2
$\begingroup$

Given an ergodic and non-singular dynamic system (definition provided here) $(X, \mathcal{B}, \mu_1, T)$ where $(X, \mathcal{B}, \mu_1)$ is a measure space and $T$ is a fixed transformation, we then will have $\mu_1$ equivalent to $T\mu$ where $T\mu_1$ is defined by $T\mu_1(A)=\mu_1[T^{-1}(A)]$ for each $A\in \mathcal{B}$. Now suppose $\mu_2$ is another measure that is equivalent to $\mu_1$ and makes $(X, \mathcal{B}, \mu_2, T)$ ergodic and non-singular. Hence, we will have $T\mu_2\sim\mu_2\sim\mu_1\sim T\mu_1$. Do we have:

$$ \frac{d\,\mu_1}{d\,\mu_2}\circ T = \frac{d\,T\mu_1}{d\,T\mu_2} $$

$\mu_1$-almost everywhere? Here the "$\mu_1$-almost everywhere" can be replaced by "$\mu_2$-almost everywhere" or any one of those four measures. By definition of $T\mu_2$, for each $A\in\mathcal{B}$, we have:

$$ \int_X\chi_A(x)d\,T\mu_2 = \int_X\chi_A(Tx)d\,\mu_2 $$

I tried to show the following equation holds for each $A\in\mathcal{B}$ but cannot:

$$ \int_A\frac{d\,\mu_1}{d\,\mu_2}(x)d\,T\mu_2 = \int_A\frac{d\,\mu_1}{d\,\mu_2}\circ T(x)d\,\mu_2 = \int_A\frac{d\,T\mu_1}{d\,T\mu_2}(x)d\,\mu_2 $$

Any hints or counter-examples will be appreciated

$\endgroup$

1 Answer 1

2
$\begingroup$

I am fairly sure this is not the case (unless I am missing something in the definition of $T$)

Consider the special case where $T$ is invertible with inverse $T_{\rm inv}$. Then, it holds for any test function $f$

\begin{aligned} \int f \,dT\mu_1 &= \int f \circ T \,d\mu_1 \\ &= \int f \circ T \,\frac{d\mu_1}{d\mu_2} \,d\mu_2 \\ &= \int f \circ T \, \frac{d\mu_1}{d\mu_2} \,d(T_{\rm inv} (T \mu_2))\\ &= \int f \,\frac{d\mu_1}{d\mu_2} \circ T_{\rm inv} \,dT\mu_2 \end{aligned} and thus it actually holds $$ \frac{dT\mu_1}{dT\mu_2} = \frac{d\mu_1}{d\mu_2} \circ T_{\rm inv}. $$

If I understand correctly, the following simple example satisfies your assumptions and shows that your claim does not hold: Take $X = \{1, 2, 3\}$, $\mu_1 = (1/6, 2/6, 3/6)$ (meaning $\mu_1 = \frac{1}{6}\delta_1 + \frac{2}{6} \delta_2 + \frac{3}{6} \delta_3$), $\mu_2 = (2/6, 3/6, 1/6)$ and $T(1) = 2, T(2) = 3, T(3) = 1$, and one quickly finds that $T$ is invertible but also that $\frac{dT\mu_1}{dT\mu_2} = \frac{d\mu_1}{d\mu_2} \circ T_{\rm inv} \neq \frac{d\mu_1}{d\mu_2} \circ T$.

$\endgroup$
3
  • 1
    $\begingroup$ Yes, and in fact an even better way to phrase the correct statement is $\frac{dT\mu_1}{dT\mu_2}\circ T=\frac{d\mu_1}{d\mu_2}$, which also holds true in case $T$ is not invertible. $\endgroup$ Commented Jun 1, 2022 at 10:06
  • $\begingroup$ Thanks @leomonsaingeon for your remark. If $T$ is not invertible, which property of $T$ is the necessary one to obtain the result you write? If $T$ is for instance constant, then the result you write cannot hold of course. $\endgroup$
    – Steve
    Commented Jun 1, 2022 at 11:32
  • $\begingroup$ @Steve Thank you for your answers! For your question, I believe at least I need $T$ to have a right inverse to have the correct equation stand even without assuming the system is ergodic. However, now if I assume the equation holds (or the equivalent one stated by leo), then I will have $\mu_1 = T\mu_1$ (?) When the system is ergodic and this equation holds, then all characteristic functions will be the constant one (?) $\endgroup$ Commented Jun 1, 2022 at 14:38

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .