| Re: [eigen] <xxx>PivLU::solve( diagonal matrices ) |
[ Thread Index |
Date Index
| More lists.tuxfamily.org/eigen Archives
]
- To: eigen@xxxxxxxxxxxxxxxxxxx
- Subject: Re: [eigen] <xxx>PivLU::solve( diagonal matrices )
- From: Benoit Jacob <jacob.benoit.1@xxxxxxxxx>
- Date: Tue, 1 Jun 2010 20:55:06 -0400
- Dkim-signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=gmail.com; s=gamma; h=domainkey-signature:mime-version:received:received:in-reply-to :references:date:message-id:subject:from:to:content-type :content-transfer-encoding; bh=t5hA/6Bv9gv0C4ks5Vly/aH5jtIxUNGNOV9FqY0++xc=; b=GrlFQJm7TePmbeni1AJ/4bxlnd0aRjt4FFFasMKH90jN1qwGbdpN5U/k1ucLITWQ1J YzHjjglcOJhu9lzXe3WehIhgQ2MpPYoDfD37bqZKdRDOV/lhmTHnsuc73eagQNwT+niu yBL3gAK217pJ1XFfIwJiz2ozf1i2C11Ozem2E=
- Domainkey-signature: a=rsa-sha1; c=nofws; d=gmail.com; s=gamma; h=mime-version:in-reply-to:references:date:message-id:subject:from:to :content-type:content-transfer-encoding; b=fOZbSLmSqSblbwybW+3fP+J4etLLQynLP+0wGoOTgugeeoohBU+KsKEtjHO8UtELKG bIxGjUva9h+XV/dF3UsfUaXrnPeLF8GxOIyJpSGP1kML/vkuVjJXzXjWt7ObVGzpOiKP vAOwBliaLZ/Ng8AQPTpicrfO1nkhwYy+Ze46w=
2010/6/1 Manoj Rajagopalan <rmanoj@xxxxxxxxx>:
> Hi,
>
> The <xxx>PivLU::solve() member seems to be implemented only for MatrixBase
> descendants.
Right.
> My PDE solver code requires calling solve on a diagonal matrix.
> One way I can think of achieving this is to call <xxx>PivLU::inverse() *
> diag-matrix. Is there a better way to achieve this?
Right now, slightly better:
lu.solve(diagmatrix.toDenseMatrix());
However, this is still not optimal. What would be really efficient
would be an optimized triangular inverse(). Then we could compute L^-1
efficiently and then use our triangular solver to compute U^-1 L^-1 =
(LU)^-1.
The usefulness of this would be far-reaching: in particular it would
give a more optimized inverse().
>
> Would an implementation of solve for DiagonalBase<> descendants as args be
> desirable, esp. since the argument is passed as const& ?
What is desirable is an optimized triangular inversion function, as
explained above :)
That would be VERY desirable (provided that it runs faster than our
current solution of applying our triangular solver to the identity
matrix).
Cheers,
Benoit
>
> Thanks,
> Manoj
>
>
>