1: \begin{abstract}

2: 

3: The problem of minimization of the least squares 

4: functional with a smooth, lower semi-continuous, convex penalizer 

5: $J(\cdot)$ is considered to be solved. 

6: Over some compact and convex subset $\Omega$ of the Hilbert

7: space $\cH,$ the regularizer is implicitly defined as 

8: $ J(\cdot) : \cC^{k}(\Omega , \cH) \rightarrow \R_{+}$

9: where $k \in \{1,2\}.$ So the cost functional associated with

10: some given linear, compact and injective forward operator 

11: $\cT :\Omega \subset \cH \rightarrow \cH,$

12: 

13: \bea

14: F_{\alpha}(\cdot , f^{\delta}) := \frac{1}{2} \norm{\cT( \cdot ) - f^{\delta}}_{\cH}^2 + \alpha J(\cdot) ,

15: \nonumber

16: \eea

17: where $f^{\delta}$ is the given perturbed data with its

18: perturbation amount $\delta$ in it.

19: Convergence of the regularized optimum solution

20: $\varphi_{\alpha(\delta)} \in \argmin F_{\alpha}(\varphi , f^{\delta})$

21: to the true solution $\varphi^{\dagger}$ is analysed

22: depending on the smoothness degree of the penalizer, 

23: {\em i.e.} the cases $k \in \{1,2\}$ in  

24: $ J(\cdot) : \cC^{k}(\Omega , \cH) \rightarrow \R_{+}.$ 

25: In both cases, we define such a regularization parameter 

26: that is in cooperation with the condition

27: 

28: \bea

29: \alpha(\delta , f^{\delta}) \in \{ \alpha > 0 \mbox{ }\vert \mbox{ }\norm{\cT\varphi_{\alpha(\delta)} - f^{\delta}}\leq \tau\delta \} ,

30: \nonumber

31: \eea

32: for some fixed $\tau \geq 1.$ In the case of $k = 2,$ we 

33: are able to evaluate the discrepancy 

34: $\norm{\cT\varphi_{\alpha(\delta)} - f^{\delta}}\leq \tau\delta$

35: with the Hessian Lipschitz constant $L_H$ of the functional 

36: $F_{\alpha}(\cdot , f^{\delta}).$

37: 

38: \bigskip

39: \textbf{Keywords.}

40: {convex regularization, Bregman divergence, Hessian Lipschitz constant, discrepancy principle.}

41: \end{abstract}

42: