Converge and freeze coarse AMR level

[drjosephbrock]

Is it possible (and/or simple) to run a simulation (in my case an Euler simulation) on a coarse mesh with several AMR levels defined. I am able to only activate the next level when my current level has reached some global residual reduction. I would like to only iterate on the current level and bypass the coarser level. In other words, I would like to refine the mesh as the solution converges and only spend time solving on the current level while freezing the previous coarse state. I am currently interested in steady-state problems, and I think this would help accelerate my simulations greatly. Any information and advice would be greatly appreciated. Thanks!

[asalmgren]

You can certainly do this by inserting some code that controls when a particular level advances the solution. But you'll need to give us more information about what code you're using and how you are handling the calling of finer levels. For example, are you using an AmrCore or AmrLevel approach? Or have you built your own?

…

On Wed, May 29, 2024 at 12:22 PM drjosephbrock ***@***.***> wrote: Is it possible (and/or simple) to run a simulation (in my case an Euler simulation) on a coarse mesh with several AMR levels defined. I am able to only activate the next level when my current level has reached some global residual reduction. I would like to only iterate on the current level and bypass the coarser level. In other words, I would like to refine the mesh as the solution converges and only spend time solving on the current level while freezing the previous coarse state. I am currently interested in steady-state problems, and I think this would help accelerate my simulations greatly. Any information and advice would be greatly appreciated. Thanks! — Reply to this email directly, view it on GitHub <#3964>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ACRE6YTXXAT6MFPWZ6HTTVTZEYTF5AVCNFSM6AAAAABIPRTJ6SVHI2DSMVQWIX3LMV43ERDJONRXK43TNFXW4OZWG42DQOBUGI> . You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>

-- Ann Almgren Senior Scientist; Dept. Head, Applied Mathematics Pronouns: she/her/hers

[drjosephbrock]

I am using AmrLevel for my approach, with my solver resembling he CNS example code from the amrex-tutorial repo. I see that using AmrCore would have direct control over the "evolve" function, whereas the AmrLevel does not. How difficult would it be to augment my approach to control level advancing? Does it make sense to convert to AmrCore?

[WeiqunZhang]

You can have a class derived from Amr and override virtual void coarseTimeStep (Real stop_time);. Then you have full control.

[asalmgren]

You could also do something like copy the old data into the new data instead of evolving it - and just wrap that in an if test on level or time or both… Ann Almgren Senior Scientist; Dept. Head, Applied Mathematics Pronouns: she/her/hers

…

On Wed, May 29, 2024 at 2:10 PM Weiqun Zhang ***@***.***> wrote: You can have a class derived from Amr and override virtual void coarseTimeStep (Real stop_time);. Then you have full control. — Reply to this email directly, view it on GitHub <#3964 (reply in thread)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ACRE6YU7WIOVC3IYTAGPNZLZEY73FAVCNFSM6AAAAABIPRTJ6SVHI2DSMVQWIX3LMV43SRDJONRXK43TNFXW4Q3PNVWWK3TUHM4TKOJYGUZDG> . You are receiving this because you commented.Message ID: ***@***.***>