Services & ResourcesWolfram ForumsMathGroup Archive

[Date Index] [Thread Index] [Author Index]

Re: Re: LinearSolve[m, b] is not equivalent to LinearSolve

• To: mathgroup at smc.vnet.net
• Subject: [mg84382] Re: [mg84356] Re: LinearSolve[m, b] is not equivalent to LinearSolve
• From: Daniel Lichtblau <danl at wolfram.com>
• Date: Thu, 20 Dec 2007 01:33:42 -0500 (EST)
• References: <fk73ks$65f$1@smc.vnet.net> <200712200505.AAA01775@smc.vnet.net>

mcmcclur at unca.edu wrote:
> On Dec 17, 7:18 pm, "Andrew Moylan" <andrew.j.moy... at gmail.com>
> wrote:
> 
>>1. LinearSolve[m, b] is not equivalent to LinearSolve[m][b].
> 
> 
> Well, I was definitely hoping someone might have clarified
> this interesting issue.  I haven't totally figured it out
> myself, but I've got a few observations.  My basic guess is
> that LinearSolve[m][b] introduces the possibility of
> compounded error via intermediate matrix factorizations that
> a direct LinearSolve[m,b] avoids.  Here's a simple example
> illustrating this.
> [...]

Strangely enough, the issue is that LinearSolve[m][b] gives a good 
result, whereas LinearSolve[m, b] gives garbage. I'm fairly certain this 
is simply a bug. It was filed as such, and will be investigated.

Daniel Lichtblau
Wolfram Research

• References:
  • Re: LinearSolve[m, b] is not equivalent to LinearSolve[m][b]
    • From: mcmcclur@unca.edu