1

Let

  • $X$ be a metric space,
  • $\mathcal C_b(X)$ the space of real-valued bounded continuous functions on $X$,
  • $\mathcal C_c(X)$ the space of real-valued continuous functions on $X$ with compact supports,
  • $\mathcal M (X)$ the space of all complex Borel measures on $X$,
  • $\mathcal M_{\mathbb R}(X)$ the space of all finite signed Borel measures on $X$, and
  • $\mathcal M_+ (X)$ the space of all finite non-negative Borel measures on $X$.

For $\nu \in \mathcal M(X)$, let $|\nu| \in \mathcal M_+ (X)$ be its variation and $[\nu] := |\nu| (X)$ its total variation norm. Then $(\mathcal M(X), [\cdot])$ is a real Banach space. For $\mu_n,\mu \in \mathcal M(X)$, we define weak convergence by $$ \mu_n \rightharpoonup \mu \overset{\text{def}}{\iff} \int_X f \mathrm d \mu_n \to \int_X f \mathrm d \mu \quad \forall f \in \mathcal C_b(X), $$ and weak$^*$ convergence by $$ \mu_n \overset{*}{\rightharpoonup} \mu \overset{\text{def}}{\iff} \int_X f \mathrm d \mu_n \to \int_X f \mathrm d \mu \quad \forall f \in \mathcal C_c (X). $$

It has been shown that

Theorem Let $\mu_n,\mu\in \mathcal{M}(X)$ such that $\mu_n \rightharpoonup \mu$. Then for any open subset $O$ of $X$, $$ |\mu|(O) \leq \liminf _{n \rightarrow \infty}\left|\mu_n\right|(O) . $$

I would like to apply the same technique to prove that

Theorem Let $X$ be locally compact separable and $\mu_n,\mu\in \mathcal{M}(X)$ such that $\mu_n \overset{*}{\rightharpoonup} \mu$. Then for any open subset $O$ of $X$, $$ |\mu|(O) \leq \liminf _{n \rightarrow \infty}\left|\mu_n\right|(O) . $$

My proof seems to be much simpler than the original one. It's likely that I made some subtle mistakes. Could you please have a check on my attempt?


Proof: We need the following lemmas, i.e.,

  • Lemma 1 Let $X$ be a locally compact separable metric space. Then $X$ is a Radon space.

  • Lemma 2 Let $X$ be a locally compact Hausdorff space. Let $K$ be a compact subset of $X$ and $U$ an open subset of $X$ such that $K \subset U$. Then there is an open subset $V$ of $X$ such that $K \subset V \subset \overline V \subset U$ and $\overline V$ is compact.

Let $\mathcal X$ be the Borel $\sigma$-algebra of $X$. By polar decomposition, there is a measurable $g:X \to [0, 2\pi)$ such that $$ |\mu| (B) = \int_B e^{i g} \mathrm d \mu \quad \forall B \in \mathcal X. $$

Fix $\varepsilon>0$. By Lemma 1, there is a compact subset $K$ of $O$ such that $|\mu| (O \setminus K) < \varepsilon$. By Lemma 2, there is an open subset $V$ of $X$ such that $\overline V$ is compact and $K \subset V \subset \overline V \subset O$. By Urysohn's lemma, there is a continuous $h:X \to [0, 1]$ such that $f(K)=1$ and $f(X \setminus V) = 0$. This implies the support of $f$ is contained in $\overline V$, so $f \in \mathcal C_c (X)$. Also, $|f| \le 1_{O}$. We have $$ \begin{align} |\mu| (O) - \operatorname{Re} \int_O fe^{i g} \mathrm d \mu &= \operatorname{Re} \int_O (1-f)e^{i g} \mathrm d \mu \\ &\le \left | \int_O (1-f)e^{i g} \mathrm d \mu \right | \\ &\le \int_O |(1-f)e^{i g}| \mathrm d |\mu| \\ &= \int_O |1-f| \mathrm d |\mu| \\ &\le |\mu| (O \setminus K) < \varepsilon. \end{align} $$

