## bbinv

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

bbinv Create a black-box operator for an inverse matrix
BB=bbinv(A) creates a black-box operator for the inverse of a matrix
A. Thus, BB*X and A\X are roughly equivalent, except that BB contains
a precomputed factorization suitable for mass evaluation.
bbinv computes a factorization, which is used to solve the
linear system. The factorization may be specified as a parameter:
'auto' - Automatic (default)
'QR' - QR
'QRP' - QR with column pivoting
'chol' - Cholesky factorization (same as 'SPD')
'LUP' - LU with partial pivoting (same as 'GEPP' and 'LU')
'LUPQ' - LU with column and row pivoting (sparse only)
'svd' - Singular Value Decomposition
'inv' - Form an explicit inverse
The following systems are supported without factorization:
'D' - Diagonal
'L' - Lower triangular
'U' - Upper triangular
The following options must be followed by a scalar:
'inorm' - Lower bound for norm(inv(A))
'thresh' - Threshold passed to LU
In addition, the following flag may be specified:
'notol' - Do not compute tolerances
Several methods may be specified. The method will be the first
applicable method in this list ('auto' specifies the first 6 methods):
D, L, U, chol, LUPQ, QRP, QR, LUP, LU, svd, and inv.
bbinv works hard to determine the accuracy of the operator. This step
is skipped if 'notol' is given as parameter. This may increase the
speed of creating the operator, but some iterative methods may not
function. Specifying 'inorm' directly may also speed up this process,
although this is almost useless for LU-factorizations.
Note: Y=BB*X satisfy (A+E)*Y=X where E is a perturbation. While E is a
important for interpreting Y, the most important issue for iterative
algorithms is linearity. For dense systems, 'QRP' is a good compromise
between accuracy and efficiency.
See also QR, LU, , , .

bbinstallbbkernel