I Derive the orthonormality condition for Legendre polynomials

  • I
  • Thread starter Thread starter hmparticle9
  • Start date Start date
hmparticle9
Messages
151
Reaction score
26
I am stuck at the gate with this one.
$$\int_{-1}^{1} P_m P_l \text{ d}x = \frac{1}{2^m m!} \frac{1}{2^l l!} \int_{-1}^{1} \bigg( \frac{d}{dx}\bigg)^m(x^2-1)^m \bigg( \frac{d}{dx}\bigg)^l(x^2-1)^l \text{ d}x$$

I don't want to look at the solutions, I just need a nudge. The Hint is that we must use integration by parts, but I am not comfortable with applying integration by parts on this expression.
 
Last edited:
Physics news on Phys.org
I know them as
$$
p_n(x):=\dfrac{1}{(b-a)^n\,n!}\,\sqrt{\dfrac{2n+1}{b-a}}\,\dfrac{d^n}{dx^n}[(x-a)(x-b)]^n\; , \;n\in \mathbb{N}_0
$$
but that shouldn't make a difference.

I'm afraid you need to use integration by parts. You can assume ##m<l## and abbreviate the factors of ##P_m## by ##C_m## for simplicity. It can be formalized as an induction, but an "and so on" will do. Same with the second part, ##\|P_m\|=1,## where the boundary conditions play a significant role. The reduction of the degree is necessary.
 
It is not that I did not want to use integration by parts, it is just that I am uncomfortable and want to become comfortable with it. :)

In the first phase of integration by parts we select an expression we want to get the anti-derivative for. Nothing jumps out, whatever I do it seems to make it really complicated.
 
For example:
$$\int_{-1}^{1} \bigg( \frac{d}{dx}\bigg)^m(x^2-1)^m \bigg( \frac{d}{dx}\bigg)^l(x^2-1)^l \text{ d}x = \int_{-1}^{1} \bigg(\frac{d}{dx}\bigg)^{m-1} \frac{d}{dx}\bigg[ (x^2-1)^m \bigg( \frac{d}{dx}\bigg)^l(x^2-1)^l \bigg] dx$$
 
hmparticle9 said:
It is not that I did not want to use integration by parts, it is just that I am uncomfortable and want to become comfortable with it. :)

In the first phase of integration by parts we select an expression we want to get the anti-derivative for. Nothing jumps out, whatever I do it seems to make it really complicated.
That's why I prefer the definition of Legendre polynomials that I have given.

Use the polynomials ##x^n\;(n\in \mathbb{N}_0)## and show
$$
\int_a^b x^np_m(x)\, dx= C_m\int_a^b x^n\,\dfrac{d^m}{dx^m}[(x-a)(x-b)]^m\, dx=0.
$$
 
I am sorry for the inconvenience. Could we stick to the definition given in my text book? (my original post). I know I am sort of cheating already by asking for help.
 
hmparticle9 said:
I am sorry for the inconvenience. Could we stick to the definition given in my text book? (my original post). I know I am sort of cheating already by asking for help.
Yes, but my hint is still valid. Prove it for the summands of ##P_l## instead of the entire polynomial.
 
  • Like
Likes hmparticle9
Okay I think I have made a breakthrough. Since the derivative acts on everything to the right of it we can say
$$\int_{-1}^{1} x^n \bigg(\frac{d}{dx} \bigg)^m (x^2 -1)^m dx = \int_{-1}^{1} n x^{n-1} \bigg(\frac{d}{dx} \bigg)^{m-1} (x^2 -1)^m dx$$
is this correct? @fresh_42. I know it does not answer your hint, but am I going in the right direction?
 
hmparticle9 said:
Okay I think I have made a breakthrough. Since the derivative acts on everything to the right of it we can say
$$\int_{-1}^{1} x^n \bigg(\frac{d}{dx} \bigg)^m (x^2 -1)^m dx = \int_{-1}^{1} n x^{n-1} \bigg(\frac{d}{dx} \bigg)^{m-1} (x^2 -1)^m dx$$
is this correct? @fresh_42
I think you missed a sign. And I wouldn't resolve the derivative of ##x^n,## just keep it as ##D^1x^n## but I'm lazy. Anyway, the first term gets ultimately to ##D^{n+1}x^n##, which makes it zero.
 
