SADDLE-POINT EQUATIONS OF A NEURAL-NETWORK WITH CORRELATED ATTRACTORS

We derive in a rigorous way, using a new technique, the saddle-point e quations of a modified Hopfield model which stores sequences of patter ns . We obtain the same equations as those of the replica-symmetry app roach without using the n --> 0 limit but assuming selfaveraging of th e Edwards-Anderson order parameter.