FAQFAQ   SearchSearch   MemberlistMemberlist   UsergroupsUsergroups 
 ProfileProfile   PreferencesPreferences   Log in to check your private messagesLog in to check your private messages   Log inLog in 
Forum index » Science and Technology » Math » num-analysis
Power Method and negative eigenvalues
Post new topic   Reply to topic Page 1 of 1 [8 Posts] View previous topic :: View next topic
Author Message
Skunk
science forum beginner


Joined: 19 Jul 2006
Posts: 2

PostPosted: Wed Jul 19, 2006 10:19 pm    Post subject: Power Method and negative eigenvalues Reply with quote

Hi,

I implemented the power method algorithm in C++ in order to find the
leading eigenvalue of a real square matrix and its associated eigenvector.
My problem is that this method compute the eigenvalue with the largest
*absolute* value and that i would it to compute the largest *positive*
eigenvalue.

Is there a possiblity to adapt the power method to find the largest
positive eigenvalue ? Any ideas i can explore or should i use another
algorithm ?

Thank you very much !
Back to top
Roy Stogner
science forum beginner


Joined: 13 Jun 2005
Posts: 38

PostPosted: Thu Jul 20, 2006 3:22 am    Post subject: Re: Power Method and negative eigenvalues Reply with quote

On Thu, 20 Jul 2006 00:19:11 +0200, Skunk wrote:

Quote:
Is there a possiblity to adapt the power method to find the largest
positive eigenvalue ? Any ideas i can explore or should i use another
algorithm ?

The first idea that comes to mind: if the largest magnitude negative
eigenvalue of A is -g, can you use the power method to find the largest
magnitude eigenvalue of (A + gI) then subtract g to get the corresponding
eigenvalue of A?
---
Roy Stogner
Back to top
Gottfried Helms
science forum Guru


Joined: 24 Mar 2005
Posts: 301

PostPosted: Thu Jul 20, 2006 7:27 am    Post subject: Re: Power Method and negative eigenvalues Reply with quote

Am 20.07.2006 00:19 schrieb Skunk:
Quote:
Hi,

I implemented the power method algorithm in C++ in order to find the
leading eigenvalue of a real square matrix and its associated eigenvector.
My problem is that this method compute the eigenvalue with the largest
*absolute* value and that i would it to compute the largest *positive*
eigenvalue.

Is there a possiblity to adapt the power method to find the largest
positive eigenvalue ? Any ideas i can explore or should i use another
algorithm ?

Thank you very much !
Would it be possible to compute the matrix-exponential EA of A

first? Then the negative eigenvalues of A transform to the
lowest eigenvalues of EA. Then find the highest eigenvalue
of EA with the power-method and take its log.

Just sketch, no warranty, that this is efficient ...

Gottfried Helms
Back to top
Jeremy Watts
science forum Guru Wannabe


Joined: 24 Mar 2005
Posts: 239

PostPosted: Thu Jul 20, 2006 8:47 am    Post subject: Re: Power Method and negative eigenvalues Reply with quote

"Skunk" <skunk@ag.com> wrote in message
news:44beafcc$0$12612$5402220f@news.sunrise.ch...
Quote:
Hi,

I implemented the power method algorithm in C++ in order to find the
leading eigenvalue of a real square matrix and its associated eigenvector.
My problem is that this method compute the eigenvalue with the largest
*absolute* value and that i would it to compute the largest *positive*
eigenvalue.

Is there a possiblity to adapt the power method to find the largest
positive eigenvalue ? Any ideas i can explore or should i use another
algorithm ?

Thank you very much !

why not just use the QR algorithm to find the whole spectrum and then search
it to find the largest positive eigenvalue?
Back to top
iandjmsmith@aol.com
science forum beginner


Joined: 13 Sep 2005
Posts: 15

PostPosted: Thu Jul 20, 2006 8:48 am    Post subject: Re: Power Method and negative eigenvalues Reply with quote

Skunk wrote:

Quote:
Hi,

I implemented the power method algorithm in C++ in order to find the
leading eigenvalue of a real square matrix and its associated eigenvector.
My problem is that this method compute the eigenvalue with the largest
*absolute* value and that i would it to compute the largest *positive*
eigenvalue.

Is there a possiblity to adapt the power method to find the largest
positive eigenvalue ? Any ideas i can explore or should i use another
algorithm ?

Thank you very much !

