-
Iterated proportional fitting procedure and infinite products of stochastic matrices
Authors:
Jean Brossard,
Christophe Leuridan
Abstract:
The iterative proportional fitting procedure, introduced in 1937 by Kruithof, aims to adjust the elements of an array to satisfy specified row and column sums. Given a rectangular non-negative matrix $X_0$ and two positive marginals $a$ and $b$, the algorithm generates a sequence of matrices $(X_n)$ starting at $X_0$, supposed to converge to a biproportional fitting, that is, to a matrix $Y$ whose…
▽ More
The iterative proportional fitting procedure, introduced in 1937 by Kruithof, aims to adjust the elements of an array to satisfy specified row and column sums. Given a rectangular non-negative matrix $X_0$ and two positive marginals $a$ and $b$, the algorithm generates a sequence of matrices $(X_n)$ starting at $X_0$, supposed to converge to a biproportional fitting, that is, to a matrix $Y$ whose marginals are $a$ and $b$ and of the form $Y=D_1X_0D_2$, for some diagonal matrices $D_1$ and $D_2$ with positive diagonal entries.
When a biproportional fitting does exist, it is unique and the sequence $(X_n)$ converges to it at an at least geometric rate. More generally, when there exists some matrix with marginal $a$ and $b$ and with support included in the support of $X_0$, the sequence $(X_n)$ converges to the unique matrix whose marginals are $a$ and $b$ and which can be written as a limit of matrices of the form $D_1X_0D_2$. In the opposite case, the sequence $(X_n)$ diverges but both subsequences $(X_{2n})$ and $(X_{2n+1})$ converge.
In the present paper, we use a new method to prove again these results and determine the two limit-points in the case of divergence. Our proof relies on a new convergence theorem for backward infinite products $\cdots M_2M_1$ of stochatic matrices $M_n$, with diagonal entries $M_n(i,i)$ bounded away from $0$ and with bounded ratios $M_n(j,i)/M_n(i,j)$. This theorem generalizes Lorenz' stabilization theorem.
We also provide an alternative proof of Touric and Nedić's theorem on backward infinite products of doubly-stochatic matrices, with diagonal entries bounded away from $0$. In both situations, we improve slightly the conclusion, since we establish not only the convergence of the sequence $(M_n \cdots M_1)$, but also its finite variation.
△ Less
Submitted 8 May, 2018; v1 submitted 29 June, 2016;
originally announced June 2016.
-
Densité des orbites des trajectoires browniennes sous l'action de la transformation de Lévy
Authors:
Jean Brossard,
Christophe Leuridan
Abstract:
Let T be a measurable transformation of a probability space $(E,\mathcal {E},π)$, preserving the measure π. Let X be a random variable with law π. Call K(\cdot, \cdot) a regular version of the conditional law of X given T(X). Fix $B\in \mathcal {E}$. We first prove that if B is reachable from π-almost every point for a Markov chain of kernel K, then the T-orbit of π-almost every point X visits B.…
▽ More
Let T be a measurable transformation of a probability space $(E,\mathcal {E},π)$, preserving the measure π. Let X be a random variable with law π. Call K(\cdot, \cdot) a regular version of the conditional law of X given T(X). Fix $B\in \mathcal {E}$. We first prove that if B is reachable from π-almost every point for a Markov chain of kernel K, then the T-orbit of π-almost every point X visits B. We then apply this result to the Lévy transform, which transforms the Brownian motion W into the Brownian motion |W| - L, where L is the local time at 0 of W. This allows us to get a new proof of Malric's theorem which states that the orbit under the Lévy transform of almost every path is dense in the Wiener space for the topology of uniform convergence on compact sets.
△ Less
Submitted 1 August, 2012;
originally announced August 2012.
-
Characterising Ocone local martingales with reflections
Authors:
Jean Brossard,
Christophe Leuridan
Abstract:
Let $M = (M_t)_{t \ge 0}$ be any continuous real-valued stochastic process such that $M_0=0$. Chaumont and Vostrikova proved that if there exists a sequence $(a_n)_{n \ge 1}$ of positive real numbers converging to 0 such that $M$ satisfies the reflection principle at levels 0, $a_n$ and $2a_n$, for each $n \ge 1$, then $M$ is an Ocone local martingale. They also asked whether the reflection princi…
▽ More
Let $M = (M_t)_{t \ge 0}$ be any continuous real-valued stochastic process such that $M_0=0$. Chaumont and Vostrikova proved that if there exists a sequence $(a_n)_{n \ge 1}$ of positive real numbers converging to 0 such that $M$ satisfies the reflection principle at levels 0, $a_n$ and $2a_n$, for each $n \ge 1$, then $M$ is an Ocone local martingale. They also asked whether the reflection principle at levels 0 and $a_n$ only (for each $n \ge 1$) is sufficient to ensure that $M$ is an Ocone local martingale. We give a positive answer to this question, using a slightly different approach, which provides the following intermediate result. Let $a$ and $b$ be two positive real numbers such that $a/(a+b)$ is not dyadic. If $M$ satisfies the reflection principle at the level 0 and at the first passage-time in $\{-a,b\}$, then $M$ is close to a local martingale in the following sense: $|\eef[M_{S \circ M}]| \le a+b$ for every stopping time $S$ in the canonical filtration of $\wwf = \{w \in \CC(\rrf_+,\rrf) : w(0)=0\}$ such that the stopped process $M_{\cdot \wedge (S \circ M)}$ is uniformly bounded.
