AN EFFICIENT WAY TO DO THE ASSEMBLY OF GLOBAL STIFFNESS MATRIX

[yosimzd2]

Hi all,
I develop a finite element code in java.
I looking for an efficiency algorithm to do the assembly process... i need to assemble the global matrix into sparse vector for using in minimum memory.
someone that develop a finite element code maybe encounter in a memory problem when he allocate the global stiffness matrix? if yes how can i overcome this problem?
thanks in advance

 

[GregLocock]

Check out wavefront analysis

Cheers

Greg Locock


New here? Try reading these, they might help FAQ731-376

http://eng-tips.com/market.cfm?

 

[yosimzd2]

what is the conection between wavefront analysis to assembly process?

 

[GregLocock]

your wavefront is controlled by the order in which you evaluate elements.

Cheers

Greg Locock


New here? Try reading these, they might help FAQ731-376

http://eng-tips.com/market.cfm?

 

[dirnad]

I've used a method that uses an additional 50-100% of the memory of the matrix itself. The matrix is sparse and each finite element is added to it one by one.

You'll need to define the sparsity pattern before using it. You can do that with a initial dummy "assembly" run which just builds a data structure representing where all the non-zeros are. I store this in an array of lists (variable length arrays). Each element in the array is one row, and each element in the list is the column number of a non-zero in that row. This probably wastes a lot of memory because of unused space in the lists so you can use a more efficient jagged array instead.