On Minimization and Maximization of Entropy in Various Disciplines

From MaRDI portal





The authors, a probabilist and a physicist, start with a Lagrange multiplyer proof of the essentially known theorem on the characterization of measures \(Q\) with minimal relative entropy \(H(Q,P)\) to a \(\sigma\)-finite reference measure under the constraint \(\int X \,dQ = c\) where \(X:\Omega \rightarrow \mathbb{R}\). They also give a variational definition of a critical point \(Q\) of \(H(Q,p)\) and give a similar explicit characterization of those. Then they give a survey over applications, interpretations and variants of these results in statistical physics, large deviation theory, statistics, information theory and financial mathematics, the main new results being in the last field. In statistical physics this problem translates into maximization of the (Boltzmann) entropy under fixed interior energy \(E\) and fixed mass \(M\). The Lagrange multiplier in their proof is the temperature \(T\). The maximizers are called stable states which turn out to be Gibbs states. The critical points correspond to metastable states. In data compression the result allows to determine the maximal information contained in a coding under fixed average cost of the coding. In financial mathematics the result is used to determine the minimal entropy martingale measure in the general discrete time asset price model satisfying natural integrability conditions. Dually to this an explicit solution of the exponential utility maximization problem is given.




Cited in
(22)








This page was built for publication: On Minimization and Maximization of Entropy in Various Disciplines

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q4830865)