  • #10
Are we also assuming that ##m \geq n##?

Also, this does not conclude the proof? I can see that ##(x^2-1)^m## is a sum of monomials, but there is the ##\big(\frac{d}{dx}\big)^m## term. Here

$$\int_{-1}^{1} P_m P_l \text{ d}x = \frac{1}{2^m m!} \frac{1}{2^l l!} \int_{-1}^{1} \bigg( \frac{d}{dx}\bigg)^m(x^2-1)^m \bigg( \frac{d}{dx}\bigg)^l(x^2-1)^l \text{ d}x$$
 
Last edited:
  • #11
hmparticle9 said:
Are we also assuming that ##m \geq n##?

Also, this does not conclude the proof? I can see that ##(x^2-1)^m## is a sum of monomials, but there is the ##\big(\frac{d}{dx}\big)^m## term.
We must show that ##\bigl\langle P_l,P_m \bigr\rangle = \delta_{lm},## so assuming that one is less than the other can be done without harm, the product is symmetric. The case ##l=m## has to be verified separately anyway. Assume ##\deg P_l(x)\leq l<m## and consider ##\bigl\langle x^l,P_m \bigr\rangle.##

Continue the integration by parts until the monomial ##x^k## vanishes, and don't bother what happens to the other term. That's why it is easier to carry the differential operator ##D=\dfrac{d}{dx}## than actually differentiating anything.
 
  • #12
My problem with what you are saying is that there are monomials ##x^k## with ##k \in [m,2m]## in the expansion of ##(x^2-1)^m## which wont vanish by taking ##m## derivatives
 
  • #13
Hold on, I have to rewrite my proof for your specific case and need to switch to a working editor.
 
  • #14
Say we start with ##n<m,## ##C_m=\dfrac{1}{2^m m!},## and ##D=\dfrac{d}{dx}## for short. Then we have with ##\displaystyle{\int_{-1}^1 uv'=[uv]_{-1}^1 -\int_{-1}^1 u'v } ##
\begin{align*}
\bigl\langle x^n , P_m \bigr\rangle &= C_m\int_{-1}^{1} x^n P_m \, dx=C_m \int_{-1}^{1} \underbrace{x^n}_{=u} \underbrace{D^m (x^2-1)^m}_{=v'}\\
&=\underbrace{C_m\left[x^nD^{m-1}(x^2-1)^m\right]_{-1}^{1}}_{=0}-C_m \int_{-1}^{1}\underbrace{Dx^n}_{=u} \underbrace{D^{m-1}(x^2-1)^m}_{=v'}\,dx\\
&=\underbrace{-C_m\left[Dx^nD^{m-2}(x^2-1)^m\right]_{-1}^{1}}_{=0}+C_m\int_{-1}^1 D^2x^nD^{m-2}(x^2-1)^m\,dx\\
&\vdots
\end{align*}
The terms on the left in brackets vanish, because ##D^{m-k}(x^2-1)^m## always contains a factor ##x+1## and a factor ##x-1## since ##x^2-1=(x+1)(x-1)## as long as ##m>0,## making it zero if we plug in the bounds. The rest is managing what happens if the degrees are zero. (I'm terribly bad with counting.)
 
