Expressing this equation in the form implicitly
defines the orthogonal permutation matrix **P** and a block diagonal **U**.
These matrices satisfy the equations , , where **I** is the unit matrix. Note that is a Kronecker
product, and that the following two
identities hold

Due to the symmetry of the matrix U, a slightly different factorization is possible, in terms of a permutation matrix which we could call

satisfies the equation but **P** and **Q** do
not commute. It is interesting to note that

Thus there are both factorizations and sum decompositions for the de Bruijn matrix, all readily obtained in a way that indicates that many more representations are possible, although the others would not be as symmetrical as the ones shown.

If we adopted a more formal definition of **P** and **Q** we would find

from which would follow

Writing the four terms of this sum according to their matrix elements, we find (where is Kronecker's delta):

a representation which can readily be generalized for any power of **B,**
and also for any number of states per cell, In particular, the
form of both the minimal equation and the characteristic equation for
the de Bruijn matrices follows. Note that the congruence in these equations
is multiplicative, not additive, making congruent to 1, not 0.

The factorized form of the de Bruijn matrices can be used to obtain the determinants and inverses of their probabilistic generalization; since the strict de Bruijn matrices are singular their zero determinant can be derived in this way, but of course they have no inverses.

E-mail:mcintosh@servidor.unam.mx