Re: Orthogonazlie with Method->"Householder"
- To: mathgroup at smc.vnet.net
- Subject: [mg97791] Re: Orthogonazlie with Method->"Householder"
- From: ADL <alberto.dilullo at tiscali.it>
- Date: Sun, 22 Mar 2009 05:44:58 -0500 (EST)
- References: <gpvh97$grj$1@smc.vnet.net>
Maris, I did the following computations also using the Rationalize
version of your matrix:
In[1]:= ClearAll[M, oM, MR, oMR];
In[2]:= M = {...}; (* deleted from the post *)
In[3]:= MR = Rationalize[M, 10^(-16)];
oMR = Orthogonalize[MR];
In[5]:= methods = {"GramSchmidt", "ModifiedGramSchmidt",
"Householder", "Reorthogonalization"};
In[6]:= (oM[#] = Orthogonalize[M, Method -> #]) & /@ methods;
In[7]:= MatrixRank[oMR]
Out[7]= 6
In[8]:= MatrixRank[oM[#]] & /@ methods
Out[8]= {6,6,12,6}
In[9]:= Chop@Max@Abs@(oM[#] - oMR) & /@ methods
Out[9]= {0,0,1.23961912632296,0}
In[10]:= $Version
Out[10]= 7.0 for Microsoft Windows (32-bit) (February 18, 2009)
>From this, I would say that the issue is not accuracy but the fact
that "Householder" is not working properly. If it is a feature, it is
a discount feature: you get 12 at the price of 6.
What do you think about starting the famous "bugs and caveats" site
with this post?
ADL
Maris Ozols ha scritto:
> I have a large numeric matrix M (see below), whose rank is 6. Clearly,
> the rank of Orthogonalize[M] must be the same (in particular, it
> should not increase). However, if I execute the following code on my
> matrix M:
>
> MatrixRank@M
> MatrixRank@Orthogonalize[M]
> MatrixRank@Orthogonalize[M,Method->"Householder"]
>
> I get 6, 6, and 12. So it appears to me that Householder's method has
> doubled the rank of M. I am using Mathematica 7.0 in Linux.
>
> Is this a bug or a feature?
>
> ~Maris Ozols~
- Follow-Ups:
- Re: Re: Orthogonazlie with Method->"Householder"
- From: Maris Ozols <marozols@gmail.com>
- Re: Re: Orthogonazlie with Method->"Householder"