HW #2 solutions

Transcription

HW #2 solutions
Math 552
Scientific Computing II
Spring 2015
SOLUTIONS: Homework Set 2
1. Apply the Gram-Schmidt orthogonalization process to the polynomials 1, x, x2 , . . .
to find the first 5 Legendre polynomials.
ANS: For a general weight
R 1function w(x) ≥ 0, the inner product of two functions
f (x) and g(x) is hf, giw = −1 f (x)g(x)w(x) dx.
The weight function w(x) = 1 generates the Legendre polynomials, or
Z 1
hf, gi =
f (x)g(x) dx.
−1
Applying the Gram-Schmidt process to {1, x, x2 , . . .} produces a set of mutually
orthogonal polynomials {L0 (x), L1 (x), L2 (x), . . .}, the first 5 of which are
L0 (x) = 1
L1 (x) = x −
0
hx, L0 i
L0 (x) = x − L0 (x)
hL0 , L0 i
2
= x
hx2 , L0 i
hx2 , L1 i
L0 (x) −
L1 (x)
hL0 , L0 i
hL1 , L1 i
2/3
0
= x2 −
L0 (x) −
L1 (x)
2
2/3
= x2 − 1/3
L2 (x) = x2 −
hx3 , L0 i
hx3 , L1 i
hx3 , L2 i
L0 (x) −
L1 (x) −
L2 (x)
hL0 , L0 i
hL1 , L1 i
hL2 , L2 i
0
2/5
0
= x3 − L0 (x) −
L1 (x) −
L2 (x)
2
2/3
8/45
= x3 − 0 ∗ 1 − 3/5 ∗ x − 0 ∗ (x2 − 1/3)
= x3 − 3/5x
L3 (x) = x3 −
hx4 , L0 i
hx4 , L1 i
hx4 , L2 i
hx4 , L3 i
L0 (x) −
L1 (x) −
L2 (x) −
L3 (x)
hL0 , L0 i
hL1 , L1 i
hL2 , L2 i
hL3 , L3 i
2/5
0
16/105
0
= x4 −
L0 (x) −
L1 (x) −
L2 (x) −
L3 (x)
2
2/3
8/45
8/175
= x4 − 1/5 ∗ 1 − 0 ∗ x − 6/7 ∗ (x2 − 1/3) − 0 ∗ (x3 − 3/5x)
= x4 − 6/7x2 + 3/35
L4 (x) = x4 −
hx5 , L0 i
hx5 , L1 i
hx5 , L2 i
hx5 , L3 i
L0 (x) −
L1 (x) −
L2 (x) −
L3 (x) −
hL0 , L0 i
hL1 , L1 i
hL2 , L2 i
hL3 , L3 i
hx5 , L4 i
L4 (x)
hL4 , L4 i
0
2/7
0
16/315
0
= x5 − L0 (x) −
L1 (x) −
L2 (x) −
L3 (x) −
L4 (x)
2
2/3
8/45
8/175
hL4 , L4 i
= x5 − 0 ∗ 1 − 3/7 ∗ x − 0 ∗ (x2 − 1/3) − 70/63 ∗ (x3 − 3/5x) − 0 ∗ (x4 − 6/7x2 + 3/35)
= x5 − 70/63x3 + 5/21x
L5 (x) = x5 −
Of course, these may be rescaled to produce, for example, an orthonormal basis in
addition to an orthogonal basis.
2. (5pts-3,1,1)The Chebyshev polynomials are defined for x ∈ [−1, 1] by Tn (x) =
cos(nθ), x = cos θ.
a) Derive the 3-term recurrence relation,
Tn+1 (x) = 2xTn (x) − Tn−1 (x) .
b) Given T0 (x) = 1 and T1 (x) = x, use the recurrence relation to find T2 (x) and
T3 (x).
c) What are the roots of T3 (x)?
ANS:
a)
2xTn (x) − Tn−1 (x) =
=
=
=
=
=
2 cos (θ) cos (nθ) − cos ((n − 1)θ)
2 cos (θ) cos (nθ) − cos (nθ − θ)
2 cos (θ) cos (nθ) − (cos (nθ) cos (θ) + sin (nθ) sin (θ))
cos (θ) cos (nθ) − sin (nθ) sin (θ)
cos ((n + 1)θ)
Tn+1 (x)
b) T2 (x) = 2xT1 (x) − T0 (x) = 2xx − 1 = 2x2 − 1 and T3 (x) = 2xT2 (x) − T1 (x) =
2x(2x2 − 1) − x = 4x3 − 3x.
p
c) The roots of T3 (x) = 4x3 − 3x = x(4x2 − 3) are x = 0, ± 3/4.
3. Consider the integral,
Z
1
2
xe−x dx .
0
Use the 4-point Gaussian-Legendre quadrature rule to approximate the integral
(after changing variables to obtain an integral over [-1,1]).
The points and weights are:
x1
x2
x3
x4
= −0.861136311594053
= −0.339981043584856
= −x2
= −x1
c1
c2
c3
c4
= 0.347854845137454
= 0.652145154862546
= c2
= c1
What is the absolute error in the approximation?
ANS: First, let’s determine the exact answer. Letting u = x2 then du = 2x dx
giving
1
Z
Z 1
1 −u e−1
1 1 −u
−x2
e du = − e =
≈ 0.316060279414279
xe
dx =
2 0
2
2e
0
0
we must make a change of variable in order to
the Rintegral as one over
R 1 express
t+1 2
1
−x2
[−1, 1]. Letting t = 2x − 1 ⇒ dt = 2dx, so 0 xe
dx = −1 t+1
e−( 2 ) 21 dt =
2
R
t+1 2
1 1
(t + 1)e−( 2 ) dt. Rewriting the integral again in therm of x, we have
4 −1
Z
1
−1
x+1 2
1
(x + 1)e−( 2 ) dx
4
Here is a MATLAB script to perform the needed computations
x = [-0.861136311594053 -0.339981043584856 ...
0.339981043584856 0.861136311594053]’;
c = [ 0.347854845137454 0.652145154862546 ...
0.652145154862546 0.347854845137454]’;
f = ’0.25*(x+1).*exp(-((x+1)/2).^2)’;
gauss_approx = dot(c,eval(f))
exact = (exp(1)-1)/(2*exp(1))
error = abs(exact-gauss_approx)
whose output is
gauss_approx
0.316058997487186
exact
0.316060279414279
error
1.28192709325514e-06
4. (5pts) We proved in class that for n ≥ 2 that Gauss-Legendre Quadrature rule
Z
1
f (x) dx ≈
(∗)
−1
n
X
cj f (xj )
j=1
is exact for any polynomial of degree ≤ 2n − 1, where {xj }nj=1 are the n distinct
roots of the Legendre polynomial pn (x) of degree n, and {cj }nj=1 the corresponding
weights. Show that indeed this is the best we can expect by proving that (∗) is
not exact for
n
Y
f (x) =
(x − xj )2 ,
j=1
a polynomial of degree 2n. (Hint: Compute the approximation for any n. Why
can’t it be correct? There are no numerics in this problem, only theory!)
ANS: Note that by construction f (xj ) = 0, indeed xj is a double root of f (x) for
1 ≤ j ≤ n. Thus,
n
X
cj f (xj ) = 0 .
j=1
Clearly f (x) is a polynomial – hence continuous – such that f (x) R≥ 0 on [−1, 1]
1
being the product of squared term, but f 6≡ 0 over the interval so −1 f (x) dx > 0
while the quadrature rule gives 0. We then have a 2n degree polynomial for which
the quadrature rule is not exact. Thus it is not exact for all polynomials up to and
including those of degree 2n.