Numerical Methods Contents - SAM
Numerical Methods Contents - SAM
Numerical Methods Contents - SAM
Create successful ePaper yourself
Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.
If we used the residual based termination criterion<br />
∥<br />
∥F(x (k) ) ∥ ≤ τ ,<br />
1.5<br />
1<br />
Divergenz des Newtonverfahrens, f(x) = arctan x<br />
5<br />
4.5<br />
4<br />
3.5<br />
then the resulting algorithm would not be affine invariant, because for F(x) = 0 and AF(x) = 0,<br />
A ∈ R n,n regular, the Newton iteration might terminate with different iterates.<br />
0.5<br />
a<br />
3<br />
2.5<br />
0<br />
x (k−1)<br />
x (k)<br />
2<br />
1.5<br />
Terminology:<br />
∆¯x (k) := DF(x (k−1) ) −1 F(x (k) ) ˆ= simplified Newton correction<br />
Reuse of LU-factorization (→ Rem. 2.2.6) of DF(x (k−1) ) ➤<br />
∆¯x (k) available<br />
with O(n 2 ) operations<br />
−0.5<br />
−1<br />
−1.5<br />
−6 −4 −2 0 2 4 6<br />
1<br />
0.5<br />
0<br />
x (k+1) −15 −10 −5 0 5 10 15<br />
red zone = {x (0) ∈ R, x (k) → 0}<br />
x<br />
Summary: The Newton Method<br />
converges asymptotically very fast: doubling of number of significant digits in each step<br />
often a very small region of convergence, which requires an initial guess<br />
rather close to the solution.<br />
Idea:<br />
we observe “overshooting” of Newton correction<br />
damping of Newton correction:<br />
With λ (k) > 0: x (k+1) := x (k) − λ (k) DF(x (k) ) −1 F(x (k) ) . (3.4.5)<br />
Ôº¿½ ¿º<br />
Terminology:<br />
λ (k) = damping factor<br />
Ôº¿½ ¿º<br />
3.4.4 Damped Newton method<br />
Example 3.4.11 (Local convergence of Newton’s<br />
method).<br />
F(x) = xe x − 1 ⇒ F ′ (−1) = 0<br />
2<br />
1.5<br />
1<br />
0.5<br />
x ↦→ xe x − 1<br />
Choice of damping factor: affine invariant natural monotonicity test [11, Ch. 3]<br />
where<br />
“maximal” λ (k) ∥<br />
> 0: ∥∆x(λ (k) ) ∥ ≤ (1 − λ(k) ∥ ∥ ∥∥∆x<br />
2 ) (k) ∥2 (3.4.6)<br />
∆x (k) := DF(x (k) ) −1 F(x (k) ) → current Newton correction ,<br />
∆x(λ (k) ) := DF(x (k) ) −1 F(x (k) + λ (k) ∆x (k) ) → tentative simplified Newton correction .<br />
Heuristics: convergence ⇔ size of Newton correction decreases<br />
x (0) < −1 ⇒ x (k) → −∞<br />
x (0) > −1 ⇒ x (k) → x ∗<br />
0<br />
−0.5<br />
✸<br />
−1<br />
Fig. 39<br />
−1.5<br />
−3 −2.5 −2 −1.5 −1 −0.5 0 0.5 1<br />
Example 3.4.12 (Region of convergence of Newton<br />
method).<br />
2<br />
1.5<br />
1<br />
F(x) = arctan(ax) , a > 0,x ∈ R<br />
with zero x ∗ = 0 .<br />
✸<br />
arctan(ax)<br />
0.5<br />
0<br />
−0.5<br />
−1<br />
−1.5<br />
a=10<br />
a=1<br />
a=0.3<br />
Fig. 40<br />
−2<br />
−15 −10 −5 0 5 10 15<br />
x<br />
Ôº¿½ ¿º<br />
Ôº¿¾¼ ¿º