Let $(X,M, \mu)$ be a $\sigma$-finite measure. Let $M' \subset M$ be a smaller $\sigma$-algebra. Let $\mu' = \mu|_M$.
Show that there exists a unique $g \in L^1(\mu')$ such that $\int_E f d\mu = \int_E g d\mu'$ for all $E\in M'$ with $\|g\|_1 \leq \|f\|_1$.
My attempt:
I want to use Radon-Nikodym theorem here. But I'm not sure how to deal with the $\mu'$, since the version I learned requires both measures to be $\sigma$-finite.
Thank you for your help.
The question is:
This result, as stated, is false. Here is a counter-example.
Consider $(X,M, \mu)$ where $X=\Bbb R$, $M$ is the Borel $\sigma$-algebra and $\mu$ is the Lebesgue measure. We know that $(X,M, \mu)$ is a $\sigma$-finite measure space.
Now, let $M'=\{\emptyset , \Bbb R\}$. Then $M'$ is a $\sigma$-algebra and $M' \subset M$. Let $\mu' = \mu|_{M'}$.
Let $f$ be any function in $L^1(\mu)$ such that $0< \int_{\Bbb R} f d\mu < +\infty$. Note that $g \in L^1(\mu')$ if and only if $g=0$.
So there is no $g \in L^1(\mu')$ such that $\int_{\Bbb R} f d\mu = \int_{\Bbb R} g d\mu'$
Remark: The key issue is that, even $\mu$ being $\sigma$-finite, $\mu'$ may not be $\sigma$-finite.
If we add to the question the additional hypothesis that $\mu'$ is $\sigma$-finite, then the result holds.
Just define on $M'$, the measure $\nu$ defined by, for all $E \in M'$, $\nu(E)= \int_E f d\mu$. Clearly, $\nu \ll \mu'$ and since now $\mu'$ is $\sigma$-finite, we can apply Radon-Nikodym theorem and conclude that there exists a unique $g \in L^1(\mu')$ such that, for all $E\in M'$, $$\int_E f d\mu = \nu(E)= \int_E g d\mu'$$
Then, let $\textrm{sign}(g) = \chi_{[g>0]} - \chi_{[g<0]}$. Since $g$ is $M'$-measurable function, so is $\textrm{sign}(g)$. So we have \begin{align*} \int_X |g| d\mu' & = \int_X \textrm{sign}(g) g d\mu'= \int_X \textrm{sign}(g) d\nu = \int_X \textrm{sign}(g) f d\mu \leqslant \int_X |f| d\mu \end{align*} So $\|g\|_1 \leq \|f\|_1$.