WaveSolver and Eigen Analysis

Numerical methods and mathematical models of Elmer
CrocoDuck
Posts: 64
Joined: 12 May 2016, 13:15
Antispam: Yes

Re: WaveSolver and Eigen Analysis

Post by CrocoDuck »

I see, that clarifies. Thanks.

As for your question:
raback wrote: 21 Aug 2020, 16:59 The same applies to "Pressure Velocity". That being said, I don't really recall what it would do in harmonic case. Does it play a role whether you have it or not? This is also taken care by the library.
I commented out and used a sort of "blank" boundary condition:

Code: Select all

Boundary Condition 1
  ! Pressure Velocity = Real 0
End
It delivered the same results.
CrocoDuck
Posts: 64
Joined: 12 May 2016, 13:15
Antispam: Yes

Re: WaveSolver and Eigen Analysis

Post by CrocoDuck »

Hey guys. If you are interested, I did a little accuracy study of the solution. I am quite satisfied. Results are here: https://computational-acoustics.gitlab. ... ed-part-2/.

Cheers!
raback
Site Admin
Posts: 3750
Joined: 22 Aug 2009, 11:57
Antispam: Yes
Location: Espoo, Finland
Contact:

Re: WaveSolver and Eigen Analysis

Post by raback »

Hi Croco

Nice work! This kind of diligent work is needed to push the code into new areas. And actually before your tests this was not possible since the wave equation hadn't been tested for eigen analysis.

A note on the output of eigenmodes. You will get the modes as timesteps in vtu file if you add in Simulation section "vtu: Eigen Analysis = Logical True". This will tell the vtu output to number the files according to eigenmodes. There could be iterative or time dependent problems where for each iteration we solve a slightly modified eigenvalue problem. So therefore this is not the default in output.

I don't remember whether the solver can deal with p-elements. However, you might want to test whether "Element = p:2" improves results.Certainly there is always a significant cost in increasing the element order but for wave problems going quadratic may pay off.

-Peter
CrocoDuck
Posts: 64
Joined: 12 May 2016, 13:15
Antispam: Yes

Re: WaveSolver and Eigen Analysis

Post by CrocoDuck »

Hi!
raback wrote: 13 Sep 2020, 01:19 Hi Croco

Nice work! This kind of diligent work is needed to push the code into new areas. And actually before your tests this was not possible since the wave equation hadn't been tested for eigen analysis.
Cool! I am glad this was interesting for you.
raback wrote: 13 Sep 2020, 01:19 A note on the output of eigenmodes. You will get the modes as timesteps in vtu file if you add in Simulation section "vtu: Eigen Analysis = Logical True". This will tell the vtu output to number the files according to eigenmodes. There could be iterative or time dependent problems where for each iteration we solve a slightly modified eigenvalue problem. So therefore this is not the default in output.
Ha! Nice, I did not know that. I will try. I definitely need to give another good read at the manuals, so many things I keep on missing.
raback wrote: 13 Sep 2020, 01:19 I don't remember whether the solver can deal with p-elements. However, you might want to test whether "Element = p:2" improves results.Certainly there is always a significant cost in increasing the element order but for wave problems going quadratic may pay off.
Cool stuff. I will try that too.
CrocoDuck
Posts: 64
Joined: 12 May 2016, 13:15
Antispam: Yes

Re: WaveSolver and Eigen Analysis

Post by CrocoDuck »

Tested both of "vtu: Eigen Analysis = Logical True" and "Element = p:2". They work like a charm, and p:2 boost the accuracy a lot. Good stuff.
raback
Site Admin
Posts: 3750
Joined: 22 Aug 2009, 11:57
Antispam: Yes
Location: Espoo, Finland
Contact:

Re: WaveSolver and Eigen Analysis

Post by raback »

Nice!

In your case can you determine that it is more efficient to get the added accuracy using the p strategy (vs. h strategy)? The theory says that for smooth problems p should be better. You can go p=3, 4, 5,... but I would suspect that the sweep spot is p=2 or p=3.

The integration rules for the p-elements may be rather costly. There might be some room for improvement here. The number of points grow now as p^3.

-Peter
Post Reply