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.

100 Polynomial Approximation of Differential Equations<br />

Figure 6.2.1 shows the behavior of the approximating polynomials Πw,nf, when w<br />

is the Legendre weight function, n = 4, 6, 8, 10, and f is defined by f(x) = |x| − 1<br />

2 .<br />

A slower rate of convergence is measured at the point x = 0, where f presents a singu-<br />

lar<strong>it</strong>y in the derivative. In figure 6.2.2, a typical discontinuous function, i.e., f(x) =<br />

−1/2 if x < 0<br />

1/2 if x ≥ 0 , is approximated by Πw,nf, w<strong>it</strong>h w ≡ 1 and n = 7, 9, 11, 13. In s<strong>it</strong>uations<br />

like this one, the Gibbs phenomenon develops (see courant and hilbert(1953),<br />

Vol.1). As a consequence, the oscillations of the approximating polynomials, in the<br />

neighborhood of the point x = 0, stay bounded but do not damp for increasing n.<br />

Next, we evaluate the rate of convergence of the lim<strong>it</strong> in (6.2.8). This depends on<br />

the regular<strong>it</strong>y of the function f. The smoother the function, the faster is the convergence.<br />

This problem has been widely studied by many authors. The classical approach relates<br />

the speed of convergence to the so called modulus of continu<strong>it</strong>y of f (see timan(1963)).<br />

A more recent approach expresses the smoothness of f in terms of <strong>it</strong>s norm in appro-<br />

priate Sobolev spaces. We prefer the latter, since <strong>it</strong> is more su<strong>it</strong>able for the analysis of<br />

differential equations. Results in this direction are given for instance in babuˇska, sz-<br />

abo and katz(1981) for the Legendre case. Extensions are considered in canuto and<br />

quarteroni(1980) and canuto and quarteroni(1982a). Other references are listed<br />

in canuto, hussaini, quarteroni and zang(1988). The next theorem summarizes<br />

the results for Jacobi type approximations. Thus, {λj = j(j + α + β + 1)}j∈N denotes<br />

the sequence of eigenvalues corresponding to the Sturm-Liouville problem (1.3.1). We<br />

recall that a and w are related by the formula a(x) = (1 − x 2 )w(x), x ∈ I =] − 1,1[.<br />

According to defin<strong>it</strong>ion (5.7.1), f ∈ H k w(I) when f and <strong>it</strong>s derivatives of order up to k<br />

belong to L 2 w(I).<br />

Theorem 6.2.4 - Let k ∈ N, then there exists a constant C > 0 such that, for any<br />

f ∈ H k w(I), one has<br />

(6.2.15) f − Πw,nf L 2 w (I) ≤ C<br />

k 1<br />

(1<br />

2 k/2<br />

− x )<br />

n<br />

dkf dxk <br />

<br />

L 2 w (I)<br />

, ∀n > k.

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

Saved successfully!

Ooh no, something went wrong!