( [ Λ , F ] T F ξ ξ − F ξ Λ T F ξ ξ ) ] . (94)
For convenience, denote the linear operator acting on F τ F_\tau F τ by
M 1 : = 2 ( I + Υ Λ ) ∂ ξ + δ Λ ∂ ξ 2 , M 1 ∗ : = − 2 ( I + Υ Λ ) ∂ ξ + δ Λ ∂ ξ 2 , \mathcal{M}_1:=2(I+\Upsilon\Lambda)\partial_\xi+\delta\Lambda\partial_\xi^2,\qquad\mathcal{M}_1^*:=-2(I+\Upsilon\Lambda)\partial_\xi+\delta\Lambda\partial_\xi^2, M 1 := 2 ( I + ΥΛ ) ∂ ξ + δ Λ ∂ ξ 2 , M 1 ∗ := − 2 ( I + ΥΛ ) ∂ ξ + δ Λ ∂ ξ 2 ,
so that,
M 1 ∗ M 1 = ( δ Λ ∂ ξ 2 ) 2 − 4 ( ( I + Υ Λ ) ∂ ξ ) 2 . \mathcal{M}_1^*\mathcal{M}_1=(\delta\Lambda\partial_\xi^2)^2-4\left((I+\Upsilon\Lambda)\partial_\xi\right)^2. M 1 ∗ M 1 = ( δ Λ ∂ ξ 2 ) 2 − 4 ( ( I + ΥΛ ) ∂ ξ ) 2 .
Multiplying (94) by M 1 ∗ \mathcal{M}_1^* M 1 ∗ and applying ( M 1 ∗ M 1 ) − 1 (\mathcal{M}_1^*\mathcal{M}_1)^{-1} ( M 1 ∗ M 1 ) − 1 (mode-by-mode on k ≠ 0 k \ne0 k = 0 ) yields the first inversion step
F τ = 1 ε ( M 1 ∗ M 1 ) − 1 M 1 ∗ [ ( I + Υ Λ ) F ξ ξ + β 4 Λ F ξ ξ ξ ξ + δ Λ F ξ ξ ξ + Λ F ] − ( M 1 ∗ M 1 ) − 1 M 1 ∗ [ 1 2 Λ ( F ξ 2 − ( Λ F ) 2 ) − [ Λ , F ξ ] F ξ + F ξ ξ Λ F + [ Λ , F ] T F − F ξ H T F + β 4 ( [ Λ , F ] T F ξ ξ ξ ξ − F ξ Λ T F ξ ξ ξ ) + δ ( [ Λ , F ] T F ξ ξ ξ − F ξ Λ T F ξ ξ ) ] . (95) \begin{aligned}
F_\tau={}& \frac{1}{\varepsilon}(\mathcal{M}_1^*\mathcal{M}_1)^{-1}\mathcal{M}_1^*\left[(I+\Upsilon\Lambda)F_{\xi\xi}+\frac{\beta}{4}\Lambda F_{\xi\xi\xi\xi}+\delta\Lambda F_{\xi\xi\xi}+\Lambda F\right] \\
&-(\mathcal{M}_1^*\mathcal{M}_1)^{-1}\mathcal{M}_1^*\left[\frac{1}{2}\Lambda\left(F_\xi^2-(\Lambda F)^2\right)-[\Lambda,F_\xi]F_\xi+F_{\xi\xi}\Lambda F+[\Lambda,F]\mathcal{T}F-F_\xi\mathcal{H}\mathcal{T}F\right. \\
&\left.\qquad\qquad+\frac{\beta}{4}\left([\Lambda,F]\mathcal{T}F_{\xi\xi\xi\xi}-F_\xi\Lambda\mathcal{T}F_{\xi\xi\xi}\right)+\delta\left([\Lambda,F]\mathcal{T}F_{\xi\xi\xi}-F_\xi\Lambda\mathcal{T}F_{\xi\xi}\right)\right]. \tag*{(95)}
\end{aligned} F τ = ε 1 ( M 1 ∗ M 1 ) − 1 M 1 ∗ [ ( I + ΥΛ ) F ξ ξ + 4 β Λ F ξ ξ ξ ξ + δ Λ F ξ ξ ξ + Λ F ] − ( M 1 ∗ M 1 ) − 1 M 1 ∗ [ 2 1 Λ ( F ξ 2 − ( Λ F ) 2 ) − [ Λ , F ξ ] F ξ + F ξ ξ Λ F + [ Λ , F ] T F − F ξ H T F + 4 β ( [ Λ , F ] T F ξ ξ ξ ξ − F ξ Λ T F ξ ξ ξ ) + δ ( [ Λ , F ] T F ξ ξ ξ − F ξ Λ T F ξ ξ ) ] . (95)
On the Fourier side (k ≠ 0 k\ne0 k = 0 ), the operator
M 1 : = 2 ( I + Υ Λ ) ∂ ξ + δ Λ ∂ ξ 2 \mathcal{M}_1:=2(I+\Upsilon\Lambda)\partial_\xi+\delta\Lambda\partial_\xi^2 M 1 := 2 ( I + ΥΛ ) ∂ ξ + δ Λ ∂ ξ 2
has symbol
m 1 ( k ) = − δ ∣ k ∣ k 2 + i 2 ( 1 + Υ ∣ k ∣ ) k . m_1(k)=-\delta|k|k^2+i\,2(1+\Upsilon|k|)k. m 1 ( k ) = − δ ∣ k ∣ k 2 + i 2 ( 1 + Υ∣ k ∣ ) k .
Therefore, ( M 1 ∗ M 1 ) − 1 M 1 ∗ (\mathcal{M}_1^*\mathcal{M}_1)^{-1}\mathcal{M}_1^* ( M 1 ∗ M 1 ) − 1 M 1 ∗ has associated symbol
m 1 ( k ) ‾ ∣ m 1 ( k ) ∣ 2 = − δ ∣ k ∣ k 2 − i 2 ( 1 + Υ ∣ k ∣ ) k δ 2 ∣ k ∣ 2 k 4 + 4 ( 1 + Υ ∣ k ∣ ) 2 k 2 . \frac{\overline{m_1(k)}}{|m_1(k)|^2}
=\frac{-\delta|k|k^2-i\,2(1+\Upsilon|k|)k}{\delta^2|k|^2k^4+4(1+\Upsilon|k|)^2k^2}. ∣ m 1 ( k ) ∣ 2 m 1 ( k ) = δ 2 ∣ k ∣ 2 k 4 + 4 ( 1 + Υ∣ k ∣ ) 2 k 2 − δ ∣ k ∣ k 2 − i 2 ( 1 + Υ∣ k ∣ ) k .
Similarly as before, we obtain the decomposition
( M 1 ∗ M 1 ) − 1 M 1 ∗ = α ( Λ , ∂ ξ ) + γ ( Λ , ∂ ξ ) H , (96) (\mathcal{M}_1^*\mathcal{M}_1)^{-1}\mathcal{M}_1^*=\alpha(\Lambda,\partial_\xi)+\gamma(\Lambda,\partial_\xi)\mathcal{H}, \tag*{(96)} ( M 1 ∗ M 1 ) − 1 M 1 ∗ = α ( Λ , ∂ ξ ) + γ ( Λ , ∂ ξ ) H , (96)
where α ( Λ , ∂ ξ ) \alpha(\Lambda,\partial_\xi) α ( Λ , ∂ ξ ) and γ ( Λ , ∂ ξ ) \gamma(\Lambda,\partial_\xi) γ ( Λ , ∂ ξ ) are real Fourier multipliers with symbols
α k : = − δ ∣ k ∣ δ 2 ∣ k ∣ 4 + 4 ( 1 + Υ ∣ k ∣ ) 2 , γ k : = 2 ( 1 + Υ ∣ k ∣ ) ∣ k ∣ ( δ 2 ∣ k ∣ 4 + 4 ( 1 + Υ ∣ k ∣ ) 2 ) , k ≠ 0. \alpha_k:=\frac{-\delta|k|}{\delta^2|k|^4+4(1+\Upsilon|k|)^2},\qquad
\gamma_k:=\frac{2(1+\Upsilon|k|)}{|k|(\delta^2|k|^4+4(1+\Upsilon|k|)^2)},\qquad k\ne0. α k := δ 2 ∣ k ∣ 4 + 4 ( 1 + Υ∣ k ∣ ) 2 − δ ∣ k ∣ , γ k := ∣ k ∣ ( δ 2 ∣ k ∣ 4 + 4 ( 1 + Υ∣ k ∣ ) 2 ) 2 ( 1 + Υ∣ k ∣ ) , k = 0.
Equivalently, at the operator level,
α ( Λ , ∂ ξ ) = − δ Λ δ 2 Λ 2 ∂ ξ 2 − 4 ( I + Υ Λ ) 2 , γ ( Λ , ∂ ξ ) = 2 ( I + Υ Λ ) Λ − 1 δ 2 Λ 2 ∂ ξ 2 − 4 ( I + Υ Λ ) 2 . \alpha(\Lambda,\partial_\xi)=-\frac{\delta\Lambda}{\delta^2\Lambda^2\partial_\xi^2-4(I+\Upsilon\Lambda)^2},\qquad
\gamma(\Lambda,\partial_\xi)=\frac{2(I+\Upsilon\Lambda)\Lambda^{-1}}{\delta^2\Lambda^2\partial_\xi^2-4(I+\Upsilon\Lambda)^2}. α ( Λ , ∂ ξ ) = − δ 2 Λ 2 ∂ ξ 2 − 4 ( I + ΥΛ ) 2 δ Λ , γ ( Λ , ∂ ξ ) = δ 2 Λ 2 ∂ ξ 2 − 4 ( I + ΥΛ ) 2 2 ( I + ΥΛ ) Λ − 1 .
With this notation, we may write
F τ = 1 ε ( α + γ H ) [ ( I + Υ Λ ) F ξ ξ + β 4 Λ F ξ ξ ξ ξ + δ Λ F ξ ξ ξ + Λ F ] − ( α + γ H ) [ 1 2 Λ ( F ξ 2 − ( Λ F ) 2 ) − [ Λ , F ξ ] F ξ + F ξ ξ Λ F + [ Λ , F ] T F − F ξ H T F + β 4 ( [ Λ , F ] T F ξ ξ ξ ξ − F ξ Λ T F ξ ξ ξ ) + δ ( [ Λ , F ] T F ξ ξ ξ − F ξ Λ T F ξ ξ ) ] . (97) \begin{aligned}
F_\tau={}&\frac{1}{\varepsilon}(\alpha+\gamma\mathcal{H})\left[(I+\Upsilon\Lambda)F_{\xi\xi}+\frac{\beta}{4}\Lambda F_{\xi\xi\xi\xi}+\delta\Lambda F_{\xi\xi\xi}+\Lambda F\right] \\
&-(\alpha+\gamma\mathcal{H})\left[\frac{1}{2}\Lambda\left(F_\xi^2-(\Lambda F)^2\right)-[\Lambda,F_\xi]F_\xi+F_{\xi\xi}\Lambda F+[\Lambda,F]\mathcal{T}F-F_\xi\mathcal{H}\mathcal{T}F\right. \\
&\left.\qquad\qquad+\frac{\beta}{4}\left([\Lambda,F]\mathcal{T}F_{\xi\xi\xi\xi}-F_\xi\Lambda\mathcal{T}F_{\xi\xi\xi}\right)+\delta\left([\Lambda,F]\mathcal{T}F_{\xi\xi\xi}-F_\xi\Lambda\mathcal{T}F_{\xi\xi}\right)\right]. \tag*{(97)}
\end{aligned} F τ = ε 1 ( α + γ H ) [ ( I + ΥΛ ) F ξ ξ + 4 β Λ F ξ ξ ξ ξ + δ Λ F ξ ξ ξ + Λ F ] − ( α + γ H ) [ 2 1 Λ ( F ξ 2 − ( Λ F ) 2 ) − [ Λ , F ξ ] F ξ + F ξ ξ Λ F + [ Λ , F ] T F − F ξ H T F + 4 β ( [ Λ , F ] T F ξ ξ ξ ξ − F ξ Λ T F ξ ξ ξ ) + δ ( [ Λ , F ] T F ξ ξ ξ − F ξ Λ T F ξ ξ ) ] . (97)
Well-posedness result for the bidirectional model
In this section we establish the well-posedness theory for (73) . In particular, we prove the following result:
Theorem 4.1 (Local well-posedness for small H 3 H^3 H 3 data). Let Υ > 0 \Upsilon> 0 Υ > 0 and δ , β > 0 \delta,\beta> 0 δ , β > 0 , and let ε ∈ R \varepsilon\in\mathbb{R} ε ∈ R be fixed. Assume that the initial data satisfy
f ( x , 0 ) = f 0 ∈ H 0 3 ( T 2 ) , ∂ t f ( x , 0 ) = f 1 ∈ H 0 1 ( T 2 ) , ∥ f 0 ∥ H 3 ≤ ε ∗ , f(x,0) = f_0 \in H^3_0(\mathbb{T}^2), \qquad\partial_t f(x,0) = f_1 \in H^1_0(\mathbb{T}^2), \qquad\|f_0\|_{H^3} \le\varepsilon_*, f ( x , 0 ) = f 0 ∈ H 0 3 ( T 2 ) , ∂ t f ( x , 0 ) = f 1 ∈ H 0 1 ( T 2 ) , ∥ f 0 ∥ H 3 ≤ ε ∗ , for some ε ∗ = ε ∗ ( Υ , δ , β , ∣ ε ∣ ) > 0 \varepsilon_* = \varepsilon_*(\Upsilon,\delta,\beta,|\varepsilon|) > 0 ε ∗ = ε ∗ ( Υ , δ , β , ∣ ε ∣ ) > 0 . Then there exists a time
T = T ( Υ , δ , β , ∣ ε ∣ , ∥ f 1 ∥ H 1 , ∥ f 0 ∥ H 3 ) > 0 T = T\left(\Upsilon,\delta,\beta,|\varepsilon|,\|f_1\|_{H^1},\|f_0\|_{H^3}\right) > 0 T = T ( Υ , δ , β , ∣ ε ∣ , ∥ f 1 ∥ H 1 , ∥ f 0 ∥ H 3 ) > 0 and a unique solution f f f of (73) on [ 0 , T ] [0,T] [ 0 , T ] such that
f ∈ L ∞ ( 0 , T ; H 0 3 ( T 2 ) ) , f t ∈ L ∞ ( 0 , T ; H 0 1 ( T 2 ) ) . f \in L^\infty(0,T;H^3_0(\mathbb{T}^2)), \qquad f_t \in L^\infty(0,T;H^1_0(\mathbb{T}^2)). f ∈ L ∞ ( 0 , T ; H 0 3 ( T 2 )) , f t ∈ L ∞ ( 0 , T ; H 0 1 ( T 2 )) .
Proof of Theorem 4.1. We begin by rewriting (73) as a first-order system in time. Set
v : = ∂ t f , N ( f , w ) : = [ Λ , f ] w + ∇ x f ⋅ R w v := \partial_t f, \qquad\mathcal{N}(f,w) := [\Lambda,f]w + \nabla_x f \cdot\mathcal{R}w v := ∂ t f , N ( f , w ) := [ Λ , f ] w + ∇ x f ⋅ R w
Then (73) is equivalent to
∂ t f = v , ( I + Υ Λ ) ∂ t v + δ Λ 3 v + ( Λ + β 4 Λ 5 ) f + N ( f , ∂ t v ) = ε Q ( v ) , (98) \begin{aligned}
\partial_t f &= v, \\
(I+\Upsilon\Lambda)\partial_t v + \delta\Lambda^3v + \left(\Lambda+\frac{\beta}{4}\Lambda^5\right)f + \mathcal{N}(f,\partial_t v) &= \varepsilon\mathcal{Q}(v), \tag*{(98)}
\end{aligned} ∂ t f ( I + ΥΛ ) ∂ t v + δ Λ 3 v + ( Λ + 4 β Λ 5 ) f + N ( f , ∂ t v ) = v , = ε Q ( v ) , (98)
where the quadratic forcing is
Q ( v ) : = 1 2 Λ ( v 2 − ∣ R v ∣ 2 ) − [ Λ , v ] v − ∇ x v ⋅ R v . (99) \mathcal{Q}(v) := \frac{1}{2}\Lambda\left(v^2-\lvert\mathcal{R}v\rvert^2\right)-[\Lambda,v]v-\nabla_xv\cdot\mathcal{R}v. \tag*{(99)} Q ( v ) := 2 1 Λ ( v 2 − ∣ R v ∣ 2 ) − [ Λ , v ] v − ∇ x v ⋅ R v . (99)
Fix T > 0 T>0 T > 0 . We work in the class of functions
X T : = { f ˉ ∈ L ∞ ( 0 , T ; H 0 3 ( T 2 ) ) : f ˉ ( 0 ) = f 0 , ∥ f ˉ ∥ L t ∞ H x 3 ≤ 2 ∥ f 0 ∥ H 3 } , \mathcal{X}_T := \left\{\bar f\in L^\infty(0,T;H^3_0(\mathbb{T}^2)):\ \bar f(0)=f_0,\ \|\bar f\|_{L^\infty_tH^3_x}\le2\|f_0\|_{H^3}\right\}, X T := { f ˉ ∈ L ∞ ( 0 , T ; H 0 3 ( T 2 )) : f ˉ ( 0 ) = f 0 , ∥ f ˉ ∥ L t ∞ H x 3 ≤ 2∥ f 0 ∥ H 3 } ,
Y T : = { v ˉ ∈ L ∞ ( 0 , T ; H 0 1 ( T 2 ) ) : v ˉ ( 0 ) = f 1 } , \mathcal{Y}_T := \left\{\bar v\in L^\infty(0,T;H^1_0(\mathbb{T}^2)):\ \bar v(0)=f_1\right\}, Y T := { v ˉ ∈ L ∞ ( 0 , T ; H 0 1 ( T 2 )) : v ˉ ( 0 ) = f 1 } ,
and construct a solution by a two-stage approximation procedure.
Step 0: Strategy of the proof. The argument is somewhat delicate: we introduce two regularizations with parameters μ \mu μ and λ \lambda λ , solve an elliptic problem to identify the time derivative, construct the evolution by a contraction mapping, and then pass to the limits using μ \mu μ – and λ \lambda λ -uniform estimates. We summarize these steps next and provide the detailed implementation afterwards.
Given ( f ˉ , v ˉ ) ∈ X T × Y T (\bar f,\bar v)\in\mathcal{X}_T\times\mathcal{Y}_T ( f ˉ , v ˉ ) ∈ X T × Y T , we solve an auxiliary λ \lambda λ -regularized elliptic equation for an unknown U λ , μ [ f ˉ , v ˉ ] U^{\lambda,\mu}[\bar f,\bar v] U λ , μ [ f ˉ , v ˉ ] (Step 1). Since N ( f , ⋅ ) \mathcal{N}(f,\cdot) N ( f , ⋅ ) is linear in its second argument, this amounts to inverting a perturbation of ( I + Υ Λ ) (I+\Upsilon\Lambda) ( I + ΥΛ ) ; the smallness of ∥ f 0 ∥ H 3 \|f_0\|_{H^3} ∥ f 0 ∥ H 3 (and hence of ∥ f ˉ ∥ L t ∞ H x 3 \|\bar f\|_{L^\infty_tH^3_x} ∥ f ˉ ∥ L t ∞ H x 3 ) ensures that the inversion is well-defined and yields a unique U λ , μ U^{\lambda,\mu} U λ , μ in the chosen Sobolev class.
With this U λ , μ U^{\lambda,\mu} U λ , μ fixed, we solve the corresponding λ \lambda λ –regularized evolution for v λ , μ v^{\lambda,\mu} v λ , μ with initial datum v 0 v_0 v 0 and choose T T T so that the map v ˉ ↦ v λ , μ \bar{v} \mapsto v^{\lambda,\mu} v ˉ ↦ v λ , μ is a contraction in Y T Y_T Y T (Step 2). In particular, the fixed point v λ , μ v^{\lambda,\mu} v λ , μ satisfies v t λ , μ = U λ , μ v_t^{\lambda,\mu}=U^{\lambda,\mu} v t λ , μ = U λ , μ by construction. We then pass to the limit λ → ∞ \lambda\to\infty λ → ∞ (with μ \mu μ fixed) using uniform-in-λ \lambda λ bounds, obtaining a function v μ v^\mu v μ solving the μ \mu μ -regularized equation (Step 3).
Second, with v μ v^\mu v μ available, we recover f μ f^\mu f μ by solving f t μ = v μ f_t^\mu=v^\mu f t μ = v μ with initial datum f 0 f_0 f 0 , and close the coupling through a fixed point argument in X T X_T X T (Step 4), now only at the μ \mu μ -regularized level. Finally, we let μ → ∞ \mu\to\infty μ → ∞ and use the corresponding uniform-in-μ \mu μ a priori bounds to pass to the limit, yielding a solution f f f of the original system on [ 0 , T ] [0,T] [ 0 , T ] with f ∈ L ∞ ( 0 , T ; H 0 3 ) f \in L^\infty(0,T;H_0^3) f ∈ L ∞ ( 0 , T ; H 0 3 ) and f t ∈ L ∞ ( 0 , T ; H 0 1 ) f_t \in L^\infty(0,T;H_0^1) f t ∈ L ∞ ( 0 , T ; H 0 1 ) (Step 5).
Step 1: Solving the elliptic problem for U U U . Fix T > 0 T>0 T > 0 and regularization parameters 0 < λ , μ ≤ 1 0<\lambda,\mu\le1 0 < λ , μ ≤ 1 . Let ( f ˉ , v ˉ ) ∈ X T × Y T (\bar{f},\bar{v})\in X_T\times Y_T ( f ˉ , v ˉ ) ∈ X T × Y T . In this step we will construct an auxiliary unknown
U = U λ , μ [ f ˉ , v ˉ ] ∈ L ∞ ( 0 , T ; H 0 1 ( T 2 ) ) , U=U^{\lambda,\mu}[\bar{f},\bar{v}]\in L^\infty(0,T;H_0^1(\mathbb{T}^2)), U = U λ , μ [ f ˉ , v ˉ ] ∈ L ∞ ( 0 , T ; H 0 1 ( T 2 )) ,
as the unique solution to the regularized elliptic problem
( I + γ Λ ) U = F λ , μ [ f ˉ , v ˉ ] − J λ N ( J μ f ˉ , J λ U ) , in ( 0 , T ) × T 2 , (100) (I+\gamma\Lambda)U=\mathcal{F}^{\lambda,\mu}[\bar{f},\bar{v}]-\mathcal{J}_\lambda\mathcal{N}(\mathcal{J}_\mu\bar{f},\mathcal{J}_\lambda U),\qquad\text{in }(0,T)\times\mathbb{T}^2, \tag*{(100)} ( I + γ Λ ) U = F λ , μ [ f ˉ , v ˉ ] − J λ N ( J μ f ˉ , J λ U ) , in ( 0 , T ) × T 2 , (100)
where the forcing is
F λ , μ [ f ˉ , v ˉ ] : = − δ J λ Λ 3 J λ v ˉ − J μ ( Λ + β 4 Λ 5 ) J μ f ˉ + ε J λ [ 1 2 Λ ( ( J λ v ˉ ) 2 − ∣ R J λ v ˉ ∣ 2 ) − [ Λ , J λ v ˉ ] J λ v ˉ − ∇ x ( J λ v ˉ ) ⋅ R ( J λ v ˉ ) ] (101) \mathcal{F}^{\lambda,\mu}[\bar{f},\bar{v}]:=-\delta\mathcal{J}_\lambda\Lambda^3\mathcal{J}_\lambda\bar{v}-\mathcal{J}_\mu\left(\Lambda+\frac{\beta}{4}\Lambda^5\right)\mathcal{J}_\mu\bar{f}
+\varepsilon\mathcal{J}_\lambda\left[\frac{1}{2}\Lambda\left((\mathcal{J}_\lambda\bar{v})^2-\lvert\mathcal{R}\mathcal{J}_\lambda\bar{v}\rvert^2\right)-[\Lambda,\mathcal{J}_\lambda\bar{v}]\mathcal{J}_\lambda\bar{v}-\nabla_x(\mathcal{J}_\lambda\bar{v})\cdot\mathcal{R}(\mathcal{J}_\lambda\bar{v})\right] \tag*{(101)} F λ , μ [ f ˉ , v ˉ ] := − δ J λ Λ 3 J λ v ˉ − J μ ( Λ + 4 β Λ 5 ) J μ f ˉ + ε J λ [ 2 1 Λ ( ( J λ v ˉ ) 2 − ∣ R J λ v ˉ ∣ 2 ) − [ Λ , J λ v ˉ ] J λ v ˉ − ∇ x ( J λ v ˉ ) ⋅ R ( J λ v ˉ ) ] (101)
and the bilinear coupling is
N ( f , g ) : = [ Λ , f ] g + ∇ x f ⋅ R g . (102) \mathcal{N}(f,g):=[\Lambda,f]g+\nabla_x f\cdot\mathcal{R}g. \tag*{(102)} N ( f , g ) := [ Λ , f ] g + ∇ x f ⋅ R g . (102)
Since Λ \Lambda Λ and each Riesz transform R j \mathcal{R}_j R j annihilate constants, and J λ , J μ \mathcal{J}_\lambda,\mathcal{J}_\mu J λ , J μ preserve the zero Fourier mode, it follows that if f ˉ , v ˉ \bar{f},\bar{v} f ˉ , v ˉ have zero spatial mean, then each term in (101) has zero mean. Similarly, after a straightforward integration by parts the bilinear term [ Λ , J μ f ˉ ] ( J λ U ) [\Lambda,\mathcal{J}_\mu\bar{f}](\mathcal{J}_\lambda U) [ Λ , J μ f ˉ ] ( J λ U ) and ∇ x ( J μ f ˉ ) ⋅ R ( J λ U ) \nabla_x(\mathcal{J}_\mu\bar{f})\cdot\mathcal{R}(\mathcal{J}_\lambda U) ∇ x ( J μ f ˉ ) ⋅ R ( J λ U ) have zero mean. Therefore, we seek U ( t ) ∈ H 0 1 ( T 2 ) U(t)\in H_0^1(\mathbb{T}^2) U ( t ) ∈ H 0 1 ( T 2 ) solving (100) .
Define the Banach space
U T : = L ∞ ( 0 , T ; H 0 1 ( T 2 ) ) , ∥ U ∥ U T : = ess sup t ∈ ( 0 , T ) ∥ U ( t ) ∥ H 1 . \mathbb{U}_T:=L^\infty(0,T;H_0^1(\mathbb{T}^2)),\qquad\|U\|_{\mathbb{U}_T}:=\operatorname*{ess\,sup}_{t\in(0,T)}\|U(t)\|_{H^1}. U T := L ∞ ( 0 , T ; H 0 1 ( T 2 )) , ∥ U ∥ U T := t ∈ ( 0 , T ) ess sup ∥ U ( t ) ∥ H 1 .
For fixed ( f ˉ , v ˉ ) (\bar{f},\bar{v}) ( f ˉ , v ˉ ) we define the operator
Φ f ˉ , v ˉ : U T → U T , \Phi_{\bar{f},\bar{v}}:\mathbb{U}_T\to\mathbb{U}_T, Φ f ˉ , v ˉ : U T → U T ,
by
( Φ f ˉ , v ˉ [ U ˉ ] ) ( t ) : = ( I + γ Λ ) − 1 ( F λ , μ [ f ˉ , v ˉ ] ( t ) − J λ N ( J μ f ˉ ( t ) , J λ U ˉ ( t ) ) ) , t ∈ ( 0 , T ) . (103) (\Phi_{\bar{f},\bar{v}}[\bar{U}])(t):=(I+\gamma\Lambda)^{-1}\left(\mathcal{F}^{\lambda,\mu}[\bar{f},\bar{v}](t)-\mathcal{J}_\lambda\mathcal{N}(\mathcal{J}_\mu\bar{f}(t),\mathcal{J}_\lambda\bar{U}(t))\right),\qquad t\in(0,T). \tag*{(103)} ( Φ f ˉ , v ˉ [ U ˉ ]) ( t ) := ( I + γ Λ ) − 1 ( F λ , μ [ f ˉ , v ˉ ] ( t ) − J λ N ( J μ f ˉ ( t ) , J λ U ˉ ( t )) ) , t ∈ ( 0 , T ) . (103)
A fixed point U = Φ f ˉ , v ˉ [ U ] U=\Phi_{\bar{f},\bar{v}}[U] U = Φ f ˉ , v ˉ [ U ] is precisely a solution to (100) in U T \mathbb{U}_T U T . Let U ˉ ∈ U T \bar{U}\in\mathbb{U}_T U ˉ ∈ U T . Using the fact that
∥ ( I + γ Λ ) − 1 G ∥ H 1 ≤ ∥ G ∥ L 2 , G ∈ H 0 1 ( T 2 ) , \|(I+\gamma\Lambda)^{-1}G\|_{H^1}\leq\|G\|_{L^2},\qquad G\in H_0^1(\mathbb{T}^2), ∥ ( I + γ Λ ) − 1 G ∥ H 1 ≤ ∥ G ∥ L 2 , G ∈ H 0 1 ( T 2 ) ,
the bilinear estimate
∥ N ( J μ f ˉ ( t ) , J λ U ˉ ( t ) ) ∥ H 1 ≤ C μ ∥ f ˉ ( t ) ∥ H 3 ∥ U ˉ ( t ) ∥ H 1 \lVert\mathbb{N}(\mathcal{J}_{\mu}\bar{f}(t),\mathcal{J}_{\lambda}\bar{U}(t))\rVert_{H^{1}} \le C_{\mu}\lVert\bar{f}(t)\rVert_{H^{3}}\lVert\bar{U}(t)\rVert_{H^{1}} ∥ N ( J μ f ˉ ( t ) , J λ U ˉ ( t )) ∥ H 1 ≤ C μ ∥ f ˉ ( t ) ∥ H 3 ∥ U ˉ ( t ) ∥ H 1
and the boundedness of J λ \mathcal{J}_{\lambda} J λ on H 1 H^{1} H 1 and J μ \mathcal{J}_{\mu} J μ on H 3 H^{3} H 3 , we obtain for a.e. t ∈ ( 0 , T ) t \in(0,T) t ∈ ( 0 , T )
∥ ( Φ f ˉ , v ˉ [ U ˉ ] ) ( t ) ∥ H 1 ≲ ∥ T λ , μ [ f ˉ , v ˉ ] ( t ) ∥ H 1 + C μ ∥ f ˉ ( t ) ∥ H 3 ∥ U ˉ ( t ) ∥ H 1 . \lVert(\Phi_{\bar{f},\bar{v}}[\bar{U}])(t)\rVert_{H^{1}} \lesssim\lVert\mathcal{T}^{\lambda,\mu}[\bar{f},\bar{v}](t)\rVert_{H^{1}}+C_{\mu}\lVert\bar{f}(t)\rVert_{H^{3}}\lVert\bar{U}(t)\rVert_{H^{1}}. ∥( Φ f ˉ , v ˉ [ U ˉ ]) ( t ) ∥ H 1 ≲ ∥ T λ , μ [ f ˉ , v ˉ ] ( t ) ∥ H 1 + C μ ∥ f ˉ ( t ) ∥ H 3 ∥ U ˉ ( t ) ∥ H 1 .
Taking the essential supremum over t t t gives
∥ Φ f ˉ , v ˉ [ U ˉ ] ∥ L ∞ ( 0 , T ; H 1 ) ≤ C ( λ , μ , f ˉ , v ˉ ) + C μ ∥ f ˉ ∥ L ∞ ( 0 , T ; H 3 ) ∥ U ˉ ∥ L ∞ ( 0 , T ; H 1 ) , (104) \lVert\Phi_{\bar{f},\bar{v}}[\bar{U}]\rVert_{L^{\infty}(0,T;H^{1})} \le C(\lambda,\mu,\bar{f},\bar{v})+C_{\mu}\lVert\bar{f}\rVert_{L^{\infty}(0,T;H^{3})}\lVert\bar{U}\rVert_{L^{\infty}(0,T;H^{1})}, \tag*{(104)} ∥ Φ f ˉ , v ˉ [ U ˉ ] ∥ L ∞ ( 0 , T ; H 1 ) ≤ C ( λ , μ , f ˉ , v ˉ ) + C μ ∥ f ˉ ∥ L ∞ ( 0 , T ; H 3 ) ∥ U ˉ ∥ L ∞ ( 0 , T ; H 1 ) , (104)
where we set
C ( λ , μ , f ˉ , v ˉ ) : = C ∥ T λ , μ [ f ˉ , v ˉ ] ∥ L ∞ ( 0 , T ; H 1 ) . (105) C(\lambda,\mu,\bar{f},\bar{v}):=C\lVert\mathcal{T}^{\lambda,\mu}[\bar{f},\bar{v}]\rVert_{L^{\infty}(0,T;H^{1})}. \tag*{(105)} C ( λ , μ , f ˉ , v ˉ ) := C ∥ T λ , μ [ f ˉ , v ˉ ] ∥ L ∞ ( 0 , T ; H 1 ) . (105)
This quantity is finite for ( f ˉ , v ˉ ) ∈ X T × Y T (\bar{f},\bar{v})\in\mathbb{X}_{T}\times\mathbb{Y}_{T} ( f ˉ , v ˉ ) ∈ X T × Y T because J λ \mathcal{J}_{\lambda} J λ and J μ \mathcal{J}_{\mu} J μ are smoothing and hence map L t ∞ H x s L^{\infty}_{t}H^{s}_{x} L t ∞ H x s into L t ∞ H s + m L^{\infty}_{t}H^{s+m} L t ∞ H s + m for any m ≥ 0 m\ge0 m ≥ 0 .
Next, let us show the contraction estimate. Let U ˉ ( 1 ) , U ˉ ( 2 ) ∈ U T \bar{U}^{(1)},\bar{U}^{(2)}\in\mathbb{U}_{T} U ˉ ( 1 ) , U ˉ ( 2 ) ∈ U T . Subtracting (103) ,
Φ f ˉ , v ˉ [ U ˉ ( 1 ) ] − Φ f ˉ , v ˉ [ U ˉ ( 2 ) ] = − ( I + T Λ ) − 1 J λ N ( J μ f ˉ , J λ ( U ˉ ( 1 ) − U ˉ ( 2 ) ) ) . \Phi_{\bar{f},\bar{v}}[\bar{U}^{(1)}]-\Phi_{\bar{f},\bar{v}}[\bar{U}^{(2)}]=-(I+\mathcal{T}\Lambda)^{-1}\mathcal{J}_{\lambda}\mathbb{N}\left(\mathcal{J}_{\mu}\bar{f},\mathcal{J}_{\lambda}(\bar{U}^{(1)}-\bar{U}^{(2)})\right). Φ f ˉ , v ˉ [ U ˉ ( 1 ) ] − Φ f ˉ , v ˉ [ U ˉ ( 2 ) ] = − ( I + T Λ ) − 1 J λ N ( J μ f ˉ , J λ ( U ˉ ( 1 ) − U ˉ ( 2 ) ) ) .
Hence, by a similar argument as before
∥ Φ f ˉ , v ˉ [ U ˉ ( 1 ) ] − Φ f ˉ , v ˉ [ U ˉ ( 2 ) ] ∥ U T ≲ ∥ f ˉ ∥ L ∞ ( 0 , T ; H 3 ) ∥ U ˉ ( 1 ) − U ˉ ( 2 ) ∥ U T . \lVert\Phi_{\bar{f},\bar{v}}[\bar{U}^{(1)}]-\Phi_{\bar{f},\bar{v}}[\bar{U}^{(2)}]\rVert_{\mathbb{U}_{T}}\lesssim\lVert\bar{f}\rVert_{L^{\infty}(0,T;H^{3})}\lVert\bar{U}^{(1)}-\bar{U}^{(2)}\rVert_{\mathbb{U}_{T}}. ∥ Φ f ˉ , v ˉ [ U ˉ ( 1 ) ] − Φ f ˉ , v ˉ [ U ˉ ( 2 ) ] ∥ U T ≲ ∥ f ˉ ∥ L ∞ ( 0 , T ; H 3 ) ∥ U ˉ ( 1 ) − U ˉ ( 2 ) ∥ U T .
