Sparse knonecker product of large and several matrices

Hi,

I'm trying to create a sparse kronecker product without actually generating the kron product and then generating the sparse matrix. Is there a way to do this? For example

E=eye(100);
I=ones(20);
K=sparse(kron(E,I));

Right now the code first generates the kron product and then keeps on the information of non-zero elements. As I need to do this lots of time because I need to generate and keep in memory different K matrices (K1, K2, ... K200, for E1,...E200, and I1,...I200), the generation of the kron product uses all my memory. Is there anyway I could go from the E's and I's to the K's without generating the full matrices?

Thanks,

Answer 1

[Matt J]

   E=speye(100);
   I=ones(20);
   K=kron(E,I);

[Matt J]

I wonder if you even need K in any kind of matrix form, whether sparse or full. See this Kronecker product manipulation tool for an alternative

http://www.mathworks.com/matlabcentral/fileexchange/25969-efficient-object-oriented-kronecker-product-manipulation

Below, I construct a KronProd operator Ko equivalent to your K matrix. Note the difference in memory consumption. Ko is less than 1 KB.

 >> e=ones(20,1);  Ko=KronProd({e,e.',1},1:3,[nan,nan,100]); 

 >> Whos K Ko

    Name   Size          Kilobytes     Class      Attributes

    K      2000x2000           641     double     sparse    
    Ko     2000x2000         0.896     KronProd

[Fernando]

Thanks!

[Fernando]

Thanks, I'll look at this.