Let f_n denote an element in a sequence of functions that converges

  • #1
1,079
89
TL;DR Summary
uniformly to ##f##, where ##f## is some function. Suppose also that each ##f_n## is Riemann-integrable. Show that ##f## is Riemann-integrable also, and that the integral of ##f## is equal to that of ##f_n## whenever ##n## is sufficiently large.
Let ##\epsilon>0##. Choose ##N\in\mathbb{N}## s.t. for each integer ##n## s.t. ##n\geq N##,

$$|\sup\{|f-f_n|(x):x\in D\}|<\frac{\epsilon}{3}$$

where ##D## denotes the intersection of the domains of ##f## and ##f_n##.

Choose a partition ##P:=\{x_0,\ldots,x_m\}## with ##x_i<x_{i+1}## where ##0\leq i\leq m-1## s.t. the following holds.

%%% EDIT: The function to be integrated has been corrected from ##f## to ##f_n##.

$$|f_n(s_i)-f_n(t_i)|\Delta x_i<\frac{\epsilon}{3}$$
$$\Delta x_i\leq 1$$
$$|f_n(t_i)\Delta x_i -\int_{x_i}^{x_{i+1}} f_n|<\frac{\epsilon}{3}$$

where ##s_i,t_i\in[x_i,x_{i+1}]## and ##\Delta x_i:=x_{i+1}-x_i##.

Hence,

$$\eqalign{%
\epsilon>&|f_n(s_i)-f(t_i)|\Delta x_i\cr
+&||f-f_n||+|f_n(s_i)\Delta x_i-\int_{x_i}^{x_{i+1}}f_n|\cr
\geq&|f_n(s_i)-f_n(t_i)|\Delta x_i\cr
+&|f(s_i)-f_n(s_i)|+|f_n(t_i)\Delta x_i - \int_{x_i}^{x_{i+1}}f_n|\cr
\geq&|(f_n(s_i)-f_n(t_i))\Delta x_i\cr
+&(f(s_i)+f_n(s_i))\Delta x_i\cr
+&(f_n(t_i)\Delta x_i-\int_{x_i}^{x_{i+1}} f_n)|\cr
=&|f(s_i)\Delta x_i-\int_{x_i}^{x_{i+1}}f_n|}$$

%%%

My main worry about this proof attempt is that it only shows that a rectangle with height equal to ##f(s_i)## and base ##\Delta x_i## has an area that is within ##\epsilon## units of the area under the graph of ##f## over the region ##[x_i,x_{i+1}]##. But I'm not sure how to expand the proof, if it is correct, in order to show that the area of the entire Riemann sum converges to the integral of ##f##.

Ideally, I would start by building the partition ##P## one point at a time so that the inequality above will hold. But the problem is that I wouldn't know beforehand how many partitions of ##[x_0,x_m]## would be needed. Maybe I could have each partition be of the same length sufficient in order to ensure convergence, then calculate it that way. But I'm a bit hesitant about implementing this idea; what if there is no uniform partition length that will ensure that the area of the rectangle converges to the area under ##f## over that partition?
 
Last edited:
  • #2
The definition of your partition assumes ##\int f## exists!

The problem is also terribly stated, the integrals are probably never going to be exactly equal.

I think you should start with a simpler question. Prove the sequence ##\int_a^b f_n(x) dx## converges to a number
 
  • #3
The definition of your partition assumes ∫f exists!
Oh. I must have forgotten to subscript that.
 
  • #4
Prove the sequence ##\int_a^b f_n(x) dx## converges to a number
Choose ##N## big enough. Let ##m## be an integer greater than ##N##. Then ##\sup\{|f_m-f|(x)\}\rightarrow0##.Since the integral of a non-negative function is non-negative, and ##m## is large enough,

$$\int|f-f_m|\rightarrow0$$

Also,

$$|\int f - \int f_m| \leq \int|f-f_m|$$

so ##\{\int f_i\}_{i\in\mathbb{N}}## converges to ##\int f##. By the triangle inequality, the marked sum in my opening post must converge to that value as well. Hence, ##f## is Riemann-integrable, I think.
 
