1: \begin{abstract}%

2: There are variety of computational algorithms need sequential

3: sweeping; sweeping based on specific order; on a structured grid,

4: e.g., preconditioning (smoothing) by SOR or ILU methods and solution

5: of eikonal equation by fast sweeping algorithm. Due to sequential

6: nature, parallel implementation of these algorithms usually leads to

7: miss of efficiency; e.g. a significant convergence rate decay.

8: Therefore, there is an interest to parallelize sequential sweeping

9: procedures, keeping the efficiency of the original method

10: simultaneously. This paper goals to parallelize sequential sweeping

11: algorithms on structured grids, with emphasis on SOR and ILU

12: preconditioners. The presented method can be accounted as an

13: overlapping domain decomposition method combined to a multi-frontal

14: sweeping procedure. The implementation of method in one and two

15: dimensions are discussed in details. The extension to higher

16: dimensions and general structured n-diagonal matrices is outlined.

17: Introducing notion of alternatively block upper-lower triangular

18: matrices, the convergence theory is established in general cases.

19: Numerical results on model problems show that, unlike related

20: alternative parallel methods, the convergence rate and efficiency of

21: the presented method is close to the original sequential method.

22: Numerical results also support successful use of the presented

23: method as a cache efficient solver in sequential computations as

24: well.%

25: \end{abstract}%