We consider the singularities of the zero-temperature renormalization-group
how for random elastic manifolds. When starting from small scales, this fl
ow goes through two particular points l* and I-c, where the average value o
f the random squared potential [U-2] turns negative (l*) and where the four
th derivative of the potential correlator becomes infinite at the origin (I
-c). The latter point sets the scale where simple perturbation theory break
s down as a consequence of the competition between many metastable states.
We show that under physically well defined circumstances l(c)<l* and thus t
he apparent renormalization of [U*] to negative values does not take place.
[S0163-1829(99)06301-8].