Re: [eigen] cache-friendly matrix inverse

[ Thread Index | Date Index | More Archives ]

Hash: SHA256

Benoit Jacob schrieb:
> It's actually something I never considered before: for me, as a
> mathematician, matrix inversion is a basic operation that "of course"
> i consider important, without even asking myself why; if you ask about
> practical uses for it, well I'm pretty sure that there must be
> situations where you want to solve Ax=b for a single given matrix A,
> but for a lot of vectors b and you don't know in advance what these
> vectors b will be, you just want to be able to solve very fast
> whenever they show up; then the best thing you can do is to precompute
> A^-1 so you only have a matrix-vector product to compute whenever a
> "b" vector shows up.

That's *exacly* what LU decomposition is for. Compute it once and use it
to solve the equation many times.
Note: the amount of operations to calculate A^-1 and an LU decomposition
is the same. *AND* using a LU to solve the equation (foreward and
backward substitution) takes also as many operations as a matrix
multiplication (i.e. exactly the same as a multiplication with the inverse)

An inverse matrix is an extremely usefull "method" in mathematics - as
long as you aren't *calculating* it in numerics :)
Version: GnuPG v1.4.9 (GNU/Linux)


Mail converted by MHonArc 2.6.19+