4
$\begingroup$

Let $P_n$ be the orthogonal Legendre polynomial with a degree of $n$, meaning it satisfies the following recursive formula: $$(n+1)P_{n+1}(x)-(2n+1)xP_n(x)+nP_{n-1}(x)=0$$ where $P_0(x) = 1$ and $P_1(x) = x$. Let $w(x)=1$ be the weight function, and the interval $[-1,1]$. $I_n(f)=\sum_{k=1}^nw_k f(x_k)$ is the integration formula. For $n\geq1$ we define $$q_n(y) =\frac{P_{2n+1}(\sqrt{y})}{\sqrt{y}}.$$

Problem: prove that $q_n$ is a polynomial of degree $n$, determine the orthogonality relations, weight function (for the dot product $\int_{-1}^1w(x)f(x)g(x)dx$) and the interval where they form a set of orthogonal polynomials. Express the weights and nodes of the formula $I_n^q =\sum_{k=1}^n v_k f(y_k)$ using $w_k$ and $x_k$ from the integration formula $I_{2n+1}$.

My attempt:

I proved that $q_n$ is a polynomial of degree $n$ using induction and the fact that Legendre polynomial of degree $n$ is even/odd if $n$ is even/odd.

Next, I determined that they are orthogonal with the weight function being $x \mapsto \sqrt{x}$ and the interval $[0,1]$, by using the fact that Legendre polynomials are orthogonal and some substitutions.

Now, given that the nodes in the formula $I_n$ are roots of the n-th orthogonal polynomial we're using, when we look at the formula $I_{2n+1}$ we're looking for roots of $P_{2n+1}$. Seeing as it's a polynomial of degree $2n+1$ that is odd, it must have $n$ positive roots (let's denote them by $x_1,\ldots x_n$, $n$ negative roots and $0$. (there's also a theorem that states that the roots of these polynomials are all different)

Now, I think I should take $y_k = x_k^2$, $k = 1,\ldots,n$.

But I'm not sure how to express $v_k$ using $w_k$.

Any hints would be appreciated!

