1: \begin{abstract}

2:     We study the solution of minimax

3:     problems $\min_x \max_y G(x) + \iprod{K(x)}{y} - F^*(y)$

4:     in finite-dimensional Hilbert spaces. The functionals $G$ and

5:     $F^*$ we assume to be convex, but the operator $K$ we allow to

6:     be non-linear.

7:     We formulate a natural extension of the modified primal-dual hybrid 

8:     gradient method (PDHGM), originally for linear $K$, due to

9:     Chambolle and Pock. We prove the local convergence of the method,

10:     provided various technical conditions are satisfied. These include

11:     in particular the Aubin property of the inverse of a monotone operator

12:     at the solution. Of particular interest to us is the case

13:     arising from Tikhonov type regularisation of inverse problems with 

14:     non-linear forward operators. Mainly we are interested in total variation 

15:     and second-order total generalised variation priors.

16:     For such problems, we show that our general local

17:     convergence result holds when the noise level of the data $f$ is low,

18:     and the regularisation parameter $\alpha$ is correspondingly small.

19:     We verify the numerical performance of the 

20:     method by applying it to problems from magnetic resonance imaging

21:     (MRI) in chemical engineering and medicine. The specific applications

22:     are in diffusion tensor imaging (DTI) and MR velocity imaging. 

23:     These numerical studies show very promising performance.

24:     

25:     \paragraph{Mathematics subject classification:}

26:         49M29, % Methods involving duality

27:         90C26, % Nonconvex programming, global optimization

28:         92C55. % Biomedical imaging and signal processing.

29: 

30:     \paragraph{Keywords:} 

31:         primal-dual, 

32:         non-linear,

33:         non-convex, 

34:         convergence, 

35:         MRI.

36:     

37: \end{abstract}

38: