Let σ be the cyclic permutation i 1 → i2 → ∙ ∙ ∙ → ip → i 1 of arbitrary length
p such that w(A, σ) 6= -∞.
As w(A, σ) 6= -∞, then aij ij+1 6= -∞ for all j. So ij and ij+1 must be of
different parities for all j . This means elements in the sequence i1 , i2 , . . . , ip , i1
alternate between even and odd. Thus p must be an even number, i.e. there are
no cyclic permutations σ of odd length of finite weight. Hence the result. □
If A is symmetric, then together with Corollary 4, this gives us:
Theorem 17. If A = (aij) ∈ Rn×n is a symmetric matrix such that aij = -∞
if i + j is even, then δk(A) 6= -∞ for all even k ≤ kmax, and δk (A) = -∞ for
all odd k.
A certain type of Hankel matrix satisfies Theorem 17. Rewriting it for this
type of matrix gives:
Theorem 18. If {gr ∈ R; r = 1,..., 2n — 1} is the sequence generating Hankel
matrix H and gr = -∞ for all odd r, then δk(H) 6= -∞ for all even k ≤ kmax
and δk(H) = -∞ for all odd k.
Combining Theorem 15 and Theorem 18 enables us to decide whether δk(H)
is finite or not for any Hankel matrix H .
Theorem 19. If {gr ∈ R; r = 1,..., 2n — 1} is the sequence generating Hankel
matrix H then
1. δk (H) = -∞ for all even k ≤ kmax,
2. δk(H) = -∞ for all odd k if gr = -∞ for all odd r, and
3. δk(H) = -∞ for all odd k ≤ kmax if gr = -∞ for some odd r.
4.3 Block diagonal matrices
Let
( A1
-∞∖
A = blockdiag(A1, A2, . . . , Ap) =
∖-∞
A2
Ap
D(Ai) is a subgraph of D(A) for every i = 1, . . . , p. Every D(Ai) is disjoint
from any D(Aj ), j 6= i. So any cycle in D(A) has nodes entirely within one
14