06.04.2013 Views

Hermite Interpolating Polynomials and Gauss-Legendre Quadrature

Hermite Interpolating Polynomials and Gauss-Legendre Quadrature

Hermite Interpolating Polynomials and Gauss-Legendre Quadrature

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

<strong>Hermite</strong> <strong>Interpolating</strong> <strong>Polynomials</strong><br />

<strong>and</strong> <strong>Gauss</strong>-<strong>Legendre</strong> <strong>Quadrature</strong><br />

M581 Supplemental Notes October 3, 2005<br />

Lagrange Interpolation.<br />

Given data discrete points {x1, . . . , xQ} in 1-D <strong>and</strong> given a function f that is defined<br />

at these points, the Lagrange interpolating polynomial is the unique polynomial L of degree<br />

Q − 1 which interpolates the function f at these points. By this we mean<br />

The st<strong>and</strong>ard form for L is<br />

L(xi) = f(xi), i = 1, . . . , Q. (1)<br />

Q<br />

L(x) = cjx<br />

j=1<br />

j−1 . (2)<br />

The coefficients of L can be determined by substituting (2) into (1) <strong>and</strong> solving the resulting<br />

Q×Q linear system for the cj. Alternatively, L one can compute the Lagrange representation<br />

where the Lagrange basis functions are defined by<br />

ℓi(x) =<br />

<br />

1≤k≤Q,k=i(x − xk)<br />

<br />

1≤k≤Q,k=i(xi − xk)<br />

Note that for each i, j = 1, . . . , Q,<br />

Q<br />

L(x) = f(xi) ℓi(x), (3)<br />

i=1<br />

= (x − x1) · · · (x − xi−1)(x − xi+1) · · · (x − xQ)<br />

(xi − x1) · · · (xi − xi−1)(xi − xi+1) · · · (xi − xQ)<br />

ℓi(xj) = δi,j,<br />

where δi,j denotes the Kronacker delta, whose value is 1 of i = j <strong>and</strong> 0 if i = j.<br />

<strong>Quadrature</strong> Rules Based on Lagrange Interpolation.<br />

By integrating (3) we obtain the Lagrange quadrature rule,<br />

where the quadrature weights are given by<br />

b<br />

Q<br />

f(x) dx ≈ f(xi) wi<br />

a<br />

i=1<br />

wi =<br />

b<br />

a<br />

(4)<br />

(5)<br />

ℓi(x) dx. (6)<br />

Example 1: The Lagrange form of the polynomial which interpolates f(x) at x1 = a <strong>and</strong><br />

x2 = b is<br />

x − b x − a<br />

L(x) = f(a) · + f(b) ·<br />

a − b b − a .<br />

1


Integrating L(x) over the interval a ≤ x ≤ b yields the trapazoidal rule<br />

b<br />

a<br />

f(x) dx ≈<br />

b<br />

b<br />

f(a) (x − b) dx/(a − b) + f(b) (x − a) dx/(b − a)<br />

a<br />

a<br />

= f(a) ·<br />

b − a<br />

2<br />

+ f(b) · b − a<br />

2 ,<br />

Here Q = 2, the quadrature points are x1 = a, x2 = b, <strong>and</strong> the quadrature weights are<br />

w1 = w2 = b−a<br />

2 .<br />

Error Analysis for Lagrange <strong>Quadrature</strong>.<br />

If L(x) is the polynomial of degree Q − 1 which interpolates f(x) at {xi} Q<br />

i=1 <strong>and</strong> if<br />

f ∈ C Q [a, b], then it can be shown that the pointwise approximation error is given by<br />

