CFD Online Logo CFD Online URL
www.cfd-online.com
[Sponsors]
Home > Forums > General Forums > Main CFD Forum

Does algebraic multigrid affected by CFL number?

Register Blogs Community New Posts Updated Threads Search

Like Tree4Likes
  • 1 Post By sbaffini
  • 1 Post By sbaffini
  • 1 Post By sbaffini
  • 1 Post By sbaffini

Reply
 
LinkBack Thread Tools Search this Thread Display Modes
Old   November 30, 2020, 12:10
Default Does algebraic multigrid affected by CFL number?
  #1
New Member
 
ibopaul's Avatar
 
ibopaul
Join Date: Nov 2020
Posts: 12
Rep Power: 6
ibopaul is on a distinguished road
Hi,

I read reference paper 'Implicit solution of Preconditioned Navier-Stokes Equations Using Algebraic Multigrid' and found that CFL number was close to infinity when AMG used in a density-based algorithm for Euler and Navier-Stokes equations.

Should CFL numbers always be large when using algebraic multigrid for density based solver?

If CFL number is 1 or less than 1, there is no acceleration while using algebraic multigrid?

Thanks in advance.

Last edited by ibopaul; December 1, 2020 at 07:07.
ibopaul is offline   Reply With Quote

Old   December 2, 2020, 07:16
Default
  #2
Senior Member
 
sbaffini's Avatar
 
Paolo Lampitella
Join Date: Mar 2009
Location: Italy
Posts: 2,192
Blog Entries: 29
Rep Power: 39
sbaffini will become famous soon enoughsbaffini will become famous soon enough
Send a message via Skype™ to sbaffini
CFL in that algorithm (Fluent one if I recall correctly) is basically an implicit under relaxation factor that affects the diagonal dominance of the system you are going to solve at each iteration. Still, it is also connected to a local time step.

What happens when you lower the CFL is that, for each iteration, you advance less but you do it more robustly (the system more and more resembles a diagonal one).

AMG is used to accelerate each of these single iterations or, better, solve the system in them. But, as with lower CFL the system becomes more diagonal and the solution advances less, the AMG acceleration becomes less useful. Indeed, with CFL 1 you could probably go full explicit and see little difference in the overall solution advancement during the iterations
ibopaul likes this.
sbaffini is offline   Reply With Quote

Old   December 2, 2020, 07:54
Default
  #3
Senior Member
 
sbaffini's Avatar
 
Paolo Lampitella
Join Date: Mar 2009
Location: Italy
Posts: 2,192
Blog Entries: 29
Rep Power: 39
sbaffini will become famous soon enoughsbaffini will become famous soon enough
Send a message via Skype™ to sbaffini
Now I saw your previous post as well. Note that the capability to work at very large CFL numbers is not only linked to the proper coding of the AMG, but also depends from all the other parts in the code, that must be implicit. And, in any case, not all problems can work at infinite CFL, that is only true for linear problems.
ibopaul likes this.
sbaffini is offline   Reply With Quote

Old   December 2, 2020, 12:20
Default
  #4
New Member
 
ibopaul's Avatar
 
ibopaul
Join Date: Nov 2020
Posts: 12
Rep Power: 6
ibopaul is on a distinguished road
Quote:
Originally Posted by sbaffini View Post
Now I saw your previous post as well. Note that the capability to work at very large CFL numbers is not only linked to the proper coding of the AMG, but also depends from all the other parts in the code, that must be implicit. And, in any case, not all problems can work at infinite CFL, that is only true for linear problems.

Thanks for the advice!
ibopaul is offline   Reply With Quote

Old   December 14, 2020, 07:42
Default May I ask you one more question?
  #5
New Member
 
ibopaul's Avatar
 
ibopaul
Join Date: Nov 2020
Posts: 12
Rep Power: 6
ibopaul is on a distinguished road
Quote:
Originally Posted by sbaffini View Post
Now I saw your previous post as well. Note that the capability to work at very large CFL numbers is not only linked to the proper coding of the AMG, but also depends from all the other parts in the code, that must be implicit. And, in any case, not all problems can work at infinite CFL, that is only true for linear problems.
Thank you for reply.

