Thread Subject: sparse matrix assignment problem

I have to create a sparse matrix Y using only some of the
elements from sparse matrices K and M.

The elements I want to include in Y are stored in the array
"freedofs".

I am doing the following to assemble Y:

Y(freedofs,freedofs)=K(freedofs,freedofs)-(omega)^2*M(freedofs,freedofs)

This instantly gives me an "Out of Memory" error.

I am able to see what the entries of the matrix should be by
evaluating

K(freedofs,freedofs)-(omega)^2*M(freedofs,freedofs)

So, apparently, it's the assignment to Y which causes the
out of memory error.

Also, I am easily able to make the Y matrix if I include all
elements of K and M by doing

Y = K-(omega)^2*M


What might be the problem in assigning values to certain
locations of Y by using Y(freedofs,freedofs).

 

"Deepak " <deepak.trivedi@gmail.com> wrote in message
<fjh0gm$f5f$1@fred.mathworks.com>...

>
> What might be the problem in assigning values to certain
> locations of Y by using Y(freedofs,freedofs).
>
>

This could be due to an overflow of linear indices. You
should absolutely avoid to program in such way that MATLAB
must use linear indice that is larger than intmax()
(=2147483647).

n=50000;
i=1:n;
j=1:n;
M=sparse(i,j,i);

>> M(double(intmax))

ans =

     0

>> M(double(intmax)+1)
??? Maximum variable size allowed by the program is exceeded.

BTW, are you working on EM integral operators?

Bruno

"Deepak " <deepak.trivedi@gmail.com> wrote in message
<fjh0gm$f5f$1@fred.mathworks.com>...
> I have to create a sparse matrix Y using only some of the
> elements from sparse matrices K and M.
>
> The elements I want to include in Y are stored in the
array
> "freedofs".
>
> I am doing the following to assemble Y:
>
> Y(freedofs,freedofs)=K(freedofs,freedofs)-(omega)^2*M
(freedofs,freedofs)
>
> This instantly gives me an "Out of Memory" error.
>
> I am able to see what the entries of the matrix should be
by
> evaluating
>
> K(freedofs,freedofs)-(omega)^2*M(freedofs,freedofs)
>
> So, apparently, it's the assignment to Y which causes the
> out of memory error.
>
> Also, I am easily able to make the Y matrix if I include
all
> elements of K and M by doing
>
> Y = K-(omega)^2*M
>
>
> What might be the problem in assigning values to certain
> locations of Y by using Y(freedofs,freedofs).
>
>

I think the problem is in your use of indices. For
instance, I suspect that you think the following will give
you free numbers:

  A = rand(5) ;
  ind = 1:3 ;
  A(ind,ind)

which does not! Try to use linear indexing, using SUB2IND:

tempind = sub2ind(size(A),ind,ind) ;
A(tempind)

hth
Jos

Thanks Jos,

I don't want free indices. In fact, I want to do exactly the
same thing that you showed in your first example, i.e.,

> A = rand(5) ;
> ind = 1:3 ;
> A(ind,ind)

Should indeed produce a 3x3 matrix A. Similarly, if I do

ind = [1 2 5]
A(ind,ind)

I intend to get a 3x3 matrix containing elements that are on
both 1st, 2nd and 5th rows and columns.

One solution to the problem is of course to use loops, but
that would considerably slow everything down.

Any other ideas?

Regards
Deepak



"Jos " <DELjos@jasenDEL.nl> wrote in message
<fjhltu$7jv$1@fred.mathworks.com>...
> "Deepak " <deepak.trivedi@gmail.com> wrote in message
> <fjh0gm$f5f$1@fred.mathworks.com>...
> > I have to create a sparse matrix Y using only some of the
> > elements from sparse matrices K and M.
> >
> > The elements I want to include in Y are stored in the
> array
> > "freedofs".
> >
> > I am doing the following to assemble Y:
> >
> > Y(freedofs,freedofs)=K(freedofs,freedofs)-(omega)^2*M
> (freedofs,freedofs)
> >
> > This instantly gives me an "Out of Memory" error.
> >
> > I am able to see what the entries of the matrix should be
> by
> > evaluating
> >
> > K(freedofs,freedofs)-(omega)^2*M(freedofs,freedofs)
> >
> > So, apparently, it's the assignment to Y which causes the
> > out of memory error.
> >
> > Also, I am easily able to make the Y matrix if I include
> all
> > elements of K and M by doing
> >
> > Y = K-(omega)^2*M
> >
> >
> > What might be the problem in assigning values to certain
> > locations of Y by using Y(freedofs,freedofs).
> >
> >
>
> I think the problem is in your use of indices. For
> instance, I suspect that you think the following will give
> you free numbers:
>
> A = rand(5) ;
> ind = 1:3 ;
> A(ind,ind)
>
> which does not! Try to use linear indexing, using SUB2IND:
>
> tempind = sub2ind(size(A),ind,ind) ;
> A(tempind)
>
> hth
> Jos

"Jos " <DELjos@jasenDEL.nl> wrote in message
<fjhltu$7jv$1@fred.mathworks.com>...


> I think the problem is in your use of indices. For
> instance, I suspect that you think the following will give
> you free numbers:
>
> A = rand(5) ;
> ind = 1:3 ;
> A(ind,ind)
>
> which does not! Try to use linear indexing, using SUB2IND:
>
> tempind = sub2ind(size(A),ind,ind) ;
> A(tempind)
>

Yes, it seems the indexes are not used as OP intended
(thought the intention was not spelled out explicitly). But
I would still suggest to avoid using sub2ind() all together
for sparse matrix for the very reason I stated above. MATLAB
provides enough built-in functions to manipulate comfortably
sparse matrix (nnz, find with two outputs, sparse function,
union with optional 'rows' argument, etc...) without sub2ind().

Bruno

"Deepak " <deepak.trivedi@gmail.com> wrote in message
<fjhnup$mem$1@fred.mathworks.com>...
>
> I don't want free indices. In fact, I want to do exactly the
> same thing that you showed in your first example, i.e.,
>
> > A = rand(5) ;
> > ind = 1:3 ;
> > A(ind,ind)
>

OK! Now it's clear.

May be this would do:

[m n]=size(K);

Ysubs=K(freedofs,freedofs)-(omega)^2*M(freedofs,freedofs);
[i j]=find(Ysubs);
Y=sparse(freedofs(i),freedofs(j),nonzeros(Ysubs), m, n);

Bruno