Past exam of the mathematics course of the University of Cambridge / 2025 / iii / Paper 335 / 3 / i / Solution

A regularization of an inverse problem consists of bounded operators $R_{\alpha }:Y\to X$ and a parameter rule $\alpha =\alpha (\delta ,y^{\delta })$ such that, whenever $\Vert y^{\delta }-y\Vert \le \delta$ and $y$ lies in the domain of $A^{†}$,as $\delta \to 0$. It is needed because a compact operator on an infinite-dimensional space has singular values tending to zero, so direct inversion divides noisy data by arbitrarily small numbers and is generally discontinuous.

Tikhonov regularization definesand its normal equation givesFor every fixed $\alpha >0$, $A^{\ast }A+\alpha I$ is bounded below by $\alpha I$, and the data-to-solution operator is bounded. Small changes in $y^{\delta }$ therefore produce small changes in $x_{\alpha }^{\delta }$.