  • #5
Using the fact that Riemann and Darboux integrability are equivalent, and that a function [itex]f : [a,b] \to \mathbb{R}[/itex] is Darboux integrable if and only if for every [itex]\epsilon > 0[/itex] there exists a partition [itex]D[/itex] of [itex][a,b][/itex] such that [itex]U(f,D) - L(f,D) < \epsilon[/itex] where [itex]U(f,D)[/itex] and [itex]L(f,D)[/itex] are the upper and lower sums of [itex]f[/itex] with respect to [itex]D[/itex]:

Let [itex]\epsilon > 0[/itex]. By uniform convergence of [itex]f_n \to f[/itex] on [itex][a,b][/itex] we have that for [itex]n[/itex] sufficiently large [tex]
\sup|f_n - f| < \frac{\epsilon}{3(b-a)}.[/tex] Since [itex]f_n[/itex] is integrable, there exists a partition [itex]D[/itex] of [itex][a,b][/itex] such that [itex]U(f_n,D) - L(f_n,D) < \frac13 \epsilon[/itex]. Hence [tex]
\begin{split}
U(f,D) - L(f,D) &< \left(U(f_n,D) + \tfrac13\epsilon\right) - \left(L(f_n,D) - \tfrac13\epsilon\right) \\
&= U(f_n,D) - L(f_n,D) + \tfrac23\epsilon \\
&< \tfrac13 \epsilon + \tfrac23 \epsilon \\
&= \epsilon\end{split}[/tex] and [itex]f[/itex] is integrable.

To show that [itex]\int_a^b f_n\,dx \to \int_a^b f\,dx[/itex], let [itex]\epsilon > 0[/itex] and [itex]n[/itex] sufficiently large that [itex]\sup |f_n - f| < \epsilon/(b-a)[/itex]. Then [tex]
\left|\int_a^b f_n(x)\,dx - \int_a^b f(x)\,dx\right| \leq \int_a^b |f_n(x) - f(x)|\,dx \leq (b-a)\sup |f_n - f| < \epsilon.[/tex]
 
  • #6
Choose ##N## big enough. Let ##m## be an integer greater than ##N##. Then ##\sup\{|f_m-f|(x)\}\rightarrow0##.Since the integral of a non-negative function is non-negative, and ##m## is large enough,

$$\int|f-f_m|\rightarrow0$$

Also,

$$|\int f - \int f_m| \leq \int|f-f_m|$$

so ##\{\int f_i\}_{i\in\mathbb{N}}## converges to ##\int f##. By the triangle inequality, the marked sum in my opening post must converge to that value as well. Hence, ##f## is Riemann-integrable, I think.
Sorry I wasn't clear. Prove it without assuming that ##\int f## exists! This was me proposing a first step in the proof.
 
  • #7
Prove it without assuming that ##\int f## exists!
When was it assumed that ##\int f## exists? All I deduced was that ##\int |f_m-f|## exists, because the index ##m## is sufficiently large, meaning that ##|f_m-f|(x)\leq0## [a fact that I now realize does not actually follow from the definition of uniform convergence], and because the function ##|f_m-f|## is non-negative. Then I used the inequality ##|\int g|\leq|\int|g|## for every integrable function ##g##.

I'm not sure if I have the energy to formulate a proper/correct proof right now. And in any case, I don't very much see the point, seeing as someone has gone through the liberty of doing that already.
 
Last edited:
  • #8
You don't know that ##|f_n-f|## is integrable, as one example ##f_n=\frac{1}{n} \mathbb{1}_{\mathbb{Q}}## uniformly converges to ##f=0##, but ##|f_n-f|## is not integrated.

(Obviously the ##f_n## are not integrable, but we know a true counterexample doesn't exist).

The idea I was thinking of was that you can prove the sequence is Cauchy.

If you're happy with the outcome of the thread no need to worry more about my proposal.
 

Suggested for: Let f_n denote an element in a sequence of functions that converges

Back
Top