Cometric Association Schemes
Last modified: June 2, 2006
Some Terminology
Most of the basic definitions can be found in the
glossary
The table describes each example in terms of these bits of information:
- the first parameter listed is the number of classes, d. The
Bose-Mesner algebra of such a scheme has dimension d+1
- the number of vertices, |X|, is listed next
- the dimension, m1, of the first eigenspace in the/a
Q-polynomial ordering
- the Krein array
{b0*,
b1*, . . . , bd-1*;
c1*, c2*, . . . ,
cd* } where the tridiagonal matrix
L1* with (k,j)-entry equal to the
Krein parameter q1,jk has diagonal entries
aj*= m1 -
bj*- cj*
and b* above the diagonal
and c* below the diagonal
- the sequence of cosines in the first
eigenspace (Qi,1/ m1 : i=0,1, . . . , d)
- a designation as to whether the
scheme is primitive, Q-bipartite, Q-antipodal, or both
- a description of the scheme, comments or references
For each association scheme in the table, there is a simple
text file which includes
- MAPLE input formats for the number of classes d,
the number of vertices verts and the valencies v[i]
(0 ≤ i ≤ d);
- MAPLE input for the second eigenmatrix Q
- MAPLE input for the first matrix of Krein parameters L1*.
This matrix named L for simplicity and in row k, column j its entry is
the Krein parameter q1,jk
- next come, in some sloppy format, the intersection matrices L_1,. . .,L_d.
The matrix Li, which appears as L_i in the file, has
entry pi,jk in row k, column j
- the next data in the file are the eigenmatrices P and Q. The
ith column of P lists the eigenvalues of Li,
which are also the eigenvalues of the adjacency matrix Ai. The
matrix Q is the inverse of P, scaled by the number of vertices
- finally, we give all Krein parameters in the matrices
Lj* (0 < j < d). The matrix
Lj* is denoted in the file as Lsj
and has, in row k, column h, the Krein parameter qj,hk
For most -- but not all -- association schemes in the table, we include
a representation by vectors. In most cases, this is a set of points in
Euclidean space whose inner products determine the relations joining the
corresponding vertices in the association scheme. (In a few cases, I think
the relations are determined by Hamming distance instead.) No effort
was made to find a representation in the smallest possible dimension
or to choose vectors with centroid at the origin.