In
http://groups.google.co.uk/group/sci.math.num-analysis/browse_frm/thread/81322353885a6d4d
there is some discussion on finding the largest positive eigen value of
a real symmetric matrix.

Ian Smith
Back to top
Skunk
science forum beginner


Joined: 19 Jul 2006
Posts: 2

PostPosted: Thu Jul 20, 2006 10:29 am    Post subject: Re: Power Method and negative eigenvalues Reply with quote

Roy Stogner wrote:
Quote:
On Thu, 20 Jul 2006 00:19:11 +0200, Skunk wrote:

Is there a possiblity to adapt the power method to find the largest
positive eigenvalue ? Any ideas i can explore or should i use another
algorithm ?

The first idea that comes to mind: if the largest magnitude negative
eigenvalue of A is -g, can you use the power method to find the largest
magnitude eigenvalue of (A + gI) then subtract g to get the corresponding
eigenvalue of A?
---
Roy Stogner

It works ! Simple and effective, many thanks ! Smile
Back to top
Peter Spellucci
science forum Guru


Joined: 29 Apr 2005
Posts: 702

PostPosted: Thu Jul 20, 2006 11:14 am    Post subject: Re: Power Method and negative eigenvalues Reply with quote

In article <44beafcc$0$12612$5402220f@news.sunrise.ch>,
Skunk <skunk@ag.com> writes:
Quote:
Hi,

I implemented the power method algorithm in C++ in order to find the
leading eigenvalue of a real square matrix and its associated eigenvector.
My problem is that this method compute the eigenvalue with the largest
*absolute* value and that i would it to compute the largest *positive*
eigenvalue.

Is there a possiblity to adapt the power method to find the largest
positive eigenvalue ? Any ideas i can explore or should i use another
algorithm ?

Thank you very much !

not necessarily the best idea but simple:

with a pure power method code: compute the maxnorm of A
max_i sum_j abs(a(i,j) = N
add N to the diagonal elements of A. Now there are no negative
eigenvalues and the largest positive one is shifted by N to the right.
this might slow down convergence .
or, with inverse iteration: subtract N, use inverse iteration with shift zero
to get the reciprocal of the absolute smallest eigenvalue of the shifted
i.e. 1/(lambda_max-N)
hth
peter
Back to top
Arnold Neumaier
science forum Guru


Joined: 24 Mar 2005
Posts: 379

PostPosted: Thu Jul 20, 2006 12:46 pm    Post subject: Re: Power Method and negative eigenvalues Reply with quote

Skunk wrote:
Quote:
Hi,

I implemented the power method algorithm in C++ in order to find the
leading eigenvalue of a real square matrix and its associated eigenvector.
My problem is that this method compute the eigenvalue with the largest
*absolute* value and that i would it to compute the largest *positive*
eigenvalue.

Is there a possiblity to adapt the power method to find the largest
positive eigenvalue ? Any ideas i can explore or should i use another
algorithm ?

Try the Lanczsos algorithm. It is much more efficient than the power method.

Arnold Neumaier
Back to top
Google

Back to top
Display posts from previous:   
Post new topic   Reply to topic Page 1 of 1 [8 Posts] View previous topic :: View next topic
The time now is Sat Nov 25, 2017 1:45 am | All times are GMT
Forum index » Science and Technology » Math » num-analysis
Jump to:  

Similar Topics
Topic Author Forum Replies Last Post
No new posts Base Conversion - Negative Base Tapas Bose num-analysis 0 Tue Aug 03, 2010 8:47 pm
No new posts Help in identifying a numerical method Don11135 num-analysis 2 Thu Jul 20, 2006 8:56 pm
No new posts troubles in determination of specific surface area(air pe... eos Chem 0 Thu Jul 20, 2006 10:05 am
No new posts troubles in determination of specific surface area(air pe... eos Chem 0 Thu Jul 20, 2006 10:02 am
No new posts possible to use Generalized Method of Moments for this pr... comtech Math 1 Thu Jul 20, 2006 12:49 am

Copyright © 2004-2005 DeniX Solutions SRL
Other DeniX Solutions sites: Electronics forum |  Medicine forum |  Unix/Linux blog |  Unix/Linux documentation |  Unix/Linux forums  |  send newsletters
 


Powered by phpBB © 2001, 2005 phpBB Group
[ Time: 0.0266s ][ Queries: 16 (0.0039s) ][ GZIP on - Debug on ]