ParStokes and ice shelves

Extension of Elmer in computational glaciology
Post Reply
Posts: 6
Joined: 26 Sep 2014, 15:30
Antispam: Yes

ParStokes and ice shelves

Post by mewo2 » 28 Apr 2015, 18:05

I've been trying to run some small test simulations of an ice shelf using ParStokes, with an eye towards scaling up to looking at the long-term evolution of Larsen C. Unfortunately, I don't seem to be able to get the parallel solver to converge to anything remotely sensible. Identical runs with the standard Navier-Stokes solver work fine, as do similar runs with grounded geometry (both ParStokes and Navier-Stokes). As far as I can tell, it's just the combination of ParStokes and ice shelf type boundary conditions which breaks down.

So, two questions:
  • Is there anything in particular that has to be done to get ParStokes to play nice with Neumann boundary conditions? As far as I can tell from the code, it should "just work", but is there something I've missed?
  • Has anybody successfully used ParStokes in an ice shelf setup? Is there any special configuration required?

Site Admin
Posts: 82
Joined: 24 Aug 2009, 12:20

Re: ParStokes and ice shelves

Post by tzwinger » 30 Apr 2015, 09:01

Dear Martin,
if your "normal" Stokes solver produced a solution, then - in principle - the ParStokes should do so, too.
Could you please indicate which part of ParStokes refuses to converge: the outer (GCR) system or one of the block-preconditioning sub-problems?
Do you have grounding-line dynamics switched on? Perhaps there is an incompatibility in these procedures, which, to my knowledge, haven't been tested with the block pre-conditioner.

Best wishes,


Posts: 6
Joined: 26 Sep 2014, 15:30
Antispam: Yes

Re: ParStokes and ice shelves

Post by mewo2 » 01 May 2015, 14:23

Initially, using the actual ice shelf geometry, I thought that the GCR was failing to converge. However, I've been trying to reduce things to a simpler test case, and when I use a simple square geometry, I get (fairly slow) convergence of the GCR, but to a nonsensical result.

I haven't been using any grounding line dynamics - the domain is fully floating, with velocities specified along the upstream and lateral boundaries, and pressure boundary conditions (using the Buoyancy USF) on the downstream and lower boundaries.

I've attached a test case you can try for yourself. The "normal" Stokes solution is generated by seq.sif and the ParStokes by par.sif.
Test case .sif and .geo
(2.47 KiB) Downloaded 451 times

Posts: 48
Joined: 15 Sep 2009, 07:44

Re: ParStokes and ice shelves

Post by mika » 05 May 2015, 09:57


In the section of Boundary Condition 3 one should have

Normal-Tangential FlowVar = Logical True

instead of the original specification

Normal-Tangential Velocity = Logical True

After this change the simple test case appears to work for me.

- Mika

Post Reply