220 likes | 603 Views
2. Overview. In previous talks about inverse problems:well-posednessworst-case errorsregularization strategies . 3. Overview. In this talk:IntroductionProjection methodsGalerkin methodsSymm's integral equationConclusions. 4. Differentiation:Inverse problem ? integration:. Example:
E N D
1. Regularization by Galerkin Methods Hans Groot
2. 2 Overview In previous talks about inverse problems:
well-posedness
worst-case errors
regularization strategies
3. 3 Overview
4. 4 Example: differentiation
5. 5 Example: differentiation
6. 6 Example: differentiation
7. 7 Inverse Problems Let:
X, Y Hilbert spaces
K : X ? Y linear, bounded, one-to-one mapping
Inverse Problem:
Given y ? Y, solve Kx = y for x ? X
8. 8 Projection Methods
9. 9 Linear System of Equations
10. 10 Regularization by Disretization
11. 11 Theorem
12. 12 Galerkin Method Galerkin method:
for all zn ? Yn
Substitute
for
13. 13 Error Estimates Approximate right-hand side y? ? Y, ?y - y? ? = ? :
Equation:
Error estimate:
Approximate right-hand side ?? ? Y, |? - ?? | = ? :
Equation:
Error estimate:
for all zn ? Yn
System of equations:
for
14. 14 Example: Least Squares Method Least squares method (Yn = K(Xn )) :
for all zn ? Xn
Substitute
for
15. 15 Example: Least Squares Method Define :
Assume:
for some c > 0, all x ? X
Then least squares method is convergent and ?Rn? = ?n
16. 16 Example: Dual Least Squares Method Dual least squares method (Xn = K* (Yn )) :
with K* : Y ? X adjoint of K
for all zn ? Yn
Substitute
for
17. 17 Define :
Assume:
?n Yn dense in Y
range K(X) dense in Y
Then dual least squares method is convergent and ?Rn? = ?n
18. 18 Application: Symm’s Integral Equation Dirichlet problem for Laplace equation:
? ? R2 bounded domain
?? analytic boundary
f ? C(??)
19. 19 Symm’s Integral Equation
Simple layer potential:
solves BVP iff ? ? C(??) satisfies Symm’s equation:
20. 20 Symm’s Integral Equation Assume ?? has parametrization
for 2?-periodic analytic function ? : [0,2?] ? R2, with
Then Symm’s equation transforms into:
with
21. 21 Application: Symm’s Integral Equation Define K : Hr(0,2?) ? Hr+1(0,2?) and g ? Hr(0,2?), r = 0 by
Define Xn = Yn = { : ?j ? C}
22. 22 Application: Symm’s Integral Equation Approximate right-hand side g? ? Y, ?g - g? ? = ? :
(Bubnov-)Galerkin method:
Least squares method:
Dual least squares method:
Error Estimate:
23. 23 Conclusions Discretisation schemes can be used as regularisation strategies
Galerkin method converges iff it provides regularisation strategy
Special cases of Galerkin methods:
least squares method
dual least squares method