4
$\begingroup$

Let $f : (-1,1)\to \mathbb{R}$ be given by

$$f(x) = \begin{cases}0, & x\in (-1,0) \\ 1, & x \in [0,1)\end{cases}$$

and suppose we want to compute the Fourier-Legendre series of $f$, that is we want to write

$$f = \sum_{n}c_n P_n$$

with

$$c_n = \dfrac{\langle f,c_n\rangle }{|P_n|^2} = \dfrac{2}{2n+1}\int_{-1}^{1}f(x)P_n(x)dx.$$

This integral, though is quite complicated. Substituting $f$ we have

$$c_n = \dfrac{2}{2n+1}\int_0^1P_n(x)dx.$$

Now, I've been trying to compute the RHS but I still couldn't. I've tried writing

$$P_n(x) = \dfrac{1}{2^nn!}\dfrac{d^n}{dx^n}[(x^2-1)^n]$$

so that

$$c_n = \dfrac{2}{(2n+1)2^n n!} \left[\dfrac{d^{n-1}}{dx^{n-1}}[(x^2-1)^n]\right]_0^1,$$

but still I can't compute the boundary terms. In that case, is there an easier way to compute those coefficients? If not, how could I finish the calculation I've started?

Thanks very much in advance.

$\endgroup$
3
  • 1
    $\begingroup$ See formula (49) in mathworld.wolfram.com/LegendrePolynomial.html, where the answer is given. $\endgroup$ – Omran Kouba Aug 29 '15 at 17:20
  • $\begingroup$ That is the answer, but how does one computes it? $\endgroup$ – Gold Aug 29 '15 at 17:28
  • 1
    $\begingroup$ Anyway, the correct normalization is given by: $$ c_n=\frac{2n+1}{2}\int_{0}^{1}P_n(x)\,dx.$$ $\endgroup$ – Jack D'Aurizio Aug 29 '15 at 17:51
3
$\begingroup$

The generating function for Legendre polynomials is given by: $$ g(x,t)=\frac{1}{\sqrt{1+t^2-2xt}}=\sum_{n\geq 0}P_n(x)\, t^n\tag{1} $$ hence integrating both sides over $(0,1)$ with respect to $x$, $$\forall t\in(0,1),\qquad \frac{\sqrt{1+t^2}-1+t}{t}=\sum_{n\geq 0}\left(\int_{0}^{1}P_n(x)\,dx\right)t^n\tag{2} $$ and to compute our coefficients it is enough to compute the Taylor series of $\sqrt{1+t^2}$ around $t=0$. That proves formula $(49)$ linked by Omran Kouba and:

$$ f(x) = \frac{1}{2}+\sum_{m\geq 0}\frac{4m+3}{4m+4}\cdot\frac{(-1)^m}{4^m}\binom{2m}{m}\cdot P_{2m+1}(x).\tag{3} $$

Equality has to be intended as convergence in $L^2(-1,1)$. Gibbs phenomenon arises: it would be integersting to compute the magnitude of the overshoot near zero.

$\endgroup$

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.