Last edited:
  • #15
First and foremost. Thanks for persevering with me :)

I understand everything you said in the above post. We proceed as you have explained until we get to the point:
$$\langle x^n | P_m \rangle = C_m \int^1_{-1} D^{n+1}x^nD^{m-n}(x^2-1)^m dx = 0$$

Since ##D^{m}(x^2 -1)^m## is a sum of monomials we can write
$$D^{m}(x^2 -1)^m = \sum_{n = 0}^{m}\gamma_n x^n$$

Putting it together thus far:
$$\int_{-1}^{1}D^m(x^2 -1)^m D^{l}(x^2 - 1)^l dx = \sum_{n=0}^{m} \gamma_n \int_{-1}^{1} x^n D^l (x^2 - 1)^l dx$$
$$= \sum_{n=0}^{m} \gamma_n \int_{-1}^{1} D^{n+1}x^n D^{l-n} (x^2 - 1)^l dx = 0$$
 
Last edited:
  • #16
Yes, and the proof for ##\|P_n\|=1## is along similar lines.

We have an integrand of the form ##\left(D^n(x^2+1)^n\right)^2## and integration by parts allows us to shuffle the power from the left to the right so that we end up with ##\int (x^2-1)^n## when we reach ##D^{2n}(x^2-1)^n## multiplied by some constant. Then we do the same thing for ##(x^2-1)^n=(x-1)^n(x+1)^n## until we reach ##(x-1)^{2n}## only, which we finally have to integrate.
 
Last edited:
  • Like
Likes hmparticle9
  • #17
According to my book (in the question statement), ##||P_n||^2 = \frac{2}{2n+1}##. Okay cool I will think about it for a bit
 
  • #18
hmparticle9 said:
According to my book (in the question statement), ##||P_n||^2 = \frac{2}{2n+1}##. Okay cool I will think about it for a bit
Maybe, I haven't controlled the constants. It just may not be zero. That's the difference between orthogonal and orthonormal.
 
Last edited:
  • #19
I have started with
$$\int_{-1}^1 D^{2n}(x^2-1)^{2n} = \int_{-1}^1 D^{2n-1}(x^2-1)^{2n-1}2x$$
I don't think that is the correct way. I am not sure that I 100% understand #16. I could use integration by parts but it seems like we are just "saying the same thing". For instance I take the anti-derivative of ##D^{2n}## and take the derivative of ##(x^2 - 1)^{2n}##.
 
  • #20
I have
\begin{align*}
&\left|\left| D^n[(x-1)(x+1)]^n \right|\right|^2\\
=\,&\,\int_{-1}^1 \left[ D^n[(x-1)(x+1)]^n \right]^2\,dx\\
=\,&-\int_{-1}^1 \left[ D^{n-1}[(x-1)(x+1)]^n \right]\cdot \left[ D^{n+1}[(x-1)(x+1)]^n \right]\,dx\\
\vdots \\
\end{align*}
shifting ##u'v## to ##uv'.## The term ##[uv]_{-1}^1## is zero for the same reason as before.

You have only to differentiate ##D^{2n}(x^2-1)^n## and integrate ##\int (x-1)^n## (or ##\int (x+1)^{2n}).## The trick is to use ##\int D^n=D^{n-1}## and use the differential operator without actually differentiating anything as long as possible.
 
  • #21
Thanks. I follow your logic till I obtain:
$$||P_m||^2 = (-1)^m \frac{1}{(2^m m!)^2} \int_{-1}^1(x^2-1)^m D^{2m} (x^2-1)^m dx$$

I am working on ##D^{2m} (x^2-1)^m##, it must be a constant.
 
  • #22
hmparticle9 said:
Thanks. I follow your logic till I obtain:
$$||P_m||^2 = (-1)^m \frac{1}{(2^m m!)^2} \int_{-1}^1(x^2-1)^m D^{2m} (x^2-1)^m dx$$

I am working on ##D^{2m} (x^2-1)^m##, it must be a constant.
Yes. You can either differentiate it for a few values of ##m## until you see the pattern, or you can use the binomial formula
$$
(x^2+1)^m=\sum_{k=0}^{m}\binom{m}{k} x^{2k}
$$
to find the recursion.
 
  • #23
Okay, so I have ##D^{2m}(x^2-1)^m = 2^m(m!)^2##. I am having more trouble with the integral.
 