Assume now that
∥ f 0 ∥ H 3 ( T 2 ) ≤ ε ∗ , (106) \lVert f_{0}\rVert_{H^{3}(\mathbb{T}^{2})}\le\varepsilon_{*}, \tag*{(106)} ∥ f 0 ∥ H 3 ( T 2 ) ≤ ε ∗ , (106)
where ε ∗ > 0 \varepsilon_{*}>0 ε ∗ > 0 will be fixed momentarily. Since f ˉ ∈ X T \bar{f}\in\mathbb{X}_{T} f ˉ ∈ X T we have ∥ f ˉ ∥ L ∞ H 3 ≤ 2 ∥ f 0 ∥ H 3 \lVert\bar{f}\rVert_{L^{\infty}H^{3}}\le2\lVert f_{0}\rVert_{H^{3}} ∥ f ˉ ∥ L ∞ H 3 ≤ 2 ∥ f 0 ∥ H 3 , and thus
∥ Φ f ˉ , v ˉ [ U ˉ ( 1 ) ] − Φ f ˉ , v ˉ [ U ˉ ( 2 ) ] ∥ L ∞ H 1 ≤ ( 2 C ∥ f 0 ∥ H 3 ) ∥ U ˉ ( 1 ) − U ˉ ( 2 ) ∥ L ∞ H 1 . \lVert\Phi_{\bar{f},\bar{v}}[\bar{U}^{(1)}]-\Phi_{\bar{f},\bar{v}}[\bar{U}^{(2)}]\rVert_{L^{\infty}H^{1}}\le(2C\lVert f_{0}\rVert_{H^{3}})\lVert\bar{U}^{(1)}-\bar{U}^{(2)}\rVert_{L^{\infty}H^{1}}. ∥ Φ f ˉ , v ˉ [ U ˉ ( 1 ) ] − Φ f ˉ , v ˉ [ U ˉ ( 2 ) ] ∥ L ∞ H 1 ≤ ( 2 C ∥ f 0 ∥ H 3 ) ∥ U ˉ ( 1 ) − U ˉ ( 2 ) ∥ L ∞ H 1 .
Choose ε ∗ > 0 \varepsilon_{*}>0 ε ∗ > 0 so small that
2 C ε ∗ ≤ 1 2 and C ε ∗ ≤ 1 2 . 2C\varepsilon_{*}\le\frac{1}{2}\qquad\text{and}\qquad C\varepsilon_{*}\le\frac{1}{2}. 2 C ε ∗ ≤ 2 1 and C ε ∗ ≤ 2 1 .
(Equivalently, ε ∗ ≤ ( 4 C ) − 1 \varepsilon_{*}\le(4C)^{-1} ε ∗ ≤ ( 4 C ) − 1 .) Then Φ f ˉ , v ˉ \Phi_{\bar{f},\bar{v}} Φ f ˉ , v ˉ is a strict contraction on U T \mathbb{U}_{T} U T , hence by Banach’s fixed-point theorem there exists a unique
U λ , μ [ f ˉ , v ˉ ] ∈ L ∞ ( 0 , T ; H 0 1 ( T 2 ) ) U^{\lambda,\mu}[\bar{f},\bar{v}]\in L^{\infty}(0,T;H^{1}_{0}(\mathbb{T}^{2})) U λ , μ [ f ˉ , v ˉ ] ∈ L ∞ ( 0 , T ; H 0 1 ( T 2 ))
such that U λ , μ [ f ˉ , v ˉ ] = Φ f ˉ , v ˉ [ U λ , μ [ f ˉ , v ˉ ] ] U^{\lambda,\mu}[\bar{f},\bar{v}]=\Phi_{\bar{f},\bar{v}}[U^{\lambda,\mu}[\bar{f},\bar{v}]] U λ , μ [ f ˉ , v ˉ ] = Φ f ˉ , v ˉ [ U λ , μ [ f ˉ , v ˉ ]] , i.e. U λ , μ [ f ˉ , v ˉ ] U^{\lambda,\mu}[\bar{f},\bar{v}] U λ , μ [ f ˉ , v ˉ ] solves (100) . Moreover, taking U ˉ = U \bar{U}=U U ˉ = U in (104) and using ∥ f ˉ ∥ L t ∞ H x 3 ≤ 2 ∥ f 0 ∥ H 3 ≤ ε ∗ \lVert\bar{f}\rVert_{L^{\infty}_{t}H^{3}_{x}}\le2\lVert f_{0}\rVert_{H^{3}}\le\varepsilon_{*} ∥ f ˉ ∥ L t ∞ H x 3 ≤ 2 ∥ f 0 ∥ H 3 ≤ ε ∗ , we obtain
∥ U ∥ L ∞ ( 0 , T ; H 1 ) ≤ C ( λ , μ , f ˉ , v ˉ ) + C ε ∗ ∥ U ∥ L ∞ ( 0 , T ; H 1 ) . \lVert U\rVert_{L^{\infty}(0,T;H^{1})}\le C(\lambda,\mu,\bar{f},\bar{v})+C\varepsilon_{*}\lVert U\rVert_{L^{\infty}(0,T;H^{1})}. ∥ U ∥ L ∞ ( 0 , T ; H 1 ) ≤ C ( λ , μ , f ˉ , v ˉ ) + C ε ∗ ∥ U ∥ L ∞ ( 0 , T ; H 1 ) .
Hence C ε ∗ ≤ 1 2 C\varepsilon_{*}\le\frac{1}{2} C ε ∗ ≤ 2 1 implies
∥ U ∥ L ∞ ( 0 , T ; H 1 ) ≤ 2 C ( λ , μ , f ˉ , v ˉ ) . (107) \lVert U\rVert_{L^{\infty}(0,T;H^{1})}\le2C(\lambda,\mu,\bar{f},\bar{v}). \tag*{(107)} ∥ U ∥ L ∞ ( 0 , T ; H 1 ) ≤ 2 C ( λ , μ , f ˉ , v ˉ ) . (107)
So, we conclude that for each fixed T > 0 T>0 T > 0 and 0 < λ , μ ≤ 1 0<\lambda,\mu\le1 0 < λ , μ ≤ 1 , and for each ( f ˉ , v ˉ ) ∈ X T × Y T (\bar{f},\bar{v})\in\mathbb{X}_{T}\times\mathbb{Y}_{T} ( f ˉ , v ˉ ) ∈ X T × Y T with mean zero and satisfying (106) , the elliptic fixed-point problem (100) admits a unique solution
U λ , μ [ f ˉ , v ˉ ] ∈ L ∞ ( 0 , T ; H 0 1 ( T 2 ) ) , U^{\lambda,\mu}[\bar{f},\bar{v}]\in L^{\infty}(0,T;H^{1}_{0}(\mathbb{T}^{2})), U λ , μ [ f ˉ , v ˉ ] ∈ L ∞ ( 0 , T ; H 0 1 ( T 2 )) ,
and this solution satisfies the bound (107) .
Step 2: Solving the equation for v v v . In Step 1 we constructed, under the smallness assumption ∥ f 0 ∥ H 3 ≤ ε ∗ \|f_0\|_{H^3} \le\varepsilon_* ∥ f 0 ∥ H 3 ≤ ε ∗ , a unique auxiliary field
U λ , μ [ f ˉ , v ˉ ] ∈ U T = L ∞ ( 0 , T ; H 0 1 ( T 2 ) ) , U^{\lambda,\mu}[\bar f,\bar v] \in U_T = L^\infty(0,T;H_0^1(\mathbb{T}^2)), U λ , μ [ f ˉ , v ˉ ] ∈ U T = L ∞ ( 0 , T ; H 0 1 ( T 2 )) ,
solving the elliptic problem in (100) . In this step we solve for a velocity v = v λ , μ v=v^{\lambda,\mu} v = v λ , μ such that v t = U λ , μ [ f ˉ , v ] v_t=U^{\lambda,\mu}[\bar f,v] v t = U λ , μ [ f ˉ , v ] .
Given v ˉ ∈ Y T \bar v \in Y_T v ˉ ∈ Y T , define
U f ˉ ( v ˉ ) : = U λ , μ [ f ˉ , v ˉ ] ∈ U T , U_{\bar f}(\bar v) := U^{\lambda,\mu}[\bar f,\bar v] \in U_T, U f ˉ ( v ˉ ) := U λ , μ [ f ˉ , v ˉ ] ∈ U T ,
the elliptic solution from Step 1 associated with ( f ˉ , v ˉ ) (\bar f,\bar v) ( f ˉ , v ˉ ) . Then we define Ψ f ˉ : Y T → Y T \Psi_{\bar f}:Y_T \to Y_T Ψ f ˉ : Y T → Y T by
( Ψ f ˉ [ v ˉ ] ) ( t ) : = v 0 + ∫ 0 t U f ˉ ( v ˉ ) ( s ) d s , t ∈ [ 0 , T ] . (108) (\Psi_{\bar f}[\bar v])(t) := v_0 + \int_0^t U_{\bar f}(\bar v)(s)\,ds,\qquad t\in[0,T]. \tag*{(108)} ( Ψ f ˉ [ v ˉ ]) ( t ) := v 0 + ∫ 0 t U f ˉ ( v ˉ ) ( s ) d s , t ∈ [ 0 , T ] . (108)
Let v ˉ ∈ Y T \bar v\in Y_T v ˉ ∈ Y T . Since U f ˉ ( v ˉ ) ∈ U T U_{\bar f}(\bar v)\in U_T U f ˉ ( v ˉ ) ∈ U T , the integral in (108) is well-defined in H 1 H^1 H 1 , and
∥ Ψ f ˉ [ v ˉ ] ( t ) ∥ H 1 ≤ ∥ v 0 ∥ H 1 + ∫ 0 t ∥ U f ˉ ( v ˉ ) ( s ) ∥ H 1 d s ≤ ∥ v 0 ∥ H 1 + T ∥ U f ˉ ( v ˉ ) ∥ L ∞ ( 0 , T ; H 1 ) . \|\Psi_{\bar f}[\bar v](t)\|_{H^1}\leq\|v_0\|_{H^1}+\int_0^t\|U_{\bar f}(\bar v)(s)\|_{H^1}\,ds\leq\|v_0\|_{H^1}+T\|U_{\bar f}(\bar v)\|_{L^\infty(0,T;H^1)}. ∥ Ψ f ˉ [ v ˉ ] ( t ) ∥ H 1 ≤ ∥ v 0 ∥ H 1 + ∫ 0 t ∥ U f ˉ ( v ˉ ) ( s ) ∥ H 1 d s ≤ ∥ v 0 ∥ H 1 + T ∥ U f ˉ ( v ˉ ) ∥ L ∞ ( 0 , T ; H 1 ) .
Hence Ψ f ˉ [ v ˉ ] ∈ Y T \Psi_{\bar f}[\bar v]\in Y_T Ψ f ˉ [ v ˉ ] ∈ Y T provided we control ∥ U f ˉ ( v ˉ ) ∥ L ∞ H 1 \|U_{\bar f}(\bar v)\|_{L^\infty H^1} ∥ U f ˉ ( v ˉ ) ∥ L ∞ H 1 . Such a bound follows from Step 1, since for each t t t the fixed point U ( t ) U(t) U ( t ) solves
( I + T Λ ) U = F λ , μ [ f ˉ , v ˉ ] − J λ ( N ( J μ f ˉ , J λ U ) ) , (I+\mathcal{T}\Lambda)U=\mathcal{F}^{\lambda,\mu}[\bar f,\bar v]-\mathcal{J}_\lambda\bigl(\mathcal{N}(\mathcal{J}_\mu\bar f,\mathcal{J}_\lambda U)\bigr), ( I + T Λ ) U = F λ , μ [ f ˉ , v ˉ ] − J λ ( N ( J μ f ˉ , J λ U ) ) ,
and Step 1 provides an estimate of the form
∥ U f ˉ ( v ˉ ) ∥ L ∞ ( 0 , T ; H 1 ) ≤ C ( λ , μ , f ˉ , v ˉ ) , (109) \|U_{\bar f}(\bar v)\|_{L^\infty(0,T;H^1)}\leq C(\lambda,\mu,\bar f,\bar v), \tag*{(109)} ∥ U f ˉ ( v ˉ ) ∥ L ∞ ( 0 , T ; H 1 ) ≤ C ( λ , μ , f ˉ , v ˉ ) , (109)
where C ( λ , μ , f ˉ , v ˉ ) C(\lambda,\mu,\bar f,\bar v) C ( λ , μ , f ˉ , v ˉ ) is finite for ( f ˉ , v ˉ ) ∈ X T × Y T (\bar f,\bar v)\in X_T\times Y_T ( f ˉ , v ˉ ) ∈ X T × Y T (and depends on λ , μ \lambda,\mu λ , μ through smoothing).
Let v ˉ ( 1 ) , v ˉ ( 2 ) ∈ Y T \bar v^{(1)},\bar v^{(2)}\in Y_T v ˉ ( 1 ) , v ˉ ( 2 ) ∈ Y T , and set
U ( i ) : = U f ˉ ( v ˉ ( i ) ) = U λ , μ [ f ˉ , v ˉ ( i ) ] ∈ U T . U^{(i)}:=U_{\bar f}(\bar v^{(i)})=U^{\lambda,\mu}[\bar f,\bar v^{(i)}]\in U_T. U ( i ) := U f ˉ ( v ˉ ( i ) ) = U λ , μ [ f ˉ , v ˉ ( i ) ] ∈ U T .
Subtracting the elliptic equations (100) satisfied by U ( 1 ) U^{(1)} U ( 1 ) and U ( 2 ) U^{(2)} U ( 2 ) yields, for a.e. t ∈ ( 0 , T ) t\in(0,T) t ∈ ( 0 , T ) ,
( I + T Λ ) ( U ( 1 ) − U ( 2 ) ) = ( F λ , μ [ f ˉ , v ˉ ( 1 ) ] − F λ , μ [ f ˉ , v ˉ ( 2 ) ] ) − J λ ( N ( J μ f ˉ , J λ U ( 1 ) ) − N ( J μ f ˉ , J λ U ( 2 ) ) ) . (110) (I+\mathcal{T}\Lambda)(U^{(1)}-U^{(2)})=
\left(\mathcal{F}^{\lambda,\mu}[\bar f,\bar v^{(1)}]-\mathcal{F}^{\lambda,\mu}[\bar f,\bar v^{(2)}]\right)
-\mathcal{J}_\lambda\left(\mathcal{N}(\mathcal{J}_\mu\bar f,\mathcal{J}_\lambda U^{(1)})-\mathcal{N}(\mathcal{J}_\mu\bar f,\mathcal{J}_\lambda U^{(2)})\right). \tag*{(110)} ( I + T Λ ) ( U ( 1 ) − U ( 2 ) ) = ( F λ , μ [ f ˉ , v ˉ ( 1 ) ] − F λ , μ [ f ˉ , v ˉ ( 2 ) ] ) − J λ ( N ( J μ f ˉ , J λ U ( 1 ) ) − N ( J μ f ˉ , J λ U ( 2 ) ) ) . (110)
The difference in the coupling term is linear in U ( 1 ) − U ( 2 ) U^{(1)}-U^{(2)} U ( 1 ) − U ( 2 ) , and by the bilinear estimate from the preliminaries (applied in H 1 H^1 H 1 using that J λ \mathcal{J}_\lambda J λ is bounded on H 1 H^1 H 1 ) we have
∥ J λ ( N ( J μ f ˉ , J λ U ( 1 ) ) − N ( J μ f ˉ , J λ U ( 2 ) ) ) ∥ H 1 ≲ ∥ f ˉ ∥ H 3 ∥ U ( 1 ) − U ( 2 ) ∥ H 1 . (111) \|\mathcal{J}_\lambda(\mathcal{N}(\mathcal{J}_\mu\bar f,\mathcal{J}_\lambda U^{(1)})-\mathcal{N}(\mathcal{J}_\mu\bar f,\mathcal{J}_\lambda U^{(2)}))\|_{H^1}\lesssim\|\bar f\|_{H^3}\|U^{(1)}-U^{(2)}\|_{H^1}. \tag*{(111)} ∥ J λ ( N ( J μ f ˉ , J λ U ( 1 ) ) − N ( J μ f ˉ , J λ U ( 2 ) )) ∥ H 1 ≲ ∥ f ˉ ∥ H 3 ∥ U ( 1 ) − U ( 2 ) ∥ H 1 . (111)
It remains to estimate the forcing difference in H 1 H^1 H 1 . Inspecting (101) , the v ˉ \bar v v ˉ -dependence is contained only in the terms
− δ J λ Λ 3 J λ v ˉ and ε J λ Q ( J λ v ˉ ) , -\delta\,\mathcal{J}_\lambda\Lambda^3\mathcal{J}_\lambda\bar v
\qquad\text{and}\qquad
\varepsilon\,\mathcal{J}_\lambda Q(\mathcal{J}_\lambda\bar v), − δ J λ Λ 3 J λ v ˉ and ε J λ Q ( J λ v ˉ ) ,
where
Q ( w ) : = 1 2 Λ ( w 2 − ∣ R w ∣ 2 ) − [ Λ , w ] w − ∇ x w ⋅ R w . Q(w):=\frac{1}{2}\Lambda\left(w^2-|\mathcal{R}w|^2\right)-[\Lambda,w]w-\nabla_xw\cdot\mathcal{R}w. Q ( w ) := 2 1 Λ ( w 2 − ∣ R w ∣ 2 ) − [ Λ , w ] w − ∇ x w ⋅ R w .
Since J λ \mathcal{J}_{\lambda} J λ is smoothing, the Fourier multiplier J λ Λ 3 J λ \mathcal{J}_{\lambda}\Lambda^{3}\mathcal{J}_{\lambda} J λ Λ 3 J λ is bounded from H 1 H^{1} H 1 to H 1 H^{1} H 1 , with operator norm depending on λ \lambda λ ; hence
∥ J λ Λ 3 J λ ( v ‾ ( 1 ) − v ‾ ( 2 ) ) ∥ H 1 ≤ C λ ∥ v ‾ ( 1 ) − v ‾ ( 2 ) ∥ H 1 . (112) \left\|\mathcal{J}_{\lambda}\Lambda^{3}\mathcal{J}_{\lambda}\left(\overline{v}^{(1)}-\overline{v}^{(2)}\right)\right\|_{H^{1}}\leq C_{\lambda}\left\|\overline{v}^{(1)}-\overline{v}^{(2)}\right\|_{H^{1}}. \tag*{(112)} J λ Λ 3 J λ ( v ( 1 ) − v ( 2 ) ) H 1 ≤ C λ v ( 1 ) − v ( 2 ) H 1 . (112)
For the quadratic term, we use that J λ : H 1 → H m \mathcal{J}_{\lambda}:H^{1}\to H^{m} J λ : H 1 → H m for every m m m , and in particular ∥ J λ h ∥ H 3 ≤ C λ ∥ h ∥ H 1 \left\|\mathcal{J}_{\lambda}h\right\|_{H^{3}}\leq C_{\lambda}\left\|h\right\|_{H^{1}} ∥ J λ h ∥ H 3 ≤ C λ ∥ h ∥ H 1 . Since H 3 ( T 2 ) ↪ W 1 , ∞ ( T 2 ) H^{3}(\mathbb{T}^{2})\hookrightarrow W^{1,\infty}(\mathbb{T}^{2}) H 3 ( T 2 ) ↪ W 1 , ∞ ( T 2 ) , standard product/commutator bounds imply that Q Q Q is locally Lipschitz from H 3 H^{3} H 3 to H 1 H^{1} H 1 : for every R > 0 R>0 R > 0 ,
∥ Q ( w ( 1 ) ) − Q ( w ( 2 ) ) ∥ H 1 ≤ C ( ∥ w ( 1 ) ∥ H 3 + ∥ w ( 2 ) ∥ H 3 ) ∥ w ( 1 ) − w ( 2 ) ∥ H 3 (113) \left\|Q\left(w^{(1)}\right)-Q\left(w^{(2)}\right)\right\|_{H^{1}}\leq C\left(\left\|w^{(1)}\right\|_{H^{3}}+\left\|w^{(2)}\right\|_{H^{3}}\right)\left\|w^{(1)}-w^{(2)}\right\|_{H^{3}} \tag*{(113)} Q ( w ( 1 ) ) − Q ( w ( 2 ) ) H 1 ≤ C ( w ( 1 ) H 3 + w ( 2 ) H 3 ) w ( 1 ) − w ( 2 ) H 3 (113)
and therefore, for v ‾ ( 1 ) , v ‾ ( 2 ) \overline{v}^{(1)},\overline{v}^{(2)} v ( 1 ) , v ( 2 ) with ∥ v ‾ ( i ) ∥ H 1 ≤ R \left\|\overline{v}^{(i)}\right\|_{H^{1}}\leq R v ( i ) H 1 ≤ R ,
∥ Q ( J λ v ‾ ( 1 ) ) − Q ( J λ v ‾ ( 2 ) ) ∥ H 1 ≤ C λ , R ∥ v ‾ ( 1 ) − v ‾ ( 2 ) ∥ H 1 , (114) \left\|Q\left(\mathcal{J}_{\lambda}\overline{v}^{(1)}\right)-Q\left(\mathcal{J}_{\lambda}\overline{v}^{(2)}\right)\right\|_{H^{1}}\leq C_{\lambda,R}\left\|\overline{v}^{(1)}-\overline{v}^{(2)}\right\|_{H^{1}}, \tag*{(114)} Q ( J λ v ( 1 ) ) − Q ( J λ v ( 2 ) ) H 1 ≤ C λ , R v ( 1 ) − v ( 2 ) H 1 , (114)
for some constant C λ , R C_{\lambda,R} C λ , R depending on λ \lambda λ and R R R . Since J λ \mathcal{J}_{\lambda} J λ is bounded on H 1 H^{1} H 1 , we then also have
∥ J λ ( Q ( J λ v ‾ ( 1 ) ) − Q ( J λ v ‾ ( 2 ) ) ) ∥ H 1 ≤ C λ , R ∥ v ‾ ( 1 ) − v ‾ ( 2 ) ∥ H 1 . (115) \left\|\mathcal{J}_{\lambda}\left(Q\left(\mathcal{J}_{\lambda}\overline{v}^{(1)}\right)-Q\left(\mathcal{J}_{\lambda}\overline{v}^{(2)}\right)\right)\right\|_{H^{1}}\leq C_{\lambda,R}\left\|\overline{v}^{(1)}-\overline{v}^{(2)}\right\|_{H^{1}}. \tag*{(115)} J λ ( Q ( J λ v ( 1 ) ) − Q ( J λ v ( 2 ) ) ) H 1 ≤ C λ , R v ( 1 ) − v ( 2 ) H 1 . (115)
Combining (112) and (115) gives
∥ F λ , μ [ f ‾ , v ‾ ( 1 ) ] − F λ , μ [ f ‾ , v ‾ ( 2 ) ] ∥ H 1 ≤ C λ , R ∥ v ‾ ( 1 ) − v ‾ ( 2 ) ∥ H 1 . (116) \left\|\mathcal{F}^{\lambda,\mu}\left[\overline{f},\overline{v}^{(1)}\right]-\mathcal{F}^{\lambda,\mu}\left[\overline{f},\overline{v}^{(2)}\right]\right\|_{H^{1}}\leq C_{\lambda,R}\left\|\overline{v}^{(1)}-\overline{v}^{(2)}\right\|_{H^{1}}. \tag*{(116)} F λ , μ [ f , v ( 1 ) ] − F λ , μ [ f , v ( 2 ) ] H 1 ≤ C λ , R v ( 1 ) − v ( 2 ) H 1 . (116)
Now apply ( I + Υ Λ ) − 1 (I+\Upsilon\Lambda)^{-1} ( I + ΥΛ ) − 1 to (110) . Using that ( I + Υ Λ ) − 1 (I+\Upsilon\Lambda)^{-1} ( I + ΥΛ ) − 1 is bounded on H 0 1 ( T 2 ) H_{0}^{1}(\mathbb{T}^{2}) H 0 1 ( T 2 ) and combining (111) and (116) , we obtain
∥ U ( 1 ) − U ( 2 ) ∥ H 1 ≤ C λ , R ∥ v ‾ ( 1 ) − v ‾ ( 2 ) ∥ H 1 + C ∥ f ‾ ∥ H 3 ∥ U ( 1 ) − U ( 2 ) ∥ H 1 . (117) \left\|U^{(1)}-U^{(2)}\right\|_{H^{1}}\leq C_{\lambda,R}\left\|\overline{v}^{(1)}-\overline{v}^{(2)}\right\|_{H^{1}}+C\left\|\overline{f}\right\|_{H^{3}}\left\|U^{(1)}-U^{(2)}\right\|_{H^{1}}. \tag*{(117)} U ( 1 ) − U ( 2 ) H 1 ≤ C λ , R v ( 1 ) − v ( 2 ) H 1 + C f H 3 U ( 1 ) − U ( 2 ) H 1 . (117)
Under the same smallness assumption as in Step 1, ∥ f ‾ ∥ L ∞ ( 0 , T ; H 3 ) ≤ 2 ∥ f 0 ∥ H 3 ≤ ϵ ∗ \left\|\overline{f}\right\|_{L^{\infty}(0,T;H^{3})}\leq2\left\|f_{0}\right\|_{H^{3}}\leq\epsilon_{*} f L ∞ ( 0 , T ; H 3 ) ≤ 2 ∥ f 0 ∥ H 3 ≤ ϵ ∗ with ϵ ∗ \epsilon_{*} ϵ ∗ chosen so that C ϵ ∗ ≤ 1 2 C\epsilon_{*}\leq\frac{1}{2} C ϵ ∗ ≤ 2 1 , we can absorb the last term in (117) and obtain the Lipschitz bound
∥ U f ‾ ( v ‾ ( 1 ) ) − U f ‾ ( v ‾ ( 2 ) ) ∥ L ∞ ( 0 , T ; H 1 ) ≤ C λ , R ∥ v ‾ ( 1 ) − v ‾ ( 2 ) ∥ L ∞ ( 0 , T ; H 1 ) , (118) \left\|U_{\overline{f}}\left(\overline{v}^{(1)}\right)-U_{\overline{f}}\left(\overline{v}^{(2)}\right)\right\|_{L^{\infty}(0,T;H^{1})}\leq C_{\lambda,R}\left\|\overline{v}^{(1)}-\overline{v}^{(2)}\right\|_{L^{\infty}(0,T;H^{1})}, \tag*{(118)} U f ( v ( 1 ) ) − U f ( v ( 2 ) ) L ∞ ( 0 , T ; H 1 ) ≤ C λ , R v ( 1 ) − v ( 2 ) L ∞ ( 0 , T ; H 1 ) , (118)
for v ‾ ( 1 ) , v ‾ ( 2 ) \overline{v}^{(1)},\overline{v}^{(2)} v ( 1 ) , v ( 2 ) in the closed ball { ∥ v ∥ Y T ≤ R } \{\|v\|_{Y_{T}}\leq R\} { ∥ v ∥ Y T ≤ R } . Let v ‾ ( 1 ) , v ‾ ( 2 ) ∈ Y T \overline{v}^{(1)},\overline{v}^{(2)}\in Y_{T} v ( 1 ) , v ( 2 ) ∈ Y T with ∥ v ‾ ( i ) ∥ Y T ≤ R \left\|\overline{v}^{(i)}\right\|_{Y_{T}}\leq R v ( i ) Y T ≤ R , and set v ( i ) = Ψ f ‾ [ v ‾ ( i ) ] v^{(i)}=\Psi_{\overline{f}}\left[\overline{v}^{(i)}\right] v ( i ) = Ψ f [ v ( i ) ] . Using (108) and (118) ,
∥ v ( 1 ) ( t ) − v ( 2 ) ( t ) ∥ H 1 ≤ ∫ 0 t ∥ U f ‾ ( v ‾ ( 1 ) ) ( s ) − U f ‾ ( v ‾ ( 2 ) ) ( s ) ∥ H 1 d s ≤ t C λ , R ∥ v ‾ ( 1 ) − v ‾ ( 2 ) ∥ Y T . \left\|v^{(1)}(t)-v^{(2)}(t)\right\|_{H^{1}}\leq\int_{0}^{t}\left\|U_{\overline{f}}\left(\overline{v}^{(1)}\right)(s)-U_{\overline{f}}\left(\overline{v}^{(2)}\right)(s)\right\|_{H^{1}}\,\mathrm{d}s\leq tC_{\lambda,R}\left\|\overline{v}^{(1)}-\overline{v}^{(2)}\right\|_{Y_{T}}. v ( 1 ) ( t ) − v ( 2 ) ( t ) H 1 ≤ ∫ 0 t U f ( v ( 1 ) ) ( s ) − U f ( v ( 2 ) ) ( s ) H 1 d s ≤ t C λ , R v ( 1 ) − v ( 2 ) Y T .
Taking the essential supremum over t ∈ ( 0 , T ) t\in(0,T) t ∈ ( 0 , T ) gives
∥ Ψ f ‾ [ v ‾ ( 1 ) ] − Ψ f ‾ [ v ‾ ( 2 ) ] ∥ Y T ≤ T C λ , R ∥ v ‾ ( 1 ) − v ‾ ( 2 ) ∥ Y T . (119) \left\|\Psi_{\overline{f}}\left[\overline{v}^{(1)}\right]-\Psi_{\overline{f}}\left[\overline{v}^{(2)}\right]\right\|_{Y_{T}}\leq TC_{\lambda,R}\left\|\overline{v}^{(1)}-\overline{v}^{(2)}\right\|_{Y_{T}}. \tag*{(119)} Ψ f [ v ( 1 ) ] − Ψ f [ v ( 2 ) ] Y T ≤ T C λ , R v ( 1 ) − v ( 2 ) Y T . (119)
Choosing T > 0 T>0 T > 0 so that T C λ , R ≤ 1 2 TC_{\lambda,R}\leq\frac{1}{2} T C λ , R ≤ 2 1 , we conclude that Ψ f ‾ \Psi_{\overline{f}} Ψ f is a strict contraction on the closed ball { ∥ v ∥ Y T ≤ R } \{\|v\|_{Y_{T}}\leq R\} { ∥ v ∥ Y T ≤ R } , provided R R R is chosen so that Ψ f ‾ \Psi_{\overline{f}} Ψ f maps the ball into itself. This follows from (109) and
∥ v ∥ Y T ≤ ∥ v 0 ∥ H 1 + T ∥ U f ‾ ( v ) ∥ L ∞ ( 0 , T ; H 1 ) . \|v\|_{Y_{T}}\leq\|v_{0}\|_{H^{1}}+T\left\|U_{\overline{f}}(v)\right\|_{L^{\infty}(0,T;H^{1})}. ∥ v ∥ Y T ≤ ∥ v 0 ∥ H 1 + T U f ( v ) L ∞ ( 0 , T ; H 1 ) .
Hence Banach’s fixed point theorem yields a unique
v λ , μ ∈ Y T such that v λ , μ = Ψ f ‾ [ v λ , μ ] . v^{\lambda,\mu}\in Y_{T}\quad\text{such that}\quad v^{\lambda,\mu}=\Psi_{\overline{f}}\left[v^{\lambda,\mu}\right]. v λ , μ ∈ Y T such that v λ , μ = Ψ f [ v λ , μ ] .
Moreover, we have
∂ t v λ , μ = U λ , μ [ f ‾ , v λ , μ ] in L ∞ ( 0 , T ; H 1 ( T 2 ) ) , (120) \partial_{t}v^{\lambda,\mu}=U^{\lambda,\mu}\left[\overline{f},v^{\lambda,\mu}\right]\quad\text{in }L^{\infty}\left(0,T;H^{1}(\mathbb{T}^{2})\right), \tag*{(120)} ∂ t v λ , μ = U λ , μ [ f , v λ , μ ] in L ∞ ( 0 , T ; H 1 ( T 2 ) ) , (120)
thereby identifying the elliptic unknown with the time derivative of v v v . In particular, substituting (120) into the elliptic identity (100) (with v ˉ = v λ , μ \bar{v}=v^{\lambda,\mu} v ˉ = v λ , μ ) shows that v λ , μ v^{\lambda,\mu} v λ , μ satisfies the λ \lambda λ –regularized evolution equation
( I + T Λ ) ∂ t v λ , μ + δ J λ Λ 3 J λ v λ , μ + J μ ( Λ + β 4 Λ 5 ) J μ f ˉ + J λ N ( J μ f ˉ , J λ ∂ t v λ , μ ) = ε J λ [ 1 2 Λ ( ( J λ v λ , μ ) 2 − ∣ R J λ v λ , μ ∣ 2 ) − [ Λ , J λ v λ , μ ] J λ v λ , μ − ∇ x ( J λ v λ , μ ) ⋅ R ( J λ v λ , μ ) ] , (121) \begin{aligned}
(I+\mathcal{T}\Lambda)\partial_t v^{\lambda,\mu}+\delta\mathcal{J}_{\lambda}\Lambda^3\mathcal{J}_{\lambda}v^{\lambda,\mu}+\mathcal{J}_{\mu}\left(\Lambda+\frac{\beta}{4}\Lambda^5\right)\mathcal{J}_{\mu}\bar{f}+\mathcal{J}_{\lambda}\mathcal{N}\left(\mathcal{J}_{\mu}\bar{f},\mathcal{J}_{\lambda}\partial_t v^{\lambda,\mu}\right) \\
&=\varepsilon\mathcal{J}_{\lambda}\left[\frac{1}{2}\Lambda\left((\mathcal{J}_{\lambda}v^{\lambda,\mu})^2-\left|\mathcal{R}\mathcal{J}_{\lambda}v^{\lambda,\mu}\right|^2\right)-\left[\Lambda,\mathcal{J}_{\lambda}v^{\lambda,\mu}\right]\mathcal{J}_{\lambda}v^{\lambda,\mu}-\nabla_x(\mathcal{J}_{\lambda}v^{\lambda,\mu})\cdot\mathcal{R}(\mathcal{J}_{\lambda}v^{\lambda,\mu})\right], \tag*{(121)}
\end{aligned} ( I + T Λ ) ∂ t v λ , μ + δ J λ Λ 3 J λ v λ , μ + J μ ( Λ + 4 β Λ 5 ) J μ f ˉ + J λ N ( J μ f ˉ , J λ ∂ t v λ , μ ) = ε J λ [ 2 1 Λ ( ( J λ v λ , μ ) 2 − R J λ v λ , μ 2 ) − [ Λ , J λ v λ , μ ] J λ v λ , μ − ∇ x ( J λ v λ , μ ) ⋅ R ( J λ v λ , μ ) ] , (121)
in L ∞ ( 0 , T ; H 1 ( T 2 ) ) L^\infty(0,T;H^1(\mathbb{T}^2)) L ∞ ( 0 , T ; H 1 ( T 2 )) .
