1: \begin{abstract}

2: An efficient way of solving 2D stability problems in fluid mechanics is

3: to use, after discretization of the equations that cast the problem in the

4: form of a generalized eigenvalue problem, the incomplete Arnoldi-Chebyshev

5: method. This method preserves the banded structure sparsity of matrices

6: of the algebraic eigenvalue problem and thus decreases memory use and

7: CPU-time consumption.

8: 

9: The errors that affect computed eigenvalues and eigenvectors are due

10: to the truncation in the discretization and to finite precision in the

11: computation of the discretized problem.  In this paper we analyze those

12: two errors and the interplay between them.  We use as a test case the

13: two-dimensional eigenvalue problem yielded by the computation of inertial

14: modes in a spherical shell.  This problem contains many difficulties

15: that make it a very good test case.  It turns out that that single

16: modes (especially most-damped modes i.e. with high spatial frequency)

17: can be very sensitive to round-off errors, even when apparently good

18: spectral convergence is achieved.  The influence of round-off errors is

19: analyzed using the spectral portrait technique and by comparison of double

20: precision and extended precision computations.  Through the analysis we

21: give practical recipes to control the truncation and round-off errors

22: on eigenvalues and eigenvectors.

23: 

24: \end{abstract}

25: