The onset of instability in first-order systems (Q1643999)

The paper is devoted to the study of ill-posedness in Sobolev spaces for the Cauchy problem for first-order quasilinear real systems of partial differential equations of the form \[ \partial_t U + \sum_{1 \leq j \leq d} A_j(t,x,U)\partial_{x_j} U=F(t,x,U),\,\,\,U(0,x)=U_0(x). \] There exist several results about Lax-Mizohata-type theorems for necessity of well-posedness of non-characteristic initial value problems for first-order systems. Necessary is the property of hyperbolicity. The authors are interested in general ill-posedness results under defect assumptions of hyperbolicity that describe the transition from hyperbolicity (real eigenvalues of the principal symbol) to ellipticity (existence of non-real eigenvalues of the principal symbol). The well-posedness or ill-posedness is inderstood in the sense of Hadamard, that is, existence and regularity of a flow. The authors show that even a weak defect of hyperbolicity leads to a strong Hadamard instability. In the following three of the results will be presented. \par To define ill-posedness the authors assume that there exist a local smooth solution $\Phi$ of the above Cauchy problem with a large Sobolev regularity, $\Phi \in C^\infty([0,T_0],H^{s}(U))$ for a positive $T_0$, an open set $U \in \mathbb{R}^d$ and a large $s$. Then the above Cauchy problem is ill-posed in a neighborhood of the reference solution $\Phi$ if either data arbitrarily close to $\Phi(0)$ fail to generate trajectories, or if trajectories $U(t)$ issued from $U(0)$ close to $\Phi(0)$ do not satisfy a suitable Hölder condition with respect to $\Phi(t)$. \par To explain some results let us introduce the characteristic polynomial of the principal part of the above system is defined as follows: \[ P(t,x,\xi,\lambda):=\det \big(\lambda\,I - A(t,x,\xi)\big). \] \par \textit{Initial ellipticity} \par Some ellipticity assumption is posed for $t=0$. It holds $P(0,\omega_0)=0$ for some vector $\omega_0=(x_0,\xi_0,\lambda_0) \in U\times (\mathbb{R}^d \setminus \{0\}) \times (\mathbb{C} \setminus \mathbb{R})$. Then the Cauchy problem for the starting system is ill-posed in a neighborhood of the reference solution $\Phi$. \par \textit{Non-semisimple defect of hyperbolicity} \par In this case the principal symbol is supposed to be hyperbolic in $t=0$. To describe situations in which some roots of $P$ are not real for $t>0$ let \[ \Gamma:=\{\omega=(x,\xi,\lambda) \in U\times (\mathbb{R}^d \setminus \{0\}) \times \mathbb{R} : P(0,\omega)=0\}.\] The authors assume for $\omega_0 \in \Gamma$ the following conditions: \[ (\partial_\lambda P)(0,\omega_0)=0,\,\,(\partial^2_\lambda P)(0,\omega_0)\neq 0,\,\, (\partial^2_\lambda P\partial_t P)(0,\omega_0)> 0. \] {Under these conditions and the assumption that the other eigenvalues of $A(0,x_0,\xi_0)$ are simple the Cauchy problem for the starting system is ill-posed in a neighborhood of the reference solution $\Phi$.} \par \textit{Semisimple defect of hyperbolicity} \par In this case the principal symbol is supposed to be hyperbolic in $t=0$. But now the authors suppose besides \[ (\partial_\lambda P)(0,\omega_0)=0,\,\,(\partial^2_\lambda P)(0,\omega_0)\neq 0,\,\, \] a time-differentiable defect of hyperbolicity of order $2$. For the vector $\omega_0=(x_0,\xi_0,\lambda_0) \in \Gamma$ the component $\lambda_0$ is a semisimple eigenvalue of $A(0,x_0,\xi_0)$. Moreover, for all vectors $\omega=(x,\xi,\lambda)$ in a neighborhood of $\omega_0$ in $\Gamma$ it holds \[ (\partial_\lambda P)(0,\omega)=(\partial_t P)(0,\omega)=0,\] where $\lambda$ is a semisimple eigenvalue of $A(0,x,\xi)$. Under these conditions and the assumption that the other eigenvalues of $A(0,x_0,\xi_0)$ are simple the Cauchy problem for the starting system is ill-posed in a neighborhood of the reference solution $\Phi$. \par These results are applied to several models as Burgers systems, van der Waals gas dynamics, Klein-Gordon Zakharov systems, as well.