Suppose a sequence of probability measures $(\mathbb P_n)_{n\in\mathbb N}$ on a Polish space $X$ satisfies the large deviations principle with a good rate function $I$ and rate $n$. Informally speaking (I'm very imprecise here!), this means that for a Borel set $A\subset X$, we have $$\mathbb P_n(A)\approx \exp\big(-n\inf_{x\in A} I(x)\big), $$ and for two sets $B\subset A\subset X$, we have $$\frac{\mathbb P_n(B)}{\mathbb P_n(A)}\approx \exp\big(-n\,(\inf_{x\in B} I(x)-\inf_{x\in A} I(x))\big). $$ The last equality looks very much like a new large deviations principle for measures conditioned on $A$. That is, the new Polish space is $A$, the new sequence of probability measures consists of $\mathbb Q_n(\cdot ):=\frac{\mathbb P_n(\cdot)}{\mathbb P_n(A)}$, and the new rate function is $$I_A(y):=I(y)-\inf_{x\in A} I(x). $$

I'm pretty sure that one can make this construction rigorous, for example when $A$ is an $I$-continuous set. However, I'm also pretty sure that I am not the first to make this observation.

My question is: has anybody seen a statement of this kind before? Does it have a name? Where can I read more about it?


Your Answer

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

Browse other questions tagged or ask your own question.