Often in mathematics, when requiring some structure/operation/property/… to exist at every finite arity?, it suffices to require only the binary (-ary) and nullary (-ary) forms, from which the others follow.
For example, in defining a category, one could use an “unbiased?” definition in which composites of all finite sequences of morphisms are directly postulated, with corresponding associativity laws, but it suffices to require only binary composites and nullary composites (i.e., identity morphisms) and some particular corresponding associativity laws.
As a special case of this, we have perhaps the prototypical example of a binary/nullary pair sufficing to generate all finite instances of some structure: the natural numbers themselves are the free monoid on one generator, and thus are freely associatively produced from that one generator (aka, ) using only binary and nullary addition.
Sometimes a nullary operation does not exist but one still wants to decompose a n-ary operation into binary operations. For example, consider the reals, , as an unbounded lattice (top, , and bottom, , do not exist) where
and
Here does not exist while
.
One approach is to compute in the extended reals, ( enlarged with and .) Here
and
In we have the nullary which gives:
.
Another approach is to define a special scheme for composition for when a nullary operator does not exist that instead uses a unary operator that is an identity map (or factored through one).
So far an unbounded lattice is the only example I can think of where an associative binary operation does not have a corresponding nullary form (or when an associative (co) product does not have a corresponding (co) terminal object.) It is unknown if whether all such cases can be handled by the general scheme computing in an enlarged context.