Step 3: Passing to the limit as λ → 0 \lambda\to0 λ → 0 . Fix 0 < μ ≤ 1 0<\mu\leq1 0 < μ ≤ 1 and let v λ v^\lambda v λ be the solutions constructed in Step 2 on [ 0 , T ] [0,T] [ 0 , T ] , where v λ ∈ L ∞ ( 0 , T ; H 0 1 ( T 2 ) ) v^\lambda\in L^\infty(0,T;H^1_0(\mathbb{T}^2)) v λ ∈ L ∞ ( 0 , T ; H 0 1 ( T 2 )) .1 We claim that, for T > 0 T>0 T > 0 chosen as in Step 2, the family { v λ } \{v^\lambda\} { v λ } satisfies the uniform estimates
sup 0 ≤ t ≤ T ∥ v λ ( t ) ∥ H 1 ≤ C , ∫ 0 T ∥ J λ v λ ( t ) ∥ H 2 2 d t ≤ C , ∫ 0 T ∥ ∂ t v λ ( t ) ∥ H − 1 / 2 2 d t ≤ C , (122) \sup_{0\leq t\leq T}\|v^\lambda(t)\|_{H^1}\leq C,\qquad\int_0^T\|\mathcal{J}_{\lambda}v^\lambda(t)\|_{H^2}^2\,dt\leq C,\qquad\int_0^T\|\partial_t v^\lambda(t)\|_{H^{-1/2}}^2\,dt\leq C, \tag*{(122)} 0 ≤ t ≤ T sup ∥ v λ ( t ) ∥ H 1 ≤ C , ∫ 0 T ∥ J λ v λ ( t ) ∥ H 2 2 d t ≤ C , ∫ 0 T ∥ ∂ t v λ ( t ) ∥ H − 1/2 2 d t ≤ C , (122)
with a constant C C C independent of λ \lambda λ . Testing the regularized equation (121) against Λ v λ \Lambda v^\lambda Λ v λ yields an inequality of the form
1 2 d d t ( ∥ v λ ∥ H 1 / 2 2 + T ∥ v λ ∥ H 1 2 ) + δ ∥ J λ v λ ∥ H 2 2 ≤ R 1 λ ( t ) , (123) \frac{1}{2}\frac{d}{dt}\left(\|v^\lambda\|_{H^{1/2}}^2+\mathcal{T}\|v^\lambda\|_{H^1}^2\right)+\delta\|\mathcal{J}_{\lambda}v^\lambda\|_{H^2}^2\leq R_1^\lambda(t), \tag*{(123)} 2 1 d t d ( ∥ v λ ∥ H 1/2 2 + T ∥ v λ ∥ H 1 2 ) + δ ∥ J λ v λ ∥ H 2 2 ≤ R 1 λ ( t ) , (123)
where R 1 λ R_1^\lambda R 1 λ collects the quadratic terms and the coupling term N ( J μ f ˉ , J λ v λ ) \mathcal{N}(\mathcal{J}_{\mu}\bar{f},\mathcal{J}_{\lambda}v^\lambda) N ( J μ f ˉ , J λ v λ ) . More precisely,
R 1 λ ( t ) = − ∫ T 2 ( J μ ( Λ + β 4 Λ 5 ) J μ f ˉ ( t ) ) Λ v λ ( t ) d x − ∫ T 2 ( J λ N ( J μ f ˉ ( t ) , J λ v λ ( t ) ) ) Λ v λ ( t ) d x + ε ∫ T 2 J λ [ 1 2 Λ ( ( J λ v λ ( t ) ) 2 − ∣ R J λ v λ ( t ) ∣ 2 ) − [ Λ , J λ v λ ( t ) ] J λ v λ ( t ) − ∇ x ( J λ v λ ( t ) ) ⋅ R ( J λ v λ ( t ) ) ] Λ v λ ( t ) d x = : M 1 ( t ) + M 2 ( t ) + M 3 ( t ) . \begin{aligned}
R_1^\lambda(t)=-\int_{\mathbb{T}^2}\left(\mathcal{J}_{\mu}\left(\Lambda+\frac{\beta}{4}\Lambda^5\right)\mathcal{J}_{\mu}\bar{f}(t)\right)\Lambda v^\lambda(t)\,dx-\int_{\mathbb{T}^2}\left(\mathcal{J}_{\lambda}\mathcal{N}\left(\mathcal{J}_{\mu}\bar{f}(t),\mathcal{J}_{\lambda}v^\lambda(t)\right)\right)\Lambda v^\lambda(t)\,dx \\
&+\varepsilon\int_{\mathbb{T}^2}\mathcal{J}_{\lambda}\left[\frac{1}{2}\Lambda\left((\mathcal{J}_{\lambda}v^\lambda(t))^2-\left|\mathcal{R}\mathcal{J}_{\lambda}v^\lambda(t)\right|^2\right)-\left[\Lambda,\mathcal{J}_{\lambda}v^\lambda(t)\right]\mathcal{J}_{\lambda}v^\lambda(t)-\nabla_x(\mathcal{J}_{\lambda}v^\lambda(t))\cdot\mathcal{R}(\mathcal{J}_{\lambda}v^\lambda(t))\right]\Lambda v^\lambda(t)\,dx \\
&=:M_1(t)+M_2(t)+M_3(t).
\end{aligned} R 1 λ ( t ) = − ∫ T 2 ( J μ ( Λ + 4 β Λ 5 ) J μ f ˉ ( t ) ) Λ v λ ( t ) d x − ∫ T 2 ( J λ N ( J μ f ˉ ( t ) , J λ v λ ( t ) ) ) Λ v λ ( t ) d x + ε ∫ T 2 J λ [ 2 1 Λ ( ( J λ v λ ( t ) ) 2 − R J λ v λ ( t ) 2 ) − [ Λ , J λ v λ ( t ) ] J λ v λ ( t ) − ∇ x ( J λ v λ ( t )) ⋅ R ( J λ v λ ( t )) ] Λ v λ ( t ) d x =: M 1 ( t ) + M 2 ( t ) + M 3 ( t ) .
Using ∥ f ˉ ∥ L ∞ H 3 ≤ 2 ∥ f 0 ∥ H 3 \|\bar{f}\|_{L^\infty H^3}\leq2\|f_0\|_{H^3} ∥ f ˉ ∥ L ∞ H 3 ≤ 2∥ f 0 ∥ H 3 and Cauchy–Schwarz together with the boundedness/smoothing of J μ \mathcal{J}_{\mu} J μ on Sobolev spaces, we have
∣ M 1 ( t ) ∣ ≲ ∥ f ˉ ( t ) ∥ H 3 ∥ v λ ( t ) ∥ H 1 ≤ C μ , β ∥ f 0 ∥ H 3 ∥ v λ ( t ) ∥ H 1 . |M_1(t)|\lesssim\|\bar{f}(t)\|_{H^3}\|v^\lambda(t)\|_{H^1}\leq C_{\mu,\beta}\|f_0\|_{H^3}\|v^\lambda(t)\|_{H^1}. ∣ M 1 ( t ) ∣ ≲ ∥ f ˉ ( t ) ∥ H 3 ∥ v λ ( t ) ∥ H 1 ≤ C μ , β ∥ f 0 ∥ H 3 ∥ v λ ( t ) ∥ H 1 .
To estimate M 3 ( t ) M_3(t) M 3 ( t ) we use that J λ \mathcal{J}_{\lambda} J λ is self-adjoint and bounded on all H s H^s H s , H 2 ( T 2 ) ↪ W 1 , ∞ ( T 2 ) H^2(\mathbb{T}^2)\hookrightarrow W^{1,\infty}(\mathbb{T}^2) H 2 ( T 2 ) ↪ W 1 , ∞ ( T 2 ) , and the boundedness of Riesz transforms on Sobolev spaces. We obtain
∣ M 3 ( t ) ∣ ≲ ∥ J λ v λ ( t ) ∥ H 2 ∥ v λ ( t ) ∥ L 2 ∥ v λ ( t ) ∥ H 1 + ∥ J λ v λ ( t ) ∥ H 2 ∥ v λ ( t ) ∥ H 1 2 . (124) |M_3(t)|\lesssim\|\mathcal{J}_{\lambda}v^\lambda(t)\|_{H^2}\|v^\lambda(t)\|_{L^2}\|v^\lambda(t)\|_{H^1}+\|\mathcal{J}_{\lambda}v^\lambda(t)\|_{H^2}\|v^\lambda(t)\|_{H^1}^2. \tag*{(124)} ∣ M 3 ( t ) ∣ ≲ ∥ J λ v λ ( t ) ∥ H 2 ∥ v λ ( t ) ∥ L 2 ∥ v λ ( t ) ∥ H 1 + ∥ J λ v λ ( t ) ∥ H 2 ∥ v λ ( t ) ∥ H 1 2 . (124)
To estimate M 2 ( t ) M_2(t) M 2 ( t ) we use that J λ \mathcal{J}_{\lambda} J λ is self-adjoint on L 2 L^2 L 2 and write
M 2 ( t ) : = − ∫ T 2 ( J λ N ( J μ f ˉ ( t ) , J λ v t λ ( t ) ) ) Λ v λ ( t ) d x = − ∫ T 2 N ( J μ f ˉ ( t ) , J λ v t λ ( t ) ) Λ J λ v λ ( t ) d x . M_2(t):=-\int_{\mathbb{T}^2}\left(\mathcal{J}_{\lambda}\mathcal{N}\left(\mathcal{J}_{\mu}\bar{f}(t),\mathcal{J}_{\lambda}v_t^\lambda(t)\right)\right)\Lambda v^\lambda(t)\,dx=-\int_{\mathbb{T}^2}\mathcal{N}\left(\mathcal{J}_{\mu}\bar{f}(t),\mathcal{J}_{\lambda}v_t^\lambda(t)\right)\Lambda\mathcal{J}_{\lambda}v^\lambda(t)\,dx. M 2 ( t ) := − ∫ T 2 ( J λ N ( J μ f ˉ ( t ) , J λ v t λ ( t ) ) ) Λ v λ ( t ) d x = − ∫ T 2 N ( J μ f ˉ ( t ) , J λ v t λ ( t ) ) Λ J λ v λ ( t ) d x .
Expanding N ( f , g ) = [ Λ , f ] g + ∇ f ⋅ R g \mathbb{N}(f,g)=[\Lambda,f]g+\nabla f\cdot\mathcal{R}g N ( f , g ) = [ Λ , f ] g + ∇ f ⋅ R g , integrating by parts and Sobolev embedding we find that
∣ M 2 ( t ) ∣ ≤ C ∥ J v λ ∥ H 2 ∥ v t λ ∥ L 2 ∥ f ˉ ∥ L ∞ − 2 ∫ T 2 f ˉ Λ J v λ Λ J v t λ d x , ≤ C ∥ J v λ ∥ H 2 ∥ v t λ ∥ L 2 2 ∥ f ˉ ∥ L ∞ + C ∥ f ˉ Λ J v λ ∥ H 1 ∥ Λ J v t λ ∥ H − 1 , ≤ C ∥ J v λ ∥ H 2 2 ∥ v t λ ∥ L 2 2 ∥ f ˉ ∥ H 2 . \begin{aligned}
|M_2(t)|\le C\|\mathcal{J}v^\lambda\|_{H^2}\|v_t^\lambda\|_{L^2}\|\bar{f}\|_{L^\infty}-2\int_{\mathbb{T}^2}\bar{f}\Lambda\mathcal{J}v^\lambda\Lambda\mathcal{J}v_t^\lambda\,dx, \\
&\le C\|\mathcal{J}v^\lambda\|_{H^2}\|v_t^\lambda\|_{L^2}^2\|\bar{f}\|_{L^\infty}+C\|\bar{f}\Lambda\mathcal{J}v^\lambda\|_{H^1}\|\Lambda\mathcal{J}v_t^\lambda\|_{H^{-1}}, \\
&\le C\|\mathcal{J}v^\lambda\|_{H^2}^2\|v_t^\lambda\|_{L^2}^2\|\bar{f}\|_{H^2}.
\end{aligned} ∣ M 2 ( t ) ∣ ≤ C ∥ J v λ ∥ H 2 ∥ v t λ ∥ L 2 ∥ f ˉ ∥ L ∞ − 2 ∫ T 2 f ˉ Λ J v λ Λ J v t λ d x , ≤ C ∥ J v λ ∥ H 2 ∥ v t λ ∥ L 2 2 ∥ f ˉ ∥ L ∞ + C ∥ f ˉ Λ J v λ ∥ H 1 ∥Λ J v t λ ∥ H − 1 , ≤ C ∥ J v λ ∥ H 2 2 ∥ v t λ ∥ L 2 2 ∥ f ˉ ∥ H 2 .
where in the second inequality we make use of the H − 1 − H 1 H^{-1}-H^1 H − 1 − H 1 duality. In particular, since ∥ f ˉ ∥ L t ∞ H x 2 ≤ ∥ f ˉ ∥ L t ∞ H x 3 ≤ 2 ∥ f 0 ∥ H 3 \|\bar{f}\|_{L^\infty_tH_x^2}\le\|\bar{f}\|_{L^\infty_tH_x^3}\le2\|f_0\|_{H^3} ∥ f ˉ ∥ L t ∞ H x 2 ≤ ∥ f ˉ ∥ L t ∞ H x 3 ≤ 2∥ f 0 ∥ H 3 ,
∣ M 2 ( t ) ∣ ≲ μ ∥ f 0 ∥ H 3 ∥ v t λ ( t ) ∥ L 2 ∥ J v λ ( t ) ∥ H 2 . |M_2(t)|\lesssim_\mu\|f_0\|_{H^3}\|v_t^\lambda(t)\|_{L^2}\|\mathcal{J}v^\lambda(t)\|_{H^2}. ∣ M 2 ( t ) ∣ ≲ μ ∥ f 0 ∥ H 3 ∥ v t λ ( t ) ∥ L 2 ∥ J v λ ( t ) ∥ H 2 .
Therefore, using Young’s inequality we find that
∣ R 1 λ ( t ) ∣ ≤ δ 4 ∥ J v λ ( t ) ∥ H 2 2 + C δ , μ , β ( 1 + ∥ f 0 ∥ H 3 4 ) ( ∥ v λ ( t ) ∥ H 1 4 + ∥ v t λ ( t ) ∥ L 2 4 + 1 ) . (125) |\mathcal{R}_1^\lambda(t)|\le\frac{\delta}{4}\|\mathcal{J}v^\lambda(t)\|_{H^2}^2+C_{\delta,\mu,\beta}\left(1+\|f_0\|_{H^3}^4\right)\left(\|v^\lambda(t)\|_{H^1}^4+\|v_t^\lambda(t)\|_{L^2}^4+1\right). \tag*{(125)} ∣ R 1 λ ( t ) ∣ ≤ 4 δ ∥ J v λ ( t ) ∥ H 2 2 + C δ , μ , β ( 1 + ∥ f 0 ∥ H 3 4 ) ( ∥ v λ ( t ) ∥ H 1 4 + ∥ v t λ ( t ) ∥ L 2 4 + 1 ) . (125)
Thus,
1 2 d d t ( ∥ v λ ∥ H 1 / 2 2 + γ ∥ v λ ∥ H 1 2 ) + δ 2 ∥ J v λ ∥ H 2 2 ≤ C δ , μ , β ( 1 + ∥ f 0 ∥ H 3 4 ) ( ∥ v λ ( t ) ∥ H 1 4 + ∥ v t λ ( t ) ∥ L 2 4 + 1 ) . (126) \frac{1}{2}\frac{d}{dt}\left(\|v^\lambda\|_{H^{1/2}}^2+\gamma\|v^\lambda\|_{H^1}^2\right)+\frac{\delta}{2}\|\mathcal{J}v^\lambda\|_{H^2}^2\le C_{\delta,\mu,\beta}\left(1+\|f_0\|_{H^3}^4\right)\left(\|v^\lambda(t)\|_{H^1}^4+\|v_t^\lambda(t)\|_{L^2}^4+1\right). \tag*{(126)} 2 1 d t d ( ∥ v λ ∥ H 1/2 2 + γ ∥ v λ ∥ H 1 2 ) + 2 δ ∥ J v λ ∥ H 2 2 ≤ C δ , μ , β ( 1 + ∥ f 0 ∥ H 3 4 ) ( ∥ v λ ( t ) ∥ H 1 4 + ∥ v t λ ( t ) ∥ L 2 4 + 1 ) . (126)
Next, we test (121) against Λ − 1 v t λ \Lambda^{-1}v_t^\lambda Λ − 1 v t λ . Using that I + Υ Λ I+\Upsilon\Lambda I + ΥΛ is self-adjoint and commutes with Λ − 1 \Lambda^{-1} Λ − 1 , and that J λ \mathcal{J}_\lambda J λ is self-adjoint and commutes with Λ \Lambda Λ , we obtain
1 2 ∥ v t λ ( t ) ∥ H ˙ − 1 / 2 2 + γ 2 ∥ v t λ ( t ) ∥ L 2 2 + δ 2 d d t ∥ J v λ ( t ) ∥ H ˙ 1 2 ≤ R 2 λ ( t ) , (127) \frac{1}{2}\|v_t^\lambda(t)\|_{\dot{H}^{-1/2}}^2+\frac{\gamma}{2}\|v_t^\lambda(t)\|_{L^2}^2+\frac{\delta}{2}\frac{d}{dt}\|\mathcal{J}v^\lambda(t)\|_{\dot{H}^1}^2\le\mathcal{R}_2^\lambda(t), \tag*{(127)} 2 1 ∥ v t λ ( t ) ∥ H ˙ − 1/2 2 + 2 γ ∥ v t λ ( t ) ∥ L 2 2 + 2 δ d t d ∥ J v λ ( t ) ∥ H ˙ 1 2 ≤ R 2 λ ( t ) , (127)
where R 2 λ \mathcal{R}_2^\lambda R 2 λ collects the quadratic terms and the coupling term N ( J f ˉ , J v λ ) \mathbb{N}(\mathcal{J}\bar{f},\mathcal{J}v^\lambda) N ( J f ˉ , J v λ ) tested against Λ − 1 v t λ \Lambda^{-1}v_t^\lambda Λ − 1 v t λ . Arguing exactly as in the estimate of R 1 λ \mathcal{R}_1^\lambda R 1 λ , we obtain the bound
R 2 λ ( t ) ≤ C ∥ v λ ( t ) ∥ H 1 4 + γ 4 ∥ v t λ ( t ) ∥ L 2 2 , (128) \mathcal{R}_2^\lambda(t)\le C\|v^\lambda(t)\|_{H^1}^4+\frac{\gamma}{4}\|v_t^\lambda(t)\|_{L^2}^2, \tag*{(128)} R 2 λ ( t ) ≤ C ∥ v λ ( t ) ∥ H 1 4 + 4 γ ∥ v t λ ( t ) ∥ L 2 2 , (128)
with C C C independent of λ \lambda λ . Absorbing the last term into the left-hand side of (127) yields
∥ v t λ ( t ) ∥ H ˙ − 1 / 2 2 + γ 2 ∥ v t λ ( t ) ∥ L 2 2 + δ d d t ∥ J v λ ( t ) ∥ H ˙ 1 2 ≤ C ∥ v λ ( t ) ∥ H 1 4 . (129) \|v_t^\lambda(t)\|_{\dot{H}^{-1/2}}^2+\frac{\gamma}{2}\|v_t^\lambda(t)\|_{L^2}^2+\delta\frac{d}{dt}\|\mathcal{J}v^\lambda(t)\|_{\dot{H}^1}^2\le C\|v^\lambda(t)\|_{H^1}^4. \tag*{(129)} ∥ v t λ ( t ) ∥ H ˙ − 1/2 2 + 2 γ ∥ v t λ ( t ) ∥ L 2 2 + δ d t d ∥ J v λ ( t ) ∥ H ˙ 1 2 ≤ C ∥ v λ ( t ) ∥ H 1 4 . (129)
Set
E 1 λ ( t ) : = ∥ v λ ( t ) ∥ H 1 / 2 2 + γ ∥ v λ ( t ) ∥ H 1 2 , Y ( t ) : = sup 0 ≤ s ≤ t ∥ v λ ( s ) ∥ H 1 2 . E_1^\lambda(t):=\|v^\lambda(t)\|_{H^{1/2}}^2+\gamma\|v^\lambda(t)\|_{H^1}^2,\qquad Y(t):=\sup_{0\le s\le t}\|v^\lambda(s)\|_{H^1}^2. E 1 λ ( t ) := ∥ v λ ( t ) ∥ H 1/2 2 + γ ∥ v λ ( t ) ∥ H 1 2 , Y ( t ) := 0 ≤ s ≤ t sup ∥ v λ ( s ) ∥ H 1 2 .
Since v λ v^\lambda v λ has zero mean, Poincaré implies E 1 λ ( t ) ∼ ∥ v λ ( t ) ∥ H 1 2 E_1^\lambda(t)\sim\|v^\lambda(t)\|_{H^1}^2 E 1 λ ( t ) ∼ ∥ v λ ( t ) ∥ H 1 2 . From (126) we have, for all t ∈ [ 0 , T ] t\in[0,T] t ∈ [ 0 , T ] ,
d d t E 1 λ ( t ) + δ 2 ∥ J v λ ( t ) ∥ H ˙ 2 2 ≤ C 0 ( ∥ v λ ( t ) ∥ H 1 4 + ∥ v t λ ( t ) ∥ L 2 4 + 1 ) , (130) \frac{d}{dt}E_1^\lambda(t)+\frac{\delta}{2}\|\mathcal{J}v^\lambda(t)\|_{\dot{H}^2}^2\le C_0\left(\|v^\lambda(t)\|_{H^1}^4+\|v_t^\lambda(t)\|_{L^2}^4+1\right), \tag*{(130)} d t d E 1 λ ( t ) + 2 δ ∥ J v λ ( t ) ∥ H ˙ 2 2 ≤ C 0 ( ∥ v λ ( t ) ∥ H 1 4 + ∥ v t λ ( t ) ∥ L 2 4 + 1 ) , (130)
where C 0 = C 0 ( Υ , δ , μ , β ) ( 1 + ∥ f 0 ∥ H 3 4 ) C_0=C_0(\Upsilon,\delta,\mu,\beta)(1+\|f_0\|_{H^3}^4) C 0 = C 0 ( Υ , δ , μ , β ) ( 1 + ∥ f 0 ∥ H 3 4 ) and is independent of λ \lambda λ . Integrating (130) on [ 0 , t ] [0,t] [ 0 , t ] gives
sup 0 ≤ s ≤ t E 1 λ ( s ) + δ 2 ∫ 0 t ∥ J v λ ( s ) ∥ H ˙ 2 2 d s ≤ E 1 λ ( 0 ) + C 0 ∫ 0 t ( ∥ v λ ( s ) ∥ H 1 4 + ∥ v t λ ( s ) ∥ L 2 4 + 1 ) d s . (131) \sup_{0\le s\le t}E_1^\lambda(s)+\frac{\delta}{2}\int_0^t\|\mathcal{J}v^\lambda(s)\|_{\dot{H}^2}^2\,ds\le E_1^\lambda(0)+C_0\int_0^t\left(\|v^\lambda(s)\|_{H^1}^4+\|v_t^\lambda(s)\|_{L^2}^4+1\right)\,ds. \tag*{(131)} 0 ≤ s ≤ t sup E 1 λ ( s ) + 2 δ ∫ 0 t ∥ J v λ ( s ) ∥ H ˙ 2 2 d s ≤ E 1 λ ( 0 ) + C 0 ∫ 0 t ( ∥ v λ ( s ) ∥ H 1 4 + ∥ v t λ ( s ) ∥ L 2 4 + 1 ) d s . (131)
From (129) we have, for all t ∈ [ 0 , T ] t\in[0,T] t ∈ [ 0 , T ] ,
∥ v t λ ( t ) ∥ H ˙ − 1 / 2 2 + γ 2 ∥ v t λ ( t ) ∥ L 2 2 + δ d d t ∥ J v λ ( t ) ∥ H ˙ 1 2 ≤ C ∥ v λ ( t ) ∥ H 1 4 . (132) \|v_t^\lambda(t)\|_{\dot{H}^{-1/2}}^2+\frac{\gamma}{2}\|v_t^\lambda(t)\|_{L^2}^2+\delta\frac{d}{dt}\|\mathcal{J}v^\lambda(t)\|_{\dot{H}^1}^2\le C\|v^\lambda(t)\|_{H^1}^4. \tag*{(132)} ∥ v t λ ( t ) ∥ H ˙ − 1/2 2 + 2 γ ∥ v t λ ( t ) ∥ L 2 2 + δ d t d ∥ J v λ ( t ) ∥ H ˙ 1 2 ≤ C ∥ v λ ( t ) ∥ H 1 4 . (132)
with C C C independent of λ \lambda λ . Integrating (132) on [ 0 , t ] [0,t] [ 0 , t ] yields
∫ 0 t ∥ v t λ ( s ) ∥ H − 1 / 2 2 d s + γ 2 ∫ 0 t ∥ v λ ( s ) ∥ L 2 2 d s + δ ∥ J λ v λ ( t ) ∥ H 1 2 ≤ δ ∥ J λ v 0 ∥ H 1 2 + C ∫ 0 t ∥ v λ ( s ) ∥ H 1 2 d s . (133) \int_{0}^{t}\|v_{t}^{\lambda}(s)\|_{H^{-1/2}}^{2}\,\mathrm{d}s+\frac{\gamma}{2}\int_{0}^{t}\|v^{\lambda}(s)\|_{L^{2}}^{2}\,\mathrm{d}s+\delta\|\mathcal{J}_{\lambda}v^{\lambda}(t)\|_{H^{1}}^{2}\leq\delta\|\mathcal{J}_{\lambda}v_{0}\|_{H^{1}}^{2}+C\int_{0}^{t}\|v^{\lambda}(s)\|_{H^{1}}^{2}\,\mathrm{d}s. \tag*{(133)} ∫ 0 t ∥ v t λ ( s ) ∥ H − 1/2 2 d s + 2 γ ∫ 0 t ∥ v λ ( s ) ∥ L 2 2 d s + δ ∥ J λ v λ ( t ) ∥ H 1 2 ≤ δ ∥ J λ v 0 ∥ H 1 2 + C ∫ 0 t ∥ v λ ( s ) ∥ H 1 2 d s . (133)
In particular, using ∥ J λ v 0 ∥ H 1 ≤ ∥ v 0 ∥ H 1 \|\mathcal{J}_{\lambda}v_{0}\|_{H^{1}}\leq\|v_{0}\|_{H^{1}} ∥ J λ v 0 ∥ H 1 ≤ ∥ v 0 ∥ H 1 and ∥ v λ ( s ) ∥ H 1 4 ≤ Y ( t ) 2 \|v^{\lambda}(s)\|_{H^{1}}^{4}\leq Y(t)^{2} ∥ v λ ( s ) ∥ H 1 4 ≤ Y ( t ) 2 for s ≤ t s\leq t s ≤ t ,
∫ 0 t ∥ v t λ ( s ) ∥ L 2 2 d s ≲ ∥ v 0 ∥ H 1 2 + t Y ( t ) 2 , ∫ 0 t ∥ v t λ ( s ) ∥ H − 1 / 2 2 d s ≲ ∥ v 0 ∥ H 1 2 + t Y ( t ) 2 , (134) \int_{0}^{t}\|v_{t}^{\lambda}(s)\|_{L^{2}}^{2}\,\mathrm{d}s\lesssim\|v_{0}\|_{H^{1}}^{2}+tY(t)^{2},\qquad\int_{0}^{t}\|v_{t}^{\lambda}(s)\|_{H^{-1/2}}^{2}\,\mathrm{d}s\lesssim\|v_{0}\|_{H^{1}}^{2}+tY(t)^{2}, \tag*{(134)} ∫ 0 t ∥ v t λ ( s ) ∥ L 2 2 d s ≲ ∥ v 0 ∥ H 1 2 + t Y ( t ) 2 , ∫ 0 t ∥ v t λ ( s ) ∥ H − 1/2 2 d s ≲ ∥ v 0 ∥ H 1 2 + t Y ( t ) 2 , (134)
uniformly in λ \lambda λ . Insert (134) and ∫ 0 t ∥ v λ ( s ) ∥ H 1 4 d s ≤ t Y ( t ) 2 \int_{0}^{t}\|v^{\lambda}(s)\|_{H^{1}}^{4}\,\mathrm{d}s\leq tY(t)^{2} ∫ 0 t ∥ v λ ( s ) ∥ H 1 4 d s ≤ t Y ( t ) 2 into (131) . Using E 1 λ ( 0 ) ∼ ∥ v 0 ∥ H 1 2 E_{1}^{\lambda}(0)\sim\|v_{0}\|_{H^{1}}^{2} E 1 λ ( 0 ) ∼ ∥ v 0 ∥ H 1 2 and E 1 λ ( s ) ∼ ∥ v λ ( s ) ∥ H 1 2 E_{1}^{\lambda}(s)\sim\|v^{\lambda}(s)\|_{H^{1}}^{2} E 1 λ ( s ) ∼ ∥ v λ ( s ) ∥ H 1 2 , we obtain
Y ( t ) + ∫ 0 t ∥ J λ v λ ( s ) ∥ H 2 2 d s ≤ C 1 ( 1 + ∥ v 0 ∥ H 1 2 ) + C 2 t Y ( t ) 2 , t ∈ [ 0 , T ] , (135) Y(t)+\int_{0}^{t}\|\mathcal{J}_{\lambda}v^{\lambda}(s)\|_{H^{2}}^{2}\,\mathrm{d}s\leq C_{1}(1+\|v_{0}\|_{H^{1}}^{2})+C_{2}tY(t)^{2},\qquad t\in[0,T], \tag*{(135)} Y ( t ) + ∫ 0 t ∥ J λ v λ ( s ) ∥ H 2 2 d s ≤ C 1 ( 1 + ∥ v 0 ∥ H 1 2 ) + C 2 t Y ( t ) 2 , t ∈ [ 0 , T ] , (135)
where C 1 , C 2 C_{1},C_{2} C 1 , C 2 are independent of λ \lambda λ (and depend on γ , δ , μ , β \gamma,\delta,\mu,\beta γ , δ , μ , β and ∥ f 0 ∥ H 3 \|f_{0}\|_{H^{3}} ∥ f 0 ∥ H 3 ). In particular we have the polynomial inequality
Y ( t ) ≤ C 1 ( 1 + ∥ v 0 ∥ H 1 2 ) + C 2 t Y ( t ) 2 , t ∈ [ 0 , T ] . (136) Y(t)\leq C_{1}(1+\|v_{0}\|_{H^{1}}^{2})+C_{2}tY(t)^{2},\qquad t\in[0,T]. \tag*{(136)} Y ( t ) ≤ C 1 ( 1 + ∥ v 0 ∥ H 1 2 ) + C 2 t Y ( t ) 2 , t ∈ [ 0 , T ] . (136)
Choose T > 0 T>0 T > 0 such that
C 2 T ( 2 C 1 ( 1 + ∥ v 0 ∥ H 1 2 ) ) ≤ 1 2 . C_{2}T\left(2C_{1}(1+\|v_{0}\|_{H^{1}}^{2})\right)\leq\frac{1}{2}. C 2 T ( 2 C 1 ( 1 + ∥ v 0 ∥ H 1 2 ) ) ≤ 2 1 .
A standard continuity/bootstrapping argument applied to (136) yields
sup 0 ≤ t ≤ T ∥ v λ ( t ) ∥ H 1 2 = sup 0 ≤ t ≤ T Y ( t ) ≤ 2 C 1 ( 1 + ∥ v 0 ∥ H 1 2 ) , (137) \sup_{0\leq t\leq T}\|v^{\lambda}(t)\|_{H^{1}}^{2}=\sup_{0\leq t\leq T}Y(t)\leq2C_{1}(1+\|v_{0}\|_{H^{1}}^{2}), \tag*{(137)} 0 ≤ t ≤ T sup ∥ v λ ( t ) ∥ H 1 2 = 0 ≤ t ≤ T sup Y ( t ) ≤ 2 C 1 ( 1 + ∥ v 0 ∥ H 1 2 ) , (137)
with a bound independent of λ \lambda λ . Plugging (137) back into (135) and (133) gives the remaining uniform bounds:
sup 0 ≤ t ≤ T ∥ v λ ( t ) ∥ H 1 ≤ C , ∫ 0 T ∥ J λ v λ ( t ) ∥ H 2 2 d t ≤ C , ∫ 0 T ∥ v t λ ( t ) ∥ H − 1 / 2 2 d t ≤ C , (138) \sup_{0\leq t\leq T}\|v^{\lambda}(t)\|_{H^{1}}\leq C,\qquad\int_{0}^{T}\|\mathcal{J}_{\lambda}v^{\lambda}(t)\|_{H^{2}}^{2}\,\mathrm{d}t\leq C,\qquad\int_{0}^{T}\|v_{t}^{\lambda}(t)\|_{H^{-1/2}}^{2}\,\mathrm{d}t\leq C, \tag*{(138)} 0 ≤ t ≤ T sup ∥ v λ ( t ) ∥ H 1 ≤ C , ∫ 0 T ∥ J λ v λ ( t ) ∥ H 2 2 d t ≤ C , ∫ 0 T ∥ v t λ ( t ) ∥ H − 1/2 2 d t ≤ C , (138)
for a constant C C C independent of λ \lambda λ .
By (138) we have the λ \lambda λ –uniform bounds
v λ bounded in L ∞ ( 0 , T ; H 1 ( T 2 ) ) , ∂ t v λ bounded in L 2 ( 0 , T ; H − 1 / 2 ( T 2 ) ) . v^{\lambda}\text{ bounded in }L^{\infty}(0,T;H^{1}(\mathbb{T}^{2})),\qquad\partial_{t}v^{\lambda}\text{ bounded in }L^{2}(0,T;H^{-1/2}(\mathbb{T}^{2})). v λ bounded in L ∞ ( 0 , T ; H 1 ( T 2 )) , ∂ t v λ bounded in L 2 ( 0 , T ; H − 1/2 ( T 2 )) .
