Noise Models for Ill-Posed Problems
The standard view of noise in ill-posed problems is that it is either deterministic and small (strongly bounded noise) or random and large (not necessarily small). Following Eggerment, LaRiccia and Nashed (2009), a new noise model is investigated, wherein the noise is weakly bounded. Roughly speaking, this means that local averages of the noise are small. A precise definition is given in a Hilbert space setting, and Tikhonov regularization of ill-posed problems with weakly bounded noise is analysed. The analysis unifies the treatment of “classical” ill-posed problems with strongly bounded noise with that of ill-posed problems with weakly bounded noise. Regularization parameter selection is discussed, and an example on numerical differentiation is presented.
KeywordsRegularization Parameter Noise Model Source Condition Tikhonov Regularization Fredholm Integral Equation
- Groetsch CW (1983) Comments on Morozov’s discrepancy principle. In: Hämmerlin G, Hoffman KN (eds) Improperly posed problems and their numerical treatment. Birkhäuser, BaselGoogle Scholar
- Groetsch CW, Scherzer O (2002) Iterative stabilization and edge detection. Contemp Math 313:129–141. American Mathematical Society, ProvidenceGoogle Scholar
- Tikhonov AN, Arsenin VYa (1977) Solutions of ill-posed problems. Wiley, New YorkGoogle Scholar