where<br />

f(x) − L(x) = 1 d<br />

Q!<br />

Qf dxQ (ξx) π(x), for some ξx ∈ (a, b), (7)<br />

Q<br />

π(x) ≡ (x − xk) = (x − x1)(x − x2) · · · (x − xQ). (8)<br />

k=1<br />

By integrating (7) we obtain the quadrature error bounds<br />

<br />

<br />

<br />

b<br />

Q<br />

<br />

<br />

<br />

f(x)dx − f(xq)wq<br />

<br />

<br />

a<br />

q=0 <br />

=<br />

<br />

1 <br />

b<br />

d<br />

<br />

Q! a<br />

Qf dxQ (ξx)<br />

≤<br />

<br />

<br />

<br />

π(x) dx<br />

<br />

1<br />

Q! max<br />

<br />

<br />

d<br />

<br />

a≤ξ≤b <br />

Q ≤<br />

<br />

f b<br />

<br />

(ξ) |π(x)| dx<br />

dxQ a<br />

(9)<br />

1<br />

Q! max<br />

<br />

<br />

d<br />

<br />

a≤ξ≤b <br />

Q <br />

f <br />

<br />

(ξ) <br />

dxQ max<br />

a≤x≤b |π(x)|<br />

b<br />

dx<br />

a<br />

≤ const (b − a) Q+1 . (10)<br />

Defining the interval length h = b − a, we obtain O(h Q+1 ) accuracy. The quadrature rule is<br />

exact (i.e., there is no quadrature error) for all polynomials of degree ≤ Q − 1 because the<br />

Qth derivative of these polynomials vanishes.<br />

Continuation of Example 1: For the trapazoidal rule, Q = 2, so the method is exact for<br />

polynomials of degree ≤ 1 <strong>and</strong> the accuracy is O(h 3 ) provided the function being integrated<br />

is twice continuously differentiable.<br />

Exercise 1. Derive Simpson’s rule, which is based on interpolation at points a, (a + b)/2,<br />

<strong>and</strong> b, <strong>and</strong> show that the accuracy is O(h 4 ), where h = b − a.<br />

<strong>Hermite</strong> Interpolation.<br />

Given a differentiable function f defined on discrete points {x1, . . . , xQ}, the <strong>Hermite</strong><br />

interpolating polynomial is the unique polynomial H of degree 2Q − 1 that interpolates f<br />

<strong>and</strong> its derivative,<br />

H(xi) = f(xi),<br />

dH<br />

dx (xi) = df<br />

dx (xi), i = 1, . . . , Q. (11)<br />

2


H(x) does have a st<strong>and</strong>ard representation which is analogous to eqn (2), but to derive<br />

quadrature rules it is convenient to use the <strong>Hermite</strong> representation<br />

where<br />

Q<br />

H(x) = f(xi) hi(x) +<br />

Q<br />

i=1<br />

i=1<br />

hi(x) =<br />

<br />

1 − 2 dℓi<br />

dx (xi)<br />

<br />

(x − xi)<br />

df<br />

dx (xi) hi(x), (12)<br />

ℓ 2 i (x) (13)<br />

hi(x) = (x − xi) ℓ 2 i (x) (14)<br />

with Lagrange basis functions ℓi defined as before in (4).<br />

Exercise 2. Verify that for each i, j = 1, . . . , Q,<br />

hi(xj) = δij,<br />

hi(xj) = 0,<br />

<strong>and</strong> hence, the interpolation conditions (11) hold.<br />

<strong>Gauss</strong>-<strong>Legendre</strong> <strong>Quadrature</strong>.<br />

By integrating (12) we obtain a <strong>Hermite</strong> quadrature rule<br />

b<br />

Q<br />

b<br />

f(x) dx ≈ f(xi) hi(x) dx +<br />

a<br />

i=1<br />

a<br />

<br />

wi<br />

dhi<br />

dx (xj) = 0, (15)<br />

dhi<br />

dx (xj) = δij, (16)<br />

Q<br />

i=1<br />

df<br />

dx (xi)<br />

b<br />

hi(x) dx<br />

a<br />

<br />

wi<br />

. (17)<br />

The key to obtaining a numerical integration scheme, called <strong>Gauss</strong>-<strong>Legendre</strong> quadrature,<br />

which is exact for all polynomials of degree ≤ degree 2Q − 1 is to pick the quadrature points<br />

to be (possibly transformed) roots of <strong>Legendre</strong> polynomials. This will cause the terms wi in<br />

the second sum in (17) to vanish.<br />

Let PQ denote the Qth <strong>Legendre</strong> polynomial. This has leading coefficient 1 <strong>and</strong> it is<br />

orthogonal on the interval [−1, 1] to <strong>Legendre</strong> polynomials Pi of lower degree,<br />

