Abstract
An efficient implementation of the shifting algorithm ([BPS]) for min-max tree partitioning is given. The complexity is reduced from ORO+ kn) to ORM + logd) + n) where a tree of n vertices, radius of R edges, and maximum degree d is partitioned into k + 1 sub trees. The improvement is mainly due to the new junction tree data structure, which suggests a succinct representation for subsets of edges, of a given tree, that preserves the interrelation between the edges on the tree.