I am currently reading an article in which the author goes to certain lengths which could be avoided if the following result were true:

Lemma (proposed): Let $T$ be an ergodic measure-preserving transformation of a probability space $(X,\mathcal{F},\mu)$, and let $(f_n)$ be a sequence of integrable functions from $X$ to $\mathbb{R}$ which satisfy the subadditivity relation $f_{n+m} \leq f_n \circ T^m + f_m$ a.e. for all integers $n,m \geq 1$. Suppose that $f_n(x) \to -\infty$ in the limit as $n \to \infty$ for $\mu$-a.e. $x \in X$. Then $\lim_{n \to \infty} \frac{1}{n}\int f_n d\mu <0$.

Via the subadditive ergodic theorem, this effectively states that if $f_n(x) \to -\infty$ almost everywhere then it must do so at an asymptotically linear rate. The supposed lemma would also be equivalent to the statement that if $\frac{1}{n} f_n(x) \to 0$ almost everywhere, then for almost every $x$ the sequence $(f_n(x))$ must return infinitely often to some neighbourhood of $0$ which is not a neighbourhood of $-\infty$. If the sequence $(f_n)$ is additive rather than just subadditive then this last formulation of the result follows from a well-known theorem of G. Atkinson, but the more general subadditive case is less clear.

If the lemma were true then several parts of the paper I am reading would be redundant, which makes me wonder whether it is in fact false. Yet it seems rather plausible. Does anyone know whether this result is true or not?

  • $\begingroup$ The supposed equivalent statement is false by the same counterexample I mentioned to Vaughan below: $f_n(x)=sqrt n$ (!) $\endgroup$ Commented Jul 19, 2011 at 6:59
  • $\begingroup$ I was quite careful with my phrasing so as not to exclude cases such as this: the example $f_n(x)=\sqrt{n}$ does indeed return infinitely often to some neighbourhood of 0 which is not a neighbourhood of $-\infty$, namely the open neighbourhood $(−1,+\infty)$. $\endgroup$
    – Ian Morris
    Commented Jul 19, 2011 at 11:09

1 Answer 1


I believe the Lemma you propose is true, via a relatively straightforward adaptation of the proof given for the additive case in Giles Atkinson, Recurrence of co-cycles and random walks, J. Lond. Math. Soc. (2) 13 (1976), 486-488. (I assume this is the result you were referencing.) This may already be known and written somewhere -- I can't speak to that. In case it's not, here's a proof.

Proof of Lemma. By the subadditive ergodic theorem, there exists a measurable function $f\colon X\to \mathbb{R}\cup\{-\infty\}$ such that $\lim \frac 1n f_n(x) = f(x)$ for $\mu$-a.e. $x\in X$ and $\lim \frac 1n \int f_n\,d\mu = \int f\,d\mu$. So our task is to show that $\int f\,d\mu < 0$.

To this end, given $x\in X$, let $M_x = \{n \mid f_n(x) \geq -1 \}$, and observe that $M_x$ is finite $\mu$-a.e. Thus writing $A_n = \{x\in X \mid \# M_x < n \}$, we see that there exists $N$ such that $\mu(A_N) > \frac 12$.

Furthermore, given $x\in X$, write $L_x = \{ k \mid T^k(x) \in A_N \}$. Since $\mu(A_N) > \frac 12$, we see that $\mu$-a.e. $x$ has $$ (*)\qquad\qquad\qquad\# L_x \cap [1,n] \geq \frac n2\qquad\qquad\qquad\qquad\qquad\quad $$ for all sufficiently large $n$. We fix such an $x$ and show that $f(x) < 0$.

Let $k_0$ be the smallest element of $L_x$. We define $k_i\in L_x$ recursively with the property that $$ f_{k_i}(x) < f_{k_0}(x)-i, $$ as follows. Let $J_i$ be the $N$ smallest elements of $L_x \cap (k_i,\infty)$. Because $T^{k_i}(x)\in A_N$, there exists $k_{i+1}\in J_i$ such that $k_{i+1} - k_i \notin M_{T^{k_i}(x)}$. In particular, we have $$ f_{k_{i+1} - k_i}(T^{k_i}(x)) < -1. $$ Now subadditivity gives $$ f_{k_{i+1}}(x) \leq f_{k_i}(x) + f_{k_{i+1} - k_i}(T^{k_i}(x)) < f_{k_0(x)} - i-1. $$ The next observation to make is that by $(*)$, we have $k_i \leq 2Ni$ for all sufficiently large $i$. Thus we have $$ \frac 1{k_i} f_{k_i}(x) \leq \frac 1{2Ni}(f_{k_0}(x) - i), $$ and sending $i\to\infty$, we obtain $f(x) \leq -\frac 1{2N}$. Since this holds for $\mu$-a.e. $x$, we have $\int f\,d\mu \leq -\frac 1{2N} < 0$, which completes the proof.

Note. It's quite important in this proof that we're dealing with negative values of $f_n$; the proof would fail if we tried to show that $f_n(x)\to+\infty$ for $\mu$-a.e. $x$ implies that $\int f\,d\mu > 0$. I'm not sure if the result is true in this case.

  • 1
    $\begingroup$ It's definitely false in the case $f_n(x)\to\infty$ as you see from the example $f_n(x)=sqrt n$. $\endgroup$ Commented Jul 19, 2011 at 6:57
  • $\begingroup$ This reminds me of a lemma of Tanny that I used recently: Suppose $T$ an ergodic measure-preserving transformation; $f$ a non-negative(not necessarily integrable) function. either (1) $f(T^nx)/n\to 0$ a.e. or (2) $\limsup f(T^nx)/n=\infty$ a.e. I eventually stumbled on a 6 line proof that looks a lot like this proof, apparently due to Feldman that appears in a paper of Lyons, Pemantle and Peres. $\endgroup$ Commented Jul 19, 2011 at 7:44
  • $\begingroup$ Thanks! I'd noticed that the result is false for $f_n(x) \to +\infty$ using exactly Anthony's example. For some reason I let this make me believe that Atkinson's argument wouldn't be applicable and didn't bother checking. I've even got Atkinson's paper on my hard drive, so I feel a little stupid ;o) $\endgroup$
    – Ian Morris
    Commented Jul 19, 2011 at 10:10
  • $\begingroup$ It occurred to me after sleeping that in the paper I'm reading, the extra assumption is made that the functions $f_n$ are essentially bounded above independently of $n$, and in this subcase we can just use the MCT applied to the sequence $\sup_{m \geq n} f_m$ to conclude that $\int f_N<0$. However the general case is of course extremely nice to know, and I think even answers an open question in that paper. $\endgroup$
    – Ian Morris
    Commented Jul 19, 2011 at 10:12

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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