Trouble with lateral friction
Posted: 24 Jun 2014, 02:34
Hi all, I am currently trying to set up a flowline marine ice sheet simulation with parameterised lateral drag.
The lateral drag code is described here:
http://elmerice.elmerfem.org/wiki/doku. ... alfriction
The basal drag law is a function of overburden pressure.
I am trying to spin up from a slab of uniform thickness. This works fine with no lateral friction, or with very low lateral friction (paramterised channel width 500km). But the non linear iterations of the Stokes solver do not converge when higher lateral friction is used (parameterised channel width 100km).
I've tried adding in the lateral friction gradually: a linear increase from zero to a parameterised channel width of 100km over 20,000 years, but the simulation failed after about 4000 years.
I've tried hard coding upper limits for the velocity that can be used to calculate lateral friction, but the solution still doesn't converge.
I've tried switching early from Picard to Newton iterations, but this doesn't help.
I attach my sif and mesh files. The bedrock is like the original MISMIP overdeepened bed. The setup is not very elegant, but it seems to work fine without the lateral friction. Any suggestions for how to get this setup to run smoothly with lateral friction would be welcome. I have mostly been running this on 16 processors, but the serial version fails quick enough (first timestep) with a parameterised channel width of 100km that it can be useful for debugging. Thanks for your help!
Rupert
The lateral drag code is described here:
http://elmerice.elmerfem.org/wiki/doku. ... alfriction
The basal drag law is a function of overburden pressure.
I am trying to spin up from a slab of uniform thickness. This works fine with no lateral friction, or with very low lateral friction (paramterised channel width 500km). But the non linear iterations of the Stokes solver do not converge when higher lateral friction is used (parameterised channel width 100km).
I've tried adding in the lateral friction gradually: a linear increase from zero to a parameterised channel width of 100km over 20,000 years, but the simulation failed after about 4000 years.
I've tried hard coding upper limits for the velocity that can be used to calculate lateral friction, but the solution still doesn't converge.
I've tried switching early from Picard to Newton iterations, but this doesn't help.
I attach my sif and mesh files. The bedrock is like the original MISMIP overdeepened bed. The setup is not very elegant, but it seems to work fine without the lateral friction. Any suggestions for how to get this setup to run smoothly with lateral friction would be welcome. I have mostly been running this on 16 processors, but the serial version fails quick enough (first timestep) with a parameterised channel width of 100km that it can be useful for debugging. Thanks for your help!
Rupert