Fix θ ∈ ( 0 , 1 2 ) \theta\in(0,\frac{1}{2}) θ ∈ ( 0 , 2 1 ) . Since the embedding H 1 ( T 2 ) ↪ H 1 − θ ( T 2 ) H^{1}(\mathbb{T}^{2})\hookrightarrow H^{1-\theta}(\mathbb{T}^{2}) H 1 ( T 2 ) ↪ H 1 − θ ( T 2 ) is compact and H 1 − θ ( T 2 ) ↪ H − 1 / 2 ( T 2 ) H^{1-\theta}(\mathbb{T}^{2})\hookrightarrow H^{-1/2}(\mathbb{T}^{2}) H 1 − θ ( T 2 ) ↪ H − 1/2 ( T 2 ) is continuous, the Aubin–Lions lemma yields, after extracting a subsequence (not relabeled), the existence of v v v such that
v λ → v strongly in L 2 ( 0 , T ; H 1 − θ ( T 2 ) ) , v λ ⇀ ∗ v in L ∞ ( 0 , T ; H 1 ( T 2 ) ) . (139) v^{\lambda}\to v\quad\text{strongly in }L^{2}(0,T;H^{1-\theta}(\mathbb{T}^{2})),\qquad v^{\lambda}\rightharpoonup^{*}v\quad\text{in }L^{\infty}(0,T;H^{1}(\mathbb{T}^{2})). \tag*{(139)} v λ → v strongly in L 2 ( 0 , T ; H 1 − θ ( T 2 )) , v λ ⇀ ∗ v in L ∞ ( 0 , T ; H 1 ( T 2 )) . (139)
Moreover, up to a further subsequence, there exists U U U such that
∂ t v λ ⇀ U weakly in L 2 ( 0 , T ; H − 1 / 2 ( T 2 ) ) . (140) \partial_{t}v^{\lambda}\rightharpoonup U\quad\text{weakly in }L^{2}(0,T;H^{-1/2}(\mathbb{T}^{2})). \tag*{(140)} ∂ t v λ ⇀ U weakly in L 2 ( 0 , T ; H − 1/2 ( T 2 )) . (140)
In particular, v ∈ L ∞ ( 0 , T ; H 1 ( T 2 ) ) ∩ W 1 , 2 ( 0 , T ; H − 1 / 2 ( T 2 ) ) v\in L^{\infty}(0,T;H^{1}(\mathbb{T}^{2}))\cap W^{1,2}(0,T;H^{-1/2}(\mathbb{T}^{2})) v ∈ L ∞ ( 0 , T ; H 1 ( T 2 )) ∩ W 1 , 2 ( 0 , T ; H − 1/2 ( T 2 )) , and hence
v ∈ C ( [ 0 , T ] ; H s ( T 2 ) ) for every s ∈ [ − 1 2 , 1 ) . (141) v\in C([0,T];H^{s}(\mathbb{T}^{2}))\qquad\text{for every }s\in\left[-\frac{1}{2},1\right). \tag*{(141)} v ∈ C ([ 0 , T ] ; H s ( T 2 )) for every s ∈ [ − 2 1 , 1 ) . (141)
Since v λ ∈ L ∞ ( 0 , T ; H 1 ) v^\lambda\in L^\infty(0,T;H^1) v λ ∈ L ∞ ( 0 , T ; H 1 ) and ∂ t v λ \partial_t v^\lambda ∂ t v λ is bounded in L 2 ( 0 , T ; H − 1 / 2 ) L^2(0,T;H^{-1/2}) L 2 ( 0 , T ; H − 1/2 ) , after extraction we have ∂ t v λ ⇀ U \partial_t v^\lambda\rightharpoonup U ∂ t v λ ⇀ U weakly in L 2 ( 0 , T ; H − 1 / 2 ) L^2(0,T;H^{-1/2}) L 2 ( 0 , T ; H − 1/2 ) for some U U U . For any φ ∈ C c ∞ ( ( 0 , T ) × T 2 ) \varphi\in C^\infty_c((0,T)\times\mathbb{T}^2) φ ∈ C c ∞ (( 0 , T ) × T 2 ) ,
∫ 0 T ⟨ U λ , φ ⟩ d t = − ∫ 0 T ⟨ v λ , ∂ t φ ⟩ d t . \int_0^T \langle U^\lambda,\varphi\rangle\,dt=-\int_0^T \langle v^\lambda,\partial_t\varphi\rangle\,dt. ∫ 0 T ⟨ U λ , φ ⟩ d t = − ∫ 0 T ⟨ v λ , ∂ t φ ⟩ d t .
We now pass to the limit λ → 0 \lambda\to0 λ → 0 in the regularized equation (121) . The linear terms converge by (4.42) and the boundedness of Fourier multipliers on Sobolev spaces. For the nonlinear terms we use the regularization and the uniform bounds (4.25)`. Consequently, the limit v μ [ f ˉ ] ∈ L ∞ ( 0 , T ; H 0 1 ( T 2 ) ) v^\mu[\bar f] \in L^\infty(0,T;H^1_0(\mathbb{T}^2)) v μ [ f ˉ ] ∈ L ∞ ( 0 , T ; H 0 1 ( T 2 )) satisfies
( I + γ Λ ) ∂ t v + δ Λ 3 v + J μ ( Λ + β 4 Λ 5 ) J μ f ˉ + N ( J μ f ˉ , ∂ t v ) = ε Q ( v ) , (142) (I+\gamma\Lambda)\partial_t v+\delta\Lambda^3v+\mathcal{J}_\mu\left(\Lambda+\frac{\beta}{4}\Lambda^5\right)\mathcal{J}_\mu\bar f+\mathcal{N}(\mathcal{J}_\mu\bar f,\partial_t v)=\varepsilon Q(v), \tag*{(142)} ( I + γ Λ ) ∂ t v + δ Λ 3 v + J μ ( Λ + 4 β Λ 5 ) J μ f ˉ + N ( J μ f ˉ , ∂ t v ) = εQ ( v ) , (142)
Step 4: Solving the equation for f f f . Fix 0 < μ ≤ 1 0<\mu\leq1 0 < μ ≤ 1 and let T > 0 T>0 T > 0 be the time obtained in Step 3. In particular, for every f ˉ ∈ X T \bar f\in X_T f ˉ ∈ X T we dispose of a function
v μ [ f ˉ ] ∈ L ∞ ( 0 , T ; H 0 1 ( T 2 ) ) ∩ W 1 , 2 ( 0 , T ; H − 1 / 2 ( T 2 ) ) , v μ = v 0 , v^\mu[\bar f]\in L^\infty(0,T;H^1_0(\mathbb{T}^2))\cap W^{1,2}(0,T;H^{-1/2}(\mathbb{T}^2)),\qquad v^\mu=v_0, v μ [ f ˉ ] ∈ L ∞ ( 0 , T ; H 0 1 ( T 2 )) ∩ W 1 , 2 ( 0 , T ; H − 1/2 ( T 2 )) , v μ = v 0 ,
which satisfies
( I + γ Λ ) ∂ t v μ [ f ˉ ] + δ Λ 3 v μ [ f ˉ ] + J μ ( Λ + β 4 Λ 5 ) J μ f ˉ + N ( J μ f ˉ , ∂ t v μ [ f ˉ ] ) = ε Q ( v μ [ f ˉ ] ) , (143) (I+\gamma\Lambda)\partial_t v^\mu[\bar f]+\delta\Lambda^3v^\mu[\bar f]+\mathcal{J}_\mu\left(\Lambda+\frac{\beta}{4}\Lambda^5\right)\mathcal{J}_\mu\bar f+\mathcal{N}(\mathcal{J}_\mu\bar f,\partial_t v^\mu[\bar f])=\varepsilon Q(v^\mu[\bar f]), \tag*{(143)} ( I + γ Λ ) ∂ t v μ [ f ˉ ] + δ Λ 3 v μ [ f ˉ ] + J μ ( Λ + 4 β Λ 5 ) J μ f ˉ + N ( J μ f ˉ , ∂ t v μ [ f ˉ ]) = εQ ( v μ [ f ˉ ]) , (143)
where
Q ( w ) : = 1 2 Λ ( w 2 − ∣ R w ∣ 2 ) − [ Λ , w ] w − ∇ ⊥ w ⋅ R w . Q(w):=\frac{1}{2}\Lambda\left(w^2-\lvert\mathcal{R}w\rvert^2\right)-[\Lambda,w]w-\nabla^\perp w\cdot\mathcal{R}w. Q ( w ) := 2 1 Λ ( w 2 − ∣ R w ∣ 2 ) − [ Λ , w ] w − ∇ ⊥ w ⋅ R w .
We recall that
X T : = { f ˉ ∈ L ∞ ( 0 , T ; H 3 ( T 2 ) ) : f ˉ ( 0 ) = f 0 , ∥ f ˉ ∥ L ∞ ( 0 , T ; H 3 ) ≤ 2 ∥ f 0 ∥ H 3 } , X_T:=\left\{\bar f\in L^\infty(0,T;H^3(\mathbb{T}^2)):\ \bar f(0)=f_0,\ \|\bar f\|_{L^\infty(0,T;H^3)}\leq2\|f_0\|_{H^3}\right\}, X T := { f ˉ ∈ L ∞ ( 0 , T ; H 3 ( T 2 )) : f ˉ ( 0 ) = f 0 , ∥ f ˉ ∥ L ∞ ( 0 , T ; H 3 ) ≤ 2∥ f 0 ∥ H 3 } ,
which records the desired H 3 H^3 H 3 control.
For f ˉ ∈ X T \bar f\in X_T f ˉ ∈ X T , define
( T μ [ f ˉ ] ) ( t ) : = f 0 + ∫ 0 t J μ J μ v μ [ f ˉ ] ( s ) d s , t ∈ [ 0 , T ] . (144) (\mathcal{T}_\mu[\bar f])(t):=f_0+\int_0^t\mathcal{J}_\mu\mathcal{J}_\mu v^\mu[\bar f](s)\,ds,\qquad t\in[0,T]. \tag*{(144)} ( T μ [ f ˉ ]) ( t ) := f 0 + ∫ 0 t J μ J μ v μ [ f ˉ ] ( s ) d s , t ∈ [ 0 , T ] . (144)
Since v μ [ f ˉ ] ∈ L ∞ ( 0 , T ; H 1 ) v^\mu[\bar f]\in L^\infty(0,T;H^1) v μ [ f ˉ ] ∈ L ∞ ( 0 , T ; H 1 ) , the map T μ [ f ˉ ] ∈ X T \mathcal{T}_\mu[\bar f]\in X_T T μ [ f ˉ ] ∈ X T and
∂ t ( T μ [ f ˉ ] ) = J μ J μ v μ [ f ˉ ] in L ∞ ( 0 , T ; H 1 ( T 2 ) ) . (145) \partial_t(\mathcal{T}_\mu[\bar f])=\mathcal{J}_\mu\mathcal{J}_\mu v^\mu[\bar f]\quad\text{in }L^\infty(0,T;H^1(\mathbb{T}^2)). \tag*{(145)} ∂ t ( T μ [ f ˉ ]) = J μ J μ v μ [ f ˉ ] in L ∞ ( 0 , T ; H 1 ( T 2 )) . (145)
Thus, if f = T μ [ f ] f=\mathcal{T}_\mu[f] f = T μ [ f ] , then f t = J μ J μ v μ [ f ] f_t=\mathcal{J}_\mu\mathcal{J}_\mu v^\mu[f] f t = J μ J μ v μ [ f ] and inserting f ˉ = f \bar f=f f ˉ = f in (4.46) yields the coupled μ \mu μ -regularized system
( I + γ Λ ) v t + δ Λ 3 v + J μ ( Λ + β 4 Λ 5 ) J μ f + N ( J μ f , v t ) = ε Q ( v ) , (146) (I+\gamma\Lambda)v_t+\delta\Lambda^3v+\mathcal{J}_\mu\left(\Lambda+\frac{\beta}{4}\Lambda^5\right)\mathcal{J}_\mu f+\mathcal{N}(\mathcal{J}_\mu f,v_t)=\varepsilon Q(v), \tag*{(146)} ( I + γ Λ ) v t + δ Λ 3 v + J μ ( Λ + 4 β Λ 5 ) J μ f + N ( J μ f , v t ) = εQ ( v ) , (146)
f t = J μ J μ v . (147) f_t=\mathcal{J}_\mu\mathcal{J}_\mu v. \tag*{(147)} f t = J μ J μ v . (147)
Let f ˉ ( 1 ) , f ˉ ( 2 ) ∈ X T \bar f^{(1)},\bar f^{(2)}\in X_T f ˉ ( 1 ) , f ˉ ( 2 ) ∈ X T , and set v ( i ) : = v μ [ f ˉ ( i ) ] v^{(i)}:=v^\mu[\bar f^{(i)}] v ( i ) := v μ [ f ˉ ( i ) ] and v ~ : = v ( 1 ) − v ( 2 ) \widetilde v:=v^{(1)}-v^{(2)} v := v ( 1 ) − v ( 2 ) . Subtracting (143) for f ˉ ( 1 ) \bar f^{(1)} f ˉ ( 1 ) and f ˉ ( 2 ) \bar f^{(2)} f ˉ ( 2 ) gives
( I + γ Λ ) ∂ t v ~ + δ Λ 3 v ~ = − J μ ( Λ + β 4 Λ 5 ) J μ ( f ˉ ( 1 ) − f ˉ ( 2 ) ) − ( N ( J μ f ˉ ( 1 ) , ∂ t v ( 1 ) ) − N ( J μ f ˉ ( 2 ) , ∂ t v ( 2 ) ) ) + ε ( Q ( v ( 1 ) ) − Q ( v ( 2 ) ) ) . (148) \begin{aligned}
(I+\gamma\Lambda)\partial_t\widetilde v+\delta\Lambda^3\widetilde v
={}&-\mathcal{J}_\mu\left(\Lambda+\frac{\beta}{4}\Lambda^5\right)\mathcal{J}_\mu(\bar f^{(1)}-\bar f^{(2)})\\
&-\left(\mathcal{N}(\mathcal{J}_\mu\bar f^{(1)},\partial_t v^{(1)})-\mathcal{N}(\mathcal{J}_\mu\bar f^{(2)},\partial_t v^{(2)})\right)\\
&+\varepsilon\left(Q(v^{(1)})-Q(v^{(2)})\right). \tag*{(148)}
\end{aligned} ( I + γ Λ ) ∂ t v + δ Λ 3 v = − J μ ( Λ + 4 β Λ 5 ) J μ ( f ˉ ( 1 ) − f ˉ ( 2 ) ) − ( N ( J μ f ˉ ( 1 ) , ∂ t v ( 1 ) ) − N ( J μ f ˉ ( 2 ) , ∂ t v ( 2 ) ) ) + ε ( Q ( v ( 1 ) ) − Q ( v ( 2 ) ) ) . (148)
Using bilinearity of N \mathcal{N} N in its second argument we split
N ( J μ f ˉ ( 1 ) , ∂ t v ( 1 ) ) − N ( J μ f ˉ ( 2 ) , ∂ t v ( 2 ) ) = N ( J μ ( f ˉ ( 1 ) − f ˉ ( 2 ) ) , ∂ t v ( 1 ) ) + N ( J μ f ˉ ( 2 ) , ∂ t v ~ ) . \mathcal{N}(\mathcal{J}_{\mu}\bar{f}^{(1)},\partial_t v^{(1)})-\mathcal{N}(\mathcal{J}_{\mu}\bar{f}^{(2)},\partial_t v^{(2)})=\mathcal{N}(\mathcal{J}_{\mu}(\bar{f}^{(1)}-\bar{f}^{(2)}),\partial_t v^{(1)})+\mathcal{N}(\mathcal{J}_{\mu}\bar{f}^{(2)},\partial_t\tilde{v}). N ( J μ f ˉ ( 1 ) , ∂ t v ( 1 ) ) − N ( J μ f ˉ ( 2 ) , ∂ t v ( 2 ) ) = N ( J μ ( f ˉ ( 1 ) − f ˉ ( 2 ) ) , ∂ t v ( 1 ) ) + N ( J μ f ˉ ( 2 ) , ∂ t v ~ ) .
Testing (148) against Λ v ~ \Lambda\tilde{v} Λ v ~ and arguing as in Step 3 (using that J μ f ˉ ( 2 ) \mathcal{J}_{\mu}\bar{f}^{(2)} J μ f ˉ ( 2 ) is smooth and bounded in W 1 , ∞ W^{1,\infty} W 1 , ∞ with constants depending on μ \mu μ and ∥ f ˉ ( 2 ) ∥ H 3 \|\bar{f}^{(2)}\|_{H^3} ∥ f ˉ ( 2 ) ∥ H 3 ) yields, for a.e. t ∈ ( 0 , T ) t\in(0,T) t ∈ ( 0 , T ) ,
d d t ( ∥ v ~ ∥ H 1 / 2 2 + γ ∥ v ~ ∥ H 1 2 ) + δ ∥ v ~ ∥ H 2 2 ≤ C μ , R ∥ J μ ( f ˉ ( 1 ) − f ˉ ( 2 ) ) ∥ H 3 2 + C μ , R ∥ v ~ ∥ H 1 2 , (149) \frac{d}{dt}\left(\|\tilde{v}\|_{H^{1/2}}^2+\gamma\|\tilde{v}\|_{H^1}^2\right)+\delta\|\tilde{v}\|_{H^2}^2\leq C_{\mu,R}\|\mathcal{J}_{\mu}(\bar{f}^{(1)}-\bar{f}^{(2)})\|_{H^3}^2+C_{\mu,R}\|\tilde{v}\|_{H^1}^2, \tag*{(149)} d t d ( ∥ v ~ ∥ H 1/2 2 + γ ∥ v ~ ∥ H 1 2 ) + δ ∥ v ~ ∥ H 2 2 ≤ C μ , R ∥ J μ ( f ˉ ( 1 ) − f ˉ ( 2 ) ) ∥ H 3 2 + C μ , R ∥ v ~ ∥ H 1 2 , (149)
where R : = sup f ˉ ∈ X T ∥ f ˉ ∥ L ∞ H 3 ≤ 2 ∥ f 0 ∥ H 3 R:=\sup_{\bar{f}\in X_T}\|\bar{f}\|_{L^\infty H^3}\leq2\|f_0\|_{H^3} R := sup f ˉ ∈ X T ∥ f ˉ ∥ L ∞ H 3 ≤ 2∥ f 0 ∥ H 3 and C μ , R C_{\mu,R} C μ , R depends on μ \mu μ , the model parameters and R R R , but not on T T T . Since v ~ ( 0 ) = 0 \tilde{v}(0)=0 v ~ ( 0 ) = 0 , Grönwall implies
∥ v μ [ f ˉ ( 1 ) ] − v μ [ f ˉ ( 2 ) ] ∥ L ∞ ( 0 , T ; H 1 ) ≤ C μ , R T ∥ f ˉ ( 1 ) − f ˉ ( 2 ) ∥ L ∞ ( 0 , T ; H 1 ) . (150) \|v^\mu[\bar{f}^{(1)}]-v^\mu[\bar{f}^{(2)}]\|_{L^\infty(0,T;H^1)}\leq C_{\mu,R}T\|\bar{f}^{(1)}-\bar{f}^{(2)}\|_{L^\infty(0,T;H^1)}. \tag*{(150)} ∥ v μ [ f ˉ ( 1 ) ] − v μ [ f ˉ ( 2 ) ] ∥ L ∞ ( 0 , T ; H 1 ) ≤ C μ , R T ∥ f ˉ ( 1 ) − f ˉ ( 2 ) ∥ L ∞ ( 0 , T ; H 1 ) . (150)
To show the contraction of T μ \mathcal{T}_{\mu} T μ in X T X_T X T , let f ˉ ( 1 ) , f ˉ ( 2 ) ∈ X T \bar{f}^{(1)},\bar{f}^{(2)}\in X_T f ˉ ( 1 ) , f ˉ ( 2 ) ∈ X T . Using (144) and (150) ,
∥ T μ [ f ˉ ( 1 ) ] − T μ [ f ˉ ( 2 ) ] ∥ X T ≤ C μ T ∥ v μ [ f ˉ ( 1 ) ] − v μ [ f ˉ ( 2 ) ] ∥ L ∞ H 1 ≤ T 2 C μ , R ∥ f ˉ ( 1 ) − f ˉ ( 2 ) ∥ L ∞ H 1 . \|\mathcal{T}_{\mu}[\bar{f}^{(1)}]-\mathcal{T}_{\mu}[\bar{f}^{(2)}]\|_{X_T}\leq C_{\mu}T\|v^\mu[\bar{f}^{(1)}]-v^\mu[\bar{f}^{(2)}]\|_{L^\infty H^1}\leq T^2C_{\mu,R}\|\bar{f}^{(1)}-\bar{f}^{(2)}\|_{L^\infty H^1}. ∥ T μ [ f ˉ ( 1 ) ] − T μ [ f ˉ ( 2 ) ] ∥ X T ≤ C μ T ∥ v μ [ f ˉ ( 1 ) ] − v μ [ f ˉ ( 2 ) ] ∥ L ∞ H 1 ≤ T 2 C μ , R ∥ f ˉ ( 1 ) − f ˉ ( 2 ) ∥ L ∞ H 1 .
Since ∥ h ∥ H 1 ≤ ∥ h ∥ H 3 \|h\|_{H^1}\leq\|h\|_{H^3} ∥ h ∥ H 1 ≤ ∥ h ∥ H 3 and we can take 0 < T μ < 1 0<T_{\mu}<1 0 < T μ < 1 , we also have
∥ T μ [ f ˉ ( 1 ) ] − T μ [ f ˉ ( 2 ) ] ∥ X T ≤ T C μ , R ∥ f ˉ ( 1 ) − f ˉ ( 2 ) ∥ X T . \|\mathcal{T}_{\mu}[\bar{f}^{(1)}]-\mathcal{T}_{\mu}[\bar{f}^{(2)}]\|_{X_T}\leq TC_{\mu,R}\|\bar{f}^{(1)}-\bar{f}^{(2)}\|_{X_T}. ∥ T μ [ f ˉ ( 1 ) ] − T μ [ f ˉ ( 2 ) ] ∥ X T ≤ T C μ , R ∥ f ˉ ( 1 ) − f ˉ ( 2 ) ∥ X T .
Choose T μ ∈ ( 0 , T ] T_{\mu}\in(0,T] T μ ∈ ( 0 , T ] such that T μ C μ , R ≤ 1 2 T_{\mu}C_{\mu,R}\leq\frac{1}{2} T μ C μ , R ≤ 2 1 . Then T μ \mathcal{T}_{\mu} T μ is a strict contraction on X T μ X_{T_{\mu}} X T μ , and Banach’s fixed point theorem yields a unique
f μ ∈ X T μ such that f μ = T μ [ f μ ] . f^\mu\in X_{T_{\mu}}\quad\text{such that}\quad f^\mu=\mathcal{T}_{\mu}[f^\mu]. f μ ∈ X T μ such that f μ = T μ [ f μ ] .
Setting v μ : = v μ [ f μ ] v^\mu:=v^\mu[f^\mu] v μ := v μ [ f μ ] , we obtain a pair ( f μ , v μ ) (f^\mu,v^\mu) ( f μ , v μ ) satisfying (146) –(147) on ( 0 , T μ ) (0,T_{\mu}) ( 0 , T μ ) , with f μ ( 0 ) = f 0 f^\mu(0)=f_0 f μ ( 0 ) = f 0 and v μ ( 0 ) = v 0 v^\mu(0)=v_0 v μ ( 0 ) = v 0 .
At this stage we know f μ ∈ L ∞ ( 0 , T μ ; H 3 ) f^\mu\in L^\infty(0,T_{\mu};H^3) f μ ∈ L ∞ ( 0 , T μ ; H 3 ) and f t μ = v μ ∈ L ∞ ( 0 , T μ ; H 1 ) f_t^\mu=v^\mu\in L^\infty(0,T_{\mu};H^1) f t μ = v μ ∈ L ∞ ( 0 , T μ ; H 1 ) . In Step 5 we prove μ \mu μ -uniform a priori estimates at the H 3 H^3 H 3 level for the coupled system (146) –(147) . In particular, there exists a time T ∗ > 0 T_*>0 T ∗ > 0 , depending only on ∥ f 0 ∥ H 3 + ∥ v 0 ∥ H 1 \|f_0\|_{H^3}+\|v_0\|_{H^1} ∥ f 0 ∥ H 3 + ∥ v 0 ∥ H 1 and the parameters ( γ , δ , β ) (\gamma,\delta,\beta) ( γ , δ , β ) but independent of μ \mu μ , such that, for T ∗ ≤ T μ T_*\leq T_{\mu} T ∗ ≤ T μ ,
sup 0 ≤ t ≤ T ∗ ∥ f μ ( t ) ∥ H 3 ≤ 2 ∥ f 0 ∥ H 3 . \sup_{0\leq t\leq T_*}\|f^\mu(t)\|_{H^3}\leq2\|f_0\|_{H^3}. 0 ≤ t ≤ T ∗ sup ∥ f μ ( t ) ∥ H 3 ≤ 2∥ f 0 ∥ H 3 .
Thus f μ ∈ X T ∗ f^\mu\in X_{T_*} f μ ∈ X T ∗ and f t μ ∈ L ∞ ( 0 , T ∗ ; H 1 ) f_t^\mu\in L^\infty(0,T_*;H^1) f t μ ∈ L ∞ ( 0 , T ∗ ; H 1 ) , which is the regularity required to pass to the limit μ → 0 \mu\to0 μ → 0 in Step 5.
Step 5: Passing to the limit as μ → 0 \mu\to0 μ → 0 . Let ( f μ , v μ ) (f^\mu,v^\mu) ( f μ , v μ ) be the μ \mu μ -regularized solution from Step 4, with v μ = f t μ v^\mu=f_t^\mu v μ = f t μ . Repeating the energy multipliers from Step 3 (now with f ˉ = f μ \bar{f}=f^\mu f ˉ = f μ and v = v μ v=v^\mu v = v μ ) yields, for a.e. t ∈ ( 0 , T ) t\in(0,T) t ∈ ( 0 , T ) ,
1 2 d d t ( ∥ f t μ ∥ H 1 / 2 2 + γ ∥ f t μ ∥ H 1 2 + ∥ f μ ∥ H 1 2 + β 4 ∥ f μ ∥ H 3 2 ) ≤ ∥ f t μ ∥ H 1 4 + C ∥ J μ f μ ∥ H 2 2 ∥ f t μ ∥ L 2 2 , (151) \frac{1}{2}\frac{d}{dt}\left(\|f_t^\mu\|_{H^{1/2}}^2+\gamma\|f_t^\mu\|_{H^1}^2+\|f^\mu\|_{H^1}^2+\frac{\beta}{4}\|f^\mu\|_{H^3}^2\right)\leq\|f_t^\mu\|_{H^1}^4+C\|\mathcal{J}_{\mu}f^\mu\|_{H^2}^2\|f_t^\mu\|_{L^2}^2, \tag*{(151)} 2 1 d t d ( ∥ f t μ ∥ H 1/2 2 + γ ∥ f t μ ∥ H 1 2 + ∥ f μ ∥ H 1 2 + 4 β ∥ f μ ∥ H 3 2 ) ≤ ∥ f t μ ∥ H 1 4 + C ∥ J μ f μ ∥ H 2 2 ∥ f t μ ∥ L 2 2 , (151)
1 2 ∥ f t t μ ∥ H − 1 / 2 2 + γ 4 ∥ f t t μ ∥ L 2 2 + δ 2 d d t ∥ f t μ ∥ H 1 2 ≤ ∥ f t μ ∥ H 1 4 . (152) \frac{1}{2}\|f_{tt}^\mu\|_{H^{-1/2}}^2+\frac{\gamma}{4}\|f_{tt}^\mu\|_{L^2}^2+\frac{\delta}{2}\frac{d}{dt}\|f_t^\mu\|_{H^1}^2\leq\|f_t^\mu\|_{H^1}^4. \tag*{(152)} 2 1 ∥ f tt μ ∥ H − 1/2 2 + 4 γ ∥ f tt μ ∥ L 2 2 + 2 δ d t d ∥ f t μ ∥ H 1 2 ≤ ∥ f t μ ∥ H 1 4 . (152)
Define
Y μ ( t ) : = ∥ f t μ ( t ) ∥ H 1 2 + Υ ∥ f t μ ( t ) ∥ H 1 2 + ∥ f μ ( t ) ∥ H 1 2 + β 4 ∥ f μ ( t ) ∥ H 3 2 . Y^\mu(t) := \|f_t^\mu(t)\|_{H^1}^2 + \Upsilon\|f_t^\mu(t)\|_{H^1}^2 + \|f^\mu(t)\|_{H^1}^2 + \frac{\beta}{4}\|f^\mu(t)\|_{H^3}^2. Y μ ( t ) := ∥ f t μ ( t ) ∥ H 1 2 + Υ∥ f t μ ( t ) ∥ H 1 2 + ∥ f μ ( t ) ∥ H 1 2 + 4 β ∥ f μ ( t ) ∥ H 3 2 .
Since J μ \mathcal{J}_\mu J μ is bounded on H s H^s H s and ∥ g ∥ H 2 ≲ ∥ g ∥ H 3 \|g\|_{H^2} \lesssim\|g\|_{H^3} ∥ g ∥ H 2 ≲ ∥ g ∥ H 3 , we have
∥ J μ f μ ∥ H 2 2 ≲ ∥ f μ ∥ H 3 2 ≲ Y μ , ∥ f t μ ∥ H 1 2 ≲ Y μ . \|\mathcal{J}_\mu f^\mu\|_{H^2}^2 \lesssim\|f^\mu\|_{H^3}^2 \lesssim Y^\mu,\qquad\|f_t^\mu\|_{H^1}^2 \lesssim Y^\mu. ∥ J μ f μ ∥ H 2 2 ≲ ∥ f μ ∥ H 3 2 ≲ Y μ , ∥ f t μ ∥ H 1 2 ≲ Y μ .
Integrating (152) on [ 0 , t ] [0,t] [ 0 , t ] and using Υ > 0 \Upsilon> 0 Υ > 0 gives
∫ 0 t ∥ f t t μ ( s ) ∥ L 2 2 d s ≲ ∥ v 0 ∥ H 1 2 + ∫ 0 t ∥ f t μ ( s ) ∥ H 1 4 d s ≲ 1 + ∫ 0 t ( Y μ ( s ) ) 2 d s . (153) \int_0^t \|f_{tt}^\mu(s)\|_{L^2}^2\,ds \lesssim\|v_0\|_{H^1}^2 + \int_0^t \|f_t^\mu(s)\|_{H^1}^4\,ds \lesssim1 + \int_0^t (Y^\mu(s))^2\,ds. \tag*{(153)} ∫ 0 t ∥ f tt μ ( s ) ∥ L 2 2 d s ≲ ∥ v 0 ∥ H 1 2 + ∫ 0 t ∥ f t μ ( s ) ∥ H 1 4 d s ≲ 1 + ∫ 0 t ( Y μ ( s ) ) 2 d s . (153)
Now integrate (151) on [ 0 , t ] [0,t] [ 0 , t ] and estimate
∫ 0 t ∥ J μ f μ ∥ H 2 2 ∥ f t t μ ∥ L 2 2 d s ≤ ( sup 0 ≤ s ≤ t Y μ ( s ) ) ∫ 0 t ∥ f t t μ ( s ) ∥ L 2 2 d s , \int_0^t \|\mathcal{J}_\mu f^\mu\|_{H^2}^2\|f_{tt}^\mu\|_{L^2}^2\,ds \leq\left(\sup_{0\leq s\leq t}Y^\mu(s)\right)\int_0^t \|f_{tt}^\mu(s)\|_{L^2}^2\,ds, ∫ 0 t ∥ J μ f μ ∥ H 2 2 ∥ f tt μ ∥ L 2 2 d s ≤ ( 0 ≤ s ≤ t sup Y μ ( s ) ) ∫ 0 t ∥ f tt μ ( s ) ∥ L 2 2 d s ,
then use (153) . Writing Z μ ( t ) : = sup 0 ≤ s ≤ t Y μ ( s ) Z^\mu(t):=\sup_{0\leq s\leq t}Y^\mu(s) Z μ ( t ) := sup 0 ≤ s ≤ t Y μ ( s ) we obtain
Z μ ( t ) ≤ C 0 + C 1 t ( Z μ ( t ) ) 2 + C 2 t ( Z μ ( t ) ) 3 , t ∈ [ 0 , T ] , (154) Z^\mu(t) \leq C_0+C_1t\left(Z^\mu(t)\right)^2+C_2t\left(Z^\mu(t)\right)^3,\qquad t\in[0,T], \tag*{(154)} Z μ ( t ) ≤ C 0 + C 1 t ( Z μ ( t ) ) 2 + C 2 t ( Z μ ( t ) ) 3 , t ∈ [ 0 , T ] , (154)
where C 0 ∼ 1 + Y μ ( 0 ) C_0\sim1+Y^\mu(0) C 0 ∼ 1 + Y μ ( 0 ) and C 1 , C 2 C_1,C_2 C 1 , C 2 depend only on the model parameters and on ∥ f 0 ∥ H 3 + ∥ v 0 ∥ H 1 \|f_0\|_{H^3}+\|v_0\|_{H^1} ∥ f 0 ∥ H 3 + ∥ v 0 ∥ H 1 , and are independent of μ \mu μ . A standard continuity argument applied to (154) yields a time T ∗ > 0 T_*>0 T ∗ > 0 (depending only on ( f 0 , v 0 ) (f_0,v_0) ( f 0 , v 0 ) , but not on μ \mu μ ) such that
sup 0 ≤ t ≤ T ∗ Y μ ( t ) ≤ 2 Y μ ( 0 ) for all 0 < μ ≤ 1. (155) \sup_{0\leq t\leq T_*}Y^\mu(t)\leq2Y^\mu(0)\qquad\text{for all }0<\mu\leq1. \tag*{(155)} 0 ≤ t ≤ T ∗ sup Y μ ( t ) ≤ 2 Y μ ( 0 ) for all 0 < μ ≤ 1. (155)
In particular, uniformly in $0<\mu\leq 1,
{ f μ } is bounded in L ∞ ( 0 , T ∗ ; H 3 ( T 2 ) ) , { f t μ } is bounded in L ∞ ( 0 , T ∗ ; H 1 ( T 2 ) ) . (156) \{f^\mu\}\text{ is bounded in }L^\infty(0,T_*;H^3(\mathbb{T}^2)),\qquad\{f_t^\mu\}\text{ is bounded in }L^\infty(0,T_*;H^1(\mathbb{T}^2)). \tag*{(156)} { f μ } is bounded in L ∞ ( 0 , T ∗ ; H 3 ( T 2 )) , { f t μ } is bounded in L ∞ ( 0 , T ∗ ; H 1 ( T 2 )) . (156)
{ f t t μ } is bounded in L 2 ( 0 , T ∗ ; H − 1 / 2 ( T 2 ) ) . (157) \{f_{tt}^\mu\}\text{ is bounded in }L^2(0,T_*;H^{-1/2}(\mathbb{T}^2)). \tag*{(157)} { f tt μ } is bounded in L 2 ( 0 , T ∗ ; H − 1/2 ( T 2 )) . (157)
By (156) the family { f μ } \{f^\mu\} { f μ } is equi-Lipschitz in time with values in H 1 H^1 H 1 and bounded in L t ∞ H x 3 L^\infty_tH_x^3 L t ∞ H x 3 . As in Step 3, Simon’s compactness theorem (using H 3 ⊂ H 2 + θ ↪ H 1 H^3\subset H^{2+\theta}\hookrightarrow H^1 H 3 ⊂ H 2 + θ ↪ H 1 ) yields, after extracting a subsequence, f μ → f f^\mu\to f f μ → f strongly in C ( [ 0 , T ∗ ] ; H 2 + θ ( T 2 ) ) C([0,T_*];H^{2+\theta}(\mathbb{T}^2)) C ([ 0 , T ∗ ] ; H 2 + θ ( T 2 )) for any θ ∈ ( 0 , 1 ) \theta\in(0,1) θ ∈ ( 0 , 1 ) , and f μ ⇀ ∗ f f^\mu\rightharpoonup^* f f μ ⇀ ∗ f in L ∞ ( 0 , T ∗ ; H 3 ) L^\infty(0,T_*;H^3) L ∞ ( 0 , T ∗ ; H 3 ) . Moreover, f t μ ⇀ ∗ f t f_t^\mu\rightharpoonup^* f_t f t μ ⇀ ∗ f t in L ∞ ( 0 , T ∗ ; H 1 ) L^\infty(0,T_*;H^1) L ∞ ( 0 , T ∗ ; H 1 ) and f t t μ ⇀ f t t f_{tt}^\mu\rightharpoonup f_{tt} f tt μ ⇀ f tt weakly in L 2 ( 0 , T ∗ ; H − 1 / 2 ) L^2(0,T_*;H^{-1/2}) L 2 ( 0 , T ∗ ; H − 1/2 ) . Using these convergences, the fact that J μ → I \mathcal{J}_\mu\to I J μ → I strongly on H s H^s H s as μ → 0 \mu\to0 μ → 0 , and the continuity of N ( ⋅ , ⋅ ) \mathcal{N}(\cdot,\cdot) N ( ⋅ , ⋅ ) and Q ( ⋅ ) \mathcal{Q}(\cdot) Q ( ⋅ ) at this regularity, we may pass to the limit μ → 0 \mu\to0 μ → 0 in the μ \mu μ -regularized second-order equation and conclude that f f f solves (73) on ( 0 , T ∗ ) × T 2 (0,T_*)\times\mathbb{T}^2 ( 0 , T ∗ ) × T 2 , with f ( ⋅ , 0 ) = f 0 f(\cdot,0)=f_0 f ( ⋅ , 0 ) = f 0 and f t ( ⋅ , 0 ) = v 0 f_t(\cdot,0)=v_0 f t ( ⋅ , 0 ) = v 0 . In particular,
f ∈ L ∞ ( 0 , T ∗ ; H 0 3 ( T 2 ) ) , f t ∈ L ∞ ( 0 , T ∗ ; H 0 1 ( T 2 ) ) . f\in L^\infty(0,T_*;H_0^3(\mathbb{T}^2)),\qquad f_t\in L^\infty(0,T_*;H_0^1(\mathbb{T}^2)). f ∈ L ∞ ( 0 , T ∗ ; H 0 3 ( T 2 )) , f t ∈ L ∞ ( 0 , T ∗ ; H 0 1 ( T 2 )) .
