Optimal control problem for the Lyapunov exponents of random matrix products (Q1584024)

The author studies an optimal problem in which the objective function is the essential supremum of the Lyapunov exponents for a dynamical system described by random matrix products when these matrices depend on a controlled Markov process \((\xi_n)\) with values in a finite or countable set \(I\). \((\xi_n)\) has transition probability \(P(a)= (P_{ij}(a):i, j\in I)\), with a control parameter \(a\). For any admissible control \((u_t)\) the \(\mathbb{R}^d\)-valued random variables \((X_n:n= 0,1,\dots)\) are given by the difference equation: \[ X_{n+1}= M(\xi_{n+1}, Y_{n+1}) X_n,\quad X_0= x\in\mathbb{R}^d,\tag{1} \] where \((Y_n)\) are i.i.d. random variables and \(M(i,y)\) are invertible \(d\times d\) matrices. \(X^u_n(x)\) is the solution of (1) associated with the control \((u_t)\). The process \((u_t)\) affects the solutions through \(P(a)\). Some variants of the Lyapunov exponent are used. A decision \(\pi_t\) at a time \(t\) is a stochastic kernel, and a sequence of decisions is called a policy. The Markov stationary policy is defined. The main result: If there exists a Markov policy such that some condition is satisfied, then there exists a stationary policy which minimizes the Lyapunov exponent of solutions of (1). In this case the spectrum of the system (1) consists of only one element.