Time dependent 3D Maxwell material viscoelastic Earth deformation

Numerical methods and mathematical models of Elmer
fpk
Posts: 10
Joined: 31 Jul 2020, 11:36
Antispam: Yes

Re: Time dependent 3D Maxwell material viscoelastic Earth deformation

Post by fpk »

tzwinger wrote: 06 Oct 2021, 20:26 Hello,

If it helps, please find attached a similar case like the 2D test-case in 3D (runs on my gaming PC with 16 GB RAM and 6 core i5). Picture shows timestep of max. deformation:
deformation.0099.jpeg
Mind, to first create mesh-file (keeping in mind, that one has to move from kilometres to metres):

Code: Select all

gmsh -3 duallayer3d.geo
ElmerGrid 14 2 duallayer3d.msh -autoclean -scale 1000 1000 1000 
ElmerGrid 2 2 duallayer3d -partdual -metiskway 6
(mind, my computer has 6 cores, change the last number accordingly to your hardware)

Best wishes,
Thomas
Thomas, thank you for 3D case example,
it is good work in my PC for "Linear System Convergence Tolerance = 1.0e-07",

but if i replace mesh by mine mesh, which is complicate at the top middle part and try use solver with:
" Linear System Solver = Direct
Linear System Direct Method = MUMPS
Linear System Symmetric = Logical True"
ten years loading computation take about 10 minutes.

Then i am using:
" Linear System Solver = Iterative
Linear System Iterative Method = GCR
Linear System Max Iterations = 6000
Linear System Preconditioning = BILU2
Linear System Convergence Tolerance = 1.0e-05"
ten years loading computation take about 3-4 minutes, at the begining of iterations it is require more 2500 iteration but then decrease to less then 200 - 150 iterations. But under "iterative solver system" i couldn't increase convergence tolerance to 1.0e-07.

Thank you again.
Rustam Bulgakov
fpk
Posts: 10
Joined: 31 Jul 2020, 11:36
Antispam: Yes

Re: Time dependent 3D Maxwell material viscoelastic Earth deformation

Post by fpk »

kevinarden wrote: 06 Oct 2021, 16:03 Youngs Modulus = 6.5844e+11
Viscosity = 2.7E+21

These are large numbers even for a 64 bit double precision computer. The problem is these values are divided by, use to divide, possibly very small numbers. Even though the 64b dp might be able to handle it it severely impacts the precision of the result, you start losing accuracy to the first decimal place. This potential accuracy issue could cause convergence issues.
Kevin, thank you for advice,
my understanding of your recommendation "to divide, possibly very small numbers" is to use instead Elmer SI system some artificial physic system which for example for 1.0e+6 more the SI system, for example instead 1kg or 1 meter put in material conditions number for 1.0e+6 higher, is this understanding correct?
fpk
Posts: 10
Joined: 31 Jul 2020, 11:36
Antispam: Yes

Re: Time dependent 3D Maxwell material viscoelastic Earth deformation

Post by fpk »

Dear All of society,

could somebody advice how to restart simulation from actual displacements which achieved in previous computation? My experience show every restart begin from initial (zero displaced cube).

I've tried:
Restart Before Initial Conditions = Logical True
Restart Variables = String "t[d:3 p:1]

but unsuccessful

Thank you in advance!
Post Reply