block_bandsolve: Solve linear system when the matrix is block-wise with a band...
In goepp/hazreg: Hazard Regularized estimation for time-to-event data

Description Usage Arguments Value Examples

Solve linear system when the matrix is block-wise with a band diagonal block

1	block_bandsolve(mat, vect, mat_as_rotated = FALSE, lim = NULL)

`mat`	Square matrix
`vect`	Vector of same length as `mat`
`lim`	Positive integer

The vector mat ^ -1 vect is returned. If mat[lim:end, lim:end] is not a band matrix or if mat is symmetric, an error is returned. If these two conditions are met, the returned value is the same as solve(mat, vect). There might be nummerical impercision errors.

## Not run: 
## With mat_as_rotated = FALSE
set.seed(0)
A <- matrix(rnorm(36), 6, 6)
B <- matrix(rnorm(24), 6, 4)
D <- diag(rnorm(4)) + rbind(cbind(0, diag(rnorm(3))), 0)
mat <- rbind(cbind(A + t(A), B),
                     cbind(t(B), D + t(D)))
vect <- rnorm(10)
lim <- 6
max(block_bandsolve(mat, vect, lim) - solve(mat, vect))

## With mat_as_rotated = TRUE
set.seed(0)
A <- matrix(rnorm(36), 6, 6)
B <- matrix(rnorm(24), 6, 4)
D <- cbind(rnorm(4), c(rnorm(3), 0))
mat_rotated <- list('A' = A + t(A),
            'B' = B,
            'D' = D)
mat <- rbind(cbind(mat_rotated$A, mat_rotated$B),
                     cbind(t(mat_rotated$B), rot2mat(mat_rotated$D)))
vect <- rnorm(10)
max(block_bandsolve(mat_rotated, vect, mat_as_rotated = TRUE) - solve(mat, vect))

## End(Not run)