Convergence in HFSS and out of memory problems

[nomee]

Convergence in HFSS.

Hello,
I have been using CST MWS for the simulations of Infinite Reflectarrays and now I have started using Ansoft HFSS comparisons or validation of my results.
Can anyone help me in understanding WHat exactly is convergence?
Also my total tetrahedra goes beyond 90k on 10 passes and when i set my deltaS=0.001, after 15 passes my system goes out of memory.(Although i have 12G ,Quad core System).
Can anyone please tell the possible reason for this?

 

[katko]

Convergence in HFSS.

You probably don't need deltaS that low. It defaults to .02, so unless you need some insane accuracy, you don't need .001. Try setting it at .01.

Convergence is just when the deltaS gets below that value. Try setting it to deltaS=.01 and then forcing it to do 2 or 3 consecutive converged passes or something rather than a really low deltaS for (I'm assuming) one pass. Did you change the refinement factor for each pass? HFSS blows up memory consumption if your geometry gets too electrically large really quickly.

The number of tetrahedra is vastly different depending on your solution frequency. I assume you've looked into the proper solution frequency some - if not, look at some of the documentation for what to set the solution frequency at. As mentioned above, the memory used shoots up with the electrical size of your simulation.

 

[nomee]

Re: Convergence in HFSS.

thanks katko,
you are right, 0.02 is enough for my case but i needed some study for deltaS as low as 0.001 and number of passes required.
How can I set consecutive converged passes? right now i am using 1 pass.
I am designing at 10GHz. should i not be setting solution frequeny to be 10GHz?

Added after 2 minutes:

I frogot to mention the frequecy sweep range is also 8GHz to 12GHz.

 

[katko]

Convergence in HFSS.

The number of converged passes setting is either in the solution setup dialogue or in the HFSS/solver settings. I don't have access/remember at the moment, but I'll verify which it is later. If you set the consecutive solved passes higher than the default of 1, often you'll get below .001 after a couple converged passes, but not always. Just depends on the problem. I always set it to at least 2.

Solution frequency sounds ok. It's all about electrical size of the entire problem, though, and HFSS may not be the most effective for this. Quick example:
I was doing a 14x14 microstrip patch array and we got some strange coupling in fabrication/prototyping, so we decided to simulate the entire array rather than use PBC/Master-slave. I had a PC with the same specs as yours and I was able to simulate the entire array using CST MWS, although it took a really long time. It was impossible to simulate the entire thing in HFSS due to the memory requirements for such a huge domain on my PC. HFSS just wasn't able to do this. We could've used a supercomputer setup to do it, though.

Also, how do you have the sweep set? Trying a fast sweep can kill the simulation if it requires too much memory, even when it actually finishes the solution at 10 GHz. If your PC can get through the initial simulation at 10 GHz, though, you can do a discrete step sweep and it should be able to handle that as long as you have enough hard drive space.