1: \begin{abstract}

2: For large scale symmetric discrete ill-posed problems, MINRES and MR-II are

3: often used iterative regularization solvers. We call a regularized solution

4: best possible if it is at least as accurate as the best regularized solution

5: obtained by the truncated singular value decomposition (TSVD) method.

6: In this paper, we analyze

7: their regularizing effects and establish the following results:

8: (i) the filtered SVD expression are derived for the regularized solutions by MINRES;

9: (ii) a hybrid MINRES that uses explicit regularization

10: within projected problems is needed to compute a best possible

11: regularized solution to a given ill-posed problem;

12: (iii) the $k$th iterate by MINRES is more accurate

13: than the $(k-1)$th iterate by MR-II until the semi-convergence

14: of MINRES, but MR-II has globally better regularizing effects than MINRES;

15: (iv) bounds are obtained for the

16: 2-norm distance between an underlying $k$-dimensional Krylov subspace and

17: the $k$-dimensional dominant eigenspace. They show that MR-II has better

18: regularizing effects for severely and moderately ill-posed

19: problems than for mildly ill-posed problems, and

20: a hybrid MR-II is needed to get a best possible regularized solution

21: for mildly ill-posed problems; (v) bounds are derived for the entries generated

22: by the symmetric Lanczos process that MR-II is based on,

23: showing how fast they decay. Numerical experiments confirm our assertions.

24: Stronger than our theory, the regularizing effects of MR-II are experimentally

25: shown to be good enough to obtain best possible regularized

26: solutions for severely and moderately ill-posed problems.

27: 

28: \smallskip

29: \noindent \textbf{Keywords:}

30: Symmetric ill-posed problem, regularization, partial regularization, full

31: regularization, semi-convergence, MR-II, MINRES, LSQR, hybrid.

32: \smallskip

33: 

34: \noindent \textbf{Mathematics Subject Classifications (2010)}:

35: 65F22, 65J20, 15A18.

36: \end{abstract}