next up previous contents index
Next: Selecting a Band Storage Up: Templates and Driver Routines Previous: Post-processing and Accuracy Checking

Band Drivers

    If the matrix ${\bf A}$ and ${\bf M}$ are stored in LAPACK band form, then one of the band drivers may be used. Band drivers are named in the form of XYbdrZ, where the first character X specifies the precision and data type,
s single precision
d double precision
c single precision complex
z double precision complex
the second character Y indicates the symmetry property of the problem,
s symmetric problem
n nonsymmetric problem
and the third character Z is a number between 1 and 6 indicating the type of the problem to be solved and the mode to be used. Each number is associated with a combination of bmat and iparam(7) settings used in that driver. Tables A.10--A.12 list the double precision band storage drivers.


 
Table A.10: Band storage drivers for symmetric eigenvalue problems
   
BAND DRIVER PROBLEM SOLVED
   
dsbdr1 Standard eigenvalue problem (bmat = 'I')
  in the regular mode (iparam(7) = 1).
dsbdr2 Standard eigenvalue problem (bmat = 'I')
  in a shift-invert mode (iparam(7) = 3).
dsbdr3 Generalized eigenvalue problem (bmat = 'G')
  in the regular inverse mode (iparam(7) = 2).
dsbdr4 Generalized eigenvalue problem (bmat = 'G')
  in a shift-invert mode (iparam(7) = 3).
dsbdr5 Generalized eigenvalue problem (bmat = 'G')
  in the Buckling mode (iparam(7) = 4).
dsbdr6 Generalized eigenvalue problem (bmat = 'G')
  in the Cayley mode (iparam(7) = 5).


 
Table A.11: Band storage drivers for non-symmetric eigenvalue problems
   
BAND DRIVER PROBLEM SOLVED
   
dnbdr1 Standard eigenvalue problem (bmat = 'I')
  in the regular mode (iparam(7) = 1).
dnbdr2 Standard eigenvalue problem (bmat = 'I')
  in a shift-invert mode (iparam(7) = 3).
dnbdr3 Generalized eigenvalue problem (bmat = 'G')
  in the regular inverse mode (iparam(7) = 2).
dnbdr4 Generalized eigenvalue problem (bmat = 'G')
  in a real shift-invert mode (iparam(7) = 3).
dnbdr5 Standard eigenvalue problem (bmat = 'I')
  in a complex shift invert mode (iparam(7) = 4).
dnbdr6 Generalized eigenvalue problem (bmat = 'G')
  in a complex shift invert mode (iparam(7) = 4).


 
Table A.12: Band storage drivers for Complex arithmetic eigenvalue problems.
   
BAND DRIVER PROBLEM SOLVED
   
znbdr1 Standard eigenvalue problem (bmat = 'I')
  in the regular mode (iparam(7) = 1).
znbdr2 Standard eigenvalue problem (bmat = 'I')
  in a shift-invert mode (iparam(7) = 3).
znbdr3 Generalized eigenvalue problem (bmat = 'G')
  in the regular inverse mode (iparam(7) = 2).
znbdr4 Generalized eigenvalue problem (bmat = 'G')
  in a shift-invert mode (iparam(7) = 3).

There are no special drivers for complex Hermitian problem.   Complex Hermitian problems can be solved by using [c,z]nbdrZ. These drivers call the band eigenvalue computation routine XYband, where the first character X (s,d) specifies the precision and data type as listed above, and the second character Y indicates the symmetry property of the problem that can be solved with this routine. Since the reverse communication interface has already been implemented in these computational routines, users only need to provide the matrix and modify a few variables in these drivers to solve their own problem. A procedure for modifying these drivers is presented below.



 
next up previous contents index
Next: Selecting a Band Storage Up: Templates and Driver Routines Previous: Post-processing and Accuracy Checking
Chao Yang
11/7/1997