Skip to main content

Section 2.3 An Extended Criterion for the Interchange of Integration and Limits of a Sequence of Functions

Remark 2.3.1.

The conditions in Theorem 2.2.7 often can’t be verified on the entity of \(E\text{,}\) but can be verified after deleting a set of small size, for example, deleting a small neighbored of one or a finite number of points. If some further uniform integrability conditions are assumed, then the conclusions of Theorem 2.2.7 still holds. The discussion below will involve some aspect of improper integrals.

Proof.

First, for any \(a' < b', a \le a' < b' < b\text{,}\) we can apply Theorem 2.2.7 on \([a', b']\) to conclude that
\begin{equation*} \int_{a'}^{b'} f(x)\, d\alpha=\lim_{n\to \infty}\int_{a'}^{b'} f_{n}(x)\, d\alpha. \end{equation*}
Next, let \(c\) be chosen according to the uniform integrability of the family \(\left\{f_{n}\right\}\) in (2.3.2) for a given \(\epsilon>0\text{.}\) If we set \(a'=c\text{,}\) then for any \(c', c < c' < b\text{,}\) we would get
\begin{equation*} \left|\int_{c}^{c'} f_{n}(x)\, d\alpha \right| < \epsilon \text{ for all } n. \end{equation*}
This then implies that
\begin{equation*} \left|\int_{c}^{c'} f(x)\, d\alpha \right| =\lim_{n\to \infty } \left|\int_{c}^{c'} f_{n}(x)\, d\alpha \right| \le \epsilon, \end{equation*}
which implies that the integral \(\int_{a}^{b} f(x)\, d\alpha\) is convergent at \(x=b\text{,}\) and
\begin{equation*} \left|\int_{c}^{b} f(x)\, d\alpha \right| \le \epsilon. \end{equation*}
Finally, using
\begin{align*} \amp \left| \int_{a}^{b} f(x)\, d\alpha - \int_{a}^{b} f_{n}(x)\, d\alpha \right| \\ \le \amp \left|\int_{c}^{b} f(x)\, d\alpha - \int_{c}^{b} f_{n}(x)\, d\alpha \right| + \left| \int_{a}^{c} f(x)\, d\alpha - \int_{a}^{c} f_{n}(x)\, d\alpha \right|\\ \le \amp 2\epsilon + \left| \int_{a}^{c} f(x)\, d\alpha - \int_{a}^{c} f_{n}(x)\, d\alpha \right| \end{align*}
and applying Theorem 2.2.7 on \([a, c]\text{,}\) we find some \(N\) such that \(\left| \int_{a}^{c} f(x)\, d\alpha - \int_{a}^{c} f_{n}(x)\, d\alpha \right| < \epsilon\) for all \(n\ge N\text{,}\) which implies
\begin{equation*} \left| \int_{a}^{b} f(x)\, d\alpha - \int_{a}^{b} f_{n}(x)\, d\alpha \right| < 3\epsilon \end{equation*}
for all \(n\ge N\text{,}\) therefore proving the claimed conclusion.

Note 2.3.3.

What is called uniform integrability is also called equi-integrability. The prefix "equi" here refers uniformity in \(n\text{,}\) as the notion of equip-continuity to be introduced soon.
Note also that the above formulation allows \(b=\infty\text{,}\) which is a case of an improper integral.
One sufficient condition for (2.3.2) is the existence of an integrable dominating function \(g\) in the sense that
  • \(|f_{n}(x)|\le g(x) \) for all \(\in [a, b), n\) sufficiently large.
  • \(\int_{a}^{b} g\, d\alpha\) is convergent.
In the case that \(d\alpha\) gives rise to the usual infinite series, namely, when \(\alpha (x) =\sum_{n} \chi_{\left\{n\le x\right\}}(x)\text{,}\) the condition (2.3.2) takes the form of
\begin{equation*} \forall \;\epsilon >0, \exists\; N \text{ such that } \forall \; N'>N, \forall \;n, \left| \sum_{m=N}^{N'} f_{n}(m) \right| < \epsilon. \end{equation*}
Namely, the tail part of the summation, \(\sum_{m=N}^{\infty} f_{n}(m)\text{,}\) can be made small uniformly in \(n\text{.}\) One sufficient condition for the above is condition of the existence of a similar dominating function: \(\exists g(m)\ge 0\) defined for \(m\in \mathbb N\) such that
  • \(|f_{n}(m)|\le g(m) \) for all sufficiently large \(n, m.\)
  • \(\sum_{m=1}^{\infty} g(m) \) is convergent.

