Author Topic: Time Integration Schemes  (Read 404 times)


  • Newbie
  • *
  • Posts: 21
Time Integration Schemes
« on: September 14, 2016, 01:12:20 PM »
Hi together,

I have a suggestion concerning the time integration schemes in the Kratos. As it is now, there is a "residual_based_bossak_displacement_scheme" and a "residual_based_newmark_displacement_scheme" implemented in the core. When you look at the code it is implemented as a generalized alpha scheme and simply the parameters are set in the way that you can get one or the other. I would suggest make only one scheme called "residual_based_generalzied_alpha_scheme" and control the type of time integration scheme by setting the parameters through the ProjectParameters file, since schemes like Newmark and Bossak can be derived from generalized alpha by simply setting the parameters, there.

For example when you select as a scheme type "Newmark" in the ProjectParameters both alpha parameters are set to zero, if you select "Bossak" alpha_f is set to zero and alpha_m can be chosen and so on.

Especiallay when treating coupled problems like FSI it might be important to know what time integration scheme and which parameters are used.

Please let me know what you think about this suggestion.

Thank you very much.



Josep Maria

  • Global Moderator
  • Jr. Member
  • *****
  • Posts: 50
Re: Time Integration Schemes
« Reply #1 on: September 14, 2016, 10:25:38 PM »
Hello Andreas,

   The schemes of the core were located in the SolidMechanicsApplication some months ago.  As you realized the implementation was the generalized alpha  scheme. In fact the only scheme used was the Bossak one,  which turned to Newmark  changing the alpha_m (which changes the time integration point for the accelerations).  The problem comes when you want to  change the time integration point for the forces (the right hand side of the equation),  not only the alpha_f needs to be modified also the forces from the previous time step have to be stored  in order to apply the correct average. Kratos solution schemes were not designed accounting for the storage of the members of the solution equations across time steps. It uses to be computationally inefficient. Therefore the way to apply a force average (integrate forces in some time between the current time step) needs for the implementation of a new element able to do that average locally and return the result as his own right hand side.

This is the reason why the generalized alpha method is not there as a general scheme class.
If you have any idea to surpass this problem or improve the implementation it will be welcome.

Kind regards

Josep Maria
« Last Edit: September 17, 2016, 06:02:01 PM by Josep Maria »