1: \begin{abstract}

2: Pole-swapping algorithms, which are generalizations of the QZ algorithm

3: for the generalized eigenvalue problem, are studied.  A new modular 

4: (and therefore more flexible) convergence theory that applies to all pole-swapping

5: algorithms is developed.  A key component of all such algorithms is a procedure that swaps two

6: adjacent eigenvalues in a triangular pencil.  An improved swapping routine is developed,

7: and its superiority over existing methods is demonstrated by a backward error analysis and 

8: numerical tests.  The modularity of the new convergence theory and the generality of the

9: pole-swapping approach shed new light on bi-directional chasing algorithms, optimally

10: packed shifts, and bulge pencils,  and allow the design of novel algorithms.

11: \end{abstract}

12: