Reduced-order models and controllers for continuous-time stochastic systems: an information theory approach

Leland derives reduced-order models and controllers for continuous-time stochastic systems using mutual information as the approximation criterion rather than classical H2/H-infinity norms. The key result is that minimizing the loss of mutual information between input and output under model reduction leads to a new balanced truncation-style procedure grounded in information theory, providing an information-theoretic justification for balanced truncation and extending it to the controller reduction problem.