Since AMG is based on implicit time stepping code, i understand that your 'implicit' word as exact differentiation of flux jacobian in implicit operator.

May I ask you one more question? As reference paper used Barth & Jespersen limiter, i tried to use them either. however, since Barth & Jespersen limiter has Minimum, Maximum functions , i realized that i can't differentiate them. (because Minimum, Maximum functions are discontinuous)

Is there any method or reference paper which shows the way how to exactly differentiate flux jacobian with limiter used?

Thanks in advance.
ibopaul is offline   Reply With Quote

Old   December 14, 2020, 07:47
Default
  #6
Senior Member
 
sbaffini's Avatar
 
Paolo Lampitella
Join Date: Mar 2009
Location: Italy
Posts: 2,192
Blog Entries: 29
Rep Power: 39
sbaffini will become famous soon enoughsbaffini will become famous soon enough
Send a message via Skype™ to sbaffini
I follow the route of the Fluent paper. In that case the implicit part only takes into account the first order part of a scheme, so no gradients and no limiters. It is not exact but it works and is easy
ibopaul likes this.
sbaffini is offline   Reply With Quote

Old   December 14, 2020, 07:53
Default Thanks for reply!
  #7
New Member
 
ibopaul's Avatar
 
ibopaul
Join Date: Nov 2020
Posts: 12
Rep Power: 6
ibopaul is on a distinguished road
Quote:
Originally Posted by sbaffini View Post
I follow the route of the Fluent paper. In that case the implicit part only takes into account the first order part of a scheme, so no gradients and no limiters. It is not exact but it works and is easy
Thanks for reply!

I will follow the Fluent's work as you mentioned. Since my code solves only Euler equations, my code's verification case will be small bump case.

Thanks again!
ibopaul is offline   Reply With Quote

Old   December 14, 2020, 08:00
Default
  #8
Senior Member
 
sbaffini's Avatar
 
Paolo Lampitella
Join Date: Mar 2009
Location: Italy
Posts: 2,192
Blog Entries: 29
Rep Power: 39
sbaffini will become famous soon enoughsbaffini will become famous soon enough
Send a message via Skype™ to sbaffini
I also suggest to test on the shock tube (as many variants as you can) and converging-diverging nozzle at least.
ibopaul likes this.
sbaffini is offline   Reply With Quote

Old   December 14, 2020, 08:06
Default Thank you for suggestion.
  #9
New Member
 
ibopaul's Avatar
 
ibopaul
Join Date: Nov 2020
Posts: 12
Rep Power: 6
ibopaul is on a distinguished road
Thank you for suggestion.

I will test the cases that you mentioned as soon as my code completed!

Thanks again!
ibopaul is offline   Reply With Quote

Reply

Tags
algebraic multigrid


Posting Rules
You may not post new threads
You may not post replies
You may not post attachments
You may not edit your posts

BB code is On
Smilies are On
[IMG] code is On
HTML code is Off
Trackbacks are Off
Pingbacks are On
Refbacks are On


Similar Threads
Thread Thread Starter Forum Replies Last Post
decomposePar no field transfert Jeanp OpenFOAM Pre-Processing 3 June 18, 2022 13:01
[snappyHexMesh] Error snappyhexmesh - Multiple outside loops avinashjagdale OpenFOAM Meshing & Mesh Conversion 53 March 8, 2019 10:42
Courant Number waseeqsiddiqui FLUENT 3 December 27, 2018 11:43
[mesh manipulation] Importing Multiple Meshes thomasnwalshiii OpenFOAM Meshing & Mesh Conversion 18 December 19, 2015 19:57
Cluster ID's not contiguous in compute-nodes domain. ??? Shogan FLUENT 1 May 28, 2014 16:03


All times are GMT -4. The time now is 01:34.