[nilesh]

Hi,
I need to implement RK4 for the external time stepping (i.e. instead of first order dual time stepping).
From what I understood, line no 6125 of solver_direct_mean.cpp has to be modified (I do not mind entering the coefficients directly as constants at this point of time instead of taking them from the config file).
However, I do not understand how to go about finding the solution (space derivatives) at the intermediate stages of the Runge kutta iteration. Please help me with this.
Thanks.

[nilesh]

Sorry, apparently that line is only for calculating the residue. I am confused by the object oriented structure of the code. Which is the function to be called for calculating the space derivative part? Does ->GetSolution() which returns SOLUTION only return the solution or also calculated it?
I am very eager (desperate) to use RK3 (or RK4) for the external loop. If any user is interested, we can collaborate and work on it.

Thanks.

[demanosalvas]

Nilesh,

RK time integration is already implemented in the code, all you need to do is select this option in your config file.

David

[nilesh]

Originally Posted by demanosalvas

Nilesh,

RK time integration is already implemented in the code, all you need to do is select this option in your config file.

David

That is for internal iterations, I wanted RK for external iterations when using dual-time stepping.

[demanosalvas]

Nilesh,

Can you elaborate a bit more on why would you like to do that?

David

[nilesh]

I wanted to use RK3 for the external time marching which is presently being done by euler explicit or 3-time level BDF method. I have done it for the the SU2_CFD module as of now (without support for mesh deformation or turbulence models) and the results look great.
Would be nice if you include it throughout the SU2 suite.

Thanks.

[economon]

Hi,

Just to make sure everything is clear.. you mean that you would like to perform time-accurate simulations with an explicit RK scheme rather than the current dual-time stepping method (1st- or 2nd-order BDF), right?

We do have a "time-stepping" mode for unsteady calculations (check the config file options), which will essentially use a global time step for the entire mesh and solve the flow in a time-accurate manner with the chosen scheme. You might try experimenting with this to see if it meets your needs. Otherwise, perhaps we can chat a little more about your implementation.

All the best,
Tom

[nilesh]

Originally Posted by economon

Hi,

Just to make sure everything is clear.. you mean that you would like to perform time-accurate simulations with an explicit RK scheme rather than the current dual-time stepping method (1st- or 2nd-order BDF), right?

We do have a "time-stepping" mode for unsteady calculations (check the config file options), which will essentially use a global time step for the entire mesh and solve the flow in a time-accurate manner with the chosen scheme. You might try experimenting with this to see if it meets your needs. Otherwise, perhaps we can chat a little more about your implementation.

All the best,
Tom

Thanks. Last time I tried it, I was unable to write files. Actually the filename of the output files did not increment with the time step and consequently the output files kept overwriting the previous ones. I will give it another try and keep you posted.

[economon]

I see.. yep, please let us know your experience, and if you are still running into issues, we should be able to modify the code quickly to resolve it.

T

[d_ray]

Originally Posted by economon

I see.. yep, please let us know your experience, and if you are still running into issues, we should be able to modify the code quickly to resolve it.

T

I have been trying to run an unsteady problem with RK3 (w/o dual stepping), as mentioned previously by Nilesh. I am facing the following issues:
1) The solution files are not being saved in an incremental manner. The solutions are overwritten into the previous file. However, the incremental storage is working for the dual time stepping scenario.
2) It is unclear how to specify the final time for the simulation using TIME_STEPPING.

I have been trying to look into the cfg files of various unsteady examples. Unfortunately all of them seem to use the dual time stepping strategy.

I am attaching my cfg file for your reference.

Thanks.