bandsolve: bandsolve
In Monneret/bandsolve: Fast solver for symetric band matrix

Description Usage Arguments Value Examples

View source: R/bandsolve.R

Main function to solve efficiently and quickly a symmetric bandlinear system. Theses systems are solved much faster than standards system, dropping from complexity O(n<c2><b3>) to O(0.5*nk<c2><b2>), where k is the number of sub diagonal.

1	bandsolve(A, b = NULL, inplace = FALSE)

`A`	Band square matrix in rotated form. The rotated form can be obtained with the function as.rotated: it's the visual rotation by 90 degrees of the matrix, where subdiagonal are discarded.
`b`	right hand side of the equation. Can be either a vector or a matrix. If not supplied, the function return the inverse of A.
`inplace`	Should results overwrite pre-existing data? Default set to false.

Solution of the linear problem.

A=diag(4)
A[2,3]=2
A[3,2]=2
R=mat2rot(A)
solve(A)
bandsolve(ref)

set.seed(100)

n=1000;
D0=rep(1.25,n)
D1=rep(-0.5,n-1)
b=rnorm(n)

## Comparison with solve

if(require(microbenchmark)){
A=diag(D0);
A[-n,-1]=A[-n,-1]+diag(D1);
A[-1,-n]=A[-1,-n]+diag(D1);
R=mat2rot(list(D0,D1))
r=microbenchmark(
SOLVE=solve(A,b),
BANDSOLVE=bandsolve(R,b=b,inplace=TRUE),times=100)
boxplot(r)
}