Mathematics Stack Exchange is a question and answer site for people studying math at any level and professionals in related fields. Join them; it only takes a minute:

Sign up
Here's how it works:
  1. Anybody can ask a question
  2. Anybody can answer
  3. The best answers are voted up and rise to the top

Let $X$ be the vector space of all the continuous complex-valued functions on $[0,1]$. Then $X$ has an inner product $$(f,g) = \int_0^1 f(t)\overline{g(t)} dt$$ to make it an inner product space. But this is not a Hilbert space.

Why isn't is complete? Which Cauchy sequence in it is not convergent?


share|cite|improve this question
In this norm, you can have a sequence of continuous functions converging to a $L^2$ function. – Tomás Nov 21 '12 at 11:35
The functions $\exp(-ax)$ as $a\to\infty$ converge pointwise on $[0,1]$ to the function $$\begin{cases}1&x=0\\0&x\ne0,\end{cases}$$ which is not continuous at $x=0$. – anon Nov 21 '12 at 11:38
anon, we do not care of pointwise convergence here. In the given norm, your sequence converges to $0$, which is continuous. However, you can easily fix your example by adding a chunk of meat on $[-1,0]$ ;). – fedja Nov 21 '12 at 11:44
So you mean, "why is this space not complete in the norm induced by this inner product?" – JohnD Dec 16 '12 at 21:38

Consider the sequence of continuous functions $f_n$ defined by $$ f_n(x) = \left\{ \begin{array}{rl} 0 &\mbox{ if $x\in \Big[0,\frac{1}{2}-\frac{1}{n}\Big]$} \\ \frac{nx}{2}-\frac{n}{4}+\frac{1}{2} &\mbox{ if $x\in\Big[\frac{1}{2}-\frac{1}{n},\frac{1}{2}+\frac{1}{n}\Big]$}\\ 1 &\mbox{ if $x\in\Big[\frac{1}{2}+\frac{1}{n},1\Big]$} \end{array} \right. $$

Let $f(x)=0$ if $x\in[0,\frac{1}{2}]$ and $f(x)=1$ if $x\in (\frac{1}{2},1]$. Im gonna prove that $f_n\rightarrow f$ in your norm. In fact, we have \begin{eqnarray} \int_0^1|f_n-f|^2 &=& \int_{\frac{1}{2}-\frac{1}{n}}^{\frac{1}{2}+\frac{1}{n}}( \frac{nx}{2}-\frac{n}{4}+\frac{1}{2}-f)^2 \nonumber \\ &=& \int_{\frac{1}{2}-\frac{1}{n}}^{\frac{1}{2}} (\frac{nx}{2}-\frac{n}{4}+\frac{1}{2})^2+\int_{\frac{1}{2}}^{\frac{1}{2}+\frac{1}{n}}(\frac{nx}{2}-\frac{n}{4}-\frac{1}{2})^2 \nonumber \\ &=& \frac{1}{12n}+\frac{7}{12n}\\ &\rightarrow& 0 \end{eqnarray}

So the sequence $f_n$ converges to a function $f$ not continuous.

share|cite|improve this answer
As remarked by fedja in the comments under the question, this sequence converges to the zero function in the given norm. In fact that is exactly what you proved, the function $f$ does not appear in your proof! And your $f$ is not an element of the vector space $X$ considered (it is discontinuous) so saying that $f_n\to f$ in the norm on $X$ just makes no sense. – Marc van Leeuwen Nov 21 '12 at 13:56
You are right. Im gonnna change it. – Tomás Nov 21 '12 at 14:06
Is it right now @MarcvanLeeuwen ? – Tomás Nov 21 '12 at 14:23
I didn't check the integrals, but the example looks OK now (and a lot like the examples in the other answers). – Marc van Leeuwen Nov 21 '12 at 15:00

I replace $[0,1]$ by $I:=[-1,1]$. We have to produce a sequence $f_n:\ I\to{\mathbb R}$ of continuous functions which is a Cauchy sequence in $L^2(I)$ but does not converge in $L^2(I)$ to an $f\in X$.

Put $$f_n(t):=\cases{n t\quad &$(|t|\leq{1\over n})$ \cr {\rm sgn}\ t & $(|t|\geq{1\over n})$ \cr}\ .$$ Let an $\epsilon >0$ be given. When $m$, $n>\frac1\epsilon$ then $|f_m(t)-f_n(t)|\leq 1$ when $|t|\leq\epsilon$, and $=0$ when $|t|>\epsilon$. Therefore $\|f_m-f_n\|^2\leq 2\epsilon$, which shows that $(f_n)_{n\geq1}$ is a Cauchy sequence in $L^2(I)$. Therefore the $f_n$ converge in $L^2(I)$ to a certain $f\in L^2(I)$.

Assume that this $f$ has a continuous representant, again denoted by $f$, and that $f(0)=:c\leq0$. Then there is an $\epsilon_0>0$ with $$f(t)\leq{1\over 2}\qquad(0\leq t\leq\epsilon_0)\ .$$ When $n>{2\over\epsilon_0}$ then $$f_n(t)=1\qquad({\epsilon_0\over2}\leq t\leq\epsilon_0)\ .$$ It follows that $|f_n(t)-f(t)|\geq{1\over2}$ on an interval of length ${\epsilon_0\over2}$, and this implies $\|f_n-f\|^2\geq{\epsilon_0\over 8}$ for all these $n$. This contradicts the already established fact $\lim_{n\to\infty} f_n=f$ in $L^2(I)$.

share|cite|improve this answer

Let $$f_n(t) := \left(1-n \cdot dist\left(\{t\}, \left[\frac{1}{4},\frac{3}{4} \right] \right) \right) \vee 0$$

Then $(f_n)_n$ is a Cauchy-sequence,

$$f_n \to 1_{\left[\frac{1}{4},\frac{3}{4} \right]} \notin C[0,1]$$

share|cite|improve this answer

Your Answer


By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.