ON THE MINIMUM EIGENVALUE SEPARATION FOR MATRICES
矩阵特征值分离度的下界

The minimum eigenvalue separation of a matrix is defined as the minimum of di-stances between distinet eigenvalues. Some lower bounds of the separations are derivedfor tridiagonal matrices in this paper. For Hermitian matrices, a Lonezos algorithm isdesigned to reduce them to tridiagonals with special forms. Two examples are givento show the improvement of the given estimation.