BBTools

bbsolveopt

This function is not yet fully documented. This is a transcript of the text-formatted help.

bbsolveopt  Create parameters for solving a linear syste,
   bbsolveopt(A) creates a set of parameters suitable for solving the
   linear system A*X=B.

   Syntax:
     OPT2=bbsolveopt(A)
     OPT2=bbsolveopt(A,REG)
     OPT2=bbsolveopt(A,OPT1)
     OPT2=bbsolveopt(A,REG,OPT1)

   bbpresolve, bbsolve, and bbregsolve are flexible tools, which can
   be controlled by parameters given in a struct. bbsolveopt is
   intended for the case where it is necessary to tune a computation.
   Instead of creating a set of parameters from scratch, bbsolveopt
   creates the required parameters which can be used as a starting point.

   Fields in OPT1 are passed unmodified in OPT2. Therefore, you may fill
   in some fields, and let bbsolveopt choose any unspecified parameters.
   These will, in general, depend on the parameters present in OPT1.

   The most useful options are the following:
      reg       Enable regularized solutions       [true | {false}]
      tol       Solution tolerance                 [scalar | {1e-6}]
      ntol      Null-space tolerance               [scalar | {tau}]
      rmaxi     Maximum number of restarts         [integer | inf | {30}]
      mem       Available memory in megabytes      [scalar]
      disp      Verbosity level                    [-1 | 0 | {1} | 2]

   TOLERANCE: an unregularized solution, A*X=B+Z, should satisfy:
     norm(A*X+Z-B) <= OPT.tol          (accuracy of the solution)
     norm(A'*Z)    <= OPT.ntol*norm(Z) (definition of null-space)

   REGULARIZATION: to compute regularized solutions, is is necessary to
   set OPT.reg=true. The following options is also respected:
      rtol      Relative tolerance                 [scalar | {0.1}]
      atol      Absolute tolerance                 [scalar | {tau}]

   For reliable results, the singular triplets must be computed with some
   accuracy, which is controlled by OPT.rtol. The appropriate value
   depends on the steepness of the filter-factors used for regularization.

   A component is also accepted if the absolute accuracy is better than
   OPT.atol. If it is known that the filter-factors will damp out singular
   values below some threshold, setting OPT.atol and OPT.ntol may save
   a substantial amount of computation.

   TUNING: It is possible to control the inner workings more precisely.
   The main reason for doing this is to reduce the memory used.
      K         Dimension of the workspace          [integer]
      rkeep     Ideal dimension of the workspace    [integer]

   These are subject to change. Please consult the manual for details.

   See also bbpresolve, bbsolve, bbregsolve.

bbsolvemembbsvdf