So $$ \begin{align} |\mu| (O) - \varepsilon &< \operatorname{Re} \int_O fe^{i g} \mathrm d \mu \\ &\overset{(1)}{=} \operatorname{Re} \lim_n \int_O fe^{i g} \mathrm d \mu_n \\ &= \lim_n \operatorname{Re} \int_O fe^{i g} \mathrm d \mu_n \\ &\le \liminf_n \left| \int_O fe^{i g} \mathrm d \mu_n \right| \\ &\le \liminf_n \int_O |fe^{i g}| \mathrm d |\mu_n | \\ &= \liminf_n \int_O |f| \mathrm d |\mu_n | \\ &\le \liminf_n \int_O \mathrm d |\mu_n | \\ &= \liminf_n |\mu_n| (O). \end{align} $$

Notice that $(1)$ is true even if $fe^{i g}$ is complex-valued. The claim then follows by taking the limit $\varepsilon \to 0^+$.

Analyst
  • 5,637
  • 1
    (+1) Seems awfully simpler than the answer in MO. After a couple of reading all seems fine, but I think I should let it sit for a while before I get back to this. – Mittens Nov 09 '22 at 18:49
  • @OliverDíaz I think my proof is what Giorgio meant in his/her comment. – Analyst Nov 09 '22 at 18:53
  • @OliverDíaz Maybe I miss something, but Gioigio only posted a comment but no answer. – Analyst Nov 09 '22 at 19:16
  • @OliverDíaz I'm sorry for bothering you too much... Have a good break. – Analyst Nov 09 '22 at 19:32
  • Hi @Analysit, I see now a little flaw (I don't know how easy is to fix it) in your argument. It is right on the $\stackrel{(1)}{=}$ step. $\mu_n$ converges weakly to $\mu$ and indeed one can use complex test functions, the issue is that although $f\in C_{00}(X)\subset C_b(X)$, $fe^{ig}$ may not; hence $\lim_n\int_O fe^{ig},d\mu_n=\int_O fe^{ig},d\mu$ is not evident. – Mittens Nov 09 '22 at 21:06
  • @OliverDíaz Please check my reasoning. We have $fe^{ig} \neq 0 \iff f \neq 0$. So $\operatorname{supp} (fe^{ig}) = \operatorname{supp} (f) \subset \overline V$ where $\overline V$ is compact. So $fe^{ig}$ is compactly supported. – Analyst Nov 09 '22 at 21:14
  • no problem with the supports, it is just the potential lack of continuity of $fe^{ig}$. Under some conditions, continuity may be relaxed, for example of the st of discontinuities are a set of measure $0$, but not in general. I think there is still a way to fix the argument baby approximating $e^ig$ by functions in $C_{00}(X)$, after all integrable functions w.r.t. $|\mu|$ are in the closure of $C_{00}(X)$ under the $L_1(|\mu|)$ norm. Ij a way, that is what Iosif's argument does. – Mittens Nov 09 '22 at 21:24
  • @OliverDíaz Ah now I got your point. Thank you so much! – Analyst Nov 09 '22 at 21:25

2 Answers2

1

Here is a short proof of the main Theorem in the OP. Let $G$ be an open set in $X$. Since $|\mu|$ is a Radon measure, for $\varepsilon>0$ there is $f\in\mathcal{C}_{00}(X)$ with $0\leq f\prec G$ such that $$|\mu|(G)-\tfrac{\varepsilon}{2}<|\mu|(f)=\int f\,d|\mu|$$ Since $|\mu|(f)=\sup\{|\mu(h)|:h\in\mathcal{C}_{00}(X;\mathbf{C}), |h|\leq f\}$ (see Proposition below), there is $g\in\mathbf{C}_{00}(X)$ such that $|g|\leq f$ and $$|\mu|(f)-\tfrac{\varepsilon}{2}<|\mu(g)|$$ Hence \begin{align} |\mu|(G)-\varepsilon&<|\mu(g)|=\lim_n|\mu_n(g)|\\ &\leq\liminf_n|\mu_n|(|g|)\\ &\leq \liminf_n|\mu_n|(G) \end{align} since $|g|\leq f\leq\mathbb{1}_G$.


