REPRESENTATION OF EVENTS IN NERVE NETS 31
inner neuron at time p + 2 , when started with suit
able states of the inner neurons at time l .
The proof will follow. By Corollary Theorem 5 below, all repre
sentable events are regular. So by Theorems 4 and 5 together, combinations
of regular events by &, V and are regular, which with Theorem 3 in
cludes Theorem U . We have not defined & and as operations on sets of
tables, so EF and E*F cannot be used after the application of & or .
PROOF OF THEOREM k . To each of the regular events which enter in the con
struction by &, V and of the given event, consider a regular expression
for the regular event. Apply to this Lemma 5 with s = 2 , and to the re
sulting terms of the second kind Lemma 6 . Thus we obtain an expression for
the given event by the operations &, V and — from components
E.F, , ..., E F where each E. is an expression for a definite event and
1 1 m m 1
F^ is a regular expression (then the definite event expressed by E^ is
non-initial and of length > 2 ) or empty. Let E| come from E^ as E'
came from E in the'proof of Theorem 3 Case 2 if F^ is regular, and be
the result of introducing an extra input neuron j l ^ + ^ to fire at the first
moment of E| if F^ is empty. Now consider (as an event on the k + m
neurons , . . ., , . . ., same combination of Ej, . . ., E^
as the given event is of ..., EmFm . If this combination of
Ej, ..., E^ when treated as a definite event in the sense of Section 5 (not
Section 6 ) of length equal to the greatest of the lengths of E] , ..., E^
is not positive, we make it so by adding "& E 1 1" where E ' m+1 refers
to the firing of a neuron ^ - +m+-| time p. Now use the method of net
construction for Theorem 1 Corollary 1 to construct a representing net for
this event on k + m or k + m + 1 neurons. Then for each i for which
F^ is regular, identify with the output neuron of a net given by
Theorem 3 representing F^; and for each i for which F^ is empty make
^k+i an inner neuron required to fire at time 1 , as in Figure 1 6 if E^
is non-initial or i = m + 1, and as in Figure 19 if E^ is initial.
7.^ PROBLEMS. Numerous problems remain open, which the limited time we have
given to this subject did not permit us to consider, although probably some
of them at least can be solved quickly.
Is there an extension of Theorem 1 Corollary 2 to all regular events
By the complete set of tables for an event we mean the set of tables
all of them initial which describes the event. By the minimal set of tables
for an event we mean the set of tables describing the event each of which has
the property that neither a proper upper segment of it, nor itself if it is
initial, as a non-initial table describes an occurrence of the event. The
complete set of tables for a regular event is regular, by Theorem 3 and the