Transient acoustic problem with large number of elements

[Schwarz2012]

Hi everyone.
I have a task and I don't know how to start.
The general view of the problem: “A large room with a work-unit inside that produces a lot of noise. Frequency up to 3.5 kHz. Need to study the propagation of acoustic waves inside this room”
The main problem is that the computational domain is so large that it can include 200-300 thousand elements.
At the moment, I am studying the possibility of using Open Source packages. Elmer attracted me because there is a good example stationary Helmholtz model using, but there is very little information on transient models.
Can Elmer solve a problem of this size?
Do I understand correctly that for such a task I can use the “Large-amplitude Wave Motion in Air” model?
What boundary conditions does this model have (there is no information in the manual)?
Are there any examples of calculations using transient models in Elmer?

[kevinarden]

https://github.com/ElmerCSC/elmerfem/tr ... ts/WaveEqu

Page 102 of the models manual
https://www.nic.funet.fi/pub/sci/physic ... Manual.pdf

[Schwarz2012]

https://github.com/ElmerCSC/elmerfem/tr ... ts/WaveEqu

Page 102 of the models manual
https://www.nic.funet.fi/pub/sci/physic ... Manual.pdf

Thanks for the answer.

[Schwarz2012]

Now I can formulate more correct.
My main question, what about variables type in Elmer? It's int32 or int64? Because, for example, if problem contain 400 millions of elements and 3-4 billions of DOFs? Can Elmer solve this problem or that is impossible?

[kevinarden]

This is a rather old post
viewtopic.php?t=4326

I suspect your answer is that the limitations are imposed by the machines RAM, disk space, and cores.

[raback]

Hi

The largest 32 bit signed integer is 2,147,483,647 so that indeed may set an upper limit. I guess there are not too many places where this should be replaced but this has not been in focus. I think the nodes are given a parallel global numbering. You will probably face other issues before that. Wave equation may require some special techniques for preconditioning (e.g. adding some damping to the preconditioning matrix). Transient version probably scales more easily but you need huge number of timesteps. For sure you need a big computer. This is not something to be run on your PC.

-Peter

[Schwarz2012]

Hi

The largest 32 bit signed integer is 2,147,483,647 so that indeed may set an upper limit. I guess there are not too many places where this should be replaced but this has not been in focus. I think the nodes are given a parallel global numbering. You will probably face other issues before that. Wave equation may require some special techniques for preconditioning (e.g. adding some damping to the preconditioning matrix). Transient version probably scales more easily but you need huge number of timesteps. For sure you need a big computer. This is not something to be run on your PC.

-Peter

Thanks for the answer!

Theoretically, if the limit is reached, will it be possible to do something about it? Is it realistic to replace int32 with int64 or is it not even worth trying?

[raback]

Hi

I would not do it before really needed. Changing all "integer" in the code "integer(kind=int64)" is probably not a good idea. It will make the code slover and probably there will a lot of places where the API to other libraries will be broken. Most libraries work with int32. Going to int64 would require delicately changing the type just in the right places.

-Peter

[Schwarz2012]

Hi

I would not do it before really needed. Changing all "integer" in the code "integer(kind=int64)" is probably not a good idea. It will make the code slover and probably there will a lot of places where the API to other libraries will be broken. Most libraries work with int32. Going to int64 would require delicately changing the type just in the right places.

-Peter

Thank you for discussion! It was really helpful and saved me from wasting time.