Step 6: Uniqueness. We now show that the solution constructed above is unique (in the natural class given by Theorem 4.1 ). Recall that we have obtained a pair ( f , v ) (f,v) ( f , v ) satisfying
( I + Λ ) v t + δ Λ 3 v + ( Λ + β 4 Λ 5 ) f + N ( f , v t ) = ε Q ( v ) , (158) (I+\Lambda)v_t+\delta\Lambda^3v+\left(\Lambda+\frac{\beta}{4}\Lambda^5\right)f+\mathbb{N}(f,v_t)=\varepsilon\mathcal{Q}(v), \tag*{(158)} ( I + Λ ) v t + δ Λ 3 v + ( Λ + 4 β Λ 5 ) f + N ( f , v t ) = ε Q ( v ) , (158)
f t = v . (159) f_t=v. \tag*{(159)} f t = v . (159)
Let ( f ( 1 ) , v ( 1 ) ) (f^{(1)},v^{(1)}) ( f ( 1 ) , v ( 1 ) ) and ( f ( 2 ) , v ( 2 ) ) (f^{(2)},v^{(2)}) ( f ( 2 ) , v ( 2 ) ) be two solutions of (158) –(159) on [ 0 , T ] [0,T] [ 0 , T ] with the same initial data. Setting
f ~ : = f ( 1 ) − f ( 2 ) , v ~ : = v ( 1 ) − v ( 2 ) , \widetilde{f}:=f^{(1)}-f^{(2)},\qquad\widetilde{v}:=v^{(1)}-v^{(2)}, f := f ( 1 ) − f ( 2 ) , v := v ( 1 ) − v ( 2 ) ,
we obtain a coupled system for ( f ~ , v ~ ) (\widetilde{f},\widetilde{v}) ( f , v ) with zero initial data. Testing the v ~ \widetilde{v} v –equation against Λ v ~ \Lambda\widetilde{v} Λ v (and, when needed, using the relation f ~ t = v ~ \widetilde{f}_t=\widetilde{v} f t = v ), and arguing as in the a priori estimates of Steps 3–5, we derive an energy inequality for a suitable quantity Z ( t ) Z(t) Z ( t ) controlling ∥ f ~ ∥ H 3 \|\widetilde{f}\|_{H^3} ∥ f ∥ H 3 and ∥ v ~ ∥ H 1 \|\widetilde{v}\|_{H^1} ∥ v ∥ H 1 . In particular, for t ∈ [ 0 , T ] t\in[0,T] t ∈ [ 0 , T ] one obtains an estimate of the form
Z ( t ) ≤ C 1 t ( Z ( t ) ) 2 + C 2 t ( Z ( t ) ) 3 , Z ( 0 ) = 0 , (160) Z(t)\leq C_1t\bigl(Z(t)\bigr)^2+C_2t\bigl(Z(t)\bigr)^3,\qquad Z(0)=0, \tag*{(160)} Z ( t ) ≤ C 1 t ( Z ( t ) ) 2 + C 2 t ( Z ( t ) ) 3 , Z ( 0 ) = 0 , (160)
where the constants C 1 , C 2 C_1,C_2 C 1 , C 2 depend only on the parameters of the model and on the a priori bounds for the two solutions on [ 0 , T ] [0,T] [ 0 , T ] . For T > 0 T>0 T > 0 chosen as in the existence argument (so that the right-hand side can be absorbed), the polynomial inequality (160) forces Z ( t ) ≡ 0 Z(t)\equiv0 Z ( t ) ≡ 0 on [ 0 , T ] [0,T] [ 0 , T ] . Consequently f ~ ≡ 0 \widetilde{f}\equiv0 f ≡ 0 and v ~ ≡ 0 \widetilde{v}\equiv0 v ≡ 0 , and hence f ( 1 ) ≡ f ( 2 ) f^{(1)}\equiv f^{(2)} f ( 1 ) ≡ f ( 2 ) and v ( 1 ) ≡ v ( 2 ) v^{(1)}\equiv v^{(2)} v ( 1 ) ≡ v ( 2 ) on [ 0 , T ] [0,T] [ 0 , T ] .
Therefore the solution is unique in the stated class, which completes the proof. □
Remark 4.2. Let us emphasize that we do not currently obtain an analogue of Theorem 4.1 for the alternative bidirectional model (79) . While one can construct regularized/approximate solutions for that equation by standard procedures, we are unable to close the corresponding a priori estimates uniformly at the level required to pass to the limit. Establishing a well-posedness theory for this second bidirectional closure therefore remains an open problem.
Well-posedness results for the unidirectional models
In this section we study the Cauchy problem for the unidirectional models derived in Subsection 3.1 as an O ( ε ) \mathcal{O}(\varepsilon) O ( ε ) approximation of the full hydroelastic system. The analysis relies on energy estimates adapted to the nonlocal operator structure of the model, together with standard multilinear and commutator bounds for the Fourier multipliers appearing in the equation.
We first establish a local well-posedness result for the unidirectional model (92) , stated as follows.
Theorem 5.1 (Local well-posedness). Let T > 0 \mathbb{T}>0 T > 0 and δ , β > 0 \delta,\beta>0 δ , β > 0 . Assume that F 0 ∈ H 2 ( T ) F_0\in H^2(\mathbb{T}) F 0 ∈ H 2 ( T ) has zero spatial mean and consider the unidirectional model (92) . Then there exist T > 0 T>0 T > 0 , depending only on ∥ F 0 ∥ H 2 ( T ) \|F_0\|_{H^2(\mathbb{T})} ∥ F 0 ∥ H 2 ( T ) and the parameters of the model, and a unique solution
F ∈ C ( [ 0 , T ] ; H 2 ( T ) ) , F ( ⋅ , 0 ) = F 0 . F\in C([0,T];H^2(\mathbb{T})),\qquad F(\cdot,0)=F_0. F ∈ C ([ 0 , T ] ; H 2 ( T )) , F ( ⋅ , 0 ) = F 0 .
Proof of Theorem 5.1. The argument combines classical a priori energy estimates with a standard regularization scheme based on mollifiers, cf. [25 ] . We first derive the required a priori bounds, then construct solutions at the regularized level and pass to the limit, and finally establish uniqueness. For clarity, we decompose the proof into several steps.
Step 1: conservation of the mean. Let ⟨ F ( τ ) ⟩ : = 1 2 π ∫ T F ( ξ , τ ) d ξ \langle F(\tau)\rangle:= \frac{1}{2\pi}\int_{\mathbb{T}} F(\xi,\tau)\,d\xi ⟨ F ( τ )⟩ := 2 π 1 ∫ T F ( ξ , τ ) d ξ . We claim that any sufficiently smooth real-valued solution of (92) satisfies
⟨ F ( τ ) ⟩ = ⟨ F 0 ⟩ for all τ ∈ [ 0 , T ] . \langle F(\tau)\rangle= \langle F_0\rangle\qquad\text{for all } \tau\in[0,T]. ⟨ F ( τ )⟩ = ⟨ F 0 ⟩ for all τ ∈ [ 0 , T ] .
Indeed, taking the spatial mean of (92) , the linear part has zero average (it is a combination of ξ \xi ξ -derivatives and Fourier multipliers annihilating constants). With the convention a ( 0 ) = 1 2 a(0)=\frac{1}{2} a ( 0 ) = 2 1 and b ( 0 ) = 0 b(0)=0 b ( 0 ) = 0 , we obtain
∂ τ ⟨ F ⟩ = − 1 2 ⟨ 2 F ξ Λ F − [ Λ , F ] F ξ ⟩ . \partial_\tau\langle F\rangle= -\frac{1}{2}\left\langle2F_\xi\Lambda F-[\Lambda,F]F_\xi\right\rangle. ∂ τ ⟨ F ⟩ = − 2 1 ⟨ 2 F ξ Λ F − [ Λ , F ] F ξ ⟩ .
Expanding the commutator,
2 F ξ Λ F − [ Λ , F ] F ξ = 2 F ξ Λ F − Λ ( F F ξ ) + F Λ F ξ , 2F_\xi\Lambda F-[\Lambda,F]F_\xi=2F_\xi\Lambda F-\Lambda(FF_\xi)+F\Lambda F_\xi, 2 F ξ Λ F − [ Λ , F ] F ξ = 2 F ξ Λ F − Λ ( F F ξ ) + F Λ F ξ ,
and using that ⟨ Λ ( ⋅ ) ⟩ = 0 \langle\Lambda(\cdot)\rangle=0 ⟨ Λ ( ⋅ )⟩ = 0 , it remains to show that ⟨ 2 F ξ Λ F + F Λ F ξ ⟩ = 0 \langle2F_\xi\Lambda F+F\Lambda F_\xi\rangle=0 ⟨ 2 F ξ Λ F + F Λ F ξ ⟩ = 0 . In one dimension Λ = H ∂ ξ \Lambda=\mathcal{H}\partial_\xi Λ = H ∂ ξ , hence Λ F = H F ξ \Lambda F=\mathcal{H}F_\xi Λ F = H F ξ and Λ F ξ = H F ξ ξ \Lambda F_\xi=\mathcal{H}F_{\xi\xi} Λ F ξ = H F ξ ξ . Therefore,
⟨ F ξ Λ F ⟩ = 1 2 π ∫ T F ξ H F ξ d ξ = 0 , ⟨ F Λ F ξ ⟩ = 1 2 π ∫ T F H F ξ ξ d ξ = − 1 2 π ∫ T F ξ H F ξ d ξ = 0 , \langle F_\xi\Lambda F\rangle=\frac{1}{2\pi}\int_{\mathbb{T}} F_\xi\mathcal{H}F_\xi\,d\xi=0,\qquad\langle F\Lambda F_\xi\rangle=\frac{1}{2\pi}\int_{\mathbb{T}} F\mathcal{H}F_{\xi\xi}\,d\xi=-\frac{1}{2\pi}\int_{\mathbb{T}} F_\xi\mathcal{H}F_\xi\,d\xi=0, ⟨ F ξ Λ F ⟩ = 2 π 1 ∫ T F ξ H F ξ d ξ = 0 , ⟨ F Λ F ξ ⟩ = 2 π 1 ∫ T F H F ξ ξ d ξ = − 2 π 1 ∫ T F ξ H F ξ d ξ = 0 ,
Hence ∂ τ ⟨ F ⟩ = 0 \partial_\tau\langle F\rangle=0 ∂ τ ⟨ F ⟩ = 0 and the mean is conserved. In particular, if ⟨ F 0 ⟩ = 0 \langle F_0\rangle=0 ⟨ F 0 ⟩ = 0 , then F ( ⋅ , τ ) F(\cdot,\tau) F ( ⋅ , τ ) remains mean-zero for all times of existence.
Step 2: a priori estimates. In order to derive the a priori estimates it is convenient to work with a symmetrized formulation of the evolution, namely
ε ( M ∗ M ) F τ = M ∗ ( ( I + γ Λ ) F ξ + β 4 Λ F ξ ξ ξ + δ Λ F ξ ξ + H F ) − ε M ∗ ( 2 F ξ Λ F − [ Λ , F ] F ξ ) . (161) \varepsilon(\mathcal{M}^{*}\mathcal{M})F_\tau=\mathcal{M}^{*}\left((I+\gamma\Lambda)F_\xi+\frac{\beta}{4}\Lambda F_{\xi\xi\xi}+\delta\Lambda F_{\xi\xi}+\mathcal{H}F\right)-\varepsilon\mathcal{M}^{*}(2F_\xi\Lambda F-[\Lambda,F]F_\xi). \tag*{(161)} ε ( M ∗ M ) F τ = M ∗ ( ( I + γ Λ ) F ξ + 4 β Λ F ξ ξ ξ + δ Λ F ξ ξ + H F ) − ε M ∗ ( 2 F ξ Λ F − [ Λ , F ] F ξ ) . (161)
where
M ∗ M = 4 ( I + γ Λ ) 2 − δ 2 Λ 2 ∂ ξ 2 , M ∗ = 2 ( I + γ Λ ) − δ Λ ∂ ξ . \mathcal{M}^{*}\mathcal{M}=4(I+\gamma\Lambda)^2-\delta^2\Lambda^2\partial_\xi^2,\qquad\mathcal{M}^{*}=2(I+\gamma\Lambda)-\delta\Lambda\partial_\xi. M ∗ M = 4 ( I + γ Λ ) 2 − δ 2 Λ 2 ∂ ξ 2 , M ∗ = 2 ( I + γ Λ ) − δ Λ ∂ ξ .
Taking the L 2 ( T ) L^2(\mathbb{T}) L 2 ( T ) inner product of (161) with F F F and integrating by parts in ξ \xi ξ , we obtain the evolution identity
d d τ ( ∥ F ∥ L 2 2 + 2 Υ ∥ F ∥ H ˙ 1 / 2 2 + Υ 2 ∥ F ∥ H ˙ 1 2 + δ 2 4 ∥ F ∥ H ˙ 2 2 ) = ∫ T ( I + Υ Λ ) ( ( I + Υ Λ ) F ξ ) F d ξ + β 4 ∫ T ( I + Υ Λ ) Λ F ξ ξ ξ F d ξ + δ ∫ T ( I + Υ Λ ) F ξ ξ F d ξ + ∫ T ( I + Υ Λ ) H F F d ξ + δ 2 ∫ T ( I + Υ Λ ) F ξ Λ F ξ d ξ + β δ 8 ∫ T Λ F ξ ξ ξ Λ F ξ d ξ + δ 2 2 ∫ T Λ F ξ ξ Λ F ξ d ξ + δ 2 ∫ T H F Λ F ξ d ξ − 2 ∫ T ( I + Υ Λ ) ( F ξ Λ F ) F d ξ + ∫ T ( I + Υ Λ ) ( [ Λ , F ] F ξ ) F d ξ − δ ∫ T F ξ ( Λ F ) ( Λ F ξ ) d ξ + δ 2 ∫ T ( [ Λ , F ] F ξ ) ( Λ F ξ ) d ξ = ∑ j = 1 12 I j . \begin{aligned}
\frac{d}{d\tau}\left(\|F\|_{L^2}^2+2\Upsilon\|F\|_{\dot{H}^{1/2}}^2+\Upsilon^2\|F\|_{\dot{H}^1}^2+\frac{\delta^2}{4}\|F\|_{\dot{H}^2}^2\right)
={}&\int_{\mathbb{T}}(I+\Upsilon\Lambda)\left((I+\Upsilon\Lambda)F_\xi\right)F\,d\xi+\frac{\beta}{4}\int_{\mathbb{T}}(I+\Upsilon\Lambda) \Lambda F_{\xi\xi\xi}F\,d\xi\\
&+\delta\int_{\mathbb{T}}(I+\Upsilon\Lambda)F_{\xi\xi}F\,d\xi+\int_{\mathbb{T}}(I+\Upsilon\Lambda)\mathcal{H}F F\,d\xi\\
&+\frac{\delta}{2}\int_{\mathbb{T}}(I+\Upsilon\Lambda)F_\xi\Lambda F_\xi\,d\xi+\frac{\beta\delta}{8}\int_{\mathbb{T}}\Lambda F_{\xi\xi\xi}\Lambda F_\xi\,d\xi\\
&+\frac{\delta^2}{2}\int_{\mathbb{T}}\Lambda F_{\xi\xi}\Lambda F_\xi\,d\xi+\frac{\delta}{2}\int_{\mathbb{T}}\mathcal{H}F\Lambda F_\xi\,d\xi\\
&-2\int_{\mathbb{T}}(I+\Upsilon\Lambda)(F_\xi\Lambda F)F\,d\xi+\int_{\mathbb{T}}(I+\Upsilon\Lambda)([\Lambda,F]F_\xi)F\,d\xi\\
&-\delta\int_{\mathbb{T}}F_\xi(\Lambda F)(\Lambda F_\xi)\,d\xi+\frac{\delta}{2}\int_{\mathbb{T}}([\Lambda,F]F_\xi)(\Lambda F_\xi)\,d\xi
=\sum_{j=1}^{12}I_j.
\end{aligned} d τ d ( ∥ F ∥ L 2 2 + 2Υ∥ F ∥ H ˙ 1/2 2 + Υ 2 ∥ F ∥ H ˙ 1 2 + 4 δ 2 ∥ F ∥ H ˙ 2 2 ) = ∫ T ( I + ΥΛ ) ( ( I + ΥΛ ) F ξ ) F d ξ + 4 β ∫ T ( I + ΥΛ ) Λ F ξ ξ ξ F d ξ + δ ∫ T ( I + ΥΛ ) F ξ ξ F d ξ + ∫ T ( I + ΥΛ ) H F F d ξ + 2 δ ∫ T ( I + ΥΛ ) F ξ Λ F ξ d ξ + 8 β δ ∫ T Λ F ξ ξ ξ Λ F ξ d ξ + 2 δ 2 ∫ T Λ F ξ ξ Λ F ξ d ξ + 2 δ ∫ T H F Λ F ξ d ξ − 2 ∫ T ( I + ΥΛ ) ( F ξ Λ F ) F d ξ + ∫ T ( I + ΥΛ ) ([ Λ , F ] F ξ ) F d ξ − δ ∫ T F ξ ( Λ F ) ( Λ F ξ ) d ξ + 2 δ ∫ T ([ Λ , F ] F ξ ) ( Λ F ξ ) d ξ = j = 1 ∑ 12 I j .
Using that A : = I + Υ Λ A:=I+\Upsilon\Lambda A := I + ΥΛ and Λ \Lambda Λ are self-adjoint on L 2 ( T ) L^2(\mathbb{T}) L 2 ( T ) , that ∂ ξ \partial_\xi ∂ ξ commutes with A A A and Λ \Lambda Λ , and that H ∗ = − H \mathcal{H}^*=-\mathcal{H} H ∗ = − H , we obtain the cancellations I 1 = I 2 = I 4 = I 7 = 0 I_1=I_2=I_4=I_7=0 I 1 = I 2 = I 4 = I 7 = 0 . Moreover, we also find that
I 6 = β δ 8 ∫ T ( Λ F ξ ) ξ ξ ( Λ F ξ ) d ξ = − β δ 8 ∥ F ∥ H ˙ 3 ( T ) 2 , I 3 = − δ ( ∥ F ∥ H ˙ 3 / 2 2 + Υ ∥ F ∥ H ˙ 2 2 ) , I 8 = − δ 2 ∥ F ∥ H ˙ 1 ( T ) 2 . I_6=\frac{\beta\delta}{8}\int_{\mathbb{T}}(\Lambda F_\xi)_{\xi\xi}(\Lambda F_\xi)\,d\xi=-\frac{\beta\delta}{8}\|F\|_{\dot{H}^3(\mathbb{T})}^2,\qquad I_3=-\delta\left(\|F\|_{\dot{H}^{3/2}}^2+\Upsilon\|F\|_{\dot{H}^2}^2\right),\qquad I_8=-\frac{\delta}{2}\|F\|_{\dot{H}^1(\mathbb{T})}^2. I 6 = 8 β δ ∫ T ( Λ F ξ ) ξ ξ ( Λ F ξ ) d ξ = − 8 β δ ∥ F ∥ H ˙ 3 ( T ) 2 , I 3 = − δ ( ∥ F ∥ H ˙ 3/2 2 + Υ∥ F ∥ H ˙ 2 2 ) , I 8 = − 2 δ ∥ F ∥ H ˙ 1 ( T ) 2 .
We now bound the nonlinear contributions I 9 , I 10 , I 11 , I 12 I_9,I_{10},I_{11},I_{12} I 9 , I 10 , I 11 , I 12 . Let A : = I + Υ Λ A:=I+\Upsilon\Lambda A := I + ΥΛ . By self-adjointness of A A A ,
I 9 = − 2 ∫ T A ( F ξ Λ F ) F d ξ = − 2 ⟨ F ξ Λ F , A F ⟩ , I 10 = ∫ T A ( [ Λ , F ] F ξ ) F d ξ = ⟨ [ Λ , F ] F ξ , A F ⟩ . I_9=-2\int_{\mathbb{T}}A(F_\xi\Lambda F)F\,d\xi=-2\langle F_\xi\Lambda F,AF\rangle,\qquad I_{10}=\int_{\mathbb{T}}A([\Lambda,F]F_\xi)F\,d\xi=\langle[\Lambda,F]F_\xi,AF\rangle. I 9 = − 2 ∫ T A ( F ξ Λ F ) F d ξ = − 2 ⟨ F ξ Λ F , A F ⟩ , I 10 = ∫ T A ([ Λ , F ] F ξ ) F d ξ = ⟨[ Λ , F ] F ξ , A F ⟩ .
Hence, integrating by parts and using Cauchy–Schwarz,
∣ I 9 ∣ ≤ ∥ F ∥ L ∞ ∥ F ∥ H ˙ 2 ( T ) (162) |I_9|\leq\|F\|_{L^\infty}\|F\|_{\dot{H}^2(\mathbb{T})} \tag*{(162)} ∣ I 9 ∣ ≤ ∥ F ∥ L ∞ ∥ F ∥ H ˙ 2 ( T ) (162)
Similarly, to bound I 10 I_{10} I 10 we can expand the commutator and use integration by parts to find that
∣ I 10 ∣ ≤ ∥ F ∥ L ∞ ∥ F ∥ H ˙ 2 ( T ) (163) |I_{10}|\leq\|F\|_{L^\infty}\|F\|_{\dot{H}^2(\mathbb{T})} \tag*{(163)} ∣ I 10 ∣ ≤ ∥ F ∥ L ∞ ∥ F ∥ H ˙ 2 ( T ) (163)
For I 11 I_{11} I 11 and I 12 I_{12} I 12 we similarly have
I 11 = δ 2 ∫ T F ξ ξ ( Λ F ) 2 d ξ , I 12 = δ 2 ∫ T ( Λ ( F F ξ ) − F Λ F ξ ) Λ F ξ d ξ . I_{11}=\frac{\delta}{2}\int_{\mathbb{T}}F_{\xi\xi}(\Lambda F)^2\,d\xi,\qquad I_{12}=\frac{\delta}{2}\int_{\mathbb{T}}\left(\Lambda(FF_\xi)-F\Lambda F_\xi\right)\Lambda F_\xi\,d\xi. I 11 = 2 δ ∫ T F ξ ξ ( Λ F ) 2 d ξ , I 12 = 2 δ ∫ T ( Λ ( F F ξ ) − F Λ F ξ ) Λ F ξ d ξ .
By means of Gagliardo-Nirenberg interpolation inequality (13)
∣ I 11 ∣ ≤ δ ∥ F ∥ H ˙ 2 ( T ) ∥ Λ F ∥ L 4 2 ≤ C δ ∥ F ∥ L ∞ ∥ F ∥ H ˙ 2 ( T ) 2 (164) |I_{11}|\leq\delta\|F\|_{\dot{H}^2(\mathbb{T})}\|\Lambda F\|_{L^4}^2\leq C\delta\|F\|_{L^\infty}\|F\|_{\dot{H}^2(\mathbb{T})}^2 \tag*{(164)} ∣ I 11 ∣ ≤ δ ∥ F ∥ H ˙ 2 ( T ) ∥Λ F ∥ L 4 2 ≤ C δ ∥ F ∥ L ∞ ∥ F ∥ H ˙ 2 ( T ) 2 (164)
On the other hand, using Cauchy-Schwarz inequality and Moser estimate (12) we find that
I 12 ≤ δ 4 ∥ ∂ ξ Λ ( F 2 ) ∥ L 2 ∥ Λ F ξ ∥ L 2 + δ 2 ∥ F ∥ L ∞ ∥ Λ F ξ ∥ L 2 2 ≤ C ∥ F ∥ L ∞ ∥ F ∥ H ˙ 2 ( T ) 2 (165) I_{12}\leq\frac{\delta}{4}\|\partial_\xi\Lambda(F^2)\|_{L^2}\|\Lambda F_\xi\|_{L^2}+\frac{\delta}{2}\|F\|_{L^\infty}\|\Lambda F_\xi\|_{L^2}^2\leq C\|F\|_{L^\infty}\|F\|_{\dot{H}^2(\mathbb{T})}^2 \tag*{(165)} I 12 ≤ 4 δ ∥ ∂ ξ Λ ( F 2 ) ∥ L 2 ∥Λ F ξ ∥ L 2 + 2 δ ∥ F ∥ L ∞ ∥Λ F ξ ∥ L 2 2 ≤ C ∥ F ∥ L ∞ ∥ F ∥ H ˙ 2 ( T ) 2 (165)
Therefore, combining all the previous estimates, we have shown that
d d τ ( ∥ F ∥ L 2 2 + 2 Υ ∥ F ∥ H ˙ 1 / 2 2 + Υ 2 ∥ F ∥ H ˙ 1 2 + δ 2 4 ∥ F ∥ H ˙ 2 2 ) + β δ 8 ∥ F ∥ H ˙ 3 ( T ) 2 ≤ C ∥ F ∥ L ∞ ∥ F ∥ H ˙ 2 ( T ) 2 (166) \frac{d}{d\tau}\left(\|F\|_{L^2}^2+2\Upsilon\|F\|_{\dot{H}^{1/2}}^2+\Upsilon^2\|F\|_{\dot{H}^1}^2+\frac{\delta^2}{4}\|F\|_{\dot{H}^2}^2\right)+\frac{\beta\delta}{8}\|F\|_{\dot{H}^3(\mathbb{T})}^2\leq C\|F\|_{L^\infty}\|F\|_{\dot{H}^2(\mathbb{T})}^2 \tag*{(166)} d τ d ( ∥ F ∥ L 2 2 + 2Υ∥ F ∥ H ˙ 1/2 2 + Υ 2 ∥ F ∥ H ˙ 1 2 + 4 δ 2 ∥ F ∥ H ˙ 2 2 ) + 8 β δ ∥ F ∥ H ˙ 3 ( T ) 2 ≤ C ∥ F ∥ L ∞ ∥ F ∥ H ˙ 2 ( T ) 2 (166)
Define
E ( τ ) : = ∥ F ∥ L 2 2 + 2 γ ∥ F ∥ H ˙ 1 / 2 2 + γ 2 ∥ F ∥ H ˙ 1 2 + δ 2 4 ∥ F ∥ H ˙ 2 2 . E(\tau) := \|F\|_{L^2}^2 + 2\gamma\|F\|_{\dot{H}^{1/2}}^2 + \gamma^2\|F\|_{\dot{H}^1}^2 + \frac{\delta^2}{4}\|F\|_{\dot{H}^2}^2. E ( τ ) := ∥ F ∥ L 2 2 + 2 γ ∥ F ∥ H ˙ 1/2 2 + γ 2 ∥ F ∥ H ˙ 1 2 + 4 δ 2 ∥ F ∥ H ˙ 2 2 .
Moreover, by means of the Sobolev embedding (11) , we have that ∥ F ∥ L ∞ ≲ ∥ F ∥ H 2 ≲ E ( τ ) 1 / 2 \|F\|_{L^\infty} \lesssim\|F\|_{H^2} \lesssim E(\tau)^{1/2} ∥ F ∥ L ∞ ≲ ∥ F ∥ H 2 ≲ E ( τ ) 1/2 . Therefore the energy inequality implies
E ′ ( τ ) + β δ 8 ∥ F ∥ H ˙ 3 2 ≤ C ∥ F ∥ L ∞ ∥ F ∥ H ˙ 2 2 ≲ C E ( τ ) 3 / 2 , E'(\tau) + \frac{\beta\delta}{8}\|F\|_{\dot{H}^3}^2 \le C\|F\|_{L^\infty}\|F\|_{\dot{H}^2}^2 \lesssim C E(\tau)^{3/2}, E ′ ( τ ) + 8 β δ ∥ F ∥ H ˙ 3 2 ≤ C ∥ F ∥ L ∞ ∥ F ∥ H ˙ 2 2 ≲ C E ( τ ) 3/2 ,
and in particular E ′ ( τ ) ≤ C E ( τ ) 3 / 2 E'(\tau) \le CE(\tau)^{3/2} E ′ ( τ ) ≤ C E ( τ ) 3/2 . Solving this differential inequality gives
E ( τ ) − 1 / 2 ≥ E ( 0 ) − 1 / 2 − C 2 τ , E(\tau)^{-1/2} \ge E(0)^{-1/2} - \frac{C}{2}\tau, E ( τ ) − 1/2 ≥ E ( 0 ) − 1/2 − 2 C τ ,
so for 0 ≤ τ ≤ T ∗ : = 1 C E ( 0 ) − 1 / 2 0 \le\tau\le T_* := \frac{1}{C}E(0)^{-1/2} 0 ≤ τ ≤ T ∗ := C 1 E ( 0 ) − 1/2 we have the uniform bound E ( τ ) ≤ 4 E ( 0 ) E(\tau) \le4E(0) E ( τ ) ≤ 4 E ( 0 ) .
Step 3: existence via mollification. We define the mollified unknown F ν F^\nu F ν as the unique solution to the regularized Cauchy problem
{ ∂ τ F ν = J ν { 1 ε ( a + b H ) ( ( I + γ Λ ) ∂ ξ ( J ν F ν ) + β 4 Λ ∂ ξ 3 ( J ν F ν ) + δ Λ ∂ ξ 2 ( J ν F ν ) + H ( J ν F ν ) ) − ( a + b H ) ( 2 ∂ ξ ( J ν F ν ) Λ ( J ν F ν ) − [ Λ , J ν F ν ] ∂ ξ ( J ν F ν ) ) } , F ν ( ⋅ , 0 ) = J ν F 0 . (167) \left\{
\begin{aligned}
\partial_\tau F^\nu&= \mathcal{J}_\nu\left\{\frac{1}{\varepsilon}(a+b\mathcal{H})\left((I+\gamma\Lambda)\partial_\xi(\mathcal{J}_\nu F^\nu)+\frac{\beta}{4}\Lambda\partial_\xi^3(\mathcal{J}_\nu F^\nu)+\delta\Lambda\partial_\xi^2(\mathcal{J}_\nu F^\nu)+\mathcal{H}(\mathcal{J}_\nu F^\nu)\right)\right.\\
&\qquad\left.-(a+b\mathcal{H})\left(2\partial_\xi(\mathcal{J}_\nu F^\nu)\Lambda(\mathcal{J}_\nu F^\nu)-[\Lambda,\mathcal{J}_\nu F^\nu]\partial_\xi(\mathcal{J}_\nu F^\nu)\right)\right\},\\
F^\nu(\cdot,0)&=\mathcal{J}_\nu F_0.
\end{aligned}
\right. \tag*{(167)} ⎩ ⎨ ⎧ ∂ τ F ν F ν ( ⋅ , 0 ) = J ν { ε 1 ( a + b H ) ( ( I + γ Λ ) ∂ ξ ( J ν F ν ) + 4 β Λ ∂ ξ 3 ( J ν F ν ) + δ Λ ∂ ξ 2 ( J ν F ν ) + H ( J ν F ν ) ) − ( a + b H ) ( 2 ∂ ξ ( J ν F ν ) Λ ( J ν F ν ) − [ Λ , J ν F ν ] ∂ ξ ( J ν F ν ) ) } , = J ν F 0 . (167)
Since J ν \mathcal{J}_\nu J ν is smoothing, the right-hand side of (167) defines a locally Lipschitz vector field on H 2 ( T ) H^2(\mathbb{T}) H 2 ( T ) . Therefore, by the Picard–Lindelöf theorem in Banach spaces, there exists T ν > 0 T_\nu>0 T ν > 0 and a unique solution
F ν ∈ C 1 ( [ 0 , T ν ] ; H 2 ( T ) ) . F^\nu\in C^1([0,T_\nu];H^2(\mathbb{T})). F ν ∈ C 1 ([ 0 , T ν ] ; H 2 ( T )) .
Because J ν \mathcal{J}_\nu J ν is self-adjoint on L 2 ( T ) L^2(\mathbb{T}) L 2 ( T ) and commutes with ∂ ξ \partial_\xi ∂ ξ , Λ \Lambda Λ , H \mathcal{H} H , ( I + γ Λ ) (I+\gamma\Lambda) ( I + γ Λ ) and the multipliers a , b a,b a , b , the a priori energy computation in Step 2 can be repeated for the mollified system (167) . The only modification is that J ν \mathcal{J}_\nu J ν may be shifted between factors using ⟨ J ν G , F ν ⟩ = ⟨ G , J ν F ν ⟩ \langle\mathcal{J}_\nu G,F^\nu\rangle=\langle G,\mathcal{J}_\nu F^\nu\rangle ⟨ J ν G , F ν ⟩ = ⟨ G , J ν F ν ⟩ . Consequently, the same cancellations and sign identities hold, and we obtain the same differential inequality for the mollified energy E ν ( τ ) E^\nu(\tau) E ν ( τ ) with constants independent of ν \nu ν . Denoting
E ν ( τ ) : = ∥ F ν ∥ L 2 2 + 2 γ ∥ F ν ∥ H ˙ 1 / 2 2 + γ 2 ∥ F ν ∥ H ˙ 1 2 + δ 2 4 ∥ F ν ∥ H ˙ 2 2 , E^\nu(\tau) := \|F^\nu\|_{L^2}^2 + 2\gamma\|F^\nu\|_{\dot{H}^{1/2}}^2 + \gamma^2\|F^\nu\|_{\dot{H}^1}^2 + \frac{\delta^2}{4}\|F^\nu\|_{\dot{H}^2}^2, E ν ( τ ) := ∥ F ν ∥ L 2 2 + 2 γ ∥ F ν ∥ H ˙ 1/2 2 + γ 2 ∥ F ν ∥ H ˙ 1 2 + 4 δ 2 ∥ F ν ∥ H ˙ 2 2 ,
we obtain, for τ ∈ [ 0 , T ν ) \tau\in[0,T_\nu) τ ∈ [ 0 , T ν ) ,
d d τ E ν ( τ ) ≤ C ( E ν ( τ ) ) 3 / 2 , (168) \frac{d}{d\tau}E^\nu(\tau)\le C\left(E^\nu(\tau)\right)^{3/2}, \tag*{(168)} d τ d E ν ( τ ) ≤ C ( E ν ( τ ) ) 3/2 , (168)
with a constant C C C independent of ν \nu ν . Solving (168) yields a time T ∗ = T ∗ ( E ( 0 ) ) > 0 T_* = T_*(E(0))>0 T ∗ = T ∗ ( E ( 0 )) > 0 such that
sup 0 ≤ τ ≤ T ∗ E ν ( τ ) ≤ 4 E ( 0 ) for all ν ≥ 1. (169) \sup_{0\le\tau\le T_*} E^\nu(\tau)\le4E(0) \qquad\text{for all }\nu\ge1. \tag*{(169)} 0 ≤ τ ≤ T ∗ sup E ν ( τ ) ≤ 4 E ( 0 ) for all ν ≥ 1. (169)
In particular, T ν ≥ T ∗ T_\nu\ge T_* T ν ≥ T ∗ and { F ν } \{F^\nu\} { F ν } is bounded in L ∞ ( [ 0 , T ∗ ] ; H 2 ( T ) ) L^\infty([0,T_*];H^2(\mathbb{T})) L ∞ ([ 0 , T ∗ ] ; H 2 ( T )) uniformly in ν \nu ν .
