Is the sequence $(x_n)$ convergent in the space $L^1(0,1)$ ?

$x_n(t)= n^2 t^n (1-t^2)$ for $n\in\mathbb{N}$.

norm: $\|x\|=\int_{(0,1)} \left|x(t)\right| \; dt$

I think it should converge to zero, by checking the graph. but I cannot show it explicitly.

Could you please help?

  • $\begingroup$ Have you tried using dominated convergence? As far as i can see, this will give you the desired convergence directly. $\endgroup$ – Ukhrir Feb 27 '14 at 12:09
  • $\begingroup$ Note $x_n\ge0$ and $(x_n)$ converges pointwise to the zero function. You can compute the required integrals explicitly, and easily. $\endgroup$ – David Mitra Feb 27 '14 at 12:20
  • $\left| (x_n)(t) \right|$ converges pointwise to the $0$ function on $[0,1]$.

  • We have \begin{align*} \left\|{x_n}\right\| &= \int_0^1 \left(n^2 t^n - n^2 t^{n+2}\right) \; dt \\ &= \frac{n^2}{n+1} - \frac{n^2}{n + 3} \\ &= \frac{2n^2}{(n+1)(n+3)} \to 2 \\ \end{align*}

Since $L^1$ convergence implies pointwise convergence almost everywhere for some subsequence, the only candidate for $x_n$ to converge to in the $L^1$ norm is the zero function. But $\left\|{x_n - 0}\right\| \to 2 \ne 0$, so $x_n$ does not converge to any function in the $L^1$ norm.

Direct proof that $L^1$ convergence implies pointwise convergence almost everywhere for some subsequence:

Let $f_n \to f$ in $L^1$. Then for any $k$, let $E_{n,k}$ be the set of points where $\left|f_n - f\right| > \frac{1}{k}$. Note that $\left\|{f_n - f}\right\|_{1} \ge \frac{1}{k} \mu(E_{n,k})$, implying that for a fixed $k$, $\mu(E_{n,k}) \to 0$.

For each $i$, choose an integer $n_i$ so that $n_i$ is an increasing sequence of integers, and $\mu(E_{n_i, i}) < \frac{1}{2^i}$. Note that this means $\mu(E_{n_i, k}) < \frac{1}{2^i}$ whenver $k \le i$.

Now consider the set $E$ of points $x$ where $f_{n_i}(x) \not \to f(x)$. We have $f_{n_i}(x) \not \to f(x)$ if and only if there is some $k$ such that $\left|f_{n_i}(x) - f(x) \right| > \frac1k$ infinitely often. Therefore, $$ E = \bigcup_{k=1}^\infty \bigcap_{N=1}^\infty \bigcup_{i=N}^\infty E_{n_i,k} $$ So \begin{align*} \mu(E) &\le \sum_{k=1}^\infty \lim_{N \to \infty} \sum_{i=N}^\infty \mu(E_{n_i,k}) \\ &= \sum_{k=1}^\infty \lim_{N \to \infty \atop N \ge k} \sum_{i=N}^\infty \mu(E_{n_i,k}) \\ &\le \sum_{k=1}^\infty \lim_{N \to \infty \atop N \ge k} \sum_{i=N}^\infty \mu(E_{n_i,i}) \quad \quad (\text{since } k \le i) \\ &\le \sum_{k=1}^\infty \lim_{N \to \infty \atop N \ge k} \sum_{i=N}^\infty \frac{1}{2^{i}} \\ &= \sum_{k=1}^\infty (0) = 0\\ \end{align*}

Therefore, $f_{n_i} \to f$ pointwise outside the set $E$ of measure $0$.

In fact, the above proof used the fact that $f_n \to f$ in measure, which is a weaker (more general) condition than convergence in $L^1$ norm. For an extended discussion of this fact (Corollary 3, Exercise 6) and others, see Terry Tao's blog post. See also this more recent mathSE question.

  • $\begingroup$ thanks. could you please explain the part: "$L^1$ convergence implies pointwise convergence almost everywhere" $\endgroup$ – 104078 Feb 27 '14 at 16:28
  • $\begingroup$ @user104078 I didn't type that quite right; I meant $L^1$ convergence implies pointwise convergence almost everywhere for some subsequence. $\endgroup$ – 6005 Feb 27 '14 at 17:00
  • $\begingroup$ @user104078 I've added a proof of that fact (which was stated incorrectly before). Let me know if that helps. $\endgroup$ – 6005 Feb 27 '14 at 17:16

$\{f_n\}$ does not converge in the $L^1$ sense.

Observation I. $\lim_{n\to\infty}\int_0^1 f_n(x)\,dx=2.$

Observation II. For every $0<a<1$, the sequence $f_n$ converges uniformly to $0$ in $[0,a]$.

So, due to II, for every $a\in(0,1)$, there exists an $n_0\in\mathbb N$, such that for $n\ge n_0$: $$ \int_a^{1}f_n(x)\,dx>3/2 \quad\text{and}\quad \int_0^a f_n(x)\,dx<1/2. $$ Then, due to I, we can find an $m>n$, such that $f_m(x)<1/2$, for $x\in[0,a_n]$, and thus $\int_{a_n}^1 f_m(x)\,dx>3/2$. Thus $$ \int_0^1 \lvert f_m-f_n\rvert\,dx\ge\int_{a_n}^1 \lvert f_m-f_n\rvert\,dx\ge \int_{a_n}^1 f_m-\int_{a_n}^1 f_n\ge \frac{3}{2}-\frac{1}{2}=1. $$

  • $\begingroup$ "For every $n$, there exists an $a_n < 1$, such that $\int_0^{a_n} f_n(x) \; dx > 3/2$" is false. For $n = 1$, we have $\int_0^1 f_1 = \int_0^1 t(1 - t^2) \; dt = \frac14$, so no subinterval has an integral bigger than $1/4$. $\endgroup$ – 6005 Feb 27 '14 at 12:40
  • $\begingroup$ @Goos: Correct. I have made the appropriate changes. $\endgroup$ – Yiorgos S. Smyrlis Feb 27 '14 at 13:18

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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