2

This following question is a question of Second Edition of Fourier Series and Boundary Value Problems by Churchill about Legendre Polynomials $P_n$:

Explain why it's true that

(a) when $-1 < x < 1$, then

$$\lim_{n \rightarrow \infty}P_n(x) = 0 \hspace{3cm} (n = 0, 1, 2, \cdots);$$

(b) when $f$ is sectional continuous on $(0,1)$, then

$$\lim_{n \rightarrow \infty} \sqrt{4n+1} \int_0^1 f(x)P_{2n}(x) dx = 0$$

$\textbf{My attempt:}$

(a)

By the theory developed in Churchill's book, we know that

$$\lim_{n \rightarrow \infty} \left( P_n(x) , P_n(x) \right) = \lim_{n \rightarrow \infty} ||P_n(x)||^2 = \lim_{n \rightarrow \infty} \frac{2}{2n+1} = 0.$$

By the continuity of the norm, we have

$$||\lim_{n \rightarrow \infty} P_n(x)||^2 \lim_{n \rightarrow \infty} ||P_n(x)||^2 = \lim_{n \rightarrow \infty} \frac{2}{2n+1} = 0,$$

which implies, by the definition of the norm, that

$\lim_{n \rightarrow \infty} P_n(x) = 0$.

(b)

By Cauchy-Schwarz Inequality,

$$\int_0^1 f(x) P_{2n}(x) dx \leq \left( \int_0^1 \left[ f(x) \right]^2 dx \right)^{\frac{1}{2}} \left( \int_0^1 \left[ P_{2n}(x) \right]^2 dx \right)^{\frac{1}{2}}$$

Using that $||P_n(x)||^2 = \frac{2}{2n+1}$, we have that $||P_{2n}(x)||^2 = \int_{-1}^1 \left[ P_{2n}(x) \right]^2 dx = 2 \left( \int_0^1 \left[ P_{2n}(x) \right]^2 dx \right) \Longrightarrow \left( \frac{2}{4n+1} \right) = 2 \left( \int_0^1 \left[ P_{2n}(x) \right]^2 dx \right)$, i.e., $\left( \int_0^1 \left[ P_{2n}(x) \right]^2 dx \right)^{\frac{1}{2}} = \frac{1}{\sqrt{4n+1}}$, then

$$\int_0^1 f(x) P_{2n}(x) dx \leq \left( \int_0^1 \left[ f(x) \right]^2 dx \right)^{\frac{1}{2}} \frac{1}{\sqrt{4n+1}}$$

$$\Longrightarrow \lim_{n \rightarrow \infty} \sqrt{4n+1} \int_0^1 f(x)P_{2n}(x) dx \leq \left( \int_0^1 \left[ f(x) \right]^2 dx \right)^{\frac{1}{2}}$$

I'm stuck here and I do not have ideas on what to do in order to prove the item (b).

I would like to know if my attempt for (a) is correct and I would like to receive a hint for solve (b).

Thanks in advance!

$\textbf{EDIT:}$

I saw how to do the exercises and I return to post the solutions if anyone looks for these exercises.

(a)

By the theory developed by Churchil in his book,

$$|P_n(x)| < \sqrt{\frac{\pi}{2n \left( 1 - x^2 \right)}},$$

then

$$\lim_{n \rightarrow \infty} |P_n(x)| \leq \lim_{n \rightarrow \infty} \sqrt{\frac{\pi}{2n \left( 1 - x^2 \right)}} = 0,$$

so that $\lim_{n \rightarrow \infty} |P_n(x)| = 0$, which implies that

$$\lim_{n \rightarrow \infty} P_n(x) = 0.$$

(b)

Considering the pair extension of $f$ and the orthonormal set $\left\{ \frac{P_n}{\sqrt{\frac{2}{2n+1}}} ; n \in \mathbb{N} \right\}$, we define $c_n := \int_{-1}^1 \frac{P_n}{\sqrt{\frac{2}{2n+1}}} f(x) dx$, then $c_n := \int_{-1}^1 \frac{P_n}{\sqrt{\frac{2}{2n+1}}} f(x) dx = \sqrt{\frac{2n+1}{2}} \int_{-1}^1 P_n(x) f(x) dx$

