1

I'm reading about Lusin's theorem in textbook Optimal Transport for Applied Mathematicians


Let us be more precise: take a topological space $X$ endowed with a finite regular measure $\mu$ (i.e. any Borel set $A \subset X$ satisfies $\mu(A)=\sup \{\mu(K): K \subset A, K \text{ compact} \}=\inf \{\mu(B): B \supset A, B \text{ open}\}$ ). The arrival space $Y$ will be supposed to be second-countable (i.e. it admits a countable family $\left(B_i\right)_i$ of open sets such that any other open set $B \subset Y$ may be expressed as a union of $B_i$; for instance, separable metric spaces are second-countable).

Theorem (weak Lusin) - Under the above assumptions on $X, Y, \mu$, if $f: X \rightarrow Y$ is measurable, then for every $\varepsilon>0$ there exists a compact set $K \subset X$ such that $\mu(X \backslash K)<\varepsilon$ and the restriction of $f$ to $K$ is continuous.

Proof - For every $i \in \mathbb{N}$, set $A_i^{+}=f^{-1}\left(B_i\right)$ and $A_i^{-}=f^{-1}\left(B_i^C\right)$. Consider compact sets $K_i^{\pm} \subset A_i^{\pm}$such that $\mu\left(A_i^{\pm} \backslash K_i^{\pm}\right)<\varepsilon 2^{-i}$. Set $K_i=K_i^{+} \cup K_i^{-}$and $K=\bigcap_i K_i$. For each $i$ we have $\mu\left(X \backslash K_i\right)<\varepsilon 2^{1-i}$. By construction, $K$ is compact and $\mu(X \backslash K)<4 \varepsilon$. To prove that $f$ is continuous on $K$ it is sufficient to check that $f^{-1}(B) \cap K$ is relatively open in $K$ for each open set $B$, and it is enough to check this for $B=B_i$. Equivalently, it is enough to prove that $f^{-1}\left(B_i^c\right) \cap K$ is closed, and this is true since it coincides with $K_i^{-} \cap K$.

Theorem (strong Lusin) - Under the same assumptions on $X$, if $f: X \rightarrow \mathbb{R}$ is measurable, then for every $\varepsilon>0$ there exists a compact set $K \subset X$ and a continuous function $g: X \rightarrow \mathbb{R}$ such that $\mu(X \backslash K)<\varepsilon$ and $f=g$ on $K$.

Proof - First apply weak Lusin's theorem, since $\mathbb{R}$ is second countable. Then we just need to extend $f_{\mid K}$ to a continuous function $g$ on the whole $X$. This is possible since $f_{\mid K}$ is uniformly continuous (as a continuous function on a compact set) and hence has a modulus of continuity $\omega:\left|f(x)-f\left(x^{\prime}\right)\right| \leq \omega\left(d\left(x, x^{\prime}\right)\right)$ (the function $\omega$ can be taken sub additive and continuous). Then define $g(x)=\inf \left\{f\left(x^{\prime}\right)+\omega\left(d\left(x, x^{\prime}\right)\right): x^{\prime} \in K\right\}$. It can be easily checked that $g$ is continuous and coincides with $f$ on $K$.


My concerns are about the proof of the weak version. First, the author assumes that $X, Y$ are topological spaces where $Y$ is second-countable.

  1. We have $$ \begin{align} \mu\left(X \backslash K_i\right) &= \mu\left(X \backslash (K_i^{+} \cup K_i^{-})\right) \\ &= \mu ( (X \setminus K_i^+) \cup (X \setminus K_i^-)) \\ &\le \mu (X \setminus K_i^+) + \mu (X \setminus K_i^-). \end{align} $$ Also, $\mu\left(A_i^{+} \backslash K_i^{+}\right)<\varepsilon 2^{-i}$ and $\mu\left(A_i^{-} \backslash K_i^{-}\right)<\varepsilon 2^{-i}$. How do we conclude $\mu\left(X \backslash K_i\right)<\varepsilon 2^{1-i}$?

  2. The union of two compact sets is compact, so $K_i$ is compact. Here $X$ is not assumed to be Hausdorff. How do we conclude $K := \bigcap_i K_i$ is compact?

  3. We have $f^{-1}\left(B_i^c\right) \cap K = K_i^{-} \cap K$. We know that $K_i^-$ is compact, but we don't know if it is closed. How do we conclude that $f^{-1}\left(B_i^c\right) \cap K$ is closed in $K$?

  4. It seems to me $\mu$ don't need to be outer regular.

Could you please elaborate on my confusion?

Akira
  • 17,367

1 Answers1

1
  1. You seem to be using that \begin{align} X \backslash (K_i^{+} \cup K_i^{-})=(X \setminus K_i^+) \cup (X \setminus K_i^-), \end{align} which is not true. You still have the inclusion, though, so your estimate still works. Note that $$ A_i^+\cup A_i^-=f^{-1}(B_i)\cup f^{-1}(B_i^c)=f^{-1}(B_i\cup B_i^c)=f^{-1}(Y)=X. $$ Then, since $A_i^+\cap A_i^-=\varnothing$, $$ X\setminus K_i=(A_i^+\setminus K_i^+)\cup (A_i^-\setminus K_i^-). $$

  2. I'm fairly sure that somewhere in the book (which I have never seen) it says that all topological spaces are assumed Hausdorff. It is not really possible to do analysis when limits are not well-defined.

  3. Same as two.

  4. You are working in a finite-measure space. An inner regular measure is then outer regular: if $\varepsilon>0$ and $E$ is measurable, by inner regularity there exists $K$ compact with $\mu(E^c\setminus K)<\varepsilon$. Then $$ \mu(K^c\setminus E)=\mu(K^c\cap E^c)=\mu(E^c\setminus K)<\varepsilon. $$

Martin Argerami
  • 205,756
  • The p.d.f. of the book is available online here. I have searched for the assumption of being Hausdorff, but could not find it in the book. – Akira Sep 27 '22 at 07:03
  • About your 4., the assumption of being inner regular means that there is a compact set $K \subset E$ such that $\mu(E \setminus K) < \varepsilon$. I wonder if why you used $\mu(E^c\setminus K)<\varepsilon$. – Akira Sep 27 '22 at 07:06
  • Because it is what I need to get an open set that contains $E$. – Martin Argerami Sep 27 '22 at 12:24
  • Ahh you apply the inner regularity for $E^c$, not $E$. It's my bad. Thank you so much! – Akira Sep 27 '22 at 12:27
  • 1
    No problem. As for the book, indeed no such thing as topological spaces being Hausdorff is mentioned. Still, a non-Hausdorff topology lets you do very little analysis (if the net ${f(x_j)}$ has two different limits, how do you decide if $f$ is continuous?) – Martin Argerami Sep 27 '22 at 12:43
  • As an additional comment about a certain level of looseness in the definitions in that book, note that the proof of strong Lusin uses uniform continuity, which means that that author is implicitly assuming that $X$ is a metric space (in particular, Hausdorff). – Martin Argerami Oct 02 '22 at 04:33
  • Could you confirm that with Tietze extension theorem, $X$ being Hausdorff normal topological space is enough for strong Lusin to hold? – Akira Oct 02 '22 at 08:19
  • 1
    Yes. It also works for $X$ locally compact Hausdorff. – Martin Argerami Oct 02 '22 at 12:44