Altair RISE

A program to recognize and reward our most engaged community members

Nominate Yourself Now!

Is there guidance on when and how one can combine simulations with varying excitations?

Hello,

I am currently planing a big simulation that needs to be repeated several times with different excitation sources(voltage sources for farfield calculations, mostly).

Is there documentation on what setups and settings to use to minimize total simulation time?

I assume that if I setup the problem as different configurations in a single model it would compute most efficiently, or are there settings I can change to improve things?

Similarly, are the settings or options different between pure MoM, MLFMM and ACA?

Find more posts tagged with

Feko

Accepted answers

All comments

Zeina

Hi @tobias_from_edag ,

You can either use multiple configs, each with a different excitation and run the model, or you can create separate models, and the run a .bat file on your machine that calls runfeko with each model to automate the runs instead of manually running them.

What do you mean exactly by the different settings ofr MoM, MLFMM and ACA? In which context?

Best,

Zeina

tobias_from_edag

Before your answer I was quite certain the the method with several models is fully unable to reuse intermediate results (like matrices or LU steps). How do I set things up that those work correctly, it does not reuse results for the default settigns on my machine.

What do you mean exactly by the different settings ofr MoM, MLFMM and ACA? In which context?

Are the settings I need to set to reuse the maximum amount of intermediate results different for the different solving methods, or are they different?

Zeina

Indeed, the method with multiple models does not reuse any previous result. Each model is run from scratch. It only allows the user to automate the run of models without having to manually run them separately.

There are no settings to reused intermediate results when you switch solvers (even if you have the same excitation, which you do not). Each model/solver will run independently.

tobias_from_edag

My main question was if there exists any kind of Documentation or Guidance on which options and setups and solvers can all which reuse of intermediate results. Can you please explicitly confirm that that doesn't exist?

I am not trying to reuse results between different solvers. I am asking if there is a difference in reuse capabilities between different MoM variants. And what those differences and reuse options are.

Can you please confirm that if I run a model with different configurations and the only change is switching 2 ports between load and source, there are is no possibility of reusing intermediate calculation results. That is each configuration is calculated separately.

Zeina

There is no such documentation.

Ok, it is clearer now that you are asking for each solver separately. I will check if there is a possible acceleration for the case you have mentioned of changing ports. Will get back to you shortly.

Zeina

@tobias_from_edag ,

All solutions don’t have to recompute the matrix and the preconditionner (for iterative solution) if the loads are global and only the sources are changed per solution configuration. For MoM this will result in only a back substitution and for iterative solutions like MLFMM this will only result in an iterative solution per change in source.

However, when changing the loads, the matrix needs to recomputed. For MoM using the Numerical Green’s function can help reduce the computational cost significantly when the source and load ports are part of the dynamic part of the matrix. (you can check this in the guide as well as this video for further details).

Zeina

@tobias_from_edag , what is the number of unknowns in your model? would MLFMM be required or would it be possible to use the MoM?

tobias_from_edag

Thanks.

what happens to the results in the case where the source overlapps a load?

That is I have 4 (or more) ports, each with a globally defined load of 50Ohm. And I have 4 configurations each of which assigns a voltage source to 1 port each. Will the calculation results be very different from the results I would get if I had locally defined loads and removed the load that overlaps the port?

The problem is very large. I expect to compute most of the frequency range with MLFMM, some low frequencies with MoM to avoid MLFMM's convergence problems, and some very high frequncies with a hybrid MLFMM/optical method.

NGF, might be useful for my problem. But I have no experience with it.

Zeina

"Will the calculation results be very different from the results I would get if I had locally defined loads and removed the load that overlaps the port? "
—> Yes. The source impedance and the antenna gain will be different with an extra load defined at the same port as the source. You can test this with a simple dipole model, where you define an extra load to the source port. The effect will be visible when you compare the antenna gain for instance. This is because there will be an extra power loss added to the gain due to the additional load.

NGF would be interesting to use for MoM runs. The references I shared should help you understand the workflow.

tobias_from_edag

So, the power emitted from the port will be split between going to the 3D and into the load? There should be some way to correct for this, I suppose.

Also, does that mean that Directivity calculations would be unaffected?

Zeina

You can correct the power loss in the gain by scripting it in PostFeko.

The directivity would be the same in both cases, so it would not be affected. Again, you can test all this using a simple dipole model.

Ovando_RBB

Hey there,

What solver are you using? this sounds like a good job for Flux

tobias_from_edag

I just tried to test such a model with identical loads and a configuration for each source. With MLFMM it still took almost an hour to compute the second source, which was only a marginal speedup.

Do you have any tips to check if something is wrong, or how to increase the speedup?

Zeina

@tobias_from_edag , can you please share the model you used to test, or at least the .out? You can also check the steps of the solution from the .out file to see how much time each took. This will allow you to compare the difference in runtime between the configs.

tobias_from_edag

2antennen_6000m_mlfmm_globloads.out

I could not see any indication that it was reusing any parts at all. I can't share the model, but I will try to repeat the test with a sharable model if you can't recommend anything to try first.

Zeina

So far you gained the precon computation time and about 4.1 mins off the solution runtime. I don't see any other way to further accelerate the computations for now

Zeina

@tobias_from_edag , I will be out of office in the coming weeks. If you need support meanwhile I recommend you create a support ticket (case) so that the teams helps out.

Zeina

@tobias_from_edag, I will be out of office in the coming weeks. If you need support meanwhile I recommend you create a support ticket (case) so that the teams helps out.