In general, given a sequence of shift-invariant measures $\eta_{n}$ on $\{0,1\}^{\mathbb{N}}$ what to do to guarantee this convergence of entropies: $$h(\eta_{n}) \rightarrow \log2?$$

Because I'm working with a kind of sequence of shif-invariant measures such that the sequence of entropies is monotone increasing (it could be constant...) and I couldn't find in the literature some good "machinery" to prove that convergence... I've tried, for instance, the $\bar{d}$ distance related to joinings (defined here joining or coupling) but, it didn't work too well...(I mean, I was not able to work with...)

So, any tips, hints?

Thanks for your attention

kind regards


There is only one measure of maximum entropy (i.e. equal to $\log 2$), the uniform measure (or 1/2,1/2-Bernoulli measure). So your problem translates into checking that $\eta_n$ converges to the uniform measure.

  • $\begingroup$ Dear @user39115, to be more specific, imagine you have a sequence of shif-invariant measures, say, $\eta_{n}$ which all of them have positive entropy, more than that, imagine the sequence of entropies is monote increasing (could b constant) and the support does not mind to me. In a traditional point view, what do I have to do to guarantee that the entropies are growing to $\log2$? What is common to do to show what I want? Are there examples in the literature? I only know thed-bar distance, but it is tough... thanks for your attention $\endgroup$ Oct 28 '15 at 14:44
  • $\begingroup$ Dear Bruno, I improved the previous long comment, hope to have understood your question. $\endgroup$ Oct 28 '15 at 14:48
  • $\begingroup$ Dear @user39115, you are right. It's just my problem. To make sure that the sequence of entropy is growing to $\log2$ seems quite difficult... it's much stronger than the weak* convergence. Because of this, I'm stuck... $\endgroup$ Oct 28 '15 at 14:56

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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