Let $(X,\Omega,\mu)$ be a $\sigma$-finite measure space and put $\mathscr{H}=L^{2}(X,\Omega,\mu)=L^2(\mu)$. If $\phi \in L^\infty(\mu)$, define $M_\phi:L^2(\mu) \to L^2(\mu)$ by $M_\phi f=\phi f$. Then $M_\phi \in \mathscr{B}(L^2(\mu))$ and $\|M_\phi\|=\|\phi\|_\infty$.

The operator $M_\phi$ is called a multiplication operator. I want to show that

$M_\phi=M_\phi^2$ if and only if $\phi$ is a characteristic function.

$\longleftarrow$ If $\phi$ is a characteristic function we have $M_\phi(M_\phi(f))=M_\phi(\phi f)=M_\phi(f)$.

How I can prove the converse?


We want to show that if $\phi \in L^\infty$ and $\phi f = \phi^2 f$ for every $f$ in $L^2$ then $\phi$ must be a characteristic function. That is, we want to show that $\phi$ takes values in $\{0,1\}$ a.e. (since $\phi$ is measurable by assumption).

Suppose that this is not the case. Then one of the measurable sets $A_1 = \{ x: \phi(x) > 1 \}, A_2 = \{ x: \phi(x) \in (0,1) \} , A_3 = \{ x: \phi(x) < 0\}$ has positive measure. Say $\mu(A_i) > 0$. Then since our measure space is $\sigma$-finite, $A_i$ has a measurable subset of positive, finite measure which we call $B_i$. Then $1_{B_i} \in L^2$. In particular, $\phi 1_{B_i} = \phi^2 1_{B_i}$ almost everywhere. But this is impossible by construction of the $B_i$.


This is really the same as what nobody said, rephrased so as to point out the following basic fact:

Lemma If $\mu$ is $\sigma$-finite then $||M_\phi||=||\phi||_\infty$.

It's clear that $||M_\phi||\le||\phi||_\infty$ for any measure. Suppose $\mu$ is $\sigma$-finite and $\alpha<||\phi||_\infty$. Let $S=\{x:|\phi(x)|\ge\alpha\}$. By definition $\mu(S)>0$. So there exists $E\subset S$ with $0<\mu(E)<\infty$. If $f=\chi_E$ then $||\phi f||_2\ge\alpha||f||_2>0$, hence $||M_\phi||\ge\alpha$.

So $||M_\phi||\ge\alpha$ for every $\alpha<||\phi||_\infty$, hence $||M_\phi||\ge||\phi||_\infty$.

And now you're done: $M_\phi^2=M_\phi$ if and only if $M_{\phi^2-\phi}=0$; the lemma says this happens if and only if $||\phi^2-\phi||_\infty=0$, which says $\phi^2=\phi$ almost everywhere.


$M_\phi^2= M_\phi$ holds iff $\phi^2 f = \phi f$ for all $f \in L^2$. Since $\mu$ is $\sigma$-finite there is a sequence of subsets of finite measure $E_n$ such that $X = \bigcup_n E_n$, so in particular $\phi^2 \chi_{E_n} = \phi \chi_{E_n}$ i.e. $\phi^2 = \phi$ on $E_n$ for every $n$. Hence $\phi^2 = \phi$ or $\phi(\phi - 1) = 0$ so $\phi$ must be $\{0,1\}$-valued.


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.