△ Less
Submitted 1 August, 2012;
originally announced August 2012.
-
Chaînes de Markov Constructives Indexées par Z
Authors:
Jean Brossard,
Christophe Leuridan
Abstract:
Nous étudions les cha\^{ı}nes de Markov $(X_n)_{n\in\mathbf{Z}}$ gouvernées par une relation de récurrence de la forme $X_{n+1}=f(X_n,V_{n+1})$, où $(V_n)_{n\in\mathbf{Z}}$ est une suite de variables aléatoires indépendantes et de même loi telle pour tout $n\in \mathbf{Z}$, $V_{n+1}$ est indépendante de la suite $((X_k,V_k))_{k\le n}$. L'objet de l'article est de donner une condition nécessaire…
▽ More
Nous étudions les cha\^{ı}nes de Markov $(X_n)_{n\in\mathbf{Z}}$ gouvernées par une relation de récurrence de la forme $X_{n+1}=f(X_n,V_{n+1})$, où $(V_n)_{n\in\mathbf{Z}}$ est une suite de variables aléatoires indépendantes et de même loi telle pour tout $n\in \mathbf{Z}$, $V_{n+1}$ est indépendante de la suite $((X_k,V_k))_{k\le n}$. L'objet de l'article est de donner une condition nécessaire et suffisante pour que les innovations $(V_n)_{n\in\mathbf{Z}}$ déterminent complètement la suite $(X_n)_{n\in \mathbf{Z}}$ et de décrire l'information manquante dans le cas contraire.
△ Less
Submitted 26 July, 2007;
originally announced July 2007.
-
Perte d'information dans les transformations du jeu de pile ou face
Authors:
Jean Brossard,
Christophe Leuridan
Abstract:
Soit $(ε_n)_{n\in\mathbf{Z}}$ un jeu de pile ou face, c'est-à-dire une suite de variables aléatoires indépendantes de loi $(δ_{-1}+δ_1)/2$, et $(H_n)_{n\in\mathbf{Z}}$ un processus à valeurs dans $\{-1,1\}$, prévisible dans la filtration naturelle de $(ε_n)_{n\in\mathbf{Z}}$. Alors $(H_nε_n)_{n\in \mathbf{Z}}$ est encore un jeu de pile ou face, dont la filtration naturelle est contenue dans cell…
▽ More
Soit $(ε_n)_{n\in\mathbf{Z}}$ un jeu de pile ou face, c'est-à-dire une suite de variables aléatoires indépendantes de loi $(δ_{-1}+δ_1)/2$, et $(H_n)_{n\in\mathbf{Z}}$ un processus à valeurs dans $\{-1,1\}$, prévisible dans la filtration naturelle de $(ε_n)_{n\in\mathbf{Z}}$. Alors $(H_nε_n)_{n\in \mathbf{Z}}$ est encore un jeu de pile ou face, dont la filtration naturelle est contenue dans celle de $(ε_n)_{n\in\mathbf{Z}}$. Le but de l'article est d'obtenir des conditions pour que ces filtrations soient égales et de décrire l'écart entre ces filtrations lorsqu'elles sont différentes. Nous nous intéressons plus particulièrement au cas des transformations homogènes, où le processus $(H_nε_n)_{n\in\mathbf{Z}}$ est une fonctionnelle de $(ε_n)_{n\in\mathbf{Z}}$ qui commute avec les translations. Nous étudions de façon approfondie les transformations homogènes de longueur finie, où $H_n$ est de la forme $φ(ε_{n-d},...,ε_{n-1})$ avec $d\in\mathbf {N}$ et $φ:\{-1;1\}^d\to\{-1;1\}$ fixés.
△ Less
Submitted 26 September, 2006; v1 submitted 21 May, 2004;
originally announced May 2004.
