On universality in penalisation problems with multiplicative weights (Q2088489): Difference between revisions

For a probability space \((\Omega,\mathcal{F},P)\) equipped with a filtration \((\mathcal{F}_s)_{s\geq0}\) and for a non-negative weight process \(\Gamma=(\Gamma_t)_{t\geq0}\), the author considers a \textit{penalisation problem}, i.e. the problem of finding a so-called \textit{penalised probability} \(P^\Gamma\) on \((\Omega,\mathcal{F})\) such that \begin{align*} \frac{\int_\Omega F_s\Gamma_t dP}{\int_\Omega \Gamma_t dP}\longrightarrow \int_\Omega F_s dP^\Gamma, \quad t\to\infty, \end{align*} for all \(s\geq0\) and all bounded \(\mathcal{F}_s\)-measurable functionals \(F_s\). One can often find a \(\sigma\)-finite measure \(\mathcal{P}\) on \((\Omega,\mathcal{F})\), which is independent from a particular weight \(\Gamma\), such that \begin{align*} P^\Gamma(A)=\frac{\int_A \Gamma_\infty d\mathcal{P}}{\int_\Omega \Gamma_\infty d\mathcal{P}},\qquad A\in\mathcal{F}, \end{align*} holds with a suitable limit \(\Gamma_\infty\) of \(\Gamma_t\) for a certain class of weights \(\Gamma\) (this class is called the \textit{universality class} of \(\mathcal{P}\)). In this paper, the author gives a general framework for characterization of universality classes in penalisation problems with multiplicative weights. The author defines \textit{unweighted measures} and their \textit{subsequent Markov property}, discusses a sufficient condition for a given weight to belong to the universality class of another given weight, presents penalisation limits for some penalisation problems with constant clock and with exponential clock, revises some known results on Brownian, Levy and Langevin penalisation in terms of introduced general framework. For the entire collection see [Zbl 1493.11005].