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

What does Gauss Seidel iteration mean?

Register Blogs Members List Search Today's Posts Mark Forums Read

Reply
 
LinkBack Thread Tools Search this Thread Display Modes
Old   July 2, 2015, 17:58
Default What does Gauss Seidel iteration mean?
  #1
Member
 
Join Date: Aug 2010
Posts: 31
Rep Power: 15
Turbomachine is on a distinguished road
Hello,

I am trying to build the model for bacterial chemotaxis as described in the paper Tyson, Leveque (2000) - Fractional step methods applied to a chemotaxis model.

Without going into the details, they are solving the the diffusion term using the trapezoidal rule and the BDF2 method (TR-BDF). I implemented these the standard way, by constructing a tridiagional matrix and then inverting it to get a solution to the implicit problem. Unfortunately I get oscillating solutions.

I have no idea what the authors mean by few sweeps of Gauss-Seidel iteration. Is that how I am supposed to invert the matrix? Isn't that very inefficient?

Here's the extract from the paper:

"We implemented TR-BDF2 by splitting the method dimensionally and then correcting for cross terms by using a few sweeps of Gauss-Seidel (or under-relaxed Jacobi) iteration. The first stage of the method, TR, is implemented by taking one step of a locally one-dimensional (LOD) method, which is just a fractional step method in which the x- and y -derivative terms are split apart, resulting in simple tridiagonal systems of equations to solve along each grid line [22]. The second stage,BDF2, is implemented first by taking another step of an LOD method, with differ-ent coefficients this time. This gives a very good initial guess for the Gauss-Seidel procedure. In principle advancing by LOD alone should be enough to maintain second order accuracy, but we have found that this can reintroduce grid-scale oscillations."

Any help greatly appreciated!!!
Turbomachine is offline   Reply With Quote

Old   July 2, 2015, 18:15
Default
  #2
Senior Member
 
Filippo Maria Denaro
Join Date: Jul 2010
Posts: 6,760
Rep Power: 71
FMDenaro has a spectacular aura aboutFMDenaro has a spectacular aura aboutFMDenaro has a spectacular aura about
The GS method belongs to the class of iterative solver for linear algebric system, it is not a inversion of the matrix.
However, for tridiagonal system the Thomas algorithm is well suited
FMDenaro is offline   Reply With Quote

Old   July 2, 2015, 18:27
Default
  #3
Member
 
Join Date: Aug 2010
Posts: 31
Rep Power: 15
Turbomachine is on a distinguished road
Quote:
Originally Posted by FMDenaro View Post
The GS method belongs to the class of iterative solver for linear algebric system, it is not a inversion of the matrix.
However, for tridiagonal system the Thomas algorithm is well suited
Thanks for the reply. From the extract of the paper it seems that first I need to apply the TR-BDF2 method, which involves 2 matrix inversions and which returns an array x. This is then the initial condition for Gauss Seidel? If yes, what is the linear system that I need to solve? It seems they use the Gauss-Seidel method purely to get rid of the oscillations.
Turbomachine is offline   Reply With Quote

Old   July 2, 2015, 18:36
Default
  #4
Senior Member
 
Filippo Maria Denaro
Join Date: Jul 2010
Posts: 6,760
Rep Power: 71
FMDenaro has a spectacular aura aboutFMDenaro has a spectacular aura aboutFMDenaro has a spectacular aura about
Quote:
Originally Posted by Turbomachine View Post
Thanks for the reply. From the extract of the paper it seems that first I need to apply the TR-BDF2 method, which involves 2 matrix inversions and which returns an array x. This is then the initial condition for Gauss Seidel? If yes, what is the linear system that I need to solve? It seems they use the Gauss-Seidel method purely to get rid of the oscillations.

I would need to know the details ...however, if the diffusion operator is a 2D Laplace operator (d2/dx2 + d2/dy2), the factorization technique introduces an error of high order term that can lead to oscillations. You have two successive tridiagonal systems.
But in no way an iterative solver can be able to avoid oscillation that are inherent to the type of approximate solution.
FMDenaro is offline   Reply With Quote

Old   July 2, 2015, 18:57
Default
  #5
Member
 
Join Date: Aug 2010
Posts: 31
Rep Power: 15
Turbomachine is on a distinguished road
Quote:
Originally Posted by FMDenaro View Post
I would need to know the details ...however, if the diffusion operator is a 2D Laplace operator (d2/dx2 + d2/dy2), the factorization technique introduces an error of high order term that can lead to oscillations. You have two successive tridiagonal systems.
But in no way an iterative solver can be able to avoid oscillation that are inherent to the type of approximate solution.
Thanks for your time.

The problem looks as follows. In fact I'm only doing 1D now.

\frac{\partial n}{\partial t} =D \frac{\partial^2 n}{\partial x^2} -  \frac{\partial}{\partial x}\left(A n \frac{1}{(1+c)^2}\frac{\partial c}{\partial x}\right)
\frac{\partial c}{\partial t} = \frac{\partial^2 c}{\partial x^2} + \frac{n^2}{1+n^2}


where D and A are constants. Because of the non-linear advection term I am using fractional timestep method to split the terms and use Clawpack to solve the advection term, TR-BDF2 method for the diffusion terms and Runge-Kutta for the reaction term.

My problem concerns the diffusion terms. Essentially, I take a solution and apply the trapezoidal rule with half time step (I-k/2h*A)U* =(I+k/2h*A)U_n, and then the BDF2 method with half time step (I-k/6h)U_n+1=1/3(4U* - U_n).

These involve two inversions. I am trying to make sense where the Gauss-Seidel iterations may come in.
Turbomachine is offline   Reply With Quote

Old   July 3, 2015, 04:40
Default
  #6
Senior Member
 
Filippo Maria Denaro
Join Date: Jul 2010
Posts: 6,760
Rep Power: 71
FMDenaro has a spectacular aura aboutFMDenaro has a spectacular aura aboutFMDenaro has a spectacular aura about
the first step is a simple Crank-Nicolson integration for 1D and the Thomas algorithm is fine
FMDenaro is offline   Reply With Quote

Reply

Tags
diffusion, diffusion equation, hyperbolic

Thread Tools Search this Thread
Search this Thread:

Advanced Search
Display Modes

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
Steady-State and Transient Solvers 70m1 OpenFOAM Running, Solving & CFD 21 May 8, 2021 08:09
Free Surface Ship Flow timfranke OpenFOAM Running, Solving & CFD 322 March 3, 2021 10:04
bounded Gauss upwind Scheme deepinheart OpenFOAM Running, Solving & CFD 1 February 23, 2015 06:57
same geometry,structured and unstructured mesh,different behaviour. sharonyue OpenFOAM Running, Solving & CFD 13 January 2, 2013 23:40
solution diverges when linear upwind interpolation scheme is used subash OpenFOAM 0 May 29, 2010 02:23


All times are GMT -4. The time now is 10:06.