Information and entropy flow in the Kalman-Bucy filter (Q2575531)

In the paper the continuous Kalman-Bucy filter (K-B) is investigated from the point of view of the information theory and the statistical mechanics. It is known that the K-B filter estimates the signal at each time \(t\), by the use of the information contained in the preceeding values on \((0,t)\). This is performed in a recursive way, calculating the conditional distribution of \(X_t\) (the signal at time \(t\)) for a given \(Y_t\) (the observed signal) on \(s\in(0,t)\). The paper, based on some papers related to the thermodynamics of computation initiated in last decades, gives an illustration of the Landauer principle according to the entropy can be increased by the erasure of information. The novelty of the paper lies in the fact of considering the whole systems as being constituted of some subsystems interacting between them. In the first analogy, the partially observed systems interacts with a heat bath. Owing to this interaction, the signal evolves towards a stationary state which maximises the entropy of the observed signal and the heat bath. This could be viewed as a second law of thermodynamics. But this law can be broken in the presence of partial observations (acting as a Maxwellian demon). The second analogy considers the joint system, comprising the signal and the filter, interacting with the heat bath and a loop is established in this situation. So, in the stationary state, the system is a type of perpetual motional machine. Here, the authors introduce the concept of interactive entropy flow to describe the various componentes of the whole system. The theory here developed concerns finite-dimensional mesoscopic systems but, as the authors indicate, it can be extended to infinite-dimensional Hilbert spaces. Finally, we think we are faced with a very important paper on the way to creaty a unifying point of view in the thermodymamic system theory.