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.

Time-Dependent Problems 233<br />

Thus, we can finally view pn(·,t) ∈ Pn, w<strong>it</strong>h pn(−1,t) = 0, ∀t ∈ [0,T], as the solution<br />

of the collocation scheme<br />

(10.3.17)<br />

∂pn<br />

∂t (ξ(n)<br />

i ,t) = −ζ ∂pn<br />

∂x (ξ(n)<br />

i ,t), 1 ≤ i ≤ n, ∀t ∈]0,T],<br />

where the ξ (n)<br />

i ’s are the zeroes of un (see section 3.1). By defining qn := pn/r, where<br />

r(x) := (1 + x), x ∈ [−1,1], we have qn(·,t) ∈ Pn−1, ∀t ∈ [0,T]. Using (10.3.17), qn<br />

also satisfies<br />

(10.3.18)<br />

∂qn<br />

∂t (ξ(n)<br />

i ,t) = −ζ<br />

<br />

∂qn<br />

∂x (ξ(n)<br />

(n)<br />

qn(ξ i ,t)<br />

i ,t) +<br />

r(ξ (n)<br />

<br />

, 1 ≤ i ≤ n, ∀t ∈]0,T].<br />

i )<br />

This is equivalent to a n × n linear system of differential equations in the unknowns<br />

qn(ξ (n)<br />

i , ·), 1 ≤ i ≤ n. The entries of the corresponding matrix are (see (7.2.5)):<br />

−ζ[d (1)<br />

ij + δij/(1 + ξ (n)<br />

i )], 1 ≤ i ≤ n, 1 ≤ j ≤ n.<br />

We can prove stabil<strong>it</strong>y when α > −1 and −1 < β ≤ 1. We first note that<br />

1−x<br />

1+x pn = −cnun + {lower degree terms}. Then, (9.3.16) yields<br />

(10.3.19)<br />

= −2ζ<br />

= −2ζ<br />

1<br />

1−x<br />

1+x<br />

−1<br />

1<br />

1−x<br />

1+x<br />

−1<br />

1<br />

d 1−x<br />

dt 1+x<br />

−1<br />

p2nw dx + c2 n(t)un2 <br />

w<br />

1<br />

1−x = 2 1+x<br />

−1<br />

∂pn<br />

∂x pnw dx + 2c ′ n(t)<br />

∂pn<br />

∂x pnw dx = ζ<br />

∂pn<br />

∂t pnw dx + 2c ′ n(t)cn(t)un 2 w<br />

1<br />

−1<br />

1<br />

p<br />

−1<br />

2 n<br />

1−x<br />

1+x unpnw dx + 2c ′ n(t)cn(t)un 2 w<br />

d<br />

<br />

1−x<br />

dx 1+xw <br />

dx ≤ 0, ∀t ∈]0,T].<br />

The last integral is negative by virtue of the restrictions on α and β. This shows that<br />

a certain weighted norm of pn(·,t) is bounded by the in<strong>it</strong>ial data for any t ∈]0,T].<br />

Further details and improvements are given in gottlieb and tadmor(1990).<br />

<br />

A proof of stabil<strong>it</strong>y in a norm which also controls the derivative ∂pn ∂x has been<br />

outlined in gottlieb and orszag(1977), section 8, for the Chebyshev case. The same<br />

proof can be extended to the other Jacobi cases, provided −1 < β ≤ 0 and α > −1.

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

Saved successfully!

Ooh no, something went wrong!