Improve efficiency of AssembleFEMatrices?

Dear all,

I am using the PDE Toolbox to perform a FEM simulation. I am using the following toolbox function to aquire the matrices I am after:

FEM = assembleFEMatrices(model,'nullspace') 

However, the only matrices I require are the global stiffness matrix (FEM.Kc) and nodal force vector (FEM.Fc).

The issue is that the assembleFEMatrices function is quite costly. I was wondering if there is a way improve the efficiency of this? Could the stiffness matrix and force vector be calculated another way, i.e. could they be calculated from the results of the 'solve' function that is less time consuming:

results = solve(model);

Or if it is possible to assemble ONLY the matrices you desire to reduce the computational expense?

Thank you for any advices/solutions that can be offered

Answer 1

[Ravi Kumar]

If you are using R2020b or newer version of MATLAB, you should be able to specify required matrices as input:

https://www.mathworks.com/help/pde/ug/assemblefematrices.html

Regards,

Ravi

[Elliot Bontoft]

Hi Ravi,

Thanks for pointing this out, i had actually completely missed this. I have updated my MATLAB release and tried to run this.

The issue I am finding is that I want to specify both the BC_method ('nullspace' or 'stiff-spring') and the matrices ('Kc' and 'Fc') arguments.

It seems when you want to specifiy the matrices you must use 'none' for the BC method.

[Elliot Bontoft]

Hi all,

Ravi was right and this was possible through using specified requested matrices in the input. You just have to manually impose the boundary conditions to build the stiffness matrix and force vector, as follows:

FEM = assembleFEMatrices(model,'HKAQFG');
[FEM.B,Or] = pdenullorth(FEM.H);  % compute the nullspace of columns of H
FEM_Fc = FEM.B'*(FEM.F + FEM.G);
FEM_Kc = FEM.B'*(FEM.K + FEM.A + FEM.Q)*FEM.B;

This is notably faster than using

FEM = assembleFEMatrices(model,'nullspace')

to access the FEM.Fc and FEM.Kc matrices

[Ravi Kumar]

Eaxctly! Selecting a subset of matrices or specifying BC appliction option are sort of orthogonal. You need to do the way you are doing. I would also add, you can assemble some of these matrices once and some that have some variation over time or solution repeatedly.