We present a family of randomized algorithms that enjoys a wide range
of time-space trade-offs in deciding undirected S-T-connectivity. Our
trade-offs cover the whole range between breadth first search and the
random walk procedure of Aleliunas at al., and achieve a time-space pr
oduct of (O) over tilde(mn) (where n is the number of vertices in the
graph, m is the number of edges, and (O) over tilde notation is used i
n order to suppress logarithmic terms). Moreover, we obtain improved t
ime-space trade-offs of (O) over tilde(n(2)) for regular graphs. A con
venient and informative way of expressing our trade-offs, that implies
the trade-offs stated above, is as (O) over tilde((Sigma(i=1)(n) d(i)
)(Sigma(i=1)(n) 1/d(i))), where d(i) is the degree of vertex i in the
input graph. In constructing our algorithms and analysing them, we bui
ld upon earlier work of Broder et al. (who achieved a time-space trade
-off of (O) over tilde(m(2))), Barnes and Feige (who achieved a time-s
pace trade-off of (O) over tilde(m(3/2)n(1/2))), and Aldous. In passin
g, we also improve previous results regarding the rate at which a rand
om walk discovers new vertices in a graph. (C) 1997 Academic Press.