Re: [eigen] inverse() method for TriangularView class? |

[ Thread Index |
Date Index
| More lists.tuxfamily.org/eigen Archives
]

*To*: eigen@xxxxxxxxxxxxxxxxxxx
*Subject*: Re: [eigen] inverse() method for TriangularView class?
*From*: Benoit Jacob <jacob.benoit.1@xxxxxxxxx>
*Date*: Mon, 31 Oct 2011 12:51:17 -0400
*Dkim-signature*: v=1; a=rsa-sha256; c=relaxed/relaxed; d=gmail.com; s=gamma; h=mime-version:in-reply-to:references:date:message-id:subject:from:to :content-type:content-transfer-encoding; bh=Ng916vQ4rwLlwg8o8tcadn1fio7fWQFXh1N07gtCrVg=; b=VpF88zIP/uMFa7DeQNxKTraka9dg/4nj/FKb2et71dXILDcLIcuLX+L56m2+2Bwq1E 5/JuZSJmBJBbH1Zmkz0GhKIh0LXODgnML2z5S0dehYjbpau0Bkp3duBRZ1EdaAFrbcOn 1TVYFOOVGrsW07+M2TmMXwxEHwmfkkXuxR/eo=

I don't think there's a big reason not to do it, it's just that
probably few people have felt a need for it so far. For most tasks,
solve() is a better tool than inverse().
How big is the speed difference between triangular.solve(rhs) and
triangular*rhs ?
If you would like this functionality to be implemented, you could file
a bug, explain your use case, and give a simple benchmark result
showing a performance advantage for triangular*rhs over
triangular.solve(rhs).
Cheers,
Benoit
2011/10/31 Douglas Bates <bates@xxxxxxxxxxxxx>:
> The TriangularView class has solve methods but not an explicit inverse
> method, which could be as simple as
> solve(DenseMatrixType::Identity(cols(), cols())) (at least I think
> that is how one would write the general form, but I'm not sure). If
> one wanted to get fancier there may be a slight advantage in taking
> account of the triangularity of the returned value and solving reduced
> linear systems for each column, as is done in the Lapack *trtri
> subroutines.
>
> Has such a method been considered, or even implemented and I have
> somehow managed to overlook it?
>
>
>