Bol Loops of Order 27

I've been working on a classification of Bol loops of small order.

Here is what I thought was a complete list of the thirteen Bol loops of order 27. Today (Sept 19, 2007) Michael Kinyon announced a couple new examples of exponent 3 with trivial centre! Now I need to go back and find a flaw in my program. Meanwhile the following list is not complete as it claims. And I haven't yet added Michael's new examples...

I would appreciate an email message () from you if you have any comments regarding this work.

I have made available

Here are 13 (right) Bol loops of order 27, including 5 groups and 8 non-associative loops. These form 9 isotopy classes of such loops. The 8 nonassociative loops all have centre (equal to the centrum) of size 3; all have exponent nine; none are Moufang; and none are G-loops.

In listing elements of the commutator (resp. associator) subloop of each of our loops, we have printed in italics any elements which are not actual commutators (resp. associators). (This phenomenon does not occur, however, among any the loops of order 27 in our list.)

8 non-associative loops, listed by number of elements of order 3

9 Isotopy Classes

There are five groups of order 27: (cyclic), (abelian C3×C9), (elementary abelian), (nonabelian of exponent 3), (nonabelian of exponent 9).

The remaining four isotopy classes each consists of two isomorphism classes of non-associative loops. These four isotopy classes are the same as the four pairs listed above, i.e. the number of elements of order 3 is an isotopy invariant. (Contrast this with the situation for Bol loops of order 16, in which the number of involutions is by no means an isotopy invariant.)

Naming of the Loops

For each of the loops of order 27, I have used a name 27.t.c.k where t is the number of elements of L of order three, c=|C(L)| and the index k=0,1,2,... indicates merely the order in which each isomorphism class of loop was first encountered by my computer.

Methodology Used

I have searched for Bol loops of order 27 by backtrack search as described elsewhere. However in this particular case the permutations comprising the partial Bol loops through which we must search (columns of the Cayley table) are chosen not from the full symmetric group of degree 27, but from a Sylow 3-subgroup thereof, a subgroup of size "only" 1594323 (see Glauberman 1968, Theorem 15). This reduces the search space considerably.

As before, we have retained a single representative of each isomorphism class of the resulting loops. To accomplish this, we have represented L by a graph which encodes all information in the Cayley table of L. I have then used Brendan McKay's software package nauty to find a "canonical" representative for each resulting graph.

I have also used the computational algebra package GAP (Graphs, Algorithms and Programming) to compute orders of left, right and full multiplication groups.


I am grateful to the Department of Mathematics and Statistics, Memorial University of Newfoundland for their hospitality while much of this study was being undertaken, and in particular to Edgar Goodaire for conversations which have stimulated me in this direction.

/ revised June, 2002