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 135<br />

We conclude this section w<strong>it</strong>h a few add<strong>it</strong>ional remarks. Given the function f ∈<br />

C1 ( Ī), we can evaluate the quant<strong>it</strong>ies f(ξ(n) j ), 1 ≤ j ≤ n. After applying the matrix<br />

Dn, we obtain the vector {(Iw,nf) ′ (ξ (n)<br />

i )}1≤i≤n. For n sufficiently large, (Iw,nf) ′ is a<br />

fairly good approximation of f ′ (see (6.6.8)).<br />

Other techniques, such as fin<strong>it</strong>e-differences, can be used to compute numerically the<br />

derivative of f. Although they generate derivative matrices w<strong>it</strong>h qu<strong>it</strong>e good structure<br />

(for instance, they are banded w<strong>it</strong>h a narrow bandwidth), the results are in general very<br />

poor when compared w<strong>it</strong>h those obtained by the procedure described above. The reason<br />

is that fin<strong>it</strong>e-differences are local methods. To compute the derivative at a given point,<br />

these methods use information pertaining to a small neighborhood of the point. In<br />

contrast, spectral methods are global methods. All n nodes in the domain I are applied<br />

in the computation. For example, when f is a polynomial in Pn−1, the values at the<br />

nodes are representative of the whole function in the domain Ī, and the exact derivative<br />

is obtained. The particular distribution of the nodes optimizes the process for a general<br />

f, and the accuracy increases rapidly w<strong>it</strong>h the regular<strong>it</strong>y of the function. The reader<br />

may find other persuasive arguments supporting these considerations in boyd(1989),<br />

p.11.<br />

7.3 Boundary cond<strong>it</strong>ions in the frequency space<br />

Clearly, the procedure of evaluating the derivative q := p ′ ∈ Pn−1 of a polynomial<br />

p ∈ Pn can be inverted except to w<strong>it</strong>hin an add<strong>it</strong>ive constant. Actually, the derivative<br />

formulas of section 7.1 do not take into consideration the coefficient c0. We need an<br />

extra cond<strong>it</strong>ion. The most popular approach is to assume that p satisfies p(ξ) = σ ∈ R,<br />

where ξ ∈ Ī. This is a boundary cond<strong>it</strong>ion when ξ ∈ ∂I, i.e., ξ is an element of the<br />

boundary of Ī. In this case, we can recover the coefficients ci, 1 ≤ i ≤ n, in terms<br />

of the coefficients dj, 0 ≤ j ≤ n − 1, of the expansion q = n−1<br />

j=0 djuj. Finally, c0 is

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

Saved successfully!

Ooh no, something went wrong!