By the uniform bound (169) and the mollified equation (167) , the sequence { F ν } \{F^\nu\} { F ν } is bounded in L ∞ ( [ 0 , T ∗ ] ; H 2 ( T ) ) L^\infty([0,T_*];H^2(\mathbb{T})) L ∞ ([ 0 , T ∗ ] ; H 2 ( T )) and { ∂ τ F ν } \{\partial_\tau F^\nu\} { ∂ τ F ν } is bounded in L ∞ ( [ 0 , T ∗ ] ; H 1 ( T ) ) L^\infty([0,T_*];H^1(\mathbb{T})) L ∞ ([ 0 , T ∗ ] ; H 1 ( T )) , uniformly in ν \nu ν . Hence, after extracting a subsequence, F ν → F F^\nu\to F F ν → F strongly in C ( [ 0 , T ∗ ] ; H 1 + θ ( T ) ) C([0,T_*];H^{1+\theta}(\mathbb{T})) C ([ 0 , T ∗ ] ; H 1 + θ ( T )) for every θ ∈ ( 0 , 1 ) \theta\in(0,1) θ ∈ ( 0 , 1 ) and weaklyin L ∞ ( [ 0 , T ∗ ] ; H 2 ( T ) ) L^\infty([0,T_*];H^2(\mathbb{T})) L ∞ ([ 0 , T ∗ ] ; H 2 ( T )) . Using the continuity of Fourier multipliers on Sobolev spaces and standard product/commutator stability at the H 2 H^2 H 2 level, we may pass to the limit in the mollified equation and conclude that F F F solves (92) on [ 0 , T ∗ ] [0,T_*] [ 0 , T ∗ ] with F ( ⋅ , 0 ) = F 0 F(\cdot,0)=F_0 F ( ⋅ , 0 ) = F 0 , and moreover F ∈ C ( [ 0 , T ∗ ] ; H 2 ( T ) ) F\in C([0,T_*];H^2(\mathbb{T})) F ∈ C ([ 0 , T ∗ ] ; H 2 ( T )) .
*Step 4: uniqueness. Let F F F and G G G be two solutions of (161) on [ 0 , T ] [0,T] [ 0 , T ] such that
F , G ∈ L ∞ ( [ 0 , T ] ; H 2 ( T ) ) , F ( ⋅ , 0 ) = G ( ⋅ , 0 ) = F 0 . F,G\in L^\infty([0,T];H^2(\mathbb{T})),\qquad F(\cdot,0)=G(\cdot,0)=F_0. F , G ∈ L ∞ ([ 0 , T ] ; H 2 ( T )) , F ( ⋅ , 0 ) = G ( ⋅ , 0 ) = F 0 .
Set W : = F − G W:=F-G W := F − G . Subtracting (161) (written for F F F and G G G ) yields
( M ∗ M ) W τ = M ∗ ( ( I + Υ Λ ) W ξ + β 4 Λ W ξ ξ + δ Λ W ξ + H W ) − M ∗ ( N ( F ) − N ( G ) ) , (170) (\mathcal{M}^*\mathcal{M})W_\tau=\mathcal{M}^*\left((I+\Upsilon\Lambda)W_\xi+\frac{\beta}{4}\Lambda W_{\xi\xi}+\delta\Lambda W_\xi+\mathcal{H}W\right)-\mathcal{M}^*(\mathcal{N}(F)-\mathcal{N}(G)), \tag*{(170)} ( M ∗ M ) W τ = M ∗ ( ( I + ΥΛ ) W ξ + 4 β Λ W ξ ξ + δ Λ W ξ + H W ) − M ∗ ( N ( F ) − N ( G )) , (170)
where N ( U ) : = 2 U ξ Λ U − [ Λ , U ] U ξ \mathcal{N}(U):=2U_\xi\Lambda U-[\Lambda,U]U_\xi N ( U ) := 2 U ξ Λ U − [ Λ , U ] U ξ . Testing (170) against W W W in L 2 ( T ) L^2(\mathbb{T}) L 2 ( T ) and repeating the energy computation of Step 2 gives
d d τ E W ( τ ) + δ β 8 ∥ W ∥ H 3 ( T ) 2 ≲ ∣ ⟨ M ∗ ( N ( F ) − N ( G ) ) , W ⟩ ∣ , (171) \frac{d}{d\tau}E_W(\tau)+\frac{\delta\beta}{8}\lVert W\rVert_{H^3(\mathbb{T})}^2\lesssim\left|\left\langle\mathcal{M}^*(\mathcal{N}(F)-\mathcal{N}(G)),W\right\rangle\right|, \tag*{(171)} d τ d E W ( τ ) + 8 δ β ∥ W ∥ H 3 ( T ) 2 ≲ ∣ ⟨ M ∗ ( N ( F ) − N ( G )) , W ⟩ ∣ , (171)
with
E W ( τ ) : = ∥ W ∥ L 2 2 + 2 Υ ∥ W ∥ H 1 / 2 2 + Υ 2 ∥ W ∥ H 1 2 + δ 2 4 ∥ W ∥ H 2 2 . E_W(\tau):=\lVert W\rVert_{L^2}^2+2\Upsilon\lVert W\rVert_{H^{1/2}}^2+\Upsilon^2\lVert W\rVert_{H^1}^2+\frac{\delta^2}{4}\lVert W\rVert_{H^2}^2. E W ( τ ) := ∥ W ∥ L 2 2 + 2Υ ∥ W ∥ H 1/2 2 + Υ 2 ∥ W ∥ H 1 2 + 4 δ 2 ∥ W ∥ H 2 2 .
Moreover, expanding with F = G + W F=G+W F = G + W we have
N ( F ) − N ( G ) = 2 W ξ Λ F + 2 G ξ Λ W − [ Λ , W ] F ξ − [ Λ , G ] W ξ . (172) \mathcal{N}(F)-\mathcal{N}(G)=2W_\xi\Lambda F+2G_\xi\Lambda W-[\Lambda,W]F_\xi-[\Lambda,G]W_\xi. \tag*{(172)} N ( F ) − N ( G ) = 2 W ξ Λ F + 2 G ξ Λ W − [ Λ , W ] F ξ − [ Λ , G ] W ξ . (172)
Using H 2 ( T ) ↪ W 1 , ∞ ( T ) H^2(\mathbb{T})\hookrightarrow W^{1,\infty}(\mathbb{T}) H 2 ( T ) ↪ W 1 , ∞ ( T ) and the standard commutator bound ∥ [ Λ , f ] g ∥ L 2 ≲ ∥ f ξ ∥ L ∞ ∥ g ∥ L 2 \lVert[\Lambda,f]g\rVert_{L^2}\lesssim\lVert f_\xi\rVert_{L^\infty}\lVert g\rVert_{L^2} ∥[ Λ , f ] g ∥ L 2 ≲ ∥ f ξ ∥ L ∞ ∥ g ∥ L 2 , the nonlinear difference is Lipschitz in the energy space, namely
∣ ⟨ M ∗ ( N ( F ) − N ( G ) ) , W ⟩ ∣ ≲ ( ∥ F ∥ H 2 + ∥ G ∥ H 2 ) E W ( τ ) . \left|\left\langle\mathcal{M}^*(\mathcal{N}(F)-\mathcal{N}(G)),W\right\rangle\right|\lesssim\left(\lVert F\rVert_{H^2}+\lVert G\rVert_{H^2}\right)E_W(\tau). ∣ ⟨ M ∗ ( N ( F ) − N ( G )) , W ⟩ ∣ ≲ ( ∥ F ∥ H 2 + ∥ G ∥ H 2 ) E W ( τ ) .
Therefore,
d d τ E W ( τ ) ≤ C ( 1 + ∥ F ∥ H 2 + ∥ G ∥ H 2 ) E W ( τ ) . \frac{d}{d\tau}E_W(\tau)\leq C\left(1+\lVert F\rVert_{H^2}+\lVert G\rVert_{H^2}\right)E_W(\tau). d τ d E W ( τ ) ≤ C ( 1 + ∥ F ∥ H 2 + ∥ G ∥ H 2 ) E W ( τ ) .
By Step 2, F F F and G G G are bounded in L ∞ ( [ 0 , T ] ; H 2 ( T ) ) L^\infty([0,T];H^2(\mathbb{T})) L ∞ ([ 0 , T ] ; H 2 ( T )) and E W ( 0 ) = 0 E_W(0)=0 E W ( 0 ) = 0 . Hence Grönwall’s inequality implies E W ( τ ) ≡ 0 E_W(\tau)\equiv0 E W ( τ ) ≡ 0 on [ 0 , T ] [0,T] [ 0 , T ] , so W ≡ 0 W\equiv0 W ≡ 0 and F = G F=G F = G . □
In the sequel, we establish global-in-time existence for the unidirectional model (92) under a smallness assumption on the initial datum. More precisely, we have the following result.
Theorem 5.2 (Global existence and decay for small data). Let Υ > 0 \Upsilon>0 Υ > 0 and δ , β > 0 \delta,\beta>0 δ , β > 0 . Assume that F 0 ∈ H 2 ( T ) F_0\in H^2(\mathbb{T}) F 0 ∈ H 2 ( T ) has zero spatial mean. There exists ε ∗ > 0 \varepsilon_*>0 ε ∗ > 0 , depending only on Υ , δ , β \Upsilon,\delta,\beta Υ , δ , β , such that if
∥ F 0 ∥ H 2 ( T ) ≤ ε ∗ , \lVert F_0\rVert_{H^2(\mathbb{T})}\leq\varepsilon_*, ∥ F 0 ∥ H 2 ( T ) ≤ ε ∗ , then the solution F F F given by Theorem 5.1 exists globally in time,
F ∈ C ( [ 0 , ∞ ) ; H 2 ( T ) ) , F\in C([0,\infty);H^2(\mathbb{T})), F ∈ C ([ 0 , ∞ ) ; H 2 ( T )) , Moreover, there exist constants c , C > 0 c,C>0 c , C > 0 , depending only on Υ , δ , β \Upsilon,\delta,\beta Υ , δ , β , such that for all τ ≥ 0 \tau\geq0 τ ≥ 0 ,
E ( τ ) ≤ C E ( 0 ) e − c τ , E ( τ ) : = ∥ F ∥ L 2 2 + 2 Υ ∥ F ∥ H 1 / 2 2 + Υ 2 ∥ F ∥ H 1 2 + δ 2 4 ∥ F ∥ H 2 2 . E(\tau)\leq C E(0)e^{-c\tau},\qquad E(\tau):=\lVert F\rVert_{L^2}^2+2\Upsilon\lVert F\rVert_{H^{1/2}}^2+\Upsilon^2\lVert F\rVert_{H^1}^2+\frac{\delta^2}{4}\lVert F\rVert_{H^2}^2. E ( τ ) ≤ C E ( 0 ) e − c τ , E ( τ ) := ∥ F ∥ L 2 2 + 2Υ ∥ F ∥ H 1/2 2 + Υ 2 ∥ F ∥ H 1 2 + 4 δ 2 ∥ F ∥ H 2 2 .
Proof of Theorem 5.2. Let F F F be the local H 2 ( T ) H^2(\mathbb{T}) H 2 ( T ) solution given by Theorem 5.1 . By Step 1 in the proof of Theorem 5.1 , the spatial mean is conserved, so ⟨ F ( τ ) ⟩ = 0 \langle F(\tau)\rangle=0 ⟨ F ( τ )⟩ = 0 for all times of existence. Recall the energy from Step 2,
E ( τ ) : = ∥ F ∥ L 2 2 + 2 Υ ∥ F ∥ H ˙ 1 / 2 2 + Υ 2 ∥ F ∥ H ˙ 1 2 + δ 2 4 ∥ F ∥ H ˙ 2 2 , E(\tau) := \|F\|_{L^2}^2 + 2\mathit{\Upsilon}\|F\|_{\dot{H}^{1/2}}^2 + \mathit{\Upsilon}^2\|F\|_{\dot{H}^1}^2 + \frac{\delta^2}{4}\|F\|_{\dot{H}^2}^2, E ( τ ) := ∥ F ∥ L 2 2 + 2 Υ ∥ F ∥ H ˙ 1/2 2 + Υ 2 ∥ F ∥ H ˙ 1 2 + 4 δ 2 ∥ F ∥ H ˙ 2 2 ,
and the differential inequality obtained there:
E ′ ( τ ) + β δ 8 ∥ F ( τ ) ∥ H ˙ 3 ( T ) 2 ≤ C ∥ F ( τ ) ∥ L ∞ ( T ) ∥ F ( τ ) ∥ H ˙ 2 ( T ) 2 . (173) E'(\tau) + \frac{\beta\delta}{8}\|F(\tau)\|_{\dot{H}^3(\mathbb{T})}^2 \le C\|F(\tau)\|_{L^\infty(\mathbb{T})}\|F(\tau)\|_{\dot{H}^2(\mathbb{T})}^2. \tag*{(173)} E ′ ( τ ) + 8 β δ ∥ F ( τ ) ∥ H ˙ 3 ( T ) 2 ≤ C ∥ F ( τ ) ∥ L ∞ ( T ) ∥ F ( τ ) ∥ H ˙ 2 ( T ) 2 . (173)
where C C C depends only on Υ , δ , β \mathit{\Upsilon},\delta,\beta Υ , δ , β .
Since ⟨ F ⟩ = 0 \langle F\rangle=0 ⟨ F ⟩ = 0 , all Fourier modes satisfy ∣ k ∣ ≥ 1 |k|\ge1 ∣ k ∣ ≥ 1 , hence ∥ F ∥ H ˙ 2 ≤ ∥ F ∥ H ˙ 3 \|F\|_{\dot{H}^2}\le\|F\|_{\dot{H}^3} ∥ F ∥ H ˙ 2 ≤ ∥ F ∥ H ˙ 3 . Moreover, in one dimension H 2 ( T ) ↪ L ∞ ( T ) H^2(\mathbb{T})\hookrightarrow L^\infty(\mathbb{T}) H 2 ( T ) ↪ L ∞ ( T ) and E ( τ ) ∼ ∥ F ( τ ) ∥ H 2 2 E(\tau)\sim\|F(\tau)\|_{H^2}^2 E ( τ ) ∼ ∥ F ( τ ) ∥ H 2 2 , so
∥ F ( τ ) ∥ L ∞ ≲ ∥ F ( τ ) ∥ H 2 ≲ E ( τ ) 1 / 2 . (174) \|F(\tau)\|_{L^\infty} \lesssim\|F(\tau)\|_{H^2} \lesssim E(\tau)^{1/2}. \tag*{(174)} ∥ F ( τ ) ∥ L ∞ ≲ ∥ F ( τ ) ∥ H 2 ≲ E ( τ ) 1/2 . (174)
Substituting these bounds into (173) yields
E ′ ( τ ) + β δ 8 ∥ F ∥ H ˙ 3 2 ≤ C 1 E ( τ ) 1 / 2 ∥ F ∥ H ˙ 3 2 , (175) E'(\tau) + \frac{\beta\delta}{8}\|F\|_{\dot{H}^3}^2 \le C_1 E(\tau)^{1/2}\|F\|_{\dot{H}^3}^2, \tag*{(175)} E ′ ( τ ) + 8 β δ ∥ F ∥ H ˙ 3 2 ≤ C 1 E ( τ ) 1/2 ∥ F ∥ H ˙ 3 2 , (175)
for some C 1 = C 1 ( Υ , δ , β ) C_1=C_1(\mathit{\Upsilon},\delta,\beta) C 1 = C 1 ( Υ , δ , β ) .
Choose ε ∗ > 0 \varepsilon_*>0 ε ∗ > 0 so that ∥ F 0 ∥ H 2 ≤ ε ∗ \|F_0\|_{H^2}\le\varepsilon_* ∥ F 0 ∥ H 2 ≤ ε ∗ implies C 1 E ( 0 ) 1 / 2 ≤ β δ 16 C_1E(0)^{1/2}\le\frac{\beta\delta}{16} C 1 E ( 0 ) 1/2 ≤ 16 β δ . Then, by continuity of E ( τ ) E(\tau) E ( τ ) , the bound C 1 E ( τ ) 1 / 2 ≤ β δ 16 C_1E(\tau)^{1/2}\le\frac{\beta\delta}{16} C 1 E ( τ ) 1/2 ≤ 16 β δ persists for as long as the solution exists, and (175) improves to
E ′ ( τ ) + β δ 16 ∥ F ( τ ) ∥ H ˙ 3 ( T ) 2 ≤ 0. (176) E'(\tau) + \frac{\beta\delta}{16}\|F(\tau)\|_{\dot{H}^3(\mathbb{T})}^2\le0. \tag*{(176)} E ′ ( τ ) + 16 β δ ∥ F ( τ ) ∥ H ˙ 3 ( T ) 2 ≤ 0. (176)
In particular, E ( τ ) E(\tau) E ( τ ) is nonincreasing and hence E ( τ ) ≤ E ( 0 ) E(\tau)\le E(0) E ( τ ) ≤ E ( 0 ) for all times of existence. Therefore ∥ F ( τ ) ∥ H 2 \|F(\tau)\|_{H^2} ∥ F ( τ ) ∥ H 2 stays bounded uniformly, and the continuation criterion from the local well-posedness theorem prevents finite-time breakdown; consequently the solution extends for all τ ≥ 0 \tau\ge0 τ ≥ 0 .
Finally, since ⟨ F ⟩ = 0 \langle F\rangle=0 ⟨ F ⟩ = 0 , we have the coercivity E ( τ ) ≤ C 0 ∥ F ( τ ) ∥ H ˙ 3 2 E(\tau)\le C_0\|F(\tau)\|_{\dot{H}^3}^2 E ( τ ) ≤ C 0 ∥ F ( τ ) ∥ H ˙ 3 2 for some C 0 = C 0 ( Υ , δ ) C_0=C_0(\mathit{\Upsilon},\delta) C 0 = C 0 ( Υ , δ ) , and combining with (176) gives
E ′ ( τ ) + c E ( τ ) ≤ 0 , c : = β δ 16 C 0 > 0 , E'(\tau)+cE(\tau)\le0,\qquad c:=\frac{\beta\delta}{16C_0}>0, E ′ ( τ ) + c E ( τ ) ≤ 0 , c := 16 C 0 β δ > 0 ,
which implies E ( τ ) ≤ E ( 0 ) e − c τ E(\tau)\le E(0)e^{-c\tau} E ( τ ) ≤ E ( 0 ) e − c τ .
We conclude this section by establishing an analogous well-posedness result for the second unidirectional model (97) . More precisely, we prove that
Theorem 5.3 (Global well-posedness for small H 3 H^3 H 3 data). Let Υ > 0 \mathit{\Upsilon}>0 Υ > 0 and δ , β > 0 \delta,\beta>0 δ , β > 0 . Assume that F 0 ∈ H 3 ( T ) F_0\in H^3(\mathbb{T}) F 0 ∈ H 3 ( T ) has zero spatial mean and consider the second unidirectional model (97) . There exists ε ∗ = ε ∗ ( Υ , δ , β ) > 0 \varepsilon_*=\varepsilon_*(\mathit{\Upsilon},\delta,\beta)>0 ε ∗ = ε ∗ ( Υ , δ , β ) > 0 such that, if
∥ F 0 ∥ H 3 ( T ) ≤ ε ∗ , \|F_0\|_{H^3(\mathbb{T})}\le\varepsilon_*, ∥ F 0 ∥ H 3 ( T ) ≤ ε ∗ , then the corresponding unique solution exists globally in time and
F ∈ C ( [ 0 , ∞ ) ; H 3 ( T ) ) , F ( ⋅ , 0 ) = F 0 . F\in C([0,\infty);H^3(\mathbb{T})),\qquad F(\cdot,0)=F_0. F ∈ C ([ 0 , ∞ ) ; H 3 ( T )) , F ( ⋅ , 0 ) = F 0 .
Remark 5.4. The smallness hypothesis in Theorem 5.3 is not only used to extend solutions globally. In fact, for the second unidirectional model (97) we do not know whether large-data local well-posedness holds in H 3 ( T ) H^3(\mathbb{T}) H 3 ( T ) . The reason is that the nonlinear terms are highly singular (in particular those multiplied by β \beta β and δ \delta δ ), and in our energy method they can be controlled only by a bootstrap/absorption argument that relies on ∥ F ∥ H 3 \lVert F\rVert_{H^3} ∥ F ∥ H 3 being small compared to the linear dissipation.
Proof of Theorem 5.3. The argument follows the same general strategy as the proof of Theorem 5.1 . In particular, one may construct solutions by a standard Friedrichs mollification scheme, obtain a time of existence independent of the regularization parameter via uniform a priori bounds, pass to the limit by compactness, and conclude uniqueness by an energy estimate on the difference of two solutions. To avoid repetition, we do not reproduce these classical approximation, compactness, and uniqueness steps here. Instead, we focus on the only new input for the present model, namely the a priori energy estimates. Once this estimate is established, the existence and uniqueness statements follow verbatim from the same scheme used in Theorem 5.1 .
Similarly as in Theorem 5.1 we can show that the mean zero assumption is preserved by the evolution. Indeed, assume ⟨ F 0 ⟩ = 0 \langle F_0\rangle= 0 ⟨ F 0 ⟩ = 0 and let F F F be a smooth solution of (97) . Taking spatial means, it suffices to note that H ^ ( 0 ) = 0 \widehat{\mathcal{H}}(0)=0 H ( 0 ) = 0 and that α ( Λ , ∂ ξ ) \alpha(\Lambda,\partial_\xi) α ( Λ , ∂ ξ ) carries a factor Λ \Lambda Λ (hence also annihilates the zero mode), so ( α + γ H ) g (\alpha+\gamma\mathcal{H})g ( α + γ H ) g is always mean-zero. Therefore ∂ τ ⟨ F ⟩ = 0 \partial_\tau\langle F\rangle=0 ∂ τ ⟨ F ⟩ = 0 and ⟨ F ( τ ) ⟩ ≡ 0 \langle F(\tau)\rangle\equiv0 ⟨ F ( τ )⟩ ≡ 0 .
For the a priori bounds it is convenient to work with the symmetrized formulation, rather than with (97) written through α + γ H \alpha+\gamma\mathcal{H} α + γ H . Throughout the proof we fix ε = 1 \varepsilon=1 ε = 1 . Set
A : = I + γ Λ , M 1 ∗ : = − 2 A ∂ ξ + δ Λ ∂ ξ 2 , M 1 ∗ M 1 = δ 2 Λ 2 ∂ ξ 4 − 4 A 2 ∂ ξ 2 . (177) A:=I+\gamma\Lambda,\qquad\mathcal{M}_1^*:=-2A\partial_\xi+\delta\Lambda\partial_\xi^2,\qquad\mathcal{M}_1^*\mathcal{M}_1=\delta^2\Lambda^2\partial_\xi^4-4A^2\partial_\xi^2. \tag*{(177)} A := I + γ Λ , M 1 ∗ := − 2 A ∂ ξ + δ Λ ∂ ξ 2 , M 1 ∗ M 1 = δ 2 Λ 2 ∂ ξ 4 − 4 A 2 ∂ ξ 2 . (177)
We may rewrite the second unidirectional model (97) in the symmetrized form
( M 1 ∗ M 1 ) F τ = M 1 ∗ ( A F ξ ξ + β 4 Λ F ξ ξ ξ + δ Λ F ξ ξ ξ + Λ F ) − M 1 ∗ N 1 ( F ) , (178) (\mathcal{M}_1^*\mathcal{M}_1)F_\tau=\mathcal{M}_1^*\left(AF_{\xi\xi}+\frac{\beta}{4}\Lambda F_{\xi\xi\xi}+\delta\Lambda F_{\xi\xi\xi}+\Lambda F\right)-\mathcal{M}_1^*\mathcal{N}_1(F), \tag*{(178)} ( M 1 ∗ M 1 ) F τ = M 1 ∗ ( A F ξ ξ + 4 β Λ F ξ ξ ξ + δ Λ F ξ ξ ξ + Λ F ) − M 1 ∗ N 1 ( F ) , (178)
where the nonlinear term N 1 ( F ) \mathcal{N}_1(F) N 1 ( F ) is given by
N 1 ( F ) : = 1 2 Λ ( F ξ 2 − ( Λ F ) 2 ) − [ Λ , F ξ ] F ξ + F ξ ξ Λ F + [ Λ , F ] T F − F ξ H ( T F ) \mathcal{N}_1(F):=\frac{1}{2}\Lambda\left(F_\xi^2-(\Lambda F)^2\right)-[\Lambda,F_\xi]F_\xi+F_{\xi\xi}\Lambda F+[\Lambda,F]\mathcal{T}F-F_\xi\mathcal{H}(\mathcal{T}F) N 1 ( F ) := 2 1 Λ ( F ξ 2 − ( Λ F ) 2 ) − [ Λ , F ξ ] F ξ + F ξ ξ Λ F + [ Λ , F ] T F − F ξ H ( T F )
+ β 4 ( [ Λ , F ] T F ξ ξ ξ − F ξ Λ ( T F ξ ξ ) ) + δ ( [ Λ , F ] T F ξ ξ − F ξ Λ ( T F ξ ) ) . (179) \qquad+\frac{\beta}{4}\left([\Lambda,F]\mathcal{T}F_{\xi\xi\xi}-F_\xi\Lambda(\mathcal{T}F_{\xi\xi})\right)+\delta\left([\Lambda,F]\mathcal{T}F_{\xi\xi}-F_\xi\Lambda(\mathcal{T}F_\xi)\right). \tag*{(179)} + 4 β ( [ Λ , F ] T F ξ ξ ξ − F ξ Λ ( T F ξ ξ ) ) + δ ( [ Λ , F ] T F ξ ξ − F ξ Λ ( T F ξ ) ) . (179)
Take the L 2 ( T ) L^2(\mathbb{T}) L 2 ( T ) inner product of (178) with F F F . Using that A A A and Λ \Lambda Λ are self-adjoint on L 2 ( T ) L^2(\mathbb{T}) L 2 ( T ) and commute with ∂ ξ \partial_\xi ∂ ξ , and integrating by parts we find that
d d τ ( 2 ∥ A F ξ ∥ L 2 2 + δ 2 2 ∥ F ∥ H 3 ( T ) 2 ) = ⟨ M 1 ∗ L 1 ( F ) , F ⟩ − ⟨ M 1 ∗ N 1 ( F ) , F ⟩ (180) \frac{d}{d\tau}\left(2\lVert AF_\xi\rVert_{L^2}^2+\frac{\delta^2}{2}\lVert F\rVert_{H^3(\mathbb{T})}^2\right)=\langle\mathcal{M}_1^*\mathcal{L}_1(F),F\rangle-\langle\mathcal{M}_1^*\mathcal{N}_1(F),F\rangle \tag*{(180)} d τ d ( 2 ∥ A F ξ ∥ L 2 2 + 2 δ 2 ∥ F ∥ H 3 ( T ) 2 ) = ⟨ M 1 ∗ L 1 ( F ) , F ⟩ − ⟨ M 1 ∗ N 1 ( F ) , F ⟩ (180)
where
L 1 ( F ) : = A F ξ ξ + β 4 Λ F ξ ξ ξ + δ Λ F ξ ξ + Λ F . \mathcal{L}_1(F):=AF_{\xi\xi}+\frac{\beta}{4}\Lambda F_{\xi\xi\xi}+\delta\Lambda F_{\xi\xi}+\Lambda F. L 1 ( F ) := A F ξ ξ + 4 β Λ F ξ ξ ξ + δ Λ F ξ ξ + Λ F .
Computing the linear terms of the right hand side of (180) we readily check that
− ⟨ M 1 ∗ L 1 ( F ) , F ⟩ = ∑ i = 1 8 J i , (181) -\langle\mathcal{M}_1^* \mathcal{L}_1(F),F\rangle=\sum_{i=1}^{8}J_i, \tag*{(181)} − ⟨ M 1 ∗ L 1 ( F ) , F ⟩ = i = 1 ∑ 8 J i , (181)
with
J 1 : = 2 ∫ T A ( F F ξ ξ ) F ξ d ξ , J 2 : = δ ∫ T Λ ( F F ξ ξ ) F ξ ξ d ξ , J 3 : = β 2 ∫ T A ( Λ F ξ ξ ξ ξ ) F ξ d ξ , J 4 : = δ β 4 ∫ T Λ ( Λ F ξ ξ ξ ξ ) F ξ ξ d ξ , J 5 : = 2 δ ∫ T A ( Λ F ξ ξ ξ ) F ξ d ξ , J 6 : = δ 2 ∫ T Λ ( Λ F ξ ξ ξ ) F ξ ξ d ξ , J 7 : = 2 ∫ T A ( Λ F ) F ξ d ξ , J 8 : = δ ∫ T Λ ( Λ F ) F ξ ξ d ξ . \begin{aligned}
J_1&:=2\int_{\mathbb{T}}A(FF_{\xi\xi})F_\xi\,\mathrm{d}\xi, &
J_2&:=\delta\int_{\mathbb{T}}\Lambda(FF_{\xi\xi})F_{\xi\xi}\,\mathrm{d}\xi,\\
J_3&:=\frac{\beta}{2}\int_{\mathbb{T}}A(\Lambda F_{\xi\xi\xi\xi})F_\xi\,\mathrm{d}\xi, &
J_4&:=\frac{\delta\beta}{4}\int_{\mathbb{T}}\Lambda(\Lambda F_{\xi\xi\xi\xi})F_{\xi\xi}\,\mathrm{d}\xi,\\
J_5&:=2\delta\int_{\mathbb{T}}A(\Lambda F_{\xi\xi\xi})F_\xi\,\mathrm{d}\xi, &
J_6&:=\delta^2\int_{\mathbb{T}}\Lambda(\Lambda F_{\xi\xi\xi})F_{\xi\xi}\,\mathrm{d}\xi,\\
J_7&:=2\int_{\mathbb{T}}A(\Lambda F)F_\xi\,\mathrm{d}\xi, &
J_8&:=\delta\int_{\mathbb{T}}\Lambda(\Lambda F)F_{\xi\xi}\,\mathrm{d}\xi.
\end{aligned} J 1 J 3 J 5 J 7 := 2 ∫ T A ( F F ξ ξ ) F ξ d ξ , := 2 β ∫ T A ( Λ F ξ ξ ξ ξ ) F ξ d ξ , := 2 δ ∫ T A ( Λ F ξ ξ ξ ) F ξ d ξ , := 2 ∫ T A ( Λ F ) F ξ d ξ , J 2 J 4 J 6 J 8 := δ ∫ T Λ ( F F ξ ξ ) F ξ ξ d ξ , := 4 δ β ∫ T Λ ( Λ F ξ ξ ξ ξ ) F ξ ξ d ξ , := δ 2 ∫ T Λ ( Λ F ξ ξ ξ ) F ξ ξ d ξ , := δ ∫ T Λ ( Λ F ) F ξ ξ d ξ .
Therefore, using once again that A A A and Λ \Lambda Λ are self-adjoint on L 2 ( T ) L^2(\mathbb{T}) L 2 ( T ) obtain that
J 1 = J 3 = J 6 = J 7 = 0. J_1=J_3=J_6=J_7=0. J 1 = J 3 = J 6 = J 7 = 0.
Moreover, we also find that
J 2 + J 5 = − δ ∥ A 1 / 2 F ∥ H 5 / 2 2 , J 4 = − δ β 4 ∥ F ∥ H 4 2 , J 8 = − δ ∥ F ∥ H 2 2 . J_2+J_5=-\delta\left\|A^{1/2}F\right\|_{H^{5/2}}^2,\qquad J_4=-\frac{\delta\beta}{4}\|F\|_{H^4}^2,\qquad J_8=-\delta\|F\|_{H^2}^2. J 2 + J 5 = − δ A 1/2 F H 5/2 2 , J 4 = − 4 δ β ∥ F ∥ H 4 2 , J 8 = − δ ∥ F ∥ H 2 2 .