Last edited:
  • #24
hmparticle9 said:
Okay, so I have ##D^{2m}(x^2-1)^m = 2^m(m!)^2##
I have ##(2m)!## Let's see.

We only need to consider the highest term of ##(x^2-1)^m## at each differentiation. So we have to solve
\begin{align*}
D^{2m}(x^2-1)^m&=D^{2m}\left(\sum_{k=0}^{m}\binom{m}{k}x^{2k}\right)\\
&=\sum_{k=0}^{m}\binom{m}{k}D^{2m}x^{2k}\\
&=\binom{m}{m}D^{2m}x^{2m}\\
&=(2m)D^{2m-1}x^{2m-1}\\
&=(2m)(2m-1)D^{2m-2}x^{2m-2}\\
&\;\vdots\\
&=(2m)(2m-1)(2m-2)\cdot\ldots\cdot 2 D^{1}x^{1}\\
&=(2m)(2m-1)(2m-2)\cdot\ldots\cdot 2\cdot 1\\
&=(2m)!
\end{align*}
 
  • Like
Likes hmparticle9
  • #25
Okay. I said
$$D^{2m}[(x^2-1)^m]$$
$$=D^{2m-1} [m(x^2-1)^{m-1}2x]$$
$$=D^{2m-2} [m(m-1)(x^2-1)^{m-2}2^2x^2 + 2m(x^2-1)^{m-1}]$$
$$=D^{2m-3} [m(m-1)(m-2)(x^2-1)^{m-3}2^3 x^3 + ...]$$
 
  • #26
You are correct. I am wrong. I just peeked at the solution. This problem has given me a major headache. Can you see where I got the incorrect value above? It "appears" that my equation for the constant above is correct.
 
  • #27
Ok. Let's do it your way, although it bears a higher risk for mistakes.
\begin{align*}
D^{2m}(x^2-1)^m&=D^{2m}\left((x^2-1)^{m-1}(x^2-1)\right)\\
&=D^{2m-1}\left((m-1)(x^2-1)^{m-2}(x^2-1)+(x^2-1)^{m-1}2x\right)\\
&=D^{2m-1}\left((m-1+2)(x^2-1)^{m-1}x\right)\\
&\ldots
\end{align*}
You failed to carry on the second factor in the product rule. It's not ##x^2,## it is ##x^2-1.##
 
Last edited:
  • #28
If you want to peek at the solution, then you could use the file I use to peek to get my guidelines from:
The last attachment is the complete manual with all questions. The whole thing has certainly typos, so you should be careful, but it is a nice document to see whether there are hints or similar problems, because it can be searched in the browser. You can find the Legendre polynomials on pages 523-524. It uses another factor so that the basis is orthonormal and not only orthogonal, and an arbitrary interval ##[a,b]## instead of ##[-1,1]## but that's the only difference.
 
  • #29
Thanks :) I try my best not to give in, but sometimes...
I am sure that
##D^{2m}[(x^2-1)^m] = D^{2m-1}[m(x^2-1)^{m-1}2x] ## ???
 
  • #30
hmparticle9 said:
Thanks :) I try my best not to give in, but sometimes...
I am sure that
##D^{2m}[(x^2-1)^m] = D^{2m-1}[m(x^2-1)^{m-1}2x] ## ???
Yes, I made a mistake.

I tried to correct it using the product rule
\begin{align*}
D^{2m}(x^2-1)^m&=D^{2m-1}(m(x^2-1)^{m-1}2x)\\
&=D^{2m-2}\left(m(m-1)(x^2-1)^{m-2}2^2x^2+2m(x^2-1)^{m-1}\right)\\
&=D^{2m-2}\left((x^2-1)^{m-2}\left(2^2m(m-1)x^2+2mx^2-2m\right)\right)
\end{align*}
but this is becoming inconvenient quickly.
 
Last edited:
  • #31
I see :) I was just keeping the leading term. When I think about it I was being a bit reckless. I appreciate your initial solution, very nice. Thank you for your help!
 
Back
Top