how about we use same method for invert(A) as for A^^-1 ?
Subject: how about we use same method for invert(A) as for A^^-1 ?
From: Robert Dodier
Date: Sat, 29 Dec 2012 17:01:05 +0000 (UTC)
I nuked the existing function invert and replaced it with one which
punts to invert_by_lu, which is much faster, as the old one used the
adjoint method. I also made A^^-1 call the new invert, so the two now
give the same result. The new invert respects the ratmx and detout
flags, as did the old one. As with the old invert, float and bigfloat
elements are not converted to CRE when ratmx=false. I revised the
description of invert and put some test cases into the test suite.
Commits 4c877473da and d7ced4040b cover that stuff.
Hope this helps,
Robert Dodier
PS.
On 2012-12-10, Robert Dodier <robert.dodier at gmail.com> wrote:
> Hello, invert(A) computes the matrix inverse via the adjoint method,
> while A^^-1 uses Gaussian elimination. Both of them can handle symbolic
> expressions and CRE in addition to numbers. Is there any reason to
> prefer the adjoint method? The adjoint method is much slower.
>
> In any event, it seems messy to have invert(A) and A^^-1 use different
> methods ....
>
> There are other methods to compute a matrix inverse -- there are, at
> least, tminverse (undocumented core function), invert_by_lu (share), and
> probably one in LAPACK.
>
> best,
>
> Robert Dodier