This is to show that

Proposition: For any $f\in\mathcal{C}^+_{00}(X)$ $$|\mu|(f)=\sup\{|\mu(h)|:h\in\mathcal{C}_{00}(X;\mathbf{C}), |h|\leq f\}$$

By the Radon-Nykodym theorem (all measures here are finite) there is $\phi\in L_1(|\mu|;\mathbb{C})$ with $|\phi|=1$ such that $d\mu=\phi\cdot\,d|\mu|$. Notice that $|\mu(f\overline{\phi})|=\big||\mu|(f)\big|=|\mu|(f)$ and $|f\overline{\phi}|\leq|f|$. The only problem is that there is not reason for $\phi$ to be continuous. Let $\varepsilon>0$. Since $K_f=\operatorname{supp}(f)$ is compact, there is a relatively compact open set $U$ such that $K_f\subset U\subset \overline{U}$ with $|\mu|(U\setminus K_f)<\varepsilon$. Appealing to Lusin's theorem, there is $g\in\mathcal{C}_{00}(X;\mathbb{C})$ with $\operatorname{supp}(g)\subset U$ and such that $$|\mu|(\{f\overline{\phi}\neq g\})<\varepsilon$$ and $\|g\|_u\leq \|f\|_u$. Suppose $g=g_r+ ig_i$, $g_r,g_i\in\mathcal{C}_{00}(X;\mathbb{R})$, and set $$h=\max(-f,\min(g_r,f))+ i\max(-f,\min(g_i,f))$$ Then, $h\in\mathcal{C}_{00}(X;\mathbb{C})$, $|h|\leq f$ and $|\mu(h)-|\mu|(f)|=|\mu(h-f\overline{\phi})|\leq 2\|f\|_u\varepsilon$

Mittens
  • 39,145
  • Thank you so much! Basically, you have generalized this result. – Analyst Nov 10 '22 at 13:56
  • I have also posted my fix below. Could you have a check on it? – Analyst Nov 10 '22 at 14:00
  • @Analyst: Everything looks fine now. – Mittens Nov 10 '22 at 15:06
  • Essentially, Lusin's theorem helps us avoid the clever construction that is the cornerstone of Iosif's answer. – Analyst Nov 10 '22 at 15:09
  • 1
    @Analyst: Sure, but by the law of the no-free lunch, one first needs to make an effort of going through Lusin's thorium instead of the effort of building things from scratch. Either way, the issue of continuity is resolved. – Mittens Nov 10 '22 at 15:12
  • I have just asked this question about a possible generalization of your lemma. – Analyst Nov 10 '22 at 16:52
  • I have not worked with the Bochner integral than much. But, the space space of elementary functions used to build it (functions of the form $\sum^N_{j=1}\phi_j\mathbf{u}j$ with $\phi_j\in C{00}(X)$ and $\mathbf{u}j\in E$ are dense in $L_1(X,E)$. I think some conditions of separability may be needed to get stronger measurability of all functions in $C{00}(X,E)$. – Mittens Nov 10 '22 at 17:00
  • If $f$ is compactly supported and continuous, then its image is compact and thus separable. Then by Pettis's theorem $f$ is strongly measurable. – Analyst Nov 10 '22 at 17:04
  • There you go then. Like I said the elementary functions $\mathcal{C}{00}(X,\mathbb{R})\oplus E$, i.e. functions of the form $\sum^n{j=1}\phi_j\mathbf{u}j$ are dense in $L_1(X,E)$ and so, $C{00}(X, E)$ is dense in $L_1(X;E)$ since $\mathcal{C}{00}(X;E)\oplus E\subset \mathcal{C}{00}(X;E)$.... I will be in work meeting in a few minutes. We can still chat tomorrow if you have more things to discuss. – Mittens Nov 10 '22 at 17:08
  • Maybe I miss your idea, but I'm interested in stronger result, i.e, $f \in \mathcal L_p$ can be approximated by $g \in \mathcal C_c$ such that $|g| \le |f|$. See you. – Analyst Nov 10 '22 at 17:10
  • 1
    The $L_p$ space can be constructed similarly to the $L_1$ space starting with elementary functions $\mathcal{C}{00}(X,\mathbb{R})\otimes E$ of the form $\Phi:=\sum^N{j=1}\phi_j\mathbf{u}_j$ under the Daniell mean $|\Phi|^*=\big||\Phi|_E\big|_p$ ($1\leq p<\infty$). As usual in this constructions, the elementary set of functions is dense in $L_p(X;E)$ since the latter is the closure of the space of elementary functions under the DAniell pseudonorm $|;|_p$ – Mittens Nov 10 '22 at 18:55
1

As @OliverDíaz mentioned in a comment, my proof was wrong because $g$ is measurable but not necessarily continuous. Below is my fix.