$\endgroup$
4
$\begingroup$

I don't see an elegant way to do this so we will apply brute force starting from this formula $$w_k=\frac{a_n}{a_{n-1}}\frac{\int_a^bw(x)p_{n-1 }(x)^2dx}{p_n^{\prime}(x_k)p_{n-1}(x_k)}$$ In our case we write it as $$v_k=\frac{b_m}{b_{m-1}}\frac{\int_0^1\sqrt y\,q_{m-1}(y)^2dy}{q_m^{\prime}(y_k)q_{m-1}(y_k)}$$ Now $b_m$ is the leading coefficient of $q_m(y)=\frac{P_{2m+1}(\sqrt y)}{\sqrt y}$ so $b_m=a_{2m+1}$, the leading coefficient of $P_{2m+1}(x)$. This is a good start! Now, $b_{m-1}=a_{2m-1}$: not quite what we want but hopefully it will work itself out. We have $$q_m^{\prime}(y_k)=\frac{P_{2m+1}^{\prime}(\sqrt{y_k})}{\sqrt{y_k}}\frac1{2\sqrt{y_k}}-\frac12\frac{P_{2m+1}(y_k)}{y_k^{3/2}}=\frac{P_{2m+1}^{\prime}(x_k)}{2x_k^2}$$ $$q_{m-1}(y_k)=\frac{P_{2m-1}(\sqrt{y_k})}{\sqrt{y_k}}=\frac{P_{2m-1}(x_k)}{x_k}$$ We can adjust this equation by taking $n=2m$ and $x=x_k$ the the three term recurrence relation to get $$(2m+1)P_{2m+1}(x_k)-(4m+1)x_kP_{2m}(x_k)+2mP_{2m-1}(x_k)=0$$ Solve for $P_{2m-1}(x_k)$ and apply to the previous equation to get $$q_{m-1}(y_k)=\frac{4m+1}{2m}P_{2m}(x_k)$$ Which is the form we want. Now, $$\begin{align}\int_0^1\sqrt y\,q_{m-1}(y)^2dy&=\int_0^1\sqrt y\,\frac{P_{2m-1}(\sqrt y)^2}{y}dy=\int_0^1x\frac{P_{2m-1}(x)^2}{x^2}\cdot2x\,dx\\ &=2\int_0^1P_{2m-1}(x)^2dx=2\int_0^{-1}P_{2m-1}(-u)^2(-du)\\ &=2\int_{-1}^0\left(-P_{2m-1}(u)\right)^2du=\int_{-1}^1P_{2m-1}(x)^2dx\end{align}$$ We go bck to the three term recurrence relation with $n=2m$, multiply by $P_{2m-1}(x)$ and integerate from $-1$ to $1$ with respect to $x$ to get $$(2m+1)\int_{-1}^1P_{2m+1}(x)P_{2m-1}(x)dx-(4m+1)\int_{-1}^1P_{2m}(x)\left[xP_{2m-1}(x)\right]dx+2m\int_{-1}^1P_{2m-1}(x)^2dx=0$$ The first integral above is $0$ by the orthogonality of the Legendre polynomials and $$xP_{2m-1}(x)=\frac{a_{2m-1}}{a_{2m}}P_{2m}(x)+R_{2m-1}(x)$$ By the division algorithm where $R_{2m-1}(x)$ is a polynomial in $x$ of degree at most $2m-1$. Thus by the orthogonality of the Legendre polynomials, $$\begin{align}\int_{-1}^1P_{2m}(x)\left[xP_{2m-1}(x)\right]dx&=\int_{-1}^1P_{2m}(x)\left[\frac{a_{2m-1}}{a_{2m}}P_{2m}(x)+R_{2m-1}(x)\right]dx\\ &=\frac{a_{2m-1}}{a_{2m}}\int_{-1}^1P_{2m}(x)^2dx\end{align}$$ So we have arrived at $$-(4m+1)\frac{a_{2m-1}}{a_{2m}}\int_{-1}^1P_{2m}(x)^2dx+2m\int_{-1}^1P_{2m-1}(x)^2dx=0$$ And we can use this result to achieve $$\int_0^1\sqrt y\,q_{m-1}(y)^2dy=\frac{(4m+1)a_{2m-1}}{2ma_{2m}}\int_{-1}^1P_{2m}(x)^2dx$$ So now we have $$\begin{align}v_k&=\frac{a_{2m+1}}{a_{2m-1}}\frac{\frac{(4m+1)a_{2m-1}}{2ma_{2m}}\int_{-1}^1P_{2m}(x)^2dx}{\frac{P_{2m+1}^{\prime}(x_k)}{2x_k^2}\frac{(4m+1)P_{2m}(x_k)}{2m}}\\ &=2x_k^2\frac{a_{2m+1}}{a_{2m}}\frac{\int_{-1}^1P_{2m+1}(x)^2dx}{P_{2m+1}^{\prime}(x_k)P_{2m}(x_k)}\\ &=2x_k^2w_k\end{align}$$ Mmm... now that we have an answer, it looks like an elegant proof should be possible, but I don't see it just now. It's getting late so I leave the reader with the above brute force approach.

EDIT: It's late but not that late. We start with the most horrible possible way of computing the weights $v_k$: $$I_j=\int_0^1\sqrt y\,y^jdy=\sum_{k=1}^mv_ky_k^j=\sum_{k=1}^mv_kx_k^{2j}$$ For $0\le j\le m-1$, where we have applied our Gauss-Jacobi quadrature formula, exact for polynomials of degree at most $2m-1$, to find $m$ linear equations for its $m$ weights. Since the matrix of coefficients is a Vandermonde matrix, the weights are uniquely determined by this system.

Now we transform variables first to get $$\begin{align}I_j&=\int_0^1\sqrt y\,y^jdy=\int_0^1x\cdot x^{2j}\cdot 2x\,dx=\int_{-1}^1x^{2j+2}dx\\ &=2\sum_{k=1}^mw_kx_k^{2j+2}=\sum_{k=1}^m(2x_k^2w_k)x_k^{2j}\end{align}$$ For $0\le j\le m-1$. We showed in our complicated proof how to fold the integral over the origin to change the interval to $[-1,1]$ and the Gauss-Legendre quadrature formula was again exact for polynomials of degree at most $4m+1$. We know already that the zeros of the Legendre polynomial of degree $2m+1$ are symmetrically disposed about the origin and that the weight corresponding to $-x_k$ is the same as that corresponding to $x_k$, due to the symmetry of the weight function and interval in Gauss-Legendre quadrature about the origin. Also the weight for the zero node doesn't enter in because it always gets multiplied by zero in every equation.

Thus we see that $v_k$ and $2x_k^2w_k$ satisfy the same nonsingular system of $m$ equations in $m$ unknowns, so they are the same.

$\endgroup$
  • $\begingroup$ That's pretty awesome. Thanks a lot! $\endgroup$ – Collapse Jun 10 at 14:58

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.