Examples for the extension.

A particular case of an integrable dominating function is a constant function when \(\alpha (b)\) is finite. Using this kind of argument one gets
\begin{equation*} \lim_{n\to \infty}\int_{0}^{1}\frac{1}{e^{nx}+1}\, dx = \int_{0}^{1}\lim_{n\to \infty} \frac{1}{e^{nx}+1}\, dx=0, \end{equation*}
even though the convergence \(\frac{1}{e^{nx}+1} \to 0\) is not uniform over \(x\in (0, 1)\text{.}\)
For the series \(\sum_{m=1}^{\infty}\frac{n}{1+n m^{2}}\text{,}\) \(g(m)=m^{-2}\) is a good dominating function, and we get
\begin{equation*} \lim_{n\to \infty} \sum_{m=1}^{\infty}\frac{n}{1+n m^{2}} = \sum_{m=1}^{\infty}\lim_{n\to \infty}\frac{n}{1+n m^{2}}= \sum_{m=1}^{\infty}\frac{1}{m^{2}}. \end{equation*}
On the other hand, consider the series \(\sum_{m=1}^{\infty}\frac{\chi_{\left\{m\le n\right\}}(m)}{n} =\sum_{m=1}^{n}\frac{1}{n}=1\text{,}\) the terms \(f_{n}(m) := \frac{\chi_{\left\{m\le n\right\}}(m)}{n} \to 0\) uniformly in \(m\) as \(n\to \infty\text{,}\) so in this situation
\begin{equation*} \lim_{n\to \infty} \sum_{m=1}^{\infty} \frac{\chi_{\left\{m\le n\right\}}(m)}{n} \ne \sum_{m=1}^{\infty} \lim_{n\to \infty} \frac{\chi_{\left\{m\le n\right\}}(m)}{n}, \end{equation*}
despite the terms in the series converging to \(0\) uniformly in \(m\) as \(n\to \infty\text{.}\)

Remark 2.3.4.

Often we encounter a situation similar to that of Theorem 2.2.7, but we work with a continuum family \(\left\{f(\cdot, y)\right\}_{y\in I}\) of functions in \(\mathcal R (\alpha)\) for the parameters \(y\) in some metric space \(I\text{,}\) instead of a sequence of functions. Then the appropriate modified conclusion should be
\begin{equation*} \lim_{y\to y_{0}} \int_{a}^{b} f(x, y)\, d\alpha = \int_{a}^{b} f(x, y_{0})\, d\alpha, \end{equation*}
and the appropriate modification of the uniform convergence condition should be
\begin{equation*} \forall \; \epsilon >0, \exists\; \delta >0 \text{ such that } |f(x, y)-f(x, y_{0})| < \epsilon \text{ for all } d(y, y_{0}) < \delta, x\in [a, b]. \end{equation*}
A similar modification for (2.3.2) can be formulated.

Exercise 2.3.5.

Identify the pointwise limit \(f(x)\) of the sequence \(\{ \frac{n^{2}}{n^{4}+x^{4}}\}\) for \(x\in [0, \infty)\text{.}\) Does it converge uniformly over \(x\in [0, \infty)\text{?}\) Does it holds that \(\int_{0}^{\infty }\frac{n^{2}}{n^{4}+x^{4}}\, dx \to \int_{0}^{\infty } f(x)\, dx\text{?}\)

Exercise 2.3.6.

Does it hold that \(\lim_{n\to \infty} \sum_{m=1}^{\infty}\frac{n}{n^{2}+ m^{2}}=0\text{?}\)