Prev: 3-D Plot Movie
Next: plot output from two for-loops in log-log scale
From: Marios Karaoulis on 4 Jun 2010 14:19
Hi,
I have a large matrix like k(20000,200000).
It is always square, symmetric and have tons of zeros (so i can make
it sparse).

this is part of the program


%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
for e=1:large_number

I=icon(1,e);
J=icon(2,e);
L=icon(3,e);

%%%%%%%%%%%%%%%%%%%%%%%%%%%
ke=[1 2 3;2 8 4;3 4 9]; %symmetric matrix, changes in every loop

%%%%%%%%%%%%%%%%


% Add [K^e] to [K]
for i=1:3
for j=1:3
k(icon(i,e),icon(j,e))=k(icon(i,e),icon(j,e))+ke(i,j);
end
end


end

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

if in my code i write
k=zeros(20000,20000)
then I get out of memory

if i use k=sparse(20000,20000), it takes forever to finish.

I am thinking of 2 solutions

1)
Since I know that k is symmetric, is any advise to store it using a
bandwidth technique (like Cuthill-McKee)? I have found a good example
in c which I can use (burnett, finite element theory), but then I need
the decomposition of this matrix.
In this book, there is a modified gaus-jordan elimination, but if I
write this algorithm in matlab it takes forever to calculate.

2) I am trying to store in every iteration the indexes of the matrix
like


%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
tmp1=1;
for e=1:large_number

I=icon(1,e);
J=icon(2,e);
L=icon(3,e);

%%%%%%%%%%%%%%%%%%%%%%%%%%%
ke=[1 2 3;2 8 4;3 4 9]; %symmetric matrix, changes in every loop

%%%%%%%%%%%%%%%%


% Add [K^e] to [K]
for i=1:3
for j=1:3
index1(tmp1)=icon(i,e);
index2(tmp1)=icon(j,e);
val(tmp1)=ke(i,j);
tmp1=tmp1+1;
end
end


end

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

how can I now create the sparse matrix k?
k=sparse(index1,index2,val,20000,20000);

This way in every element in, it will just store a value. I want to
store the value it has plus the value it already has.

Any suggestions?

 | 
Pages: 1
Prev: 3-D Plot Movie
Next: plot output from two for-loops in log-log scale