Multigrid
 

Can anyone provide any insight for me into multigrid solvers.

Fluent includes what it calls a multigrid solver but I have now been told that multigrid is very state-of-the-art CFD and that what Fluent calls multigrid is actually just an overlay method.

If anyone can shed light on this issue for me, I would be grateful.

Regards Althea

Re: Multigrid
 

Hi, MG can be divided into algebraic and geometric, you can visit http:\\www.mgnet.org\ The above webtsite has some tutorial materials and repints.

Just from my experience, MG can achieve high efficeincy of your solvers, but you have to spend a lot of time in code programming and fine-tuning.

The MG text book usually stress that MG can be made as a blackbox for users, but sometimes this kind of blackbox usually leads to low efficiency.

I am not familiar with Fluent, I don't known what kind of MG it is.

About CFD, there is a saying that you only need to apply MG for pressure poisson equation.

We have implemented MG in a finite element solver for electromganetic problem, we can get good performance after fine-tuning, but in some complicated cases, we have to change our MG algorithm time by time, maybe this is due to that there are some advanced mathematics hidden behind the superfacial simple principle.

Good luck

Hongliang

Re: Multigrid
 

The implicit solver in Fluent uses algebraic multigrid to solve the equation system. This kind of multigrid has nothing to do with the grid, it is just a mathematical way of speeding up the solution of the equation system. With the explicit solver you also have the option to use true "grid based" multigrid.

Multigrid isn't very new. It has been around for a long time and many commercial codes now has it implemented, both in the form of algebraic multigrid and "true" multigrid.

Re: Multigrid
 

Multigrid is a technique used to dampen low frequency numerical errors that appear early on in the solution process. By solving the difference equations on progressively coarser grids, the low frequency errors are reduced quicker than if the calculation proceeds solely on the fine grid. Of course, if you have a fairly coarse grid to start with, then the use of multigrid may not improve the rate of convergence by much.

Jonathan

Re: Multigrid
 

Thanks Jonathan,

That's what I understood by multigrid and that's basically what it says in the manuals (version 4.4).

That's also how I have described what Fluent does in a document on a simulation. However, someone who read it picked up on this and said Fluent doesn't use multigrid really, but just an overlay method and that multigrid is a much more complex state of the art method. This is what I would like clarification on.

Any ideas or comments (from anyone).

Regards Althea

Re: Multigrid
 

(1). I can only tell you my experience with the old versions of the code. (2). Over a year ago, when I was still using the code (UNS, Rampant), it was hard to get converged solution by using multi-grid options. (3). So, normally, I ran the codes with no multi-grid speed-up. This was especially a problem with Rampant code, which also was hard to get it started. (4).Both UNS and Rampant use un-structured mesh. I don't know how the multi-grid is implimented inside the code especially for 3-D meshes, because I simply don't use it. This could be mesh and problem dependent, so, it is hard for me to make any conclusion.

Re: Multigrid
 

Hi,

The linear equation solver used for the segregated and the coupled implicit solvers in Fluent is an algebraic multigrid method. Multigrid methods can be divided into to types, geometric, where the problem is discretized on several successively coarser grids, and algebraic, which creates smaller Ax=b linear equation systems using only information in the original fine level Ax=b system. Algebraic multigrid using piecepise constant interpolation and coarsening by grouping is sometimes reffered to as additive correction multigrid. Fluent's AMG solver uses this method.

I don't know what you mean by an "overlay method", perhaps you could elaborate?

Joe

how to use multigrid
 

Hi,

I have a steady problem and using sst transition model in Fluent. But there is a problem with convergence. The convergence is obtained too late.

So, I want to speed up my simulations with multigrid but I don't know so much about how to do.

can anybody help?

Thanks

Hi,Bugra,

Do you solve it?

The multigrid solver is already on by default.

You can achieve faster convergence simply by increasing your under-relaxation factors.

If you want a more aggressive multigrid routine, go to advanced options and make the options less restrictive, higher coarsening level, higher tolerances, etc. But in general, a more aggressive multigrid routine is also more unstable.