A perturbation theory for the Drazin inverse AD is developed. This theory i
s based on the Jordan canonical form of A. Norm estimates of \ \ (A + E)(#)
-A(D) \ \ are derived when A(D) and A + E are of the same rank and \ \E \
\ is small. A hard problem by Campbell and Meyer [Linear Algebra Appl. 10 (
1975) 77-83] is partially solved. (C) 2002 Elsevier Science Inc. All rights
reserved.