Dependence of solutions and eigenvalues of measure differential equations on measures (Q1941135)

This interesting and well-written paper is concerned with systems of linear integral equations of the form \[ \begin{aligned}y(t)&=y_0+\int_{[0,t]} z(s)\,{\mathrm d}s,\quad t\in[0,1],\cr z(t)&=\begin{cases} z_0, & t=0,\cr z_0-\int_{[0,t]} y(s)\,{\mathrm d}\mu(s), & t\in(0,1].\end{cases}\end{aligned}\tag{1} \] The function \(\mu:[0,1]\to\mathbb R\) (or \(\mu:[0,1]\to\mathbb C\)) in the Stieltjes integral is an element of the space \(\mathcal M_0\) of bounded variation functions, which are right-continuous on \((0,1)\) and satisfy \(\mu(0+)=0\). Note that the second-order linear differential equation \(y''+q(t)y=0\) is equivalent to \(y'=z\), \(z'=-q(t)y\), and hence represents a special case of (1) corresponding to \(\mu(s)=\int_{[0,s]}q(x)\,{\mathrm d}x\). The authors write (1) in the abbreviated form \[ {\mathrm d}y^\bullet+y{\mathrm d}\mu(t)=0 \] and use the term ``second-order linear measure differential equation'', which stems from the fact that \(\mu\in \mathcal M_0\) defines a measure on \([0,1]\). For every initial condition \((y_0,z_0)\), problem (1) has a unique solution \[ t\mapsto (y(t,y_0,z_0,\mu),z(t,y_0,z_0,\mu)),\quad t\in[0,1]. \] Its first component \(y\) is absolutely continuous, has a derivative a.e.~in \([0,1]\), and its right derivative exists and equals \(z\) everywhere in \((0,1)\). The second component \(z\), which is called the generalized right-derivative of \(y\) and is denoted by \(y^\bullet\), can have jumps corresponding to the jumps of \(\mu\). The goal is to study the dependence of solutions on the choice of \(\mu\). One possible approach to this problem comes from the theory of linear generalized ordinary differential equations (see, e.g., [\textit{M. Tvrdý}, Mem. Differ. Equ. Math. Phys. 25, 1--104 (2002; Zbl 1081.34504); \textit{Z. Halas} and \textit{M. Tvrdý}, Funct. Differ. Equ. 16, No. 2, 299--313 (2009; Zbl 1181.34011)]). When the space \(\mathcal M_0\) is equipped with the supremum norm \(\|\cdot\|_\infty\), it is known that the solutions depend continuously on \(\mu\), i.e., \(\|y(\cdot,y_0,z_0,\mu_n)-y(\cdot,y_0,z_0,\mu_0)\|_\infty\to 0\) and \(\|y^\bullet(\cdot,y_0,z_0,\mu_n)-y^\bullet(\cdot,y_0,z_0,\mu_0)\|_\infty\to0\) whenever \(\|\mu_n-\mu_0\|_\infty\to 0\). The authors of the present paper argue that instead of the uniform convergence \(\|\mu_n-\mu_0\|_\infty\to 0\), it is sometimes more natural to work with weak\(^*\) convergence, where \(\mu_n\to\mu_0\) means that \[ \lim_{n\to\infty}\int_0^1f{\mathrm d}\mu_n=\int_0^1f{\mathrm d}\mu_0 \] for every continuous function \(f\) on \([0,1]\). (For example, when \(a\to 0+\), the Dirac measures \(\delta_a\) converge to \(\delta_0\) in the weak\(^*\) topology, but not in the supremum norm.) Their results read as follows. \begin{itemize} \item[1)] If \(\mu_n\to\mu_0\), then \(y(\cdot,y_0,z_0,\mu_n)\to y(\cdot,y_0,z_0,\mu_0)\) in the supremum norm and \(y^\bullet(\cdot,y_0,z_0,\mu_n)\to y^\bullet(\cdot,y_0,z_0,\mu_0)\) in the weak\(^*\) topology. \item[2)] When \(\mathcal M_0\) is equipped with the total variation norm, then, for every \(t\in[0,1]\), the mapping \(\mu\mapsto (y(t,y_0,z_0,\mu),y^\bullet(t,y_0,z_0,\mu))\), \(\mu\in \mathcal M_0\), is continuously Fréchet-differentiable. \end{itemize} The next part is devoted to eigenvalue problems of the form \[ {\mathrm d}y^\bullet+\lambda y{\mathrm d}t+y{\mathrm d}\mu(t)=0,\quad t\in[0,1], \] subject to the Dirichlet conditions \(y(0)=y(1)=0\), or the Neumann conditions \(y^\bullet(0)=y^\bullet(1)=0\). It is shown that both problems have increasing sequences of real eigenvalues \(\{\lambda_m(\mu)\}_m\), where \(\lambda_m(\mu)\to\infty\) for \(m\to\infty\). Moreover, the mappings \(\mu\mapsto \lambda_m(\mu)\) are continuous in the weak\(^*\) topology, and continuously Fréchet-differentiable with respect to the total variation norm in \(\mathcal M_0\). The final section contains an application of the previous results to extremal eigenvalue problems for Sturm-Liouville operators with integrable potentials.