Anisotropic estimates for the two-dimensional Kuramoto-Sivashinsky equation (Q481248)

The authors are concerned with the famous Kuramoto-Sivashinsky (K-S) equation, with periodic boundary conditions, in a two-dimensional rectangle \(\Omega=(0,L_1)\times (0,L_2)\), with \(L_2\leq L_1\) and \(L_1\geq 2\pi\). More specifically they analyze the system satisfied by the gradient of the solution to (K-S), which reads as follows: \[ \begin{cases} D_tu_1+\Delta^2u_1+\Delta u_1+u_1D_xu_1+u_2D_xu_2,\\D_tu_2+\Delta^2u_2+\Delta u_2+u_1D_yu_1+u_2D_yu_2,\\ D_yu_1=D_xu_2\end{cases}\eqno(1) \] In [\textit{L. Molinet}, J. Dyn Differ. Equ. 12, No. 3, 533--566 (2000; Zbl 0970.35132)] the author proved the global existence of the solution the Cauchy problem \((u_1(0,\cdot),u_2(0,\cdot))=(u_{0,1},u_{0,2})\) associated with (1) with periodic boundary conditions, provided that \(L_2\leq CL_1^{-67/35}\), for some positive constant \(C\), and \(\| u_{0,j}\|_{L^2(\Omega)}\) \((j=1,2)\) are small enough. Here, the authors extend such a result and prove that there exists a positive constant \(C_0\geq 1\) such that, if \(L_2\leq C_0^{-1}L_1^{-22/25}\) and \(u_{0,1}, u_{0,2}\in L^2_{\mathrm{per}}(\Omega)\) have zero average in \(\Omega\), \(\mathrm{curl}(u_{0,1},u_{0,2})=0\) and both \(\| u_{0,1}\|_{L^2(\Omega)}\) and \(\| u_{0,2}\|_{L^2(\Omega)}\) do not exceed \(C_0^{-1}L_2^{-2}\), then the solution to \((1)\) such that \(u_j(0,\cdot)=u_{0,j}\) \((j=1,2)\) is global in time. Further, estimates of the long-time behaviour of the \(L^2(\Omega)\)-norms of \(u_1\) and \(u_2\) are also provided. The core of the proof are the differential inequalities \[ \begin{aligned} &\frac{1}{2}\frac{d}{dt}\| u_2(t,\cdot)\|_{L^2(\Omega)}^2+\frac{1}{2}\|\Delta u_2(t,\cdot)\|_{L^2(\Omega)}^2\\ &\leq C(\|\Phi\|_{L^2(\Omega)},\| N[u_1(t,\cdot)]\|_{L^2(\Omega)},\sqrt{F[u_1(t,\cdot)]})\|\Delta u_2(t,\cdot)\|_{L^2(\Omega)}^2,\end{aligned} \] \[ \begin{aligned} &\frac{1}{2}\frac{d}{dt}\| N[u_1(t,\cdot)]\|_{L^2(\Omega)}^2+\frac{1}{2}\|\Delta N[u_1(t,\cdot)]\|_{L^2(\Omega)}^2\\ &\leq C(\|\Phi\|_{L^2(\Omega)},\| u_2(t,\cdot)\|_{L^2(\Omega)},\sqrt{F(u_1)})\|\Delta N[u_1(t,\cdot)]\|_{L^2(\Omega)}^2,\end{aligned} \] and \[ \begin{aligned} \frac{1}{2}\frac{d}{dt}F[u_1(t,\cdot)]\leq &-\frac{1}{2}F[u_1(t,\cdot)]+C(\|\Phi(\cdot+b^\ast)\|_{W^{2,2}(\Omega)})\\ &+C\sqrt{F[u_1(t,\cdot)]}(\|\Delta N[u_1(t,\cdot)]\|_{L^2(\Omega)}^2+\|\Delta u_2(t,\cdot)\|_{L^2(\Omega)}^2),\end{aligned} \] whose proofs rely on some integrations by parts and the Poincaré and Agmon inequalities. Here, \(\Phi\) is a suitable function depending only on \(x\) and with null average over \([0,L_1]\), \(N[f]\) is the difference between \(f\) and its average \(M[f]\) with respect to the variable \(y\). Finally, \[ F[u_1(t,\cdot)]=\inf_{b\in [0,L_1]}\int_{\Omega}|(M[u_1(t,\cdot)])(x)-\Phi(x+b)|^2dxdy \] and \(b^\ast\) is the point where the infimum is attained. The previous differential inequalities are used to prove, by contradiction, that, if the estimate \(F[u_1(t,\cdot)]+\| N[u_1(t,\cdot)]\|_{L^2(\Omega)}^2+\| u_2(t,\cdot)\|_{L^2(\Omega)}^2\leq CL_2^{-4}\) is satisfied for a suitable constant \(C\) at time \(t=0\), then it is satisfied in all the interval where the solution exists, provided that \(L_2\leq C_0^{-1}L_1^{-22/25}\). This allows to conclude the proof. It is worth stressing that, even if the proof is technical, the authors succeeded in making it friendly to the reader, including all the necessary details.