If $n = 2k (k \in \mathbb{N})$:

$$c_{2k} = 2 \sqrt{\frac{4k+1}{2}} \int_0^1 P_{2k}(x) f(x) dx.$$

If $n = 2k+1 (k \in \mathbb{N})$:

$$c_{2k+1} = 0.$$

Observe that $c_n = \left( f, \frac{P_n}{||P_n||} \right)$, i.e., $c_n$ is the coefficient of generalized Fourier Series of $f$, then we can apply the Bessel's inequality:

$\sum_{k=1}^{\infty} \left[ 2 \sqrt{\frac{4k+1}{2}} \int_0^1 P_{2k} (x) f(x) dx \right]^2 = \sum_{k=1}^{\infty} \left[ \sqrt{\frac{4k+1}{2}} \int_{-1}^1 P_{2k} (x) f(x) dx \right]^2 \leq \int_{-1}^1 f(x)^2dx$

$\int_{-1}^1 f(x)^2dx < \infty$, then $\sum_{k=1}^{\infty} \left[ 2 \sqrt{\frac{4k+1}{2}} \int_0^1 P_{2k} (x) f(x) dx \right]^2 < \infty,$ therefore $\lim_{k \rightarrow \infty} \left[ 2 \sqrt{\frac{4k+1}{2}} \int_0^1 P_{2k} (x) f(x) dx \right] = 0$, then $\lim_{k \rightarrow \infty} \left[ \sqrt{4k+1} \int_0^1 P_{2k} (x) f(x) dx \right] = 0.$ $\square$

George
  • 3,817
  • 2
  • 15
  • 36
  • I am not familiar with the mentioned terminology: what does it mean for a function $f$ to be sectional continuous on $(0,1)$? – Jack D'Aurizio Nov 08 '17 at 00:24
  • 1
    It means that exists a finite number of points in $(0,1)$ where $f$ is discontinuous. – George Nov 08 '17 at 00:26
  • Many thanks for the clarification. I would also point out that your proof of $(a)$ is not correct: $|f_n|_2\to 0$ does not imply $f_n(x)\to 0$ pointwise, in general. Consider $f_n=n e^{-n^6 x^2}$, for instance. – Jack D'Aurizio Nov 08 '17 at 00:26
  • 1
    For a proof of $\forall x\in(-1,1), \lim_{n\to +\infty} P_n(x)=0$, see page 37 of my notes. – Jack D'Aurizio Nov 08 '17 at 00:31
  • Then $(b)$ is essentially a minor variation on the Riemann-Lebesgue lemma. Aside from $P_n(x)\to 0$ as $n\to +\infty$, it might be useful to prove that $|P_n(x)|\leq 1$ and that $P_n$ has $n$ real zeroes in $(-1,1)$ (this is a simple consequence of Rodrigues formula, for instance). – Jack D'Aurizio Nov 08 '17 at 00:37
  • Is there an elementary proof for inequality on page 37? It doesn't seem trivial even after reading the MSE that you mentioned in your notes.

    About (b), how exactly the item (b) is a variation on the Riemann-Lebesgue lemma? I can't see this.

    – George Nov 08 '17 at 01:06
  • You are stating that if a function is piecewise-continuous then the coefficients of its Fourier(-Legendre) series expansion decay faster than $\frac{1}{n^\alpha}$. But probably a more accurate reference would have been this one. – Jack D'Aurizio Nov 08 '17 at 01:16
  • Anyway, let us make an explicit mention of the Tricomi bound for $|P_n(x)|$ : https://math.stackexchange.com/questions/417999/local-maxima-of-legendre-polynomials – Jack D'Aurizio Nov 08 '17 at 01:18
  • @JackD'Aurizio, I got it the $(a)$ with an inequality given by Churchill and I saw a proof of $(b)$ using the Bessel's inequality. I edited my post and put the proofs. – George Nov 09 '17 at 12:03
  • Everything is fine now. – Jack D'Aurizio Nov 09 '17 at 12:09
  • 1
    @JackD'Aurizio, thanks a lot for your help! :) – George Nov 09 '17 at 13:47

0 Answers0