scientific article; zbMATH DE number 194156
This doctoral thesis concerns the ergodic-theoretic study of stationary stochastic processes \(\{X_ i\}_ i\) taking values in a countable Polish space S with metric \(\sigma\) and Borel sets \(\Lambda\), which have the mixing property called very weak Bernoulli with rate of order 1/n. The relevant definitions are as follows: For each positive integer n, define the Wasserstein distance \(\rho_ n(\mu,\nu)\) between two Borel probability measures \(\mu\), \(\nu\) on the Cartesian-product space \(S^ n\) by \[ \rho_ n(\mu,\nu)\equiv \inf E\{n^{-1}\sum^{n}_{i=1}\sigma (X_ i,Y_ i)\} \] where the inf is taken over all random 2n-vectors (\b{X}\({}_ n,\underline Y_ n)\) where \b{X}\({}_ n\) and \b{Y}\({}_ n\) are defined with values in \(S^ n\) on the same probability space (\(\Omega\),\({\mathcal F},P)\) in such a way that their probability laws \(P\circ \underline X_ n^{-1}\) and \(P\circ \underline Y_ n^{-1}\) are respectively equal to \(\mu\) and \(\nu\). (Here \(E\{\) \(\cdot \}\) denotes the expectation operator with respect to P.) The process \(\{X_ i\}_ i\) on (\(\Omega\),\({\mathcal F},P)\) with values in S is said to be very weak Bernoulli with rate \(\epsilon\) (n) for a sequence \(\epsilon\) (n) of constants decreasing to 0 as \(n\to \infty\) if for all integers n and m there exist measurable subsets D(n,m) of \(\sigma^ 0_{-m}\equiv \sigma (X_{-m},...,X_{-1},X_ 0)\) such that \(P\{D(n,m)\}\geq 1-\epsilon (n)\) and if \b{X}\({}_ n\equiv (X_ 1,...,X_ n)\), then \[ for\quad all\quad A\in \sigma^ 0_{-m},\quad \rho_ n(P\circ \underline X_ n^{-1}, P(\underline X_ n^{-1}(\cdot)| A))\leq \epsilon (n). \] A third definition, apparently due to the author and restricted to the case of countable S, is that \(\{X_ i\}_ i\) on (\(\Omega\),\({\mathcal F},P)\) is quasi-Markovian (with constant M) for \(0\leq M<\infty\), if for every \(x\in S\) and \b{i}\(\in S^ k\) such that \(P\{X_ 0=x\), \(X_{-1}=x_ 1,...,X_{-k}=x_ k\}>0\), \[ \sup_{n\geq 1}n \rho_ n(P(\underline X_ n^{-1}(\cdot)| X_ 0=x),\quad P(\underline X_ n^{- 1}(\cdot)| X_ 0=x,\quad X_{-1}=x_ 1,...,X_{-k}=x_ k))\leq M. \] Countable-state Markov chains \(\{X_ i\}_ i\), as well as the continuous factors of finite-state Markov chains (called ``sophic systems), are all examples of quasi-Markovian processes. The main results of the thesis under review apply only to stationary processes \(\{X_ i\}_ i\) with values in countable S. The first result, which connects the very weak Bernoulli property with the idea of quasi- Markovian processes, states that \(\{X_ i\}_ i\) is very weak Bernoulli with rate \(\epsilon\) (n) of order 1/n (abbreviated from now on as VWB \({\mathcal O}(1/n))\) if and only if for all \((x_ 0,x_ 1,...,x_ m)\in S^{m+1}\) such that \(P\{X_ 0=x,X_{-1}=x_ 1,|...,X_{-m}=x_ m\}>0,\) \[ \sup_{n\geq 1}n \rho_ n(P(\underline X_ n^{- 1}(\cdot)| X_ 0=x),\quad P(\underline X_ n^{-1}(\cdot)| X_ 0=x,\quad X_{-1}=x_ 1,...,X_{-k}=x_ k))\leq M. \] When S is finite, this is refined to yield \(\{X_ i\}_ i\) VWB \({\mathcal O}(1/n)\) if and only if \(\{X_ i\}_ i\) is quasi-Markovian and mixing, which holds if and only if \(\{X_ i\}_ i\) is quasi-Markovian and weakly mixing. These theorems are proved by means of careful combinatorial analysis, and the proofs are remarkably self-contained. The author applies these results to study (image measures for) finitary factors of stationary countable-state Markov chains \(\{X_ i\}_ i\) for which \(P\{X_ n=x\), \(X_{n-1}\neq x,...,X_ 1\neq x| X_ 0=x\}\) decreases exponentially fast in n for each fixed \(x\in S\). Some special factors of interest are the ``IEP Codes, the k'th coordinate of which is the number of transitions prior to time k since an underlying Markov process \(\{X_ i\}_ i\) last visited a specified state x. By investigating such factors, the author shows that \[ for\quad S\quad finite,\quad \{X_ i\}_ i\quad VWB\quad {\mathcal O}(1/n)\begin{matrix} \Leftrightarrow \\ \nrightarrow \end{matrix} \{X_ i\}_ i\quad sophic. \] Many special results are proved for IEP-Codes, including criteria for the VWB \({\mathcal O}(1/n)\) property. Finally, the author related VWB \({\mathcal O}(1/n)\) to other mixing conditions, showing that in general \(\phi\)-mixing implies VWB \({\mathcal O}(1/n)\), and if the metric \(\sigma\) (x,y) is bounded away from 0 for all unequal pairs of elements x,y of S, then \(\{X_ i\}_ i\) which is VWB \({\mathcal O}(1/n)\) must also be weak Bernoulli. The thesis concludes with a number of open problems.
- scientific article; zbMATH DE number 1775217
- Almost-sure waiting time results for weak and very weak Bernoulli processes
- On the rate of Poisson process approximation to a Bernoulli process
- Quite weak Bernoulli with exponential rate and percolation for random fields
- On the boundedness of Bernoulli processes
- On the rate of Poisson process approximation to a Bernoulli process
- On the rate of Poisson approximation to Bernoulli partial sum processes
- On the boundedness of Bernoulli processes over thin sets
- Stationary measures for randomly chosen maps
- Veršik processes and very weak Bernoulli processes with summable rates are independent
- scientific article; zbMATH DE number 4034800 (Why is no real title available?)
- Measures of Dependence For Processes in metric spaces
- Mixing Properties of a Class of Bernoulli-Processes
- scientific article; zbMATH DE number 1775217 (Why is no real title available?)
- Finitely determined processes in metric spaces
- How sampling reveals a process
This page was built for publication:
Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q4040948)