I want to prove that LDCT(Lebesgue Dominated Convergence Theorem) continues to hold if I replace the hypothesis $f_n \to f$ (convergence pointwise) with $f_n\to f$ (convergence in measure): $$\int fd\lambda=\lim_{n\to\infty}\int f_nd\lambda.$$

- 2,178

- 7,614
-
This is not exactly an extension, since convergence in measure does not imply convergence almost everywhere (only that there is an almost everywhere convergent subsequence). – tomasz Mar 09 '14 at 11:25
2 Answers
Here is another proof:
I'll prove that $\displaystyle a_n:=\int_\Omega f_nd\mu\longrightarrow l:=\int_\Omega fd\mu$ as $n$ tends to $\infty$.
By a very useful fact in Analysis, it's enough to prove that for each subsequence of $\{a_n\}$ like $\{a_{n_k}\}$, there is a subsequence of $\{a_{n_k}\}$ like $\{a_{n_{k_l}}\}$ which converges to $l$.
Now, given a subsequence $\{a_{n_k}\}$, we have $$f_{n_k}\xrightarrow[]{\;\mu\;}f.\tag{I}$$
By this fact, there exists a subsequence of $\{f_{n_k}\}$ like $\{f_{n_{k_l}}\}$ which $$f_{n_{k_l}}\xrightarrow{\;a.e\;}f.\tag{II}$$ (Note that for $\text{(II)}$ we should have assumed that $\Omega$ is of finite measure, however we can get rid of it as @Leo Lerena pointed in the comments.)
Since $\{f_{n_{k_l}}\}$ is dominated by function $g\in\mathcal{L}^1(\Omega,\mu)$, by the original version of $LDCT$, $$\int_\Omega f_{n_{k_l}}\xrightarrow{\;\;}\int_\Omega f.\tag{III}$$ That is : $$a_{n_{k_l}}\xrightarrow{n\rightarrow\infty}l \tag*{$\square$.}$$
Note
The technique of using subsequences, rather than sequences, is one of the most powerful tools of proof !

- 2,178
-
-
2
-
2As a comment, this works only when $E$ is of finite measure but it can easily be extended to the case when it isn't. – hedphelym Apr 30 '18 at 16:30
-
2Actually $\text{(II)}$ is true for $\textbf{every}$ measure space. The only tool we use to prove $\text{(II)}$ is the Borel-Cantelli Lemma, which is true for $\textbf{every}$ measure space. c.f. here . – Sam Wong Dec 02 '20 at 13:56
-
1Why does $\int_\Omega f_{n_{k_l}}\xrightarrow{;a.e;}\int_\Omega f$ ? Shouldn't it just be $\int_\Omega f_{n_{k_l}}\rightarrow\int_\Omega f$ as $l\to\infty$? – no lemon no melon Nov 21 '23 at 08:00
-
1
Call $(X,\cal F,\mu)$ the involved measure space. Let $g$ integrable such that $|f_n(x)|\leqslant g(x)$ for almost every $x$.
As $g$ is integrable, denote $X':=\{g\neq 0\}=\bigcup_{n\geqslant 1}\{x,|g(x)|>n^{-1}\}$. Then $X'$ with the induced measure is $\sigma$-finite. Applying this version of dominated convergence theorem, we get that $$\int_{X'}fd\mu=\lim_{n\to +\infty}\int_{X'}f_nd\mu.$$ As $X\setminus X'=\{g=0\}\subset \{f=0\}\cup\bigcap_{n\geqslant 1}\{f_n=0\}$, we have $\int_{X'}fd\mu=\int_Xfd\mu$.
So fore each $n$, $\int_{X\setminus X'}fd\mu=\int_{X\setminus X'}f_nd\mu=0$, giving the wanted result.

- 172,925
-
-
Yes (and in an arbitrary measured space when we have a non-negative measure). – Davide Giraudo Nov 17 '12 at 22:47