11.08.2013 Views

Untitled - Cdm.unimo.it

Untitled - Cdm.unimo.it

Untitled - Cdm.unimo.it

SHOW MORE
SHOW LESS

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

Derivative Matrices 137<br />

Next, consider second-order derivatives. Now, two add<strong>it</strong>ional cond<strong>it</strong>ions are re-<br />

quired to select a unique prim<strong>it</strong>ive function. Since there are several possibil<strong>it</strong>ies, we<br />

only consider one example. Let A, B, σ1, σ2, be real constants and q ∈ Pn−2, n ≥ 2.<br />

In the ultraspherical case, one is concerned w<strong>it</strong>h finding p ∈ Pn such that<br />

⎧<br />

(7.3.5)<br />

−p ′′ + Πw,n−2(Ap ′ + Bp) = q,<br />

⎪⎨<br />

p(−1) = σ1,<br />

⎪⎩<br />

p(1) = σ2.<br />

For instance, when n = 5, problem (7.3.5) leads to a linear system of the form<br />

(7.3.6)<br />

⎡<br />

B ∗ ∗ ∗ ∗<br />

⎤⎡<br />

∗<br />

⎢ 0<br />

⎢ 0<br />

⎢ 0<br />

⎣<br />

∗<br />

B<br />

0<br />

0<br />

∗<br />

∗<br />

B<br />

0<br />

∗<br />

∗<br />

∗<br />

B<br />

∗<br />

∗<br />

∗<br />

∗<br />

∗<br />

∗ ⎥⎢<br />

⎥⎢<br />

∗ ⎥⎢<br />

⎥⎢<br />

∗ ⎥⎢<br />

⎦⎣<br />

∗<br />

∗ ∗ ∗ ∗ ∗ ∗<br />

where the dk’s are the Fourier coefficients of q. Again, this can be solved w<strong>it</strong>h a direct<br />

procedure w<strong>it</strong>h a cost proportional to n 2 .<br />

From (1.3.2) and (1.3.3), in the ultraspherical case we have uk(1) = (−1) k uk(−1),<br />

k ∈ N. This implies<br />

(7.3.7) 2<br />

n<br />

k=0<br />

k even<br />

ck uk(1) = σ1 + σ2, 2<br />

c0<br />

c1<br />

c2<br />

c3<br />

c4<br />

c5<br />

⎤<br />

⎥<br />

⎦<br />

n<br />

=<br />

k=1<br />

k odd<br />

⎡<br />

⎢<br />

⎣<br />

d0<br />

d1<br />

d2<br />

d3<br />

σ1<br />

σ2<br />

⎤<br />

⎥<br />

⎥,<br />

⎥<br />

⎦<br />

ck uk(1) = σ1 − σ2.<br />

Therefore, when A = 0, (7.3.6) can be decoupled into the subsystems<br />

(7.3.8)<br />

⎡<br />

B<br />

⎣ 0<br />

∗<br />

B<br />

⎤⎡<br />

∗<br />

∗⎦<br />

⎣<br />

∗ ∗ ∗<br />

c0<br />

⎤ ⎡<br />

c2 ⎦ = ⎣ d0<br />

d2<br />

⎤<br />

⎦,<br />

⎡<br />

B<br />

⎣ 0<br />

∗<br />

B<br />

⎤⎡<br />

∗<br />

∗ ⎦⎣<br />

∗ ∗ ∗<br />

c1<br />

⎤ ⎡<br />

c3 ⎦ =<br />

c4<br />

σ1 + σ2<br />

c5<br />

⎣ d1<br />

d3<br />

σ1 − σ2<br />

This is because c (2)<br />

i , 0 ≤ i ≤ n − 2, is a linear combination of cj, i + 2 ≤ j ≤ n, where<br />

i+j is even (see for instance (7.1.9) and (7.1.10)). Following this idea, a fast algor<strong>it</strong>hm<br />

for the solution of (7.3.5) is presented in clenshaw(1957), for the Chebyshev case.<br />

Further hints are given in canuto, hussaini, quarteroni and zang(1988), p.129,<br />

and boyd(1989), p.380.<br />

⎤<br />

⎦.

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

Saved successfully!

Ooh no, something went wrong!