MathOverflow is a question and answer site for professional mathematicians. Join them; it only takes a minute:

Sign up
Here's how it works:
  1. Anybody can ask a question
  2. Anybody can answer
  3. The best answers are voted up and rise to the top

Let $E$ be a compact subset of $\mathbb{R}^n$. Let the density function $\phi(x,y)$ be Lipschitz continuous and such that $$ \int\limits_E \phi(x,y)dy=1 $$ for all $x\in E$. Let us consider the non-increasing sequence of non-empty compact sets $A_n$ such that for all $x\in A_{n+1}$ we have $$ \int\limits_{A_n} \phi(x,y)dy=1. $$ Since $A_n$ are compacts, there exists a non-empty limit set $A = \bigcap\limits_n A_n$

Do we have for all $x\in A$ that $$ \int\limits_A \phi(x,y)dy = 1? $$

share|cite|improve this question
What do you mean by "Since $A_n$ are compacts, there is a limit $A=\lim_n A_n$"? The set of compact sets is by no means compact itself. – fedja Jan 23 '11 at 13:34
Could you make the question more precise? From the hypothesis it seems that taking $A_n=E$ for all $n$ would satisfy the condition. Also what do you mean by a limit set? In the Hausdorff metric? the set of points that belong to all $A_n$ for $n$ greater than some $n_0$? Also what is $A_0$? What I think this might mean is this: think of $\phi$ as the transition kernel for a MC. Let $A_0=E$; Let $A_1$ be the support of the random variable $X_1$ given that $X_0$ has some distribution with support $E$. Now let $A_2$ be the support of $X_2$ etc. Is this right? – Anthony Quas Jan 23 '11 at 17:24
I've just mentioned that I forgot to write that the sequence $A_n$ is non-increasing. Since the limit is understood as a set which belongs to all $A_n$. – Ilya Jan 23 '11 at 21:21
up vote 1 down vote accepted

Anthony's interpretation of the question is that $A_n$ is the support of $X_n$, where $(X_n)$ is any Markov chain of transition kernel $\phi$ such that the distribution of $X_0$ has support $E$. If this interpretation is correct, the result holds. To see this, note that, under the assumption that $A_1\subset E=A_0$, the sequence $(A_n)$ is nonincreasing. Let $A$ denote the intersection of all the sets $A_n$ and let $x\in A$. Since $x\in A_n$ for every $n$, $\phi(x,A_{n+1})=1$ [writing $\phi(x,B)$ for the integral of $\phi(x,\cdot)$ over $B$]. Since $A_{n+1}$ decreases to $A$, by bounded convergence, $\phi(x,A_{n+1})$ decreases to $\phi(x,A)$, and this proves that $\phi(x,A)=1$. (Thus compactness would be irrelevant. Note that the indices $n$ and $n+1$ are mixed up in the question.)

share|cite|improve this answer
I edited the topic. Please note 1. $A_n$ is an arbitrary non-increasing sequence of non-empty compact sets. I need a compactness to be sure that the limit set is non-empty. 2. for all $x\in A_{n+1}$ we have $\phi(x,A_n) = 1$. That is all we have. Is it sufficient to hold that for all $x\in A$ we have $\phi(x,A) = 1$. – Ilya Jan 23 '11 at 21:26
Still seems like basic measure theory to me... Fix $x\in A$ and consider the nonnegative functions $f_n$ and $f$ defined on $E$ by $f_n(y)=\phi(x,y)$ if $y\in A_n$, $f_n(y)=0$ otherwise, and by $f(y)=\phi(x,y)$ if $y\in A$, $f(y)=0$ otherwise. You know that $f_n\to f$ pointwise and you want to prove that the integrals of $f_n$ converge to the integral of $f$. Does that ring a bell? (By the way, if $A$ is empty, every statement beginning by "For every $x\in A$, one has..." is true. Ergo exit compactness.) – Did Jan 23 '11 at 21:52
Thank you for these nice comment about empty set, but I need compactness to say also that the set $A$ is non-empty, it is important. Could we make the proof smaller? Say, for all $x\in A$ and all $n$ we have $K(x,A_n) = 1$ (here I use $K(x,A_n)$ rather then $\phi(x,A_n)$ to stress that it is a measure). I think that by continuity of measure we have $$ K(x,A) = \lim\limits_{n\to\infty}K(x,A_n) = 1 $$ because the sequence $A_n$ is non-increasing and $A$ is an intersection of all these sets. – Ilya Jan 24 '11 at 8:26

Your Answer


By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.