Recently, a new technique called the method of approximations has been
developed for proving lower bounds on the size of circuits computing
certain Boolean functions. To obtain a lower bound on the size complex
ity size(f) of a certain function f by the method, an appropriate legi
timate model M for the function f is chosen, and then a lower bound on
the distance rho(f, M) from f to M is derived. The lower bound on rho
(f, M) becomes a lower bound on size( f) in view of the fact that size
(f) greater than or equal to rho(f, M). Razborov gave a legitimate mon
otone model, M(mon)(F-max), and showed that rho(f, M(mon)(F-max)) = Om
ega(size(1/3)(f)), so there remains a gap between the size size(f) and
the distance rho(f, M). Employing his method, the following statement
s are established: (i) Razborov's model M(mon)(F-max) is generalized t
o obtain model M(F-max), and it is established that rho(f, M(F-max)) =
Omega(size(1/2)(f)). (ii) Allowing the underlying graphs of circuits
to have cycles, a new notion of apparently more powerful circuits, cal
led loop circuits, is introduced, and it is proved that rho(f, M(F-max
))=Theta(size(loop)(f)), where size(loop)(f) denotes the size complexi
ty of f based on loop circuits. (C) 1995 Academic Press, Inc.