1: \begin{abstract}

2: Unmatched pairs of forward and back projectors are common in X-ray CT computations

3: for large-scale problems; they are caused by

4: the need for fast algorithms that best utilize the computer hardware, and it

5: is an interesting and challenging task to develop fast and easy-to-use

6: algorithms for these cases.

7: Our approach is to use preconditioned GMRES, in the form of the

8: AB- and BA-GMRES algorithms, to handle the unmatched normal equations

9: associated with an unmatched pair.

10: These algorithms are simple to implement, they rely only on computations

11: with the available forward and back projectors, and they do not require the

12: tuning of any algorithm parameters.

13: We show that these algorithms are equivalent to well-known LSQR and

14: LSMR algorithms in the case of a matched projector.

15: Our numerical experiments demonstrate that AB- and BA-GMRES exhibit

16: a desired semi-convergence behavior that is comparable with

17: LSQR/LSMR and that standard stopping rules work well.

18: Hence, AB- and BA-GMRES are suited for

19: large-scale CT reconstruction problems with noisy data and

20: unmatched projector pairs.

21: \end{abstract}

22: