A time-varying analogue of Jensen's formula (Q5943199)

For a function \(f(z)\) of a complex variable having poles \(\{ p_i\}\) and zeros \(\{ z_j\}\) outside the unit disc and no poles and zeros on the boundary, the Jensen formula is \[ (2\pi)^{-1}\int_{-\pi }^\pi\log |f(e^{i\omega })|d\omega =\log |f(0)|+\sum_i\log |z_i|-\sum_i\log |p_i|. \] The analogue of this formula for an \(m\times m\) matrix \(F(z)\) is obtained by substitution of \(\det F(e^{i\omega })\) and \(\det F(0)\) into the place of \(f(e^{i\omega })\) and \(f(0)\), respectively, and can be easily deduced from the transfer function representation of \(F(z)\) in the case where \(F(z)\) has a discrete time-invariant system realization with a finite-dimensional state space. In the paper under review, a time-varying analogue of Jensen's formula is presented. The transfer \(m\times m\) matrix function is replaced by the input-output operator, which is a bounded operator in \(l_2^{m\times m}\). Poles and zeros of the transfer function are replaced by elements of the spectrum of two weighted shifts, one associated with the system, the other with its inverse. In the general case, a time-varying analogue of Jensen's formula appears in the form of a two-sided inequality which, however, for the regular (in some definite sense) case turns into equality. The key role is played by exponent dichotomies and inner-outer factorizations of block lower triangular operators in \(l_2^{m\times m}\).