Let $\mathcal X$ be the Borel $\sigma$-algebra of $X$. By polar decomposition, there is a measurable $g:X \to [0, 2\pi)$ such that $$ |\mu| (B) = \int_B e^{i g} \mathrm d \mu \quad \forall B \in \mathcal X. $$

Fix $\varepsilon>0$. By Lemma 1, there is a compact subset $K$ of $O$ such that $|\mu| (O \setminus K) < \varepsilon$. By Lemma 2, there is an open subset $V$ of $X$ such that $\overline V$ is compact and $K \subset V \subset \overline V \subset O$. By Urysohn's lemma, there is a continuous map $f:X \to [0, 1]$ such that $f(K)=1$ and $f(X \setminus V) = 0$. This implies the support of $f$ is contained in $\overline V$, so $f \in \mathcal C_c (X)$. Also, $|f| \le 1_{O}$.

Lusin's theorem Let $X, Y$ be topological spaces where $X$ is Hausdorff and $Y$ second-countable. Let $\mu$ be a finite Radon measure on $X$ and $f:X \to Y$ measurable. Fix $\varepsilon>0$.

  • weak version There is a compact subset $K$ of $X$ such that $\mu(X \setminus K) < \varepsilon$ and $f_{\restriction K}$ is continuous.
  • strong version Assume moreover that $X$ is normal (or locally compact) and $Y=\mathbb R$. Then there is a compact subset $K$ of $X$ and a continuous map $g:X \to \mathbb R$ such that $\mu(X \setminus K) < \varepsilon$, $f_{\restriction K} = g_{\restriction K}$, and $\sup_{x\in K} |f(x)| = \sup_{x\in X} |g(x)|$.

Note The strong version is obtained from weak version and Tietze extension theorem.

By strong Lusin's theorem there is a compact subset $L$ of $X$ and a continuous map $h:X \to \mathbb R$ such that $|\mu| (X \setminus L) < \varepsilon$, $h_{\restriction L} = g_{\restriction L}$, and $\sup_{x\in L} |g(x)| = \sup_{x\in X} |h(x)| \le 2\pi$. Notice that $f \in \mathcal C_c(X, \mathbb R)$ and $fe^{ih} \in \mathcal C_c(X, \mathbb C)$ are both supported on $O$. We have $$ \begin{align} |\mu| (O) - \operatorname{Re} \int_O fe^{i h} \mathrm d \mu &= \operatorname{Re} \int_X (1_Oe^{ig}-fe^{i h}) \mathrm d \mu \\ &= \operatorname{Re} \int_X \big [ (1_O-f)e^{ig}+f (e^{ig}-e^{i h}) \big ] \mathrm d \mu \\ &\le \int_X |1_O-f| \mathrm d |\mu| + \int_X |f| \cdot |e^{ig}-e^{i h}| \mathrm d |\mu| \\ &= \int_{O \setminus K} |1_O-f| \mathrm d |\mu| + \int_{X \setminus L} |f| \cdot |e^{ig}-e^{i h}| \mathrm d |\mu| \\ &\le |\mu| (O \setminus K) + 2 |\mu|(X \setminus L) \\ &\le \varepsilon + 2 \varepsilon = 3 \varepsilon. \end{align} $$

So $$ \begin{align} |\mu| (O) - 3\varepsilon &< \operatorname{Re} \int_O fe^{i h} \mathrm d \mu \\ &\overset{(1)}{=} \operatorname{Re} \lim_n \int_O fe^{i h} \mathrm d \mu_n \\ &= \lim_n \operatorname{Re} \int_O fe^{i h} \mathrm d \mu_n \\ &\le \liminf_n \left| \int_O fe^{i h} \mathrm d \mu_n \right| \\ &\le \liminf_n \int_O |fe^{i h}| \mathrm d |\mu_n | \\ &= \liminf_n \int_O |f| \mathrm d |\mu_n | \\ &\le \liminf_n \int_O \mathrm d |\mu_n | \\ &= \liminf_n |\mu_n| (O). \end{align} $$

Notice that $(1)$ is true even if $fe^{i h}$ is complex-valued. The claim then follows by taking the limit $\varepsilon \to 0^+$.

Analyst
  • 5,637