1<br />

PQ(x) Pi(x) dx = 0, i = 1, . . . , Q − 1.<br />

−1<br />

Since any polynomial can be expressed as a linear combination of <strong>Legendre</strong> polynomials,<br />

1<br />

PQ(x)ρ(x) dx for any polynomial ρ of degree < Q. (18)<br />

−1<br />

From (14) one can factor<br />

hi(x) = (x − xi)ℓi(x) ℓi(x) = π(x) × Cℓi(x), (19)<br />

where π is the degree Q polynomial defined in (8) <strong>and</strong> C is some constant. Note that the<br />

degree of ℓi is Q − 1. By selecting the xis to be the roots of the Qth <strong>Legendre</strong> polynomial<br />

PQ, we obtain π(x) = PQ(x), <strong>and</strong> from (18) <strong>and</strong> (19) we obtain<br />

wi =<br />

1<br />

−1<br />

hi(x) dx = 0.<br />

3


Consequently<br />

where<br />

1<br />

Q<br />

f(x) dx ≈ f(xi) wi<br />

−1<br />

i=1<br />

wi =<br />

1<br />

−1<br />

hi(x) dx<br />

whenever the quadrature points xi are taken to be the roots of the <strong>Legendre</strong> polynomial of<br />

degree Q.<br />

Exercise 3. Obtain a corresponding quadrature rule for an arbitrary interval [a, b], by<br />

deriving the linear transformation that maps [a, b] onto [−1, 1] <strong>and</strong> then applying the change<br />

of variables formula for integrals. Provide explicit formulas for the quadrature points <strong>and</strong><br />

weights for [a, b] in terms of the quadrature points <strong>and</strong> weights for [−1, 1].<br />

Error Analysis for <strong>Gauss</strong>-<strong>Legendre</strong> <strong>Quadrature</strong>.<br />

If f ∈ C 2Q [a, b] <strong>and</strong> we select <strong>Gauss</strong>-<strong>Legendre</strong> quadrature points for interpolation, then<br />

f(x) = H(x) + 1 d<br />

(2Q)!<br />

Qf dxQ (ξx) π 2 (x),<br />

where H is the degree 2Q − 1 <strong>Hermite</strong> interpolating polynomial, ξx is some point in the<br />

interval (a, b), <strong>and</strong> π is defined in (8). From the derivations above we obtain<br />

b<br />

a<br />

f(x) dx =<br />

=<br />

b<br />

a<br />

Q<br />

i=1<br />

H(x) dx + 1<br />

b d<br />

(2Q)! a<br />

Qf dxQ (ξx) π 2 (x) dx<br />

f(xi) wi + O(h 2Q+1 ),<br />

with h = b − a. The error term derivation follows as in the case of Lagrange quadrature<br />

error.<br />

The following table summarizes the first two <strong>Gauss</strong>ian quadrature rules for the interval<br />

[−1, 1].<br />

Q = 1 Q = 2<br />

w1 = 2 w1 = w2 = 1<br />

x1 = 0 x1, x2 = ∓ 1<br />

√ 3<br />

3 nd order accuracy 5 th order accuracy<br />

Exercise 4. Derive the <strong>Gauss</strong>ian quadrature rules in the table above. Then give corresponding<br />

rules (i.e., points <strong>and</strong> weights) for the interval [0, 1].<br />

4

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!