[santiagomarquezd]

Hi all, I'm doing a comparison between PCG and AMG. The problem is to solve with laplancianFoam and all Dirichlet BC's with both methods. I solved asking for the same absolute residuals of 1E-06 and then solved the problem using a direct solver in octave extracting the matrix and RHS with gdbOF.
The max absolute error is about 5E-5 with PCG and 2E-2 with AMG for a problem of 1000x1000 cells. This is the difference, same residuals in PCG and AMG lead to different quality solution. Next are the logs:

PCG:

diagonalPCG: Solving for T, Initial residual = 1, Final residual = 9.99042e-07, No Iterations 2231
ExecutionTime = 223.14 s ClockTime = 223 s

AMG:

GAMG: Solving for T, Initial residual = 1, Final residual = 5.69744e-07, No Iterations 12
ExecutionTime = 57.95 s ClockTime = 58 s

So that:

1. Is the definition of residuals different between PCG and AMG? or
2. I'm setting something bad in AMG and it is not converging at all

Any clues?

Thx in advance.

[santiagomarquezd]

I found that you have to ask for lower residuals in AMG than PCG to obtain the similar max error in the solution.

Any clues?