calculation of a sparse matrix

I have a very large matrix, with 3,26,041 rows and columns. I tried doing the simple sparse function for calculating it. But it returns an output as "out of memory" . I have a 64 bit system with 4gb ram. Is there anyway that i can calculate it or do i need a better system ? Can anyone please help, I am new to matlab and hence require urgent help..

[the cyclist]

Can you give a hint at the code your wrote?

Do you know roughly now many non-zero element in the array?

[Sushrut Naik]

m = floor((s^2*10^3)/b);

n = floor((d^2*10^3)/b);

A = zeros(m*n,m*n);

index_ij = 0; % initilising the value of variable index_ij as zero

for i = 1:1:m

for j = 1:1:n

index_ij = index_ij+1;

index_kl = 0; % initilising the value of variable index_kl as zero

for k = 1:1:m

for l = 1:1:n

index_kl = index_kl+1;

%xij=[(i-1)+.5]*delta_x;

%yij=[(j-1)+.5]*delta_y;

%xkl=[(k-1)+.5]*delta_x;

%ykl=[(l-1)+.5]*delta_y;

%A(index_ij,index_kl) = [Y_s/(2*pi)]*exp[-[[(a(i,j)-b(k,l))^2+(c(i,j)-d(k,l))^2]/(2*sigma^2)]];

A(index_ij,index_kl) = Y_s/(2*pi)*exp(-(((i-k)*delta_x)^2+((j-l)*delta_y)^2)/(2*sigma_s^2));

% value of the constant terms

% z(k,l) = e(k,l) * 2;

end

end

end

end

the code is as above. i cannot approximate the number of non zero elements, since the matrix A that i am calculating is not being shown. I wish to get the sparse of the matrix A in the above code

Answer 1

[the cyclist]

Sorry if I am saying what you already know, but the way you have defined A, it is decidedly non-sparse. You should not use the pre-allocation to zeros.

You should initialize the matrix A using the "sparse" command, which will define the indexing into A, without allocating memory for the elements. Then you can fill it in with your loops. Hopefully, there are few enough non-zero elements in the matrix to fit into memory.

[Sushrut Naik]

so you mean to say that instead of declaring it as zeros in the start i should declare it as a sparse matrix. I am new to matlab, probably the questions might be pretty basic.

What i intend to do is get the sparse matrix for the matrix A that i would get through the code listed above. But it prompts me saying that it is out of memory. It surely has zeros in it. The main values are observed along the diagonal of the matrix. I initially had 1600 x 1600 matrix, but then i added another constraint which made it swell to the size of 3,26,041 and now i wish to have a sparse matrix of the same.

[the cyclist]

Yes, I mean you should declare A as sparse instead of preallocating with zeros.

The matrix A = zeros(1000,1000) takes up 8,000,000 bytes. It will remain that size no matter how many non-zero elements you fill in. (The zeros and non-zeros take up the same memory.)

The matrix A = sparse(1000,1000) takes up 8,024 bytes; then it takes up more memory for each element you add to it. If final matrix is truly sparse (i.e. very few non-zeros), then it uses far less memory when you fill it.

The essence is that if you define it as sparse, then MATLAB doesn't waste memory storing all those zeros, which it ordinarily would if it was a (default) double-precision array.

[As a side comment, it would be good if you accept this answer, if you feel it helped, so that future questioners might find it more easily.]

[Sushrut Naik]

hey thanks..sorry for the delay...i will check it out.

thank you once again..