The routine wrongly returned 1/lambda instead of lambda, as the leading eigenvalue of the adjacency matrix.
mentioned in issue #125 (closed)