Matirx inverse

Matirx inverse

[Sushmita Roy]

Hi,
I need to perform an inverse operation on a matrix
and I am using the  gsl_linalg_LU_invert function
to do so. However, this is taking really long
as the the size of my matrix is 6500X6500.
Is there a faster way to do this?

Thank you,
Sushmita

Re: Matirx inverse

[Joerg Frochte]

Hi, 

* Sushmita Roy <sroy@cs.unm.edu> [031124 17:42] schrieb:

> I need to perform an inverse operation on a matrix
> and I am using the  gsl_linalg_LU_invert function
> to do so. However, this is taking really long
> as the the size of my matrix is 6500X6500.
> Is there a faster way to do this?

I do not know any way to speed up in gsl, but 
maybe this information can help you anyway. 

Is this matrix dense or sparse ?

If it is sparse you could use SuperLU
 
http://crd.lbl.gov/~xiaoye/SuperLU/

If it is dense, you can try to solve a linear 
system of equations 6500 times to get the rows of 
A^-1 by using e.g. the optimized version of 
lapack provided by atlas. 

http://math-atlas.sourceforge.net/

Good luck,

Joerg

--
http://www.uni-essen.de/ingmath/people/frochte.html

complex lInear algebra

[Z F]

Hello,


GSL provides vectors and matrices in several types: long, int, double,
complex, long double and long complex, etc.

However, it seems to me that the only operations supported by gsl on
these objects are +-*. While these are the only operations needed to
define algebra, the linear algebra section does not seem to support
anything other than double matrix.

I am wondering if there is a special reason for this. Without linear
algebra operations, allocating complex matrices is not very
interesting...

Am I missing something?

Thank you

Lazar

__________________________________
Do you Yahoo!?
Free Pop-Up Blocker - Get it now
http://companion.yahoo.com/

Re: complex lInear algebra

[Brian Gough]

Z F writes:
 > I am wondering if there is a special reason for this. Without linear
 > algebra operations, allocating complex matrices is not very
 > interesting...
 > 
 > Am I missing something?

See here  http://www.gnu.org/software/gsl/#TOCvolunteer
with 'complex linear algebra' added to the list

-- 
Brian Gough

Re: complex lInear algebra

[Brian Gough]

Z F writes:
 > OK, me and my brother will try to work on that. We will start with
 > the BLAS support. We need some convention for the naming of the
 > functions. The current one is:
 > S  single real
 > D  double real
 > C  single complex
 > Z  double complex
 > We propose to add
 > K  long double real
 > Q  long double complex

These are not part of the BLAS standard -- so I would not want to add
them to gsl itself.

The sort of algorithms I am interested in would be complex QR and
complex SVD.  The "GSL Design Document" at
http://sources.redhat.com/gsl/ contains information on contributing.

regards,

-- 
Brian Gough

p.s. Can you post any development related messages to the list
gsl-discuss@sources.redhat.com rather than me directly -- thanks.

Re: complex lInear algebra

[Z F]

Hello Brian,

Well, the situation is the following.

For our research we need long complex SVD. To do that we need long
complex BLAS and complex SVD.

We do not see the logic in not expanding the BLAS standard to handle
longs and thus the proposition holds:

Add data types to BLAS
K  long double real
Q  long double complex

and the functions which operate on them as the first step. Second step
is to program complex SVD algorithm and the rest of them.

Since we have to do it anyway, we would like this to be available to
others too (and ourselves in future).

Roman and Lazar

--- Brian Gough <bjg@network-theory.co.uk> wrote:
> Z F writes:
>  > OK, me and my brother will try to work on that. We will start with
>  > the BLAS support. We need some convention for the naming of the
>  > functions. The current one is:
>  > S  single real
>  > D  double real
>  > C  single complex
>  > Z  double complex
>  > We propose to add
>  > K  long double real
>  > Q  long double complex
> 
> These are not part of the BLAS standard -- so I would not want to add
> them to gsl itself.
> 
> The sort of algorithms I am interested in would be complex QR and
> complex SVD.  The "GSL Design Document" at
> http://sources.redhat.com/gsl/ contains information on contributing.
> 
> regards,
> 
> -- 
> Brian Gough
> 
> p.s. Can you post any development related messages to the list
> gsl-discuss@sources.redhat.com rather than me directly -- thanks.


__________________________________
Do you Yahoo!?
New Yahoo! Photos - easier uploading and sharing.
http://photos.yahoo.com/

Re: complex lInear algebra

[Z F]

Hello Brian,

Well, the situation is the following.

For our research we need long complex SVD. To do that we need long
complex BLAS and complex SVD.

We do not see the logic in not expanding the BLAS standard to handle
longs and thus the proposition holds:

Add data types to BLAS
K  long double real
Q  long double complex

and the functions which operate on them as the first step. Second step
is to program complex SVD algorithm and the rest of them.

Since we have to do it anyway, we would like this to be available to
others too (and ourselves in future).

Roman and Lazar

--- Brian Gough <bjg@network-theory.co.uk> wrote:
> Z F writes:
>  > OK, me and my brother will try to work on that. We will start with
>  > the BLAS support. We need some convention for the naming of the
>  > functions. The current one is:
>  > S  single real
>  > D  double real
>  > C  single complex
>  > Z  double complex
>  > We propose to add
>  > K  long double real
>  > Q  long double complex
> 
> These are not part of the BLAS standard -- so I would not want to add
> them to gsl itself.
> 
> The sort of algorithms I am interested in would be complex QR and
> complex SVD.  The "GSL Design Document" at
> http://sources.redhat.com/gsl/ contains information on contributing.
> 
> regards,
> 
> -- 
> Brian Gough
> 
> p.s. Can you post any development related messages to the list
> gsl-discuss@sources.redhat.com rather than me directly -- thanks.


__________________________________
Do you Yahoo!?
New Yahoo! Photos - easier uploading and sharing.
http://photos.yahoo.com/