Show that these polynomials are orthonormal in the $L^2$ scalar product

95 Views Asked by At

I try to show, that the polynomials $$\varphi_k(x) = \dfrac{k!}{(2k)!} \dfrac{d^k}{dx^k} (x^2-1)^k$$ are orthonormal regarding to the $L^2$-scalar product over $[-1,1]$.

I think i have to show, that $$\int_{-1}^{1}\varphi_k(x) \varphi_j(x) =\left\{\begin{matrix} 1 ,\ \text{if} \ j=k& \\ 0, \ \text{if} \ j \ne k& \end{matrix}\right.$$

Is this correct?

I tried to do this. but didn't succeed. I thought i have to do partial integration but, i don't get it. Can anyone help me?

Thank you very much.

1

There are 1 best solutions below

14
On

The polynomials are orthogonal but NOT orthoNORMAL

It seems that the polynomials as stated in your question are orthogonal but not orthonormal. First of all $\varphi_0\equiv 1$ has norm $\sqrt2\neq 1$. $k=1$ doesn't work either: $$ \varphi_1(x) = \frac12\frac\partial{\partial x}(x^2-1) = x $$ Then $$ \int_{-1}^1 \varphi_1(x) \varphi_1(x) \,{\rm d}x = \int_{-1}^1 x^2 \,{\rm d}x = \frac23 \neq 1 $$ and so on for all the other $k\in\mathbb N$.

I would rather state that $$\tag{Th} \varphi_k(x) ~=~ \frac{\sqrt{k+1/2}}{2^kk!} \frac{{\rm d}^k}{{\rm d}x^k} (x^2-1)^k $$ is the correct normalization.

A deeper look

First of all, for the sake of simplicity of notation, let us write $\partial_k$ in place of $\frac{\partial^k}{\partial x^k}$.

We want to normalize the functions $\psi_k=\partial_k(x^2-1)^k$ as $$ \varphi_k(x) = \alpha_k\psi_k(x) \qquad \exists\alpha_k\in\mathbb R \text{ to be determined} $$ so that the polynomials $\{\varphi_k\}_{k\in\mathbb N}$ are orthonormal.

Now, the interval $[-1,+1]$ is symmetric, and therefore the integral of odd functions over $[-1,+1]$ is $0$. Besides, if $f(x)$ is odd, then $f'(x)$ is even, and similarly if $f(x)$ is even, then $f'(x)$ is odd.

It will be useful to introduce the following notation: let $$ \sigma(f) = \begin{cases} +1 & \text{if $f$ is even}\\ -1 & \text{if $f$ is odd} \end{cases} $$ (the cases $f\equiv 0$ or $f$ neither even nor odd are not relevant, so let us not be fussy about it). Note that $$ \sigma(fg)=\sigma(f)\sigma(g) \quad\text{and}\quad \sigma(\partial_k f)=(-1)^k\sigma(f) $$ Now, $(x^2-1)^k$ is clearly an even function. For the above property, $$\tag{$\sigma$} \sigma(\psi_k) = (-1)^k \quad\text{and therefore}\quad \sigma(\psi_k\psi_j) = (-1)^{k+j} $$ It follows that if one between $k$ and $j$ is even and the other is odd, then $\langle\psi_k,\psi_j\rangle=0$.

It remains to check the cases $k,j$ both even and $k,j$ both odd. Assume wlog $1\leq j\leq k$ (the case when $j=0$ is trivial); integrating by parts, \begin{align} \langle\psi_k,\psi_j\rangle = & \int_{-1}^1 \partial_k(x^2-1)^k\,\, \partial_j(x^2-1)^j {\rm d}x \\ = & \partial_{k-1}(x^2-1)^k \partial_j(x^2-1)^j \Bigg|_{-1}^1 - \int_{-1}^1 \partial_{k-1}(x^2-1)^k\,\, \partial_{j+1}(x^2-1)^j {\rm d}x \end{align} Iterating $j$ times, we get \begin{align}\tag{1} \langle\psi_k,\psi_j\rangle = & \sum_{\ell=1}^{j} (-1)^{\ell+1} \partial_{k-\ell}(x^2-1)^k \partial_{j+\ell-1}(x^2-1)^j \Bigg|_{-1}^1 \\ & +\,\, (-1)^j \int_{-1}^1 \partial_{k-j}(x^2-1)^k\,\, \partial_{2j}(x^2-1)^j {\rm d}x \end{align} Note that $$ (x^2-1)^j = x^{2j} + \text{(polyn. of degree $2j-2$)} $$ therefore $$\tag{2} \partial_{2j}(x^2-1)^j=(2j)! $$ Besides, $(x^2-1)^k=(x-1)^k(x+1)^k$ has two roots $\pm1$ both of multiplicity $k$. Therefore $\pm1$ are both roots of the $\ell$-th derivatives $\partial_\ell(x^2-1)^k$, for $0\leq\ell<k$, that is, $$\tag{3} \partial_\ell\,(x^2-1)^k\big|_{x=\pm 1} ~=~ 0 \quad 0\leq\ell<k $$ It follows that every addend of the sum in $(1)$ is $0$.

From this fact and from $(2)$ we get $$\tag{4} \langle\psi_k,\psi_j\rangle = (-1)^j (2j)! \int_{-1}^1 \partial_{k-j}(x^2-1)^k {\rm d}x $$ If $k\neq j$, that is, if $k>j$ for our assumption, then $k-j-1\geq 0$ yielding $$ \langle\psi_k,\psi_j\rangle = (2j)! \partial_{k-j-1}(x^2-1)^k \Bigg|_{-1}^1 $$ Since $0\leq k-j-1<k$, from $(3)$ it follows that $$ \langle\psi_k,\psi_j\rangle = 0 \qquad j\neq k $$ So far we proved the orthogonality.

Let us focus on the case $j=k$. In light of $(4)$ (with $j=k$), $$ \langle\psi_k,\psi_j\rangle ~=~ (-1)^k(2k)!\int_{-1}^1(x^2-1)^k\,{\rm d}x $$ Iterating integration by parts, you can easily check that $$ \int_{-1}^1(x^2-1)^k = (-1)^k2^{2k+1}\frac{k!^2}{(2k+1)!} $$ so that $$ \langle\psi_k,\psi_j\rangle ~=~ (-1)^k2^{2k+1}\frac{k!^2}{(2k+1)!}(2k)!(-1)^k ~=~ 2^{2k}\frac{k!^2}{k+1/2} $$ Then, being $\varphi_k=\alpha_k\psi_k$ we have $$ \langle\varphi_k,\varphi_k\rangle ~=~ 2^{2k}\frac{k!^2}{k+1/2}\alpha_k^2 $$ In order to be orthonormal, the above quantity has to equal $1$, that is, $$ \alpha_k = \frac{\sqrt{k+1/2}}{2^kk!} $$ which proves my thesis (Th).