In a previous post, I discussed the Legendre polynomials. We now know that if is the n’th Legendre polynomial then:
1. If .
2. If has degree then .
3. has n simple roots in which are symmetric about 0.
4. is an odd function if is odd and an even function if is even.
5. form an orthogonal basis for the vector pace of polynomials of degree or less.
To make these notes complete, I’d like to review the Lagrange Interpolating polynomial:
Given a set of points where the are all distinct, the Lagrange polynomial through these points (called “nodes”) is defined to be:
It is easy to see that because the associated coefficient for the term is 1 and the other coefficients are zero. We call the coefficient polynomials
It can be shown by differentiation that if is a function that has continuous derivatives on some open interval containing and if is the Lagrange polynomial running through the nodes where all of the then the maximum of is bounded by where . So if the n+1’st derivative of is zero, then the interpolation is exact.
So, we introduce some notation: let be the Legendre polynomial, let be the roots, and , the j’th coefficient polynomial for the Lagrange polynomial through the nodes .
Now let be any polynomial of degree less than . First: the Lagrange polynomial through the nodes represents exactly as the error term is a multiple of the derivative of which is zero.
Therefore
Now here is the key: only depends on the Legendre polynomial being used and NOT on .
Hence we have a quadrature formula that is exact for all polynomials of degree less than .
But, thanks to the division algorithm, we can do even better. We show that this quadrature scheme is exact for all polynomials whose degree are less than . This is where the orthogonality of the Legendre polynomials will be used.
Let be a polynomial of degree less than . Then by the division algorithm, where BOTH and have degree less than .
Now note the following: ; that is where the fact that the are the roots of the Legendre polynomials matters.
Then if we integrate: . Now because of property 2 above, .
Now because the degree of is less than ,
This means: this quadrature scheme using the Legendre polynomials of degree m is EXACT for all polynomials of degree less than 2m.
Of course, one must know the roots and the corresponding but these have been tabulated.
Why this is exciting:
Let’s try this technique on: with, say, n = 5. We know that the exact answer is . But using the quadrature method with a spreadsheet:
The roots are in column A, rows 5-9. The weights (coefficients; the ) are in column B, rows 5-9. In column F we apply the function to the roots in column A; the products of the weights with are in column G, and then we do a simple sum to get a very accurate approximation to the integral.
The accuracy: this is due to the quadrature method being EXACT on the first 9 terms of the series for . The inexactness comes from the higher order Taylor terms.
Note: to convert , one uses the change of variable: to convert this to