Thus, noticing that
∥ A F ξ ∥ L 2 2 = ∥ F ∥ H 1 2 + 2 γ ∥ F ∥ H 3 / 2 2 + γ 2 ∥ F ∥ H 2 2 (182) \|AF_{\xi}\|_{L^2}^2=\|F\|_{H^1}^2+2\gamma\|F\|_{H^{3/2}}^2+\gamma^2\|F\|_{H^2}^2 \tag*{(182)} ∥ A F ξ ∥ L 2 2 = ∥ F ∥ H 1 2 + 2 γ ∥ F ∥ H 3/2 2 + γ 2 ∥ F ∥ H 2 2 (182)
and denoting by
E ( τ ) = ∥ F ∥ H 1 2 + 2 γ ∥ F ∥ H 3 / 2 2 + γ 2 ∥ F ∥ H 2 2 + δ 2 4 ∥ F ∥ H 3 ( T ) 2 \mathcal{E}(\tau)=\|F\|_{H^1}^2+2\gamma\|F\|_{H^{3/2}}^2+\gamma^2\|F\|_{H^2}^2+\frac{\delta^2}{4}\|F\|_{H^3(\mathbb{T})}^2 E ( τ ) = ∥ F ∥ H 1 2 + 2 γ ∥ F ∥ H 3/2 2 + γ 2 ∥ F ∥ H 2 2 + 4 δ 2 ∥ F ∥ H 3 ( T ) 2
we find the energy estimate
d d τ E ( τ ) + δ 2 ∥ A 1 / 2 F ∥ H 5 / 2 2 + δ β 8 ∥ F ∥ H 4 2 + δ 2 ∥ F ∥ H 2 2 = − 1 2 ⟨ M 1 ∗ N 1 ( F ) , F ⟩ (183) \frac{\mathrm{d}}{\mathrm{d}\tau}\mathcal{E}(\tau)+\frac{\delta}{2}\left\|A^{1/2}F\right\|_{H^{5/2}}^2+\frac{\delta\beta}{8}\|F\|_{H^4}^2+\frac{\delta}{2}\|F\|_{H^2}^2=-\frac{1}{2}\langle\mathcal{M}_1^*\mathcal{N}_1(F),F\rangle \tag*{(183)} d τ d E ( τ ) + 2 δ A 1/2 F H 5/2 2 + 8 δ β ∥ F ∥ H 4 2 + 2 δ ∥ F ∥ H 2 2 = − 2 1 ⟨ M 1 ∗ N 1 ( F ) , F ⟩ (183)
Next, we need to estimate the non-linear terms. More precisely, we have to control
K j = 2 ∫ T A ∂ ξ ( N 1 , j ( F ) ) F d ξ − δ ∫ T Λ ∂ ξ 2 ( N 1 , j ( F ) ) F d ξ . K_j=2\int_{\mathbb{T}}A\partial_\xi(\mathcal{N}_{1,j}(F))F\,\mathrm{d}\xi-\delta\int_{\mathbb{T}}\Lambda\partial_\xi^2(\mathcal{N}_{1,j}(F))F\,\mathrm{d}\xi. K j = 2 ∫ T A ∂ ξ ( N 1 , j ( F )) F d ξ − δ ∫ T Λ ∂ ξ 2 ( N 1 , j ( F )) F d ξ .
where
K 1 = 2 ∫ T A ∂ ξ [ 1 2 Λ ( F ξ 2 − ( Λ F ) 2 ) ] F d ξ − δ ∫ T Λ ∂ ξ 2 [ 1 2 Λ ( F ξ 2 − ( Λ F ) 2 ) ] F d ξ , K 2 = 2 ∫ T A ∂ ξ [ − [ Λ , F ξ ] F ξ ] F d ξ − δ ∫ T Λ ∂ ξ 2 [ − [ Λ , F ξ ] F ξ ] F d ξ , K 3 = 2 ∫ T A ∂ ξ [ F ξ ξ Λ F ] F d ξ − δ ∫ T Λ ∂ ξ 2 [ F ξ ξ Λ F ] F d ξ , K 4 = 2 ∫ T A ∂ ξ [ [ Λ , F ] T F ] F d ξ − δ ∫ T Λ ∂ ξ 2 [ [ Λ , F ] T F ] F d ξ , K 5 = 2 ∫ T A ∂ ξ [ − F ξ H ( T F ) ] F d ξ − δ ∫ T Λ ∂ ξ 2 [ − F ξ H ( T F ) ] F d ξ , K 6 = 2 ∫ T A ∂ ξ [ β 4 ( [ Λ , F ] T F ξ ξ ξ ξ − F ξ Λ ( T F ξ ξ ξ ) ) ] F d ξ − δ ∫ T Λ ∂ ξ 2 [ β 4 ( [ Λ , F ] T F ξ ξ ξ ξ − F ξ Λ ( T F ξ ξ ξ ) ) ] F d ξ , K 7 = 2 ∫ T A ∂ ξ [ δ ( [ Λ , F ] T F ξ ξ ξ − F ξ Λ ( T F ξ ξ ) ) ] F d ξ − δ ∫ T Λ ∂ ξ 2 [ δ ( [ Λ , F ] T F ξ ξ ξ − F ξ Λ ( T F ξ ξ ) ) ] F d ξ . \begin{aligned}
K_1 &= 2\int_{\mathbb{T}} A\,\partial_\xi\left[\frac{1}{2}\Lambda\left(F_\xi^2-(\Lambda F)^2\right)\right]F\,\mathrm{d}\xi-\delta\int_{\mathbb{T}}\Lambda\partial_\xi^2\left[\frac{1}{2}\Lambda\left(F_\xi^2-(\Lambda F)^2\right)\right]F\,\mathrm{d}\xi,\\
K_2 &= 2\int_{\mathbb{T}} A\,\partial_\xi\left[-[\Lambda,F_\xi]F_\xi\right]F\,\mathrm{d}\xi-\delta\int_{\mathbb{T}}\Lambda\partial_\xi^2\left[-[\Lambda,F_\xi]F_\xi\right]F\,\mathrm{d}\xi,\\
K_3 &= 2\int_{\mathbb{T}} A\,\partial_\xi[F_{\xi\xi}\Lambda F]F\,\mathrm{d}\xi-\delta\int_{\mathbb{T}}\Lambda\partial_\xi^2[F_{\xi\xi}\Lambda F]F\,\mathrm{d}\xi,\\
K_4 &= 2\int_{\mathbb{T}} A\,\partial_\xi[[\Lambda,F]\mathcal{T}F]F\,\mathrm{d}\xi-\delta\int_{\mathbb{T}}\Lambda\partial_\xi^2[[\Lambda,F]\mathcal{T}F]F\,\mathrm{d}\xi,\\
K_5 &= 2\int_{\mathbb{T}} A\,\partial_\xi[-F_\xi\mathcal{H}(\mathcal{T}F)]F\,\mathrm{d}\xi-\delta\int_{\mathbb{T}}\Lambda\partial_\xi^2[-F_\xi\mathcal{H}(\mathcal{T}F)]F\,\mathrm{d}\xi,\\
K_6 &= 2\int_{\mathbb{T}} A\,\partial_\xi\left[\frac{\beta}{4}\left([\Lambda,F]\mathcal{T}F_{\xi\xi\xi\xi}-F_\xi\Lambda(\mathcal{T}F_{\xi\xi\xi})\right)\right]F\,\mathrm{d}\xi\\
&\quad-\delta\int_{\mathbb{T}}\Lambda\partial_\xi^2\left[\frac{\beta}{4}\left([\Lambda,F]\mathcal{T}F_{\xi\xi\xi\xi}-F_\xi\Lambda(\mathcal{T}F_{\xi\xi\xi})\right)\right]F\,\mathrm{d}\xi,\\
K_7 &= 2\int_{\mathbb{T}} A\,\partial_\xi\left[\delta\left([\Lambda,F]\mathcal{T}F_{\xi\xi\xi}-F_\xi\Lambda(\mathcal{T}F_{\xi\xi})\right)\right]F\,\mathrm{d}\xi\\
&\quad-\delta\int_{\mathbb{T}}\Lambda\partial_\xi^2\left[\delta\left([\Lambda,F]\mathcal{T}F_{\xi\xi\xi}-F_\xi\Lambda(\mathcal{T}F_{\xi\xi})\right)\right]F\,\mathrm{d}\xi.
\end{aligned} K 1 K 2 K 3 K 4 K 5 K 6 K 7 = 2 ∫ T A ∂ ξ [ 2 1 Λ ( F ξ 2 − ( Λ F ) 2 ) ] F d ξ − δ ∫ T Λ ∂ ξ 2 [ 2 1 Λ ( F ξ 2 − ( Λ F ) 2 ) ] F d ξ , = 2 ∫ T A ∂ ξ [ − [ Λ , F ξ ] F ξ ] F d ξ − δ ∫ T Λ ∂ ξ 2 [ − [ Λ , F ξ ] F ξ ] F d ξ , = 2 ∫ T A ∂ ξ [ F ξ ξ Λ F ] F d ξ − δ ∫ T Λ ∂ ξ 2 [ F ξ ξ Λ F ] F d ξ , = 2 ∫ T A ∂ ξ [[ Λ , F ] T F ] F d ξ − δ ∫ T Λ ∂ ξ 2 [[ Λ , F ] T F ] F d ξ , = 2 ∫ T A ∂ ξ [ − F ξ H ( T F )] F d ξ − δ ∫ T Λ ∂ ξ 2 [ − F ξ H ( T F )] F d ξ , = 2 ∫ T A ∂ ξ [ 4 β ( [ Λ , F ] T F ξ ξ ξ ξ − F ξ Λ ( T F ξ ξ ξ ) ) ] F d ξ − δ ∫ T Λ ∂ ξ 2 [ 4 β ( [ Λ , F ] T F ξ ξ ξ ξ − F ξ Λ ( T F ξ ξ ξ ) ) ] F d ξ , = 2 ∫ T A ∂ ξ [ δ ( [ Λ , F ] T F ξ ξ ξ − F ξ Λ ( T F ξ ξ ) ) ] F d ξ − δ ∫ T Λ ∂ ξ 2 [ δ ( [ Λ , F ] T F ξ ξ ξ − F ξ Λ ( T F ξ ξ ) ) ] F d ξ .
Using the algebra property (12) , estimate (1.15) and Sobolev embedding (1.11)` we find that
∣ K 1 ∣ ≲ ( ∥ F ξ ∥ H ˙ 1 2 + ∥ Λ F ∥ H ˙ 1 2 ) ( ∥ A F ξ ∥ L 2 + ∥ F ∥ H ˙ 3 ) ≲ E 3 2 ( τ ) , ∣ K 2 ∣ ≲ ( ∥ F ξ ∥ H ˙ 1 2 + ∥ F ξ ∥ L ∞ ∥ A F ξ ∥ L 2 ) ( ∥ A F ξ ∥ L 2 + ∥ F ∥ H ˙ 3 ) ≲ E 3 2 ( τ ) , ∣ K 3 ∣ ≲ ( ∥ F ∥ H ˙ 2 ∥ Δ F ∥ L ∞ ) ( ∥ A F ξ ∥ L 2 + ∥ F ∥ H ˙ 3 ) ≲ E 3 2 ( τ ) , ∣ K 4 ∣ ≲ ( ∥ F ∥ H ˙ 1 ∥ T F ∥ H ˙ 1 + ∥ F ∥ L ∞ ∥ Λ T F ∥ L 2 ) ( ∥ A F ξ ∥ L 2 + ∥ F ∥ H ˙ 3 ) ≲ E 3 2 ( τ ) , ∣ K 5 ∣ ≲ ( ∥ F ξ ∥ L ∞ ∥ H T F ∥ L 2 ) ( ∥ A F ξ ∥ L 2 + ∥ F ∥ H ˙ 3 ) ≲ E 3 2 ( τ ) . (184) \begin{aligned}
|K_1| &\lesssim\left(\|F_\xi\|_{\dot{H}^1}^2+\|\Lambda F\|_{\dot{H}^1}^2\right)\left(\|AF_\xi\|_{L^2}+\|F\|_{\dot{H}^3}\right)\lesssim\mathcal{E}^{\frac{3}{2}}(\tau),\\
|K_2| &\lesssim\left(\|F_\xi\|_{\dot{H}^1}^2+\|F_\xi\|_{L^\infty}\|AF_\xi\|_{L^2}\right)\left(\|AF_\xi\|_{L^2}+\|F\|_{\dot{H}^3}\right)\lesssim\mathcal{E}^{\frac{3}{2}}(\tau),\\
|K_3| &\lesssim\left(\|F\|_{\dot{H}^2}\|\Delta F\|_{L^\infty}\right)\left(\|AF_\xi\|_{L^2}+\|F\|_{\dot{H}^3}\right)\lesssim\mathcal{E}^{\frac{3}{2}}(\tau),\\
|K_4| &\lesssim\left(\|F\|_{\dot{H}^1}\|\mathcal{T}F\|_{\dot{H}^1}+\|F\|_{L^\infty}\|\Lambda\mathcal{T}F\|_{L^2}\right)\left(\|AF_\xi\|_{L^2}+\|F\|_{\dot{H}^3}\right)\lesssim\mathcal{E}^{\frac{3}{2}}(\tau),\\
|K_5| &\lesssim\left(\|F_\xi\|_{L^\infty}\|\mathcal{H}\mathcal{T}F\|_{L^2}\right)\left(\|AF_\xi\|_{L^2}+\|F\|_{\dot{H}^3}\right)\lesssim\mathcal{E}^{\frac{3}{2}}(\tau). \tag*{(184)}
\end{aligned} ∣ K 1 ∣ ∣ K 2 ∣ ∣ K 3 ∣ ∣ K 4 ∣ ∣ K 5 ∣ ≲ ( ∥ F ξ ∥ H ˙ 1 2 + ∥Λ F ∥ H ˙ 1 2 ) ( ∥ A F ξ ∥ L 2 + ∥ F ∥ H ˙ 3 ) ≲ E 2 3 ( τ ) , ≲ ( ∥ F ξ ∥ H ˙ 1 2 + ∥ F ξ ∥ L ∞ ∥ A F ξ ∥ L 2 ) ( ∥ A F ξ ∥ L 2 + ∥ F ∥ H ˙ 3 ) ≲ E 2 3 ( τ ) , ≲ ( ∥ F ∥ H ˙ 2 ∥Δ F ∥ L ∞ ) ( ∥ A F ξ ∥ L 2 + ∥ F ∥ H ˙ 3 ) ≲ E 2 3 ( τ ) , ≲ ( ∥ F ∥ H ˙ 1 ∥ T F ∥ H ˙ 1 + ∥ F ∥ L ∞ ∥Λ T F ∥ L 2 ) ( ∥ A F ξ ∥ L 2 + ∥ F ∥ H ˙ 3 ) ≲ E 2 3 ( τ ) , ≲ ( ∥ F ξ ∥ L ∞ ∥ H T F ∥ L 2 ) ( ∥ A F ξ ∥ L 2 + ∥ F ∥ H ˙ 3 ) ≲ E 2 3 ( τ ) . (184)
To conclude a priori estimates, we are left with K 6 , K 7 K_6,K_7 K 6 , K 7 which are the more singular terms. Expanding the commutator and integrating by parts, we readily check that
K 6 = 2 ∫ T [ β 4 ( Λ ( T F F ξ ξ ξ ξ ) − F Λ ( T F ξ ξ ξ ) − F ξ Λ ( T F ξ ξ ) ) ] A ∂ ξ F d ξ − δ ∫ T [ β 4 ( Λ ( T F F ξ ξ ξ ξ ) − F Λ ( T F ξ ξ ξ ) − F ξ Λ ( T F ξ ξ ) ) ] Λ ∂ ξ 2 F d ξ . \begin{aligned}
K_6={}&2\int_{\mathbb{T}}\left[\frac{\beta}{4}\left(\Lambda(\mathcal{T}F F_{\xi\xi\xi\xi})-F\Lambda(\mathcal{T}F_{\xi\xi\xi})-F_\xi\Lambda(\mathcal{T}F_{\xi\xi})\right)\right]A\,\partial_\xi F\,\mathrm{d}\xi\\
&-\delta\int_{\mathbb{T}}\left[\frac{\beta}{4}\left(\Lambda(\mathcal{T}F F_{\xi\xi\xi\xi})-F\Lambda(\mathcal{T}F_{\xi\xi\xi})-F_\xi\Lambda(\mathcal{T}F_{\xi\xi})\right)\right]\Lambda\partial_\xi^2F\,\mathrm{d}\xi.
\end{aligned} K 6 = 2 ∫ T [ 4 β ( Λ ( T F F ξ ξ ξ ξ ) − F Λ ( T F ξ ξ ξ ) − F ξ Λ ( T F ξ ξ ) ) ] A ∂ ξ F d ξ − δ ∫ T [ 4 β ( Λ ( T F F ξ ξ ξ ξ ) − F Λ ( T F ξ ξ ξ ) − F ξ Λ ( T F ξ ξ ) ) ] Λ ∂ ξ 2 F d ξ .
Rather than estimating these integrals directly, we first integrate by parts once more in the first two contributions. This reveals the appropriate derivative distribution needed to close the estimate.
Indeed, we can write K 6 = K 6 , A + K 6 , δ K_6=K_{6,\mathrm{A}}+K_{6,\delta} K 6 = K 6 , A + K 6 , δ where
K 6 , A = − β 2 ∫ T F T F ξ ξ ξ Λ ∂ ξ F d ξ − β 2 ∫ T F ξ ξ ξ Λ ( F ∂ ξ F ) d ξ − β 2 ∫ T F ξ Λ ( T F ξ ξ ) A ∂ ξ F d ξ K_{6,\mathrm{A}}=-\frac{\beta}{2}\int_{\mathbb{T}}F\,TF_{\xi\xi\xi}\Lambda\partial_{\xi}F\,d\xi-\frac{\beta}{2}\int_{\mathbb{T}}F_{\xi\xi\xi}\Lambda(F\partial_{\xi}F)\,d\xi-\frac{\beta}{2}\int_{\mathbb{T}}F_{\xi}\Lambda(TF_{\xi\xi})A\partial_{\xi}F\,d\xi K 6 , A = − 2 β ∫ T F T F ξ ξ ξ Λ ∂ ξ F d ξ − 2 β ∫ T F ξ ξ ξ Λ ( F ∂ ξ F ) d ξ − 2 β ∫ T F ξ Λ ( T F ξ ξ ) A ∂ ξ F d ξ
and
K 6 , δ = β δ 4 ∫ T F T F ξ ξ ξ Λ 4 F d ξ + β δ 4 ∫ T T F ξ ξ ξ Λ ( F Λ ∂ ξ 2 F ) d ξ + β δ 4 ∫ T F ξ Λ ( T F ξ ξ ) Λ ∂ ξ 2 F d ξ . K_{6,\delta}=\frac{\beta\delta}{4}\int_{\mathbb{T}}F\,TF_{\xi\xi\xi}\Lambda^4F\,d\xi+\frac{\beta\delta}{4}\int_{\mathbb{T}}TF_{\xi\xi\xi}\Lambda(F\Lambda\partial_{\xi}^2F)\,d\xi+\frac{\beta\delta}{4}\int_{\mathbb{T}}F_{\xi}\Lambda(TF_{\xi\xi})\Lambda\partial_{\xi}^2F\,d\xi. K 6 , δ = 4 β δ ∫ T F T F ξ ξ ξ Λ 4 F d ξ + 4 β δ ∫ T T F ξ ξ ξ Λ ( F Λ ∂ ξ 2 F ) d ξ + 4 β δ ∫ T F ξ Λ ( T F ξ ξ ) Λ ∂ ξ 2 F d ξ .
Therefore, using Cauchy-Schwarz, the algebra property (12) and estimate (15) we conclude that
∣ K 6 , A ∣ ≲ ∥ F ∥ L ∞ ∥ T F ∥ H ˙ 4 ∥ A F ξ ∥ L 2 + ∥ T F ∥ H ˙ 4 ∥ F ∥ H ˙ 1 ∥ A F ξ ∥ H ˙ 1 + ∥ F ξ ξ ∥ L ∞ ∥ T F ∥ H ˙ 4 ∥ A F ξ ∥ L 2 |K_{6,\mathrm{A}}|\lesssim\|F\|_{L^\infty}\|TF\|_{\dot H^4}\|A F_{\xi}\|_{L^2}+\|TF\|_{\dot H^4}\|F\|_{\dot H^1}\|A F_{\xi}\|_{\dot H^1}+\|F_{\xi\xi}\|_{L^\infty}\|TF\|_{\dot H^4}\|A F_{\xi}\|_{L^2} ∣ K 6 , A ∣ ≲ ∥ F ∥ L ∞ ∥ T F ∥ H ˙ 4 ∥ A F ξ ∥ L 2 + ∥ T F ∥ H ˙ 4 ∥ F ∥ H ˙ 1 ∥ A F ξ ∥ H ˙ 1 + ∥ F ξ ξ ∥ L ∞ ∥ T F ∥ H ˙ 4 ∥ A F ξ ∥ L 2
and
∣ K 6 , δ ∣ ≲ ∥ F ∥ L ∞ ∥ T F ∥ H ˙ 4 2 + ∥ T F ∥ H ˙ 4 2 ∥ F ∥ H ˙ 1 + ∥ F ξ ξ ∥ L ∞ ∥ T F ∥ H ˙ 4 ∥ F ∥ H ˙ 3 . |K_{6,\delta}|\lesssim\|F\|_{L^\infty}\|TF\|_{\dot H^4}^2+\|TF\|_{\dot H^4}^2\|F\|_{\dot H^1}+\|F_{\xi\xi}\|_{L^\infty}\|TF\|_{\dot H^4}\|F\|_{\dot H^3}. ∣ K 6 , δ ∣ ≲ ∥ F ∥ L ∞ ∥ T F ∥ H ˙ 4 2 + ∥ T F ∥ H ˙ 4 2 ∥ F ∥ H ˙ 1 + ∥ F ξ ξ ∥ L ∞ ∥ T F ∥ H ˙ 4 ∥ F ∥ H ˙ 3 .
Combining both estimates, recalling (182) and using the Sobolev embedding (11) yields
∣ K 6 ∣ ≤ ∣ K 6 , A ∣ + ∣ K 6 , δ ∣ ≲ E ( τ ) ∥ F ∥ H ˙ 4 2 (185) |K_6|\leq|K_{6,\mathrm{A}}|+|K_{6,\delta}|\lesssim\sqrt{\mathcal{E}(\tau)}\|F\|_{\dot H^4}^2 \tag*{(185)} ∣ K 6 ∣ ≤ ∣ K 6 , A ∣ + ∣ K 6 , δ ∣ ≲ E ( τ ) ∥ F ∥ H ˙ 4 2 (185)
The same estimate holds for the term K 7 K_7 K 7 . More precisely, by the same argument we have that
∣ K 7 ∣ ≲ E ( τ ) ∥ F ∥ H ˙ 4 2 (186) |K_7|\lesssim\sqrt{\mathcal{E}(\tau)}\|F\|_{\dot H^4}^2 \tag*{(186)} ∣ K 7 ∣ ≲ E ( τ ) ∥ F ∥ H ˙ 4 2 (186)
Therefore, collecting estimates (183) , (184) , (185) and (186) we have shown that
d d τ E ( τ ) + δ 2 ∥ A 1 / 2 F ∥ H ˙ 5 / 2 2 + δ β 8 ∥ F ∥ H ˙ 4 2 + δ 2 ∥ F ∥ H ˙ 2 2 ≤ C 1 E ( τ ) 3 / 2 + C 2 E ( τ ) ∥ F ∥ H ˙ 4 2 (187) \frac{d}{d\tau}\mathcal{E}(\tau)+\frac{\delta}{2}\left\|A^{1/2}F\right\|_{\dot H^{5/2}}^2+\frac{\delta\beta}{8}\|F\|_{\dot H^4}^2+\frac{\delta}{2}\|F\|_{\dot H^2}^2\leq C_1\mathcal{E}(\tau)^{3/2}+C_2\sqrt{\mathcal{E}(\tau)}\|F\|_{\dot H^4}^2 \tag*{(187)} d τ d E ( τ ) + 2 δ A 1/2 F H ˙ 5/2 2 + 8 δ β ∥ F ∥ H ˙ 4 2 + 2 δ ∥ F ∥ H ˙ 2 2 ≤ C 1 E ( τ ) 3/2 + C 2 E ( τ ) ∥ F ∥ H ˙ 4 2 (187)
Define the bootstrap time
T ∗ : = sup { T > 0 : sup 0 ≤ τ ≤ T E ( τ ) ≤ 4 E ( 0 ) } . T^*:=\sup\left\{T>0:\ \sup_{0\leq\tau\leq T}\mathcal{E}(\tau)\leq4\mathcal{E}(0)\right\}. T ∗ := sup { T > 0 : 0 ≤ τ ≤ T sup E ( τ ) ≤ 4 E ( 0 ) } .
On [ 0 , T ∗ ] [0,T^*] [ 0 , T ∗ ] we have E ( τ ) ≤ 2 E ( 0 ) \sqrt{\mathcal{E}(\tau)}\leq2\sqrt{\mathcal{E}(0)} E ( τ ) ≤ 2 E ( 0 ) , hence
C 2 E ( τ ) ∥ F ∥ H ˙ 4 2 ≤ 2 C 2 E ( 0 ) ∥ F ∥ H ˙ 4 2 . C_2\sqrt{\mathcal{E}(\tau)}\|F\|_{\dot H^4}^2\leq2C_2\sqrt{\mathcal{E}(0)}\|F\|_{\dot H^4}^2. C 2 E ( τ ) ∥ F ∥ H ˙ 4 2 ≤ 2 C 2 E ( 0 ) ∥ F ∥ H ˙ 4 2 .
Assume the smallness condition
E ( 0 ) ≤ ε ∗ : = δ β 32 C 2 . (188) \sqrt{\mathcal{E}(0)}\leq\varepsilon_*:=\frac{\delta\beta}{32C_2}. \tag*{(188)} E ( 0 ) ≤ ε ∗ := 32 C 2 δ β . (188)
Then on [ 0 , T ∗ ] [0,T^*] [ 0 , T ∗ ] the last term in (187) is absorbed by δ β 8 ∥ F ∥ H ˙ 4 2 \frac{\delta\beta}{8}\|F\|_{\dot H^4}^2 8 δ β ∥ F ∥ H ˙ 4 2 , and we obtain
E ′ ( τ ) + δ 2 ∥ A 1 / 2 F ∥ H ˙ 5 / 2 2 + δ β 16 ∥ F ∥ H ˙ 4 2 + δ 2 ∥ F ∥ H ˙ 2 2 ≤ C 1 E ( τ ) 3 / 2 , τ ∈ [ 0 , T ∗ ] . (189) \mathcal{E}'(\tau)+\frac{\delta}{2}\left\|A^{1/2}F\right\|_{\dot H^{5/2}}^2+\frac{\delta\beta}{16}\|F\|_{\dot H^4}^2+\frac{\delta}{2}\|F\|_{\dot H^2}^2\leq C_1\mathcal{E}(\tau)^{3/2},\qquad\tau\in[0,T^*]. \tag*{(189)} E ′ ( τ ) + 2 δ A 1/2 F H ˙ 5/2 2 + 16 δ β ∥ F ∥ H ˙ 4 2 + 2 δ ∥ F ∥ H ˙ 2 2 ≤ C 1 E ( τ ) 3/2 , τ ∈ [ 0 , T ∗ ] . (189)
Dropping the nonnegative dissipation terms yields E ′ ( τ ) ≤ C 1 E ( τ ) 3 / 2 \mathcal{E}'(\tau)\leq C_1\mathcal{E}(\tau)^{3/2} E ′ ( τ ) ≤ C 1 E ( τ ) 3/2 , so that
E ( τ ) − 1 / 2 ≥ E ( 0 ) − 1 / 2 − C 1 2 τ , τ ∈ [ 0 , T ∗ ] . \mathcal{E}(\tau)^{-1/2}\geq\mathcal{E}(0)^{-1/2}-\frac{C_1}{2}\tau,\qquad\tau\in[0,T^*]. E ( τ ) − 1/2 ≥ E ( 0 ) − 1/2 − 2 C 1 τ , τ ∈ [ 0 , T ∗ ] .
In particular, E ( τ ) ≤ 4 E ( 0 ) \mathcal{E}(\tau)\leq4\mathcal{E}(0) E ( τ ) ≤ 4 E ( 0 ) on [ 0 , T ∗ ] [0,T^*] [ 0 , T ∗ ] , which improves the bootstrap assumption and forces T ∗ = ∞ T^*=\infty T ∗ = ∞ . Hence sup τ ≥ 0 E ( τ ) ≤ 4 E ( 0 ) \sup_{\tau\geq0}\mathcal{E}(\tau)\leq4\mathcal{E}(0) sup τ ≥ 0 E ( τ ) ≤ 4 E ( 0 ) and the corresponding H 3 H^3 H 3 solution extends globally in time. Uniqueness follows by repeating the same energy estimate on the difference of two solutions as in the proof of Theorem 5.1
Geometric formulas for graph surfaces
For convenience we collect here the basic geometric identities associated with the free surface
Γ ( t ) = { ( x 1 , x 2 , η ( x , t ) ) : x = ( x 1 , x 2 ) ∈ L S 1 × L S 1 } , \Gamma(t)=\{(x_1,x_2,\eta(x,t)):x=(x_1,x_2)\in LS^1\times LS^1\}, Γ ( t ) = {( x 1 , x 2 , η ( x , t )) : x = ( x 1 , x 2 ) ∈ L S 1 × L S 1 } ,
expressing all quantities directly in terms of the height function η ( x , t ) \eta(x,t) η ( x , t ) , where x = ( x 1 , x 2 ) x=(x_1,x_2) x = ( x 1 , x 2 ) denotes the horizontal variables.
The surface is parametrised by
X ( x , t ) = ( x 1 , x 2 , η ( x , t ) ) , X(x,t)=(x_1,x_2,\eta(x,t)), X ( x , t ) = ( x 1 , x 2 , η ( x , t )) ,
so that differentiation with respect to x x x yields the tangent vectors
∂ x i X = e i + ( ∂ x i η ) e 3 , i = 1 , 2. \partial_{x_i}X=e_i+(\partial_{x_i}\eta)e_3,\qquad i=1,2. ∂ x i X = e i + ( ∂ x i η ) e 3 , i = 1 , 2.
The first fundamental form is then
g i j = ∂ x i X ⋅ ∂ x j X = δ i j + ∂ x i η ∂ x j η , g_{ij}=\partial_{x_i}X\cdot\partial_{x_j}X=\delta_{ij}+\partial_{x_i}\eta\,\partial_{x_j}\eta, g ij = ∂ x i X ⋅ ∂ x j X = δ ij + ∂ x i η ∂ x j η ,
with determinant
∣ g ∣ = 1 + ∣ ∇ x η ∣ 2 , α : = 1 + ∣ ∇ x η ∣ 2 , |g|=1+|\nabla_x\eta|^2,\qquad\alpha:=1+|\nabla_x\eta|^2, ∣ g ∣ = 1 + ∣ ∇ x η ∣ 2 , α := 1 + ∣ ∇ x η ∣ 2 ,
and inverse metric g i j = α i j = ( g i j ) − 1 g^{ij}=\alpha^{ij}=(g_{ij})^{-1} g ij = α ij = ( g ij ) − 1 . The corresponding upward-pointing unit normal takes the familiar form
n ( x , t ) = 1 α ( − η x 1 − η x 2 1 ) , n(x,t)=\frac{1}{\sqrt{\alpha}}\begin{pmatrix}-\eta_{x_1}\\-\eta_{x_2}\\1\end{pmatrix}, n ( x , t ) = α 1 − η x 1 − η x 2 1 ,
and the induced surface area element is α d x \sqrt{\alpha}\,dx α d x .
Differentiating the parametrisation twice, ∂ x i x j X = ( 0 , 0 , ∂ x i x j η ) \partial_{x_i x_j}X=(0,0,\partial_{x_i x_j}\eta) ∂ x i x j X = ( 0 , 0 , ∂ x i x j η ) , and projecting onto the normal yields the second fundamental form,
b i j = ∂ x i x j X ⋅ n = η x i x j α . b_{ij}=\partial_{x_i x_j}X\cdot n=\frac{\eta_{x_i x_j}}{\sqrt{\alpha}}. b ij = ∂ x i x j X ⋅ n = α η x i x j .
In terms of g i j g^{ij} g ij and b i j b_{ij} b ij , the mean curvature of the graph is
H ( η ) = 1 2 g i j b i j = 1 2 α ( α 11 η x 1 x 1 + 2 α 12 η x 1 x 2 + α 22 η x 2 x 2 ) , H(\eta)=\frac{1}{2}g^{ij}b_{ij}=\frac{1}{2\sqrt{\alpha}}\left(\alpha^{11}\eta_{x_1x_1}+2\alpha^{12}\eta_{x_1x_2}+\alpha^{22}\eta_{x_2x_2}\right), H ( η ) = 2 1 g ij b ij = 2 α 1 ( α 11 η x 1 x 1 + 2 α 12 η x 1 x 2 + α 22 η x 2 x 2 ) ,
while the Gauss curvature is given by the classical identity
K ( η ) = det ( b i j ) det ( g i j ) = η x 1 x 1 η x 2 x 2 − ( η x 1 x 2 ) 2 ( 1 + ∣ ∇ x η ∣ 2 ) 2 . K(\eta)=\frac{\det(b_{ij})}{\det(g_{ij})}=\frac{\eta_{x_1x_1}\eta_{x_2x_2}-(\eta_{x_1x_2})^2}{(1+|\nabla_x\eta|^2)^2}. K ( η ) = det ( g ij ) det ( b ij ) = ( 1 + ∣ ∇ x η ∣ 2 ) 2 η x 1 x 1 η x 2 x 2 − ( η x 1 x 2 ) 2 .
Finally, for any scalar function f = f ( x , t ) f=f(x,t) f = f ( x , t ) defined on the surface, the Laplace–Beltrami operator reduces to the divergence-form expression
Δ Γ f = 1 α ∂ x i ( α α i j ∂ x j f ) , \Delta_\Gamma f=\frac{1}{\sqrt{\alpha}}\partial_{x_i}\left(\sqrt{\alpha}\,\alpha^{ij}\partial_{x_j}f\right), Δ Γ f = α 1 ∂ x i ( α α ij ∂ x j f ) ,
where all derivatives are horizontal. These formulas permit the elastic operator E ( η ) \mathcal{E}(\eta) E ( η ) to be expressed entirely in terms of η \eta η and its derivatives on the reference torus.
Nondimensionalisation of the Fluid–Structure System
In this appendix we provide a complete nondimensionalisation of the system governing the coupled evolution of the fluid potential ϕ \phi ϕ and the plate displacement η \eta η .
Choice of characteristic scales
Let L L L denote the typical horizontal wavelength and H H H the typical vertical amplitude of the elastic sheet. We write x = ( x 1 , x 2 ) x=(x_1,x_2) x = ( x 1 , x 2 ) for the horizontal variables and x 3 x_3 x 3 for the vertical coordinate. We introduce the dimensionless spatial and temporal variables
x = L x ~ , x 3 = L x ~ 3 , t = L g t ~ , x=L\widetilde{x}, \qquad x_3=L\widetilde{x}_3, \qquad t=\sqrt{\frac{L}{g}}\,\widetilde{t}, x = L x , x 3 = L x 3 , t = g L t ,
and the nondimensional surface displacement, potential, and surface potential
η ( x , t ) = H η ~ ( x ~ , t ~ ) , ϕ ( x , x 3 , t ) = H g L ϕ ~ ( x ~ , x ~ 3 , t ~ ) , ψ ( x , t ) = H g L ψ ~ ( x ~ , t ~ ) . \eta(x,t)=H\widetilde{\eta}(\widetilde{x},\widetilde{t}), \qquad\phi(x,x_3,t)=H\sqrt{gL}\,\widetilde{\phi}(\widetilde{x},\widetilde{x}_3,\widetilde{t}), \qquad\psi(x,t)=H\sqrt{gL}\,\widetilde{\psi}(\widetilde{x},\widetilde{t}). η ( x , t ) = H η ( x , t ) , ϕ ( x , x 3 , t ) = H g L ϕ ( x , x 3 , t ) , ψ ( x , t ) = H g L ψ ( x , t ) .
We denote the dimensionless steepness parameter by
ε : = H L . \varepsilon:=\frac{H}{L}. ε := L H .
In what follows all derivatives with respect to x ~ \widetilde{x} x are written with tildes. Transformations of derivatives follow from
∂ t = g L ∂ ~ t , ∇ x = 1 L ∇ ~ x , Δ x = 1 L 2 Δ ~ x , \partial_t=\sqrt{\frac{g}{L}}\,\widetilde{\partial}_t, \qquad\nabla_x=\frac{1}{L}\widetilde{\nabla}_x, \qquad\Delta_x=\frac{1}{L^2}\widetilde{\Delta}_x, ∂ t = L g ∂ t , ∇ x = L 1 ∇ x , Δ x = L 2 1 Δ x ,
and therefore
∂ t η = H g L ∂ ~ t η ~ , ∂ t 2 η = H g L ∂ ~ t 2 η ~ . \partial_t\eta=H\sqrt{\frac{g}{L}}\,\widetilde{\partial}_t\widetilde{\eta}, \qquad\partial_t^2\eta=H\frac{g}{L}\,\widetilde{\partial}_t^2\widetilde{\eta}. ∂ t η = H L g ∂ t η , ∂ t 2 η = H L g ∂ t 2 η .
Scaling of the bending operator and dynamic boundary conditions
The elastic operator in the model is
E ( η ) = B 2 Δ Γ H ( η ) + B H ( η ) 3 − B H ( η ) K ( η ) , \mathcal{E}(\eta)=\frac{B}{2}\Delta_{\Gamma}H(\eta)+BH(\eta)^3-BH(\eta)K(\eta), E ( η ) = 2 B Δ Γ H ( η ) + B H ( η ) 3 − B H ( η ) K ( η ) ,
where B > 0 B>0 B > 0 denotes the bending stiffness.2
We now express each geometric quantity more explicitly in terms of the dimensionless variables. Recall that
η ( x , t ) = H η ~ ( x ~ , t ~ ) , x i = L x ~ i , i = 1 , 2 , \eta(x,t)=H\widetilde{\eta}(\widetilde{x},\widetilde{t}), \qquad x_i=L\widetilde{x}_i,\quad i=1,2, η ( x , t ) = H η ( x , t ) , x i = L x i , i = 1 , 2 ,
so that
∂ x i η = ε ∂ x ~ i η ~ , ∂ x i x j 2 η = H L 2 ∂ x ~ i x ~ j 2 η ~ , ε = H L . \partial_{x_i}\eta=\varepsilon\partial_{\widetilde{x}_i}\widetilde{\eta}, \qquad\partial_{x_i x_j}^2\eta=\frac{H}{L^2}\partial_{\widetilde{x}_i\widetilde{x}_j}^2\widetilde{\eta}, \qquad\varepsilon=\frac{H}{L}. ∂ x i η = ε ∂ x i η , ∂ x i x j 2 η = L 2 H ∂ x i x j 2 η , ε = L H .
