Elmer Discussion Forum

Are all the time discretization schemes implemented in Elmer FEM semi-implicit? Also, where can I find the list of time discretization schemes available in Elmer FEM?

I reviewed TimeIntegrators.F90 code in the src folder as well as Chapter 6/Time discretization in Elmer Solver manual.

Thanks for any formation you can provide.

Hi,

Unfortunately the documentation of the time stepping algorithms is not fully up to date. It seems that there are some options which are not mentioned in the documentation, so a documentation update would be needed. These include at least some Runge-Kutta methods and a fractional step method. I don't know much about these undocumented options as I haven't used them myself.

Whether one gets a fully implicit or semi-implicit algorithm may depend on how the different iteration controls of the solver are used. Usually fully implicit time stepping is employed by seeking an accurate nonlinear solution before proceeding to the next time step. However, it seems that the Runge-Kutta methods of Elmer can be explicit. In addition, as far as I understand, the predictor-corrector scheme can be defined to combine explicit and semi-implicit schemes.

-- Mika

Thank you for the information. Is the documentation that you are referring to - Chapter 6/Elmer Solver Manual?

I am assuming the schemes that are not in the documentation are implemented in the source code. Would that code be TimeIntegrate.F90? If yes, then would the schemes listed in that code be a comprehensive list?

asandip wrote: ↑12 Jul 2020, 04:23 Is the documentation that you are referring to - Chapter 6/Elmer Solver Manual?

Yes

asandip wrote: ↑12 Jul 2020, 04:23 Would that code be TimeIntegrate.F90?

Yes, this is the common place for time stepping subroutines. The subroutine names used there are quite descriptive:
However, the subroutine Newmark2ndOrder seems to be a nonactive option since it is not called anywhere in the code.

-- Mika

Hello
I know Crank Nicholson is half explicit and half implicit. BDF first order is fully implicit.
So what keyword setting should I use in a solver section to make its time integration fully explicit ?
Thanks

I found out that

you keep write

Timestepping Method = runge-kutta

inside a solver section

but there are implicit Runge Kutta methods. I am not sure if the default is explicit in Elmer. Also I wanted to set it to lowest order - just a simple Euler iteration.

Hi,

Explicit Euler is chosen by "explicit euler".

Note that for Elmer this may be quite suboptimal since the machinery assumes a linear system that is solved. Explicit methods often try to avoid this. Explicit method + lumped mass matrix needs no solution. In that case a lot could be done to improve the speed. However, usually implicit methods are used so this has not been addressed.

-Peter

To find the most accurate and up-to-date information regarding time discretization schemes in Elmer FEM, I recommend checking the official Elmer documentation, release notes, or directly reaching out to the Elmer FEM community and developers through their forums or communication channels. The Elmer website and user forums are valuable resources for obtaining the latest information and seeking assistance with specific questions or concerns.