11.08.2013 Views

Untitled - Cdm.unimo.it

Untitled - Cdm.unimo.it

Untitled - Cdm.unimo.it

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

216 Polynomial Approximation of Differential Equations<br />

(9.8.3) −[F(x,y1) − F(x,y2)](y1 − y2) ≥ ǫ (y1 − y2) 2 , ∀x ∈ [−1,1], ∀y1,y2 ∈ R.<br />

For a regular F this is equivalent to requiring − ∂F<br />

∂y<br />

≥ ǫ in [−1,1] × R.<br />

It is clear that (9.8.2) is not equivalent to a linear system and the unknown vector<br />

¯v := {pn(η (n)<br />

i )}0≤i≤n has to be computed by an <strong>it</strong>erative approach. We can adopt the<br />

Richardson method (see section 7.6) by defining the sequence ¯v (k) ≡ (v (k)<br />

0 , · · · ,v(k) n ),<br />

k ∈ N, such that<br />

(9.8.4) ¯v (k+1)<br />

:= (I − θD)¯v (k) + θ ¯w (k) , k ∈ N,<br />

where ¯v (0) is the in<strong>it</strong>ial guess. In (9.8.4), D is the matrix corresponding to the system<br />

(7.4.1) and ¯w (k) ≡ σ,F(η (n)<br />

1 ,v (k)<br />

1<br />

), · · · ,F(η(n) n ,v (k)<br />

n ) , k ∈ N. If the <strong>it</strong>erates converge<br />

for some θ > 0, then one gets limk→+∞ v (k)<br />

0 = σ, limk→+∞ v (k) (n)<br />

i = pn(η i ), 1 ≤ i ≤ n.<br />

A faster convergence is realized by using precond<strong>it</strong>ioners (see sections 8.3 and 8.5).<br />

Second-order nonlinear problems can be also considered. An example is given by<br />

the equation<br />

(9.8.5)<br />

⎧<br />

⎨<br />

−U ′′ (x) + F(x,U(x),U ′ (x)) = 0 x ∈] − 1,1[,<br />

⎩<br />

U(−1) = σ1, U(1) = σ2,<br />

where F : [−1,1] × R 2 → R, σ1 ∈ R, σ2 ∈ R, are given.<br />

Spectral type approximations of (9.8.5) are studied in maday and quarteroni(1982)<br />

in the case F(x,y,z) := 1<br />

ǫ yz − f(x), σ1 = σ2 = 0, where ǫ > 0 and the function<br />

f : [−1,1] → R is given. A general trick to analyze the solution of problem (9.8.5) and<br />

<strong>it</strong>s approximation is to wr<strong>it</strong>e the equation in the form U = LF(x,U,U ′ ), where L is<br />

the linear operator which associates to any function f the solution of problem (9.1.4)<br />

w<strong>it</strong>h σ1 = σ2 = 0. Then, the investigation proceeds by using fixed-point theorems and<br />

the compactness of the operator L. More details can be found in brezzi, rappaz and<br />

raviart(1980). We further examine the above example in section 10.4.

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

Saved successfully!

Ooh no, something went wrong!