Foundations of Constructive Probability Theory

TL;DR

This book develops a constructive foundation for probability theory and stochastic processes, emphasizing explicit constructions, continuity, and new theorems, offering a rigorous and computationally meaningful approach.

Contribution

It introduces a fully constructive framework for probability and stochastic processes, including new theorems and strengthened convergence results, bridging classical and constructive methods.

Findings

Constructive treatment of stochastic processes including Brownian motion

New maximal inequality for $L_p$-martingales with explicit convergence rates

Proof that certain distribution conditions imply right Hölder continuity

Abstract

We provide a systematic, thorough treatment of the foundations of probability theory and stochastic processes along the lines of E. Bishop's constructive analysis. Every existence result presented shall be a construction; and the input data, the construction procedure, and the output objects shall be regarded as integral parts of the theorem. A brief description of this approach is in Part I of this book. Part II develops basic topics in probability theory in this constructive framework, expanding on [Bishop and Bridges 1985, Springer], and in terms familiar to probabilists. Part III, the main part of the book, builds on Part II to provide a new constructive treatment of stochastic processes, in the spirit and style of Kolmogorov's constructive methods for Brownian motion. Topics include a Daniell-Kolmogorov-Skorokhod construction of random fields, measurable random fields, a.u. continuous processes, a.u. c\`adl\`ag processes, martingales, a.u. c\`adl\`ag and strongly Markov processes, and Feller processes. This text also contains some new theorems in classical probability theory. Each construction theorem is accompanied by a metrical continuity theorem. For example, the construction of Markov processes from semigroups is shown to be metrically continuous, which strengthens the sequential weak convergence in the classical approach. Another new result is a maximal inequality for $L_p$-martingales for $p \ge 1$. In addition to providing explicit rates of convergence, this maximal inequality also provides a unified proof of a.u. convergence of martingales, which previously required separate proofs for the cases $p>1$ and $p=1$. A third new result is a proof that a familiar condition on the triple-joint distributions implies that a process is not only a.u. c\`adl\`ag, but also right Hoelder, in a sense made precise in the text.