Equational Theories

12 Weak central groupoids

In this chapter we study weak central groupoids Definition 2.30,

x=(yx)(x(zy)).
1

The first observation is that this law is equivalent to its dual:

Lemma 12.1 1485 equivalent to 2162
#

Definition 2.30 is equivalent to the dual law

x=((yz)x)(xy)
2

(equation 2162).

Proof

Given a weak central groupoid G, define a directed graph with vertices in G by declaring xy if and only if y=xz for some z. There is an equivalent characterization of this graph:

Lemma 12.2 Equivalent characterization of graph

One has xy if and only if x=wy for some w.

Proof

Define a good path in G to be a path of the form

xxyy

for some x,yG (we allow loops). By the above lemma, this is a path in G. The following claims are clear from definition and the above lemma:

  • If x,yG then there is exactly one good path xzy from x to y.

  • Any edge xy in the directed graph is the initial segment of some good path xyz.

  • Any edge xy in the directed graph is the final segment of some good path wxy.

Slightly more non-trivial is

Lemma 12.3 Claim 4

If abcdea is a 5-cycle in the directed graph, and abc and cde are good paths, then bcd is also good.

Proof

Conversely, we have

Lemma 12.4 Reversing the claims
#

Let G be a directed graph, with some paths of length two in the graph designated as “good”, in such a way that Claims 1-4 hold. Then there is a weak central groupoid structure on the vertices of G such that the good paths are precisely the paths of the form xxyy.

Proof

This gives us a graph-theoretical route to construct weak central groupoids. We first introduce a weaker version of Claim 1:

  • If x,yG then there is at least one good path xzy from x to y.

Let us call a relaxed weak central groupoid a directed graph with some paths of length 2 designated as “good” that obeys Claims 1’, 2, 3, 4.

We also define a partial weak central groupoid to be a directed graph with some paths of length 2 that obeys Claim 4 as well as the following opposite weakening of Claim 1:

  • If x,yG then there is at most one good path xzy from x to y.

If we can upgrade Claim 1” to Claim 1, and we also have Claim 2 and Claim 3, then we call this a complete weak central groupoid, and by the previous proposition this is in correspondence with Equation 1.

Let G0 be a relaxed weak central groupoid. A partial extension of G0 is a partial weak central groupoid G with a “projection map” π:GG0, which is a homomorphism in the sense that the image π(x)π(y) of any edge xy in G is an edge in G0, the image π(x)π(y)π(z) of any good path xyz in G is a good path in G0, and the image π(x)π(y)π(z) of any bad path xyz in G is a bad path in G0. Note that Claim 4 for G is then automatic from Claim 4 of the base G0. The extension is complete if the partial weak central groupoid is complete.

We have the following convenient completion property:

Proposition 12.5 Completion property

Let G0 be a directed graph obeying claims 1’, 2, 3, 4. Then any finite partial extension of G0 with carrier G0×N (and projection map π(a,n)=a) can be completed to a complete extension.

Proof

Definition 2.30 does not imply any of the following laws:

  • Equation 3457: xx=x((xx)y).

  • Equation 2087: x=((yx)x)(xx).

  • Equation 2124: x=((yy)x)(xx).

  • Equation 3511: xy=x((xy)x).

Proof

12.1 Twisting a weak central groupoid

Occasionally, an equational law is preserved under a “twist” operation in which one replaces the magma operation xy by xy:=TxUy for some automorphisms T,U of the magma G that obey additional relations. In the case of the weak central groupoid law Equation 1, we see that

(yx)(x(zy))=(T2yTUx)(UTx(UTUzU3y))

so if T is an automorphism of order 5 and U=T1 (so that T2=U3), we conclude that this twisted magma is also a weak central groupoid. This can be used to generate further counterexamples. For instance, let M2 be the order two weak central groupoid with carrier F2 and with the NAND operation xy:=1xy; this can easily be verified to be a weak central groupoid. It does not have any nontrivial automorphisms, but its fifth power M25 has a cyclic shift T of order 5: T((xi)iZ/5Z)=(xi+1)iZ/5Z. If we twist M25 by T and T1, we obtain a weak central groupoid M with carrier F25 and magma operation

(xi)iZ/5Z(yi)iZ/5Z=(1xi+1yi1)iZ/5Z.

In particular, if x=(xi)iZ/5Z, then

xx=(1xi+1xi1)iZ/5Z

and

(xx)(xx)=(1(1xi+2xi)(1xixi2))iZ/5Z

which by the laws of boolean algebra simplify to

(xx)(xx)=(xi(xi2+xi+xi+2))iZ/5Z

from which one can easily refute equation 151,

x=(xx)(xx).

Informally, the reason for this is that equation 151 has a different semigroup twist symmetries: T2=1,T=U1 instead of T5=1,T=U1.