Inverse problems for a class of measure dependent dynamical systems (Q5316233)

The authors consider the \(n\)-dimensional generalized delay system: \[ x'(t)=g(x(t),x_t,P_1,P_2,\ldots,P_{n_p}), \quad t >0 \tag{1} \] \[ x_0(t)=\Phi(t),\quad t\in Q =[-r,0], \quad x(0)=\Phi(0), \tag{2} \] \(x_t\) and \(\Phi\) denoting, respectively, the function \(\theta\to x(t + \theta)\), \(-r\leq \theta \leq 0\) and a piecewise absolutely continuous function defined on the interval \(Q\). Finally, \(P_i\), \(i=1,\ldots,n_p\), are elements of \({\mathcal P}(Q)\), the space of probability measures on \((Q,S)\), \(S\) being Borel \(\sigma\)-algebra related to \(Q\). First the authors prove the well-posedness for the direct problem (1), (2). Then they deal with the problem of determining all the probability vector measures \(\pi=(P_1,P_2,\ldots,P_{n_p})\) in the class \[ \Pi_{ad}=\prod_{j=1}^{n_p}{\mathcal P}_{ad}(Q) \] minimizing the following functional related to the solution \(x\) of problem (1), (2): \[ J(\pi,{\mathbf d})=\sum_{i=1}^{n_t} | Cx(t_i,\pi)-d_i| ^2. \] Here \({\mathbf d}=(d_1,d_2,\ldots,d_{n_d})\), \(n_d\leq n\), are observed data, \(C\in \mathbb R^{n_d\times n}\) is an observation matrix and \(t_1\), \(i=1,\ldots,n_t\), are sampled times. Of course, the authors show that the set of solutions to the inverse problem is not empty and prove a continuous dependence result involving data \({\mathbf d}\). Approximation methods are discussed both for the direct and the inverse problem. The paper ends with an application of the previous results to a model related to HIV infection dynamics.