Finding Other Eigenvectors

In the preceding example (17) we have found four left-hand Perron eigenvectors and three right-hand Perron eigenvectors. This means that there is at least one right-hand Perron eigenvector missing since the dimension of the right-hand eigenspace is equal to dimension of the left-hand eigenspace. It turns out that the missing right-hand Perron eigenvector must have negative and positive parts. Moreover the nonzero parts of this Perron eigenvector has zeros in places where the semipositive Perron eigenvectors have positive parts. Therefore the missing eigenvector can not be part of the nonnegative eigenspace. In general eigenvectors like these exist when a basic class points to two or more other basic classes, either directly or indirectly. In our example this occurs with the right-hand eigenvector at class 6. This class is pointing directly at classes 2 and 5. Both of which are basic.

The eigenvector equation corresponding to this picture is
(25)

Now, take w2=b2y2, w5=b5y5 and let q6' satisfy q6'A66*q6'»0 then
(26)

Therefore
(27)

Both the numerator and the denominator are positive. So, b2 and b5 must opposite in sign. To obtain w6 we may proceed as follows. Rewrite equation (25) in the form
(28)

with b2 and b5 satisfying equation (27). Define
(29)

where a6s is scalar, a6r' a row vector and a6c a colum vector. The matrix A66(-1) is obtained by deleting e.g. the first row and column from the matrix A66. Partition w6 accordingly by writing w6= (w6s ; w6c')'. Observe that since A66 is irreducible, the dominant eigenvalue of A66(-1) is smaller than λ. Hence (λI-A66(-1))-1>0. This inverse is not necessarily positive since the matrix might be reducible.

Let G=(0;I66(-1)), then
(30)

thus
(31)

MAIN