Modify the Problem Dependent Variables

  To set up the proper storage and to arrange for effective use of the IRLM various parameters will have to be set. These variables include:

<<354>>
• [n] The dimension of the problem.
• [nev] The number of eigenvalues needed.
• [ncv] The length of the Lanczos factorization. This represents the maximum number of Lanczos vectors used.
• [which] The eigenvalues of interest.
• [info] Set to 0 for a randomly generated starting vector. If the user decides to use another starting vector, this value should be set to 1, and the starting vector should be provided in the array resid.
• [sigma] The shift used if a spectral transformation is employed.

The variable nev may be set to be a value larger than the number of eigenvalues desired to avoid splitting a eigenvalue cluster. The only restriction is that nev must be less than ncv. The recommended initial choice of ncv is to set ncv = 2 nev. The user is encouraged to experiment with both nev and ncv. The possible choices for the input variable which are listed in Table A.3.

 

which EIGENVALUES

'LA' Largest (algebraic) eigenvalues.

'SA' Smallest (algebraic) eigenvalues.

'LM' Largest eigenvalues in magnitude.

'SM' Smallest eigenvalues in magnitude.

'BE' Eigenvalue at both ends. When nev is odd, compute one more from the high end than from the low end.

When a spectral transformation is employed, only the selections 'BE', 'LA' or 'SA' should be used.

Once the above variables are modified, the storage declarations

      integer          maxn, maxnev, maxncv, ldv
      parameter       (maxn=256, maxnev=10, maxncv=25, ldv=maxn )

should be adjusted so that the conditions

n maxn,

nev maxnev,

ncv maxncv,

nev + 1 ncv

are satisfied.