The induced metric and its determinant are
g i j = δ i j + ∂ x i η ∂ x j η = δ i j + ε 2 ∂ x ~ i η ~ ∂ x ~ j η ~ , ∣ g ∣ = 1 + ∣ ∇ x η ∣ 2 = 1 + ε 2 ∣ ∇ ~ x η ~ ∣ 2 = : α ~ , g_{ij}=\delta_{ij}+\partial_{x_i}\eta\partial_{x_j}\eta=\delta_{ij}+\varepsilon^2\partial_{\widetilde{x}_i}\widetilde{\eta}\partial_{\widetilde{x}_j}\widetilde{\eta}, \qquad|g|=1+|\nabla_x\eta|^2=1+\varepsilon^2|\widetilde{\nabla}_x\widetilde{\eta}|^2=:\widetilde{\alpha}, g ij = δ ij + ∂ x i η ∂ x j η = δ ij + ε 2 ∂ x i η ∂ x j η , ∣ g ∣ = 1 + ∣ ∇ x η ∣ 2 = 1 + ε 2 ∣ ∇ x η ∣ 2 =: α ,
and the inverse metric entries are
α 11 = 1 + ε 2 ( ∂ x ~ 2 η ~ ) 2 α ~ , α 22 = 1 + ε 2 ( ∂ x ~ 1 η ~ ) 2 α ~ , α 12 = − ε 2 ∂ x ~ 1 η ~ ∂ x ~ 2 η ~ α ~ . \alpha^{11}=\frac{1+\varepsilon^2(\partial_{\widetilde{x}_2}\widetilde{\eta})^2}{\widetilde{\alpha}}, \qquad\alpha^{22}=\frac{1+\varepsilon^2(\partial_{\widetilde{x}_1}\widetilde{\eta})^2}{\widetilde{\alpha}}, \qquad\alpha^{12}=-\frac{\varepsilon^2\partial_{\widetilde{x}_1}\widetilde{\eta}\partial_{\widetilde{x}_2}\widetilde{\eta}}{\widetilde{\alpha}}. α 11 = α 1 + ε 2 ( ∂ x 2 η ) 2 , α 22 = α 1 + ε 2 ( ∂ x 1 η ) 2 , α 12 = − α ε 2 ∂ x 1 η ∂ x 2 η .
Mean curvature in terms of η ~ \tilde{\eta} η ~ . Using the formula (see Appendix § 6 )
H ( η ) = 1 2 1 + ∣ ∇ x η ∣ 2 ( α 11 ∂ x 1 x 1 2 η + 2 α 12 ∂ x 1 x 2 2 η + α 22 ∂ x 2 x 2 2 η ) H(\eta)=\frac{1}{2\sqrt{1+|\nabla_x\eta|^2}}\left(\alpha^{11}\partial^2_{x_1x_1}\eta+2\alpha^{12}\partial^2_{x_1x_2}\eta+\alpha^{22}\partial^2_{x_2x_2}\eta\right) H ( η ) = 2 1 + ∣ ∇ x η ∣ 2 1 ( α 11 ∂ x 1 x 1 2 η + 2 α 12 ∂ x 1 x 2 2 η + α 22 ∂ x 2 x 2 2 η )
and substituting ∂ x i x j 2 η = H L 2 ∂ x ~ i x ~ j 2 η ~ \partial^2_{x_i x_j}\eta=\frac{H}{L^2}\partial^2_{\tilde{x}_i\tilde{x}_j}\tilde{\eta} ∂ x i x j 2 η = L 2 H ∂ x ~ i x ~ j 2 η ~ and 1 + ∣ ∇ x η ∣ 2 = α ~ 1+|\nabla_x\eta|^2=\tilde{\alpha} 1 + ∣ ∇ x η ∣ 2 = α ~ , we obtain
H ( η ) = H L 2 1 2 α ~ ( α 11 ∂ x ~ 1 x ~ 1 2 η ~ + 2 α 12 ∂ x ~ 1 x ~ 2 2 η ~ + α 22 ∂ x ~ 2 x ~ 2 2 η ~ ) H(\eta)=\frac{H}{L^2}\frac{1}{2\sqrt{\tilde{\alpha}}}\left(\alpha^{11}\partial^2_{\tilde{x}_1\tilde{x}_1}\tilde{\eta}+2\alpha^{12}\partial^2_{\tilde{x}_1\tilde{x}_2}\tilde{\eta}+\alpha^{22}\partial^2_{\tilde{x}_2\tilde{x}_2}\tilde{\eta}\right) H ( η ) = L 2 H 2 α ~ 1 ( α 11 ∂ x ~ 1 x ~ 1 2 η ~ + 2 α 12 ∂ x ~ 1 x ~ 2 2 η ~ + α 22 ∂ x ~ 2 x ~ 2 2 η ~ )
It is convenient to introduce the dimensionless functional
H ( η ~ ; ε ) : = 1 2 α ~ ( α 11 ∂ x ~ 1 x ~ 1 2 η ~ + 2 α 12 ∂ x ~ 1 x ~ 2 2 η ~ + α 22 ∂ x ~ 2 x ~ 2 2 η ~ ) , \mathcal{H}(\tilde{\eta};\varepsilon):=\frac{1}{2\sqrt{\tilde{\alpha}}}\left(\alpha^{11}\partial^2_{\tilde{x}_1\tilde{x}_1}\tilde{\eta}+2\alpha^{12}\partial^2_{\tilde{x}_1\tilde{x}_2}\tilde{\eta}+\alpha^{22}\partial^2_{\tilde{x}_2\tilde{x}_2}\tilde{\eta}\right), H ( η ~ ; ε ) := 2 α ~ 1 ( α 11 ∂ x ~ 1 x ~ 1 2 η ~ + 2 α 12 ∂ x ~ 1 x ~ 2 2 η ~ + α 22 ∂ x ~ 2 x ~ 2 2 η ~ ) ,
so that
H ( η ) = H L 2 H ( η ~ ; ε ) . H(\eta)=\frac{H}{L^2}\mathcal{H}(\tilde{\eta};\varepsilon). H ( η ) = L 2 H H ( η ~ ; ε ) .
Gaussian curvature in terms of η ~ \tilde{\eta} η ~ . From
K ( η ) = ∂ x 1 x 1 2 η ∂ x 2 x 2 2 η − ( ∂ x 1 x 2 2 η ) 2 ( 1 + ∣ ∇ x η ∣ 2 ) 2 , K(\eta)=\frac{\partial^2_{x_1x_1}\eta\partial^2_{x_2x_2}\eta-(\partial^2_{x_1x_2}\eta)^2}{(1+|\nabla_x\eta|^2)^2}, K ( η ) = ( 1 + ∣ ∇ x η ∣ 2 ) 2 ∂ x 1 x 1 2 η ∂ x 2 x 2 2 η − ( ∂ x 1 x 2 2 η ) 2 ,
we obtain
∂ x 1 x 1 2 η ∂ x 2 x 2 2 η − ( ∂ x 1 x 2 2 η ) 2 = H 2 L 4 ( ∂ x ~ 1 x ~ 1 2 η ~ ∂ x ~ 2 x ~ 2 2 η ~ − ( ∂ x ~ 1 x ~ 2 2 η ~ ) 2 ) , ( 1 + ∣ ∇ x η ∣ 2 ) 2 = α ~ 2 . \partial^2_{x_1x_1}\eta\partial^2_{x_2x_2}\eta-(\partial^2_{x_1x_2}\eta)^2=\frac{H^2}{L^4}\left(\partial^2_{\tilde{x}_1\tilde{x}_1}\tilde{\eta}\partial^2_{\tilde{x}_2\tilde{x}_2}\tilde{\eta}-(\partial^2_{\tilde{x}_1\tilde{x}_2}\tilde{\eta})^2\right),\qquad(1+|\nabla_x\eta|^2)^2=\tilde{\alpha}^2. ∂ x 1 x 1 2 η ∂ x 2 x 2 2 η − ( ∂ x 1 x 2 2 η ) 2 = L 4 H 2 ( ∂ x ~ 1 x ~ 1 2 η ~ ∂ x ~ 2 x ~ 2 2 η ~ − ( ∂ x ~ 1 x ~ 2 2 η ~ ) 2 ) , ( 1 + ∣ ∇ x η ∣ 2 ) 2 = α ~ 2 .
Thus
K ( η ) = H 2 L 4 ∂ x ~ 1 x ~ 1 2 η ~ ∂ x ~ 2 x ~ 2 2 η ~ − ( ∂ x ~ 1 x ~ 2 2 η ~ ) 2 α ~ 2 = H 2 L 4 K ( η ~ ; ε ) , K(\eta)=\frac{H^2}{L^4}\frac{\partial^2_{\tilde{x}_1\tilde{x}_1}\tilde{\eta}\partial^2_{\tilde{x}_2\tilde{x}_2}\tilde{\eta}-(\partial^2_{\tilde{x}_1\tilde{x}_2}\tilde{\eta})^2}{\tilde{\alpha}^2}=\frac{H^2}{L^4}\mathcal{K}(\tilde{\eta};\varepsilon), K ( η ) = L 4 H 2 α ~ 2 ∂ x ~ 1 x ~ 1 2 η ~ ∂ x ~ 2 x ~ 2 2 η ~ − ( ∂ x ~ 1 x ~ 2 2 η ~ ) 2 = L 4 H 2 K ( η ~ ; ε ) ,
where
K ( η ~ ; ε ) = ∂ x ~ 1 x ~ 1 2 η ~ ∂ x ~ 2 x ~ 2 2 η ~ − ( ∂ x ~ 1 x ~ 2 2 η ~ ) 2 α ~ 2 . \mathcal{K}(\tilde{\eta};\varepsilon)=\frac{\partial^2_{\tilde{x}_1\tilde{x}_1}\tilde{\eta}\partial^2_{\tilde{x}_2\tilde{x}_2}\tilde{\eta}-(\partial^2_{\tilde{x}_1\tilde{x}_2}\tilde{\eta})^2}{\tilde{\alpha}^2}. K ( η ~ ; ε ) = α ~ 2 ∂ x ~ 1 x ~ 1 2 η ~ ∂ x ~ 2 x ~ 2 2 η ~ − ( ∂ x ~ 1 x ~ 2 2 η ~ ) 2 .
Laplace–Beltrami of H ( η ) H(\eta) H ( η ) . Recall
Δ Γ f = 1 1 + ∣ ∇ x η ∣ 2 ∂ x i ( 1 + ∣ ∇ x η ∣ 2 α i j ∂ x j f ) , \Delta_\Gamma f=\frac{1}{\sqrt{1+|\nabla_x\eta|^2}}\partial_{x_i}\left(\sqrt{1+|\nabla_x\eta|^2}\,\alpha^{ij}\partial_{x_j}f\right), Δ Γ f = 1 + ∣ ∇ x η ∣ 2 1 ∂ x i ( 1 + ∣ ∇ x η ∣ 2 α ij ∂ x j f ) ,
where ∂ x i \partial_{x_i} ∂ x i acts on the horizontal variables x = ( x 1 , x 2 ) x=(x_1,x_2) x = ( x 1 , x 2 ) . For f = H ( η ) f=H(\eta) f = H ( η ) , with H ( η ) = H L 2 H ( η ~ ; ε ) H(\eta)=\frac{H}{L^2}\mathcal{H}(\tilde{\eta};\varepsilon) H ( η ) = L 2 H H ( η ~ ; ε ) , we have
∂ x j H = H L 2 ∂ x j H = H L 3 ∂ x ~ j H ( η ~ ; ε ) . \partial_{x_j}H=\frac{H}{L^2}\partial_{x_j}\mathcal{H}=\frac{H}{L^3}\partial_{\tilde{x}_j}\mathcal{H}(\tilde{\eta};\varepsilon). ∂ x j H = L 2 H ∂ x j H = L 3 H ∂ x ~ j H ( η ~ ; ε ) .
Hence
1 + ∣ ∇ x η ∣ 2 α i j ∂ x j H = α ~ α i j H L 3 ∂ x ~ j H , \sqrt{1+|\nabla_x\eta|^2}\,\alpha^{ij}\partial_{x_j}H=\sqrt{\tilde{\alpha}}\,\alpha^{ij}\frac{H}{L^3}\partial_{\tilde{x}_j}\mathcal{H}, 1 + ∣ ∇ x η ∣ 2 α ij ∂ x j H = α ~ α ij L 3 H ∂ x ~ j H ,
and
∂ x i ( 1 + ∣ ∇ x η ∣ 2 α i j ∂ x j H ) = 1 L ∂ x ~ i ( α ~ α i j H L 3 ∂ x ~ j H ) = H L 4 ∂ x ~ i ( α ~ α i j ∂ x ~ j H ) . \partial_{x_i}\left(\sqrt{1+|\nabla_x\eta|^2}\,\alpha^{ij}\partial_{x_j}H\right)=\frac{1}{L}\partial_{\tilde{x}_i}\left(\sqrt{\tilde{\alpha}}\,\alpha^{ij}\frac{H}{L^3}\partial_{\tilde{x}_j}\mathcal{H}\right)=\frac{H}{L^4}\partial_{\tilde{x}_i}\left(\sqrt{\tilde{\alpha}}\,\alpha^{ij}\partial_{\tilde{x}_j}\mathcal{H}\right). ∂ x i ( 1 + ∣ ∇ x η ∣ 2 α ij ∂ x j H ) = L 1 ∂ x ~ i ( α ~ α ij L 3 H ∂ x ~ j H ) = L 4 H ∂ x ~ i ( α ~ α ij ∂ x ~ j H ) .
Dividing by 1 + ∣ ∇ x η ∣ 2 = α ~ \sqrt{1+|\nabla_x\eta|^2}=\sqrt{\tilde{\alpha}} 1 + ∣ ∇ x η ∣ 2 = α ~ we arrive at
Δ Γ H ( η ) = H L 4 L Γ ( η ~ ; ε ) , \Delta_\Gamma H(\eta)=\frac{H}{L^4}\mathcal{L}_\Gamma(\tilde{\eta};\varepsilon), Δ Γ H ( η ) = L 4 H L Γ ( η ~ ; ε ) ,
where
L Γ ( η ~ ; ε ) : = 1 α ~ ∂ x ~ i ( α ~ α ~ i j ∂ x ~ j H ( η ~ ; ε ) ) . \mathcal{L}_{\Gamma}(\tilde{\eta};\varepsilon) := \frac{1}{\sqrt{\tilde{\alpha}}}\,\partial_{\tilde{x}_i}\left(\sqrt{\tilde{\alpha}}\,\tilde{\alpha}^{ij}\partial_{\tilde{x}_j}\mathcal{H}(\tilde{\eta};\varepsilon)\right). L Γ ( η ~ ; ε ) := α ~ 1 ∂ x ~ i ( α ~ α ~ ij ∂ x ~ j H ( η ~ ; ε ) ) .
Putting everything together. With the above notation,
H ( η ) = H L 2 H ( η ~ ; ε ) , K ( η ) = H 2 L 4 K ( η ~ ; ε ) , Δ Γ H ( η ) = H L 4 L Γ ( η ~ ; ε ) . H(\eta)=\frac{H}{L^2}\mathcal{H}(\tilde{\eta};\varepsilon),\qquad K(\eta)=\frac{H^2}{L^4}\mathcal{K}(\tilde{\eta};\varepsilon),\qquad\Delta_{\Gamma}H(\eta)=\frac{H}{L^4}\mathcal{L}_{\Gamma}(\tilde{\eta};\varepsilon). H ( η ) = L 2 H H ( η ~ ; ε ) , K ( η ) = L 4 H 2 K ( η ~ ; ε ) , Δ Γ H ( η ) = L 4 H L Γ ( η ~ ; ε ) .
Therefore,
B 2 Δ Γ H ( η ) = B H 2 L 4 L Γ ( η ~ ; ε ) , \frac{B}{2}\Delta_{\Gamma}H(\eta)=\frac{BH}{2L^4}\mathcal{L}_{\Gamma}(\tilde{\eta};\varepsilon), 2 B Δ Γ H ( η ) = 2 L 4 B H L Γ ( η ~ ; ε ) ,
B H ( η ) 3 = B ( H L 2 H ) 3 = B H L 4 ε 2 H ( η ~ ; ε ) 3 , BH(\eta)^3=B\left(\frac{H}{L^2}\mathcal{H}\right)^3=\frac{BH}{L^4}\varepsilon^2\mathcal{H}(\tilde{\eta};\varepsilon)^3, B H ( η ) 3 = B ( L 2 H H ) 3 = L 4 B H ε 2 H ( η ~ ; ε ) 3 ,
− B H ( η ) K ( η ) = − B ( H L 2 H ) ( H 2 L 4 K ) = − B H L 4 ε 2 H ( η ~ ; ε ) K ( η ~ ; ε ) . -BH(\eta)K(\eta)=-B\left(\frac{H}{L^2}\mathcal{H}\right)\left(\frac{H^2}{L^4}\mathcal{K}\right)=-\frac{BH}{L^4}\varepsilon^2\mathcal{H}(\tilde{\eta};\varepsilon)\mathcal{K}(\tilde{\eta};\varepsilon). − B H ( η ) K ( η ) = − B ( L 2 H H ) ( L 4 H 2 K ) = − L 4 B H ε 2 H ( η ~ ; ε ) K ( η ~ ; ε ) .
Factoring out the common prefactor B H L 4 \frac{BH}{L^4} L 4 B H we obtain
E ( η ) = B H L 4 E ( η ~ ; ε ) \mathcal{E}(\eta)=\frac{BH}{L^4}\mathcal{E}(\tilde{\eta};\varepsilon) E ( η ) = L 4 B H E ( η ~ ; ε )
where the dimensionless operator E \mathcal{E} E is given explicitly by
E ( η ~ ; ε ) = 1 2 L Γ ( η ~ ; ε ) + ε 2 ( H ( η ~ ; ε ) 3 − H ( η ~ ; ε ) K ( η ~ ; ε ) ) , \mathcal{E}(\tilde{\eta};\varepsilon)=\frac{1}{2}\mathcal{L}_{\Gamma}(\tilde{\eta};\varepsilon)+\varepsilon^2\left(\mathcal{H}(\tilde{\eta};\varepsilon)^3-\mathcal{H}(\tilde{\eta};\varepsilon)\mathcal{K}(\tilde{\eta};\varepsilon)\right), E ( η ~ ; ε ) = 2 1 L Γ ( η ~ ; ε ) + ε 2 ( H ( η ~ ; ε ) 3 − H ( η ~ ; ε ) K ( η ~ ; ε ) ) ,
with
H ( η ~ ; ε ) = 1 2 α ~ ( α 11 ∂ x ~ 1 x ~ 1 2 η ~ + 2 α 12 ∂ x ~ 1 x ~ 2 2 η ~ + α 22 ∂ x ~ 2 x ~ 2 2 η ~ ) , \mathcal{H}(\tilde{\eta};\varepsilon)=\frac{1}{2\sqrt{\tilde{\alpha}}}\left(\alpha^{11}\partial_{\tilde{x}_1\tilde{x}_1}^2\tilde{\eta}+2\alpha^{12}\partial_{\tilde{x}_1\tilde{x}_2}^2\tilde{\eta}+\alpha^{22}\partial_{\tilde{x}_2\tilde{x}_2}^2\tilde{\eta}\right), H ( η ~ ; ε ) = 2 α ~ 1 ( α 11 ∂ x ~ 1 x ~ 1 2 η ~ + 2 α 12 ∂ x ~ 1 x ~ 2 2 η ~ + α 22 ∂ x ~ 2 x ~ 2 2 η ~ ) ,
K ( η ~ ; ε ) = ∂ x ~ 1 x ~ 1 2 η ~ ∂ x ~ 2 x ~ 2 2 η ~ − ( ∂ x ~ 1 x ~ 2 2 η ~ ) 2 α ~ 2 , \mathcal{K}(\tilde{\eta};\varepsilon)=\frac{\partial_{\tilde{x}_1\tilde{x}_1}^2\tilde{\eta}\,\partial_{\tilde{x}_2\tilde{x}_2}^2\tilde{\eta}-\left(\partial_{\tilde{x}_1\tilde{x}_2}^2\tilde{\eta}\right)^2}{\tilde{\alpha}^2}, K ( η ~ ; ε ) = α ~ 2 ∂ x ~ 1 x ~ 1 2 η ~ ∂ x ~ 2 x ~ 2 2 η ~ − ( ∂ x ~ 1 x ~ 2 2 η ~ ) 2 ,
L Γ ( η ~ ; ε ) = 1 α ~ ∂ x ~ i ( α ~ α ~ i j ∂ x ~ j H ( η ~ ; ε ) ) , \mathcal{L}_{\Gamma}(\tilde{\eta};\varepsilon)=\frac{1}{\sqrt{\tilde{\alpha}}}\,\partial_{\tilde{x}_i}\left(\sqrt{\tilde{\alpha}}\,\tilde{\alpha}^{ij}\partial_{\tilde{x}_j}\mathcal{H}(\tilde{\eta};\varepsilon)\right), L Γ ( η ~ ; ε ) = α ~ 1 ∂ x ~ i ( α ~ α ~ ij ∂ x ~ j H ( η ~ ; ε ) ) ,
and
α ~ = 1 + ε 2 ∣ ∇ x ~ η ~ ∣ 2 , α i j = α i j ( ε , ∇ x ~ η ~ ) . \tilde{\alpha}=1+\varepsilon^2\lvert\nabla_{\tilde{x}}\tilde{\eta}\rvert^2,\qquad\alpha^{ij}=\alpha^{ij}(\varepsilon,\nabla_{\tilde{x}}\tilde{\eta}). α ~ = 1 + ε 2 ∣ ∇ x ~ η ~ ∣ 2 , α ij = α ij ( ε , ∇ x ~ η ~ ) .
In particular, E \mathcal{E} E depends only on η ~ \tilde{\eta} η ~ and its first and second derivatives; no additional geometric unknowns are introduced.3
Remark B.1. In many asymptotic regimes one assumes small steepness, ε ≪ 1 \varepsilon\ll1 ε ≪ 1 . Expanding
α ~ = 1 + ε 2 ∣ ∇ x ~ η ~ ∣ 2 = 1 + O ( ε 2 ) , α i j = δ i j + O ( ε 2 ) , \tilde{\alpha}=1+\varepsilon^2\lvert\nabla_{\tilde{x}}\tilde{\eta}\rvert^2=1+O(\varepsilon^2),\qquad\alpha^{ij}=\delta_{ij}+O(\varepsilon^2), α ~ = 1 + ε 2 ∣ ∇ x ~ η ~ ∣ 2 = 1 + O ( ε 2 ) , α ij = δ ij + O ( ε 2 ) ,
we obtain, to leading order,
H ( η ~ ; ε ) = 1 2 Δ x ~ η ~ + O ( ε 2 ∣ ∇ x ~ η ~ ∣ ∣ ∇ x ~ 2 η ~ ∣ ) , \mathcal{H}(\tilde{\eta};\varepsilon)=\frac{1}{2}\Delta_{\tilde{x}}\tilde{\eta}+O\left(\varepsilon^2\lvert\nabla_{\tilde{x}}\tilde{\eta}\rvert\lvert\nabla_{\tilde{x}}^2\tilde{\eta}\rvert\right), H ( η ~ ; ε ) = 2 1 Δ x ~ η ~ + O ( ε 2 ∣ ∇ x ~ η ~ ∣ ∣ ∇ x ~ 2 η ~ ∣ ) ,
and therefore
L Γ ( η ~ ; ε ) = Δ x ~ H ( η ~ ; ε ) + O ( ε 2 ) = 1 2 Δ x ~ 2 η ~ + O ( ε 2 P ( η ~ ) ) . \mathcal{L}_{\Gamma}(\tilde{\eta};\varepsilon)=\Delta_{\tilde{x}}\mathcal{H}(\tilde{\eta};\varepsilon)+O(\varepsilon^2)=\frac{1}{2}\Delta_{\tilde{x}}^2\tilde{\eta}+O\left(\varepsilon^2\mathcal{P}(\tilde{\eta})\right). L Γ ( η ~ ; ε ) = Δ x ~ H ( η ~ ; ε ) + O ( ε 2 ) = 2 1 Δ x ~ 2 η ~ + O ( ε 2 P ( η ~ ) ) .
where P ( η ~ ) \mathcal{P}(\tilde{\eta}) P ( η ~ ) denotes a polynomial expression in ∇ ~ x η ~ \widetilde{\nabla}_{x}\tilde{\eta} ∇ x η ~ , ∇ ~ x 2 η ~ \widetilde{\nabla}_{x}^{2}\tilde{\eta} ∇ x 2 η ~ , ∇ ~ x 3 η ~ \widetilde{\nabla}_{x}^{3}\tilde{\eta} ∇ x 3 η ~ , and ∇ ~ x 4 η ~ \widetilde{\nabla}_{x}^{4}\tilde{\eta} ∇ x 4 η ~ . Consequently
E ( η ~ ; ε ) = 1 4 Δ ~ x 2 η ~ + O ( ε 2 P ( η ~ ) ) , \mathcal{E}(\tilde{\eta};\varepsilon)=\frac{1}{4}\widetilde{\Delta}_{x}^{2}\tilde{\eta}+O(\varepsilon^{2}\mathcal{P}(\tilde{\eta})), E ( η ~ ; ε ) = 4 1 Δ x 2 η ~ + O ( ε 2 P ( η ~ )) ,
so that the principal part of the geometric bending operator is indeed biharmonic. This justifies the appearance of operators of the type Δ x 2 η \Delta_{x}^{2}\eta Δ x 2 η as leading-order models in small-slope asymptotic regimes, with the fully nonlinear curvature effects encoded in higher-order corrections.
We now turn to the nondimensionalisation of the dynamic boundary condition for the surface potential ψ \psi ψ . In dimensional variables this equation reads, on the free surface,
∂ t ψ = ρ s h ρ f ∂ t 2 η + 1 ρ f E ( η ) + γ ρ f Δ x ∂ t η + ∂ x 3 ϕ ( ∇ ϕ ⋅ 1 + ∣ ∇ x η ∣ 2 n ) − 1 2 ∣ ∇ ϕ ∣ 2 − g η , on Γ ( t ) . \partial_{t}\psi=\frac{\rho_{s}h}{\rho_{f}}\partial_{t}^{2}\eta+\frac{1}{\rho_{f}}\mathcal{E}(\eta)+\frac{\gamma}{\rho_{f}}\Delta_{x}\partial_{t}\eta+\partial_{x_{3}}\phi\left(\nabla\phi\cdot\sqrt{1+|\nabla_{x}\eta|^{2}}\,n\right)-\frac{1}{2}|\nabla\phi|^{2}-g\eta,\quad\text{on }\Gamma(t). ∂ t ψ = ρ f ρ s h ∂ t 2 η + ρ f 1 E ( η ) + ρ f γ Δ x ∂ t η + ∂ x 3 ϕ ( ∇ ϕ ⋅ 1 + ∣ ∇ x η ∣ 2 n ) − 2 1 ∣∇ ϕ ∣ 2 − g η , on Γ ( t ) .
Using the scalings introduced above, each term may be rewritten in nondimensional form. For the inertial term we obtain
ρ s h ρ f ∂ t 2 η = ρ s h ρ f H g L ∂ ~ t 2 η ~ = H g ( ρ s h ρ f L ) ∂ ~ t 2 η ~ . \frac{\rho_{s}h}{\rho_{f}}\partial_{t}^{2}\eta=\frac{\rho_{s}h}{\rho_{f}}\frac{Hg}{L}\widetilde{\partial}_{t}^{2}\tilde{\eta}=Hg\left(\frac{\rho_{s}h}{\rho_{f}L}\right)\widetilde{\partial}_{t}^{2}\tilde{\eta}. ρ f ρ s h ∂ t 2 η = ρ f ρ s h L H g ∂ t 2 η ~ = H g ( ρ f L ρ s h ) ∂ t 2 η ~ .
For the elastic contribution, we use
1 ρ f E ( η ) = 1 ρ f B H L 4 E ( η ~ ; ε ) = H g ( B ρ f g L 3 ) ε E ( η ~ ; ε ) . \frac{1}{\rho_{f}}\mathcal{E}(\eta)=\frac{1}{\rho_{f}}\frac{BH}{L^{4}}\mathcal{E}(\tilde{\eta};\varepsilon)=Hg\left(\frac{B}{\rho_{f}gL^{3}}\right)\varepsilon\mathcal{E}(\tilde{\eta};\varepsilon). ρ f 1 E ( η ) = ρ f 1 L 4 B H E ( η ~ ; ε ) = H g ( ρ f g L 3 B ) ε E ( η ~ ; ε ) .
The damping term scales as
γ ρ f Δ x ∂ t η = γ ρ f 1 L 2 Δ ~ x ( H g L ∂ ~ t η ~ ) = H g ( γ ρ f g L 3 ) Δ ~ x ∂ ~ t η ~ . \frac{\gamma}{\rho_{f}}\Delta_{x}\partial_{t}\eta=\frac{\gamma}{\rho_{f}}\frac{1}{L^{2}}\widetilde{\Delta}_{x}\left(H\sqrt{\frac{g}{L}}\widetilde{\partial}_{t}\tilde{\eta}\right)=Hg\left(\frac{\gamma}{\rho_{f}\sqrt{gL^{3}}}\right)\widetilde{\Delta}_{x}\widetilde{\partial}_{t}\tilde{\eta}. ρ f γ Δ x ∂ t η = ρ f γ L 2 1 Δ x ( H L g ∂ t η ~ ) = H g ( ρ f g L 3 γ ) Δ x ∂ t η ~ .
For the Bernoulli terms we use
∇ ϕ = g L H 1 L ∇ ~ ϕ ~ = H g ∇ ~ ϕ ~ , ∂ x 3 ϕ = H g ∂ x ~ 3 ϕ ~ , \nabla\phi=\sqrt{gL}H\frac{1}{L}\widetilde{\nabla}\tilde{\phi}=H\sqrt{g}\widetilde{\nabla}\tilde{\phi},\qquad\partial_{x_{3}}\phi=H\sqrt{g}\partial_{\tilde{x}_{3}}\tilde{\phi}, ∇ ϕ = g L H L 1 ∇ ϕ ~ = H g ∇ ϕ ~ , ∂ x 3 ϕ = H g ∂ x ~ 3 ϕ ~ ,
and
∣ ∇ ϕ ∣ 2 = H g ∣ ∇ ~ ϕ ~ ∣ 2 , g η = H g η ~ . |\nabla\phi|^{2}=Hg|\widetilde{\nabla}\tilde{\phi}|^{2},\qquad g\eta=Hg\tilde{\eta}. ∣∇ ϕ ∣ 2 = H g ∣ ∇ ϕ ~ ∣ 2 , g η = H g η ~ .
Moreover, on the free surface the kinematic condition implies
∇ ϕ ⋅ 1 + ∣ ∇ x η ∣ 2 n η = H g ∇ ~ ϕ ~ ⋅ 1 + ε 2 ∣ ∇ ~ x η ~ ∣ 2 n ~ , \nabla\phi\cdot\sqrt{1+|\nabla_{x}\eta|^{2}}\,n_{\eta}=H\sqrt{g}\widetilde{\nabla}\tilde{\phi}\cdot\sqrt{1+\varepsilon^{2}|\widetilde{\nabla}_{x}\tilde{\eta}|^{2}}\,\tilde{n}, ∇ ϕ ⋅ 1 + ∣ ∇ x η ∣ 2 n η = H g ∇ ϕ ~ ⋅ 1 + ε 2 ∣ ∇ x η ~ ∣ 2 n ~ ,
where n ~ \tilde{n} n ~ is the rescaled unit normal. Altogether, the nonlinear Bernoulli term becomes
∂ x 3 ϕ ( ∇ ϕ ⋅ 1 + ∣ ∇ x η ∣ 2 n η ) = H g ε ∂ x ~ 3 ϕ ~ ( ∇ ~ ϕ ~ ⋅ 1 + ε 2 ∣ ∇ ~ x η ~ ∣ 2 n ~ ) , \partial_{x_{3}}\phi\left(\nabla\phi\cdot\sqrt{1+|\nabla_{x}\eta|^{2}}\,n_{\eta}\right)=Hg\varepsilon\partial_{\tilde{x}_{3}}\tilde{\phi}\left(\widetilde{\nabla}\tilde{\phi}\cdot\sqrt{1+\varepsilon^{2}|\widetilde{\nabla}_{x}\tilde{\eta}|^{2}}\,\tilde{n}\right), ∂ x 3 ϕ ( ∇ ϕ ⋅ 1 + ∣ ∇ x η ∣ 2 n η ) = H g ε ∂ x ~ 3 ϕ ~ ( ∇ ϕ ~ ⋅ 1 + ε 2 ∣ ∇ x η ~ ∣ 2 n ~ ) ,
and the quadratic velocity contribution is
1 2 ∣ ∇ ϕ ∣ 2 = H g 2 ∣ ∇ ~ ϕ ~ ∣ 2 . \frac{1}{2}|\nabla\phi|^{2}=\frac{Hg}{2}|\widetilde{\nabla}\tilde{\phi}|^{2}. 2 1 ∣∇ ϕ ∣ 2 = 2 H g ∣ ∇ ϕ ~ ∣ 2 .
Dividing the entire dynamic boundary condition by the factor H g Hg H g , we arrive at the nondimensional evolution equation
∂ ~ t ψ ~ = − Υ ∂ ~ t 2 η ~ − β ε E ( η ~ ; ε ) + δ Δ ~ x ∂ ~ t η ~ + ε ∂ x ~ 3 ϕ ~ ( ∇ ~ ϕ ~ ⋅ 1 + ε 2 ∣ ∇ ~ x η ~ ∣ 2 n ~ ) − ε 2 ∣ ∇ ~ ϕ ~ ∣ 2 − η ~ , on Γ ( t ) \widetilde{\partial}_{t}\tilde{\psi}=-\Upsilon\widetilde{\partial}_{t}^{2}\tilde{\eta}-\beta\varepsilon\mathcal{E}(\tilde{\eta};\varepsilon)+\delta\widetilde{\Delta}_{x}\widetilde{\partial}_{t}\tilde{\eta}+\varepsilon\partial_{\tilde{x}_{3}}\tilde{\phi}\left(\widetilde{\nabla}\tilde{\phi}\cdot\sqrt{1+\varepsilon^{2}|\widetilde{\nabla}_{x}\tilde{\eta}|^{2}}\,\tilde{n}\right)-\frac{\varepsilon}{2}|\widetilde{\nabla}\tilde{\phi}|^{2}-\tilde{\eta},\quad\text{on }\Gamma(t) ∂ t ψ ~ = − Υ ∂ t 2 η ~ − β ε E ( η ~ ; ε ) + δ Δ x ∂ t η ~ + ε ∂ x ~ 3 ϕ ~ ( ∇ ϕ ~ ⋅ 1 + ε 2 ∣ ∇ x η ~ ∣ 2 n ~ ) − 2 ε ∣ ∇ ϕ ~ ∣ 2 − η ~ , on Γ ( t )
where we have introduced the dimensionless parameters
Υ : = ρ s h ρ f L , β : = B ρ f g L 3 , δ : = γ ρ f g L 3 , ε : = H L . \Upsilon:= \frac{\rho_s h}{\rho_f L}, \qquad\beta:= \frac{B}{\rho_f g L^3}, \qquad\delta:= \frac{\gamma}{\rho_f \sqrt{gL^3}}, \qquad\varepsilon:= \frac{H}{L}. Υ := ρ f L ρ s h , β := ρ f g L 3 B , δ := ρ f g L 3 γ , ε := L H .
Acknowledgements
D.A.-O. has been partially supported by grant RYC2023-045563-I (MICIU/AEI/10.13039/501100011033 and ESF+). D.A.-O. and R.G.-B. are supported by the project “Análisis Matemático Aplicado y Ecuaciones Diferenciales” Grant PID2022-141187NB-I00 funded by MCIN/AEI/10.13039/501100011033/FEDER, UE. J. S. Ziebell acknowledges that part of this work was carried out during a research visit to the University of Cantabria.
References