[CIG-SHORT] Dynamic faulting under gravity

Sun Nov 25 15:44:25 PST 2012

Birendra,

A condition number of 12 is very large and is likely a significant 
factor in the poor rate of convergence. You have two general options for 
trying to resolve very small scale geometric features: (1) create a mesh 
that resolves those features (and hopefully coarsens with distance) or 
(2) switch to adaptive mesh refinement methods.

For (1) you should switch to an unstructured mesh. Using a structured 
mesh would be inappropriate because it would use so much memory. For (2) 
I would recommend switching to deal.II.

The refine meshing feature in PyLith refines an entire mesh uniformly.

Regards,
Brad

On 11/25/2012 03:36 PM, Birendra jha wrote:
> Hi Charles
>
> Max. is around 12. I guess it has to be close to 1 or 2. Is that the only reason for nonconvergence of the nonlinear solver?
>
> CUBIT> quality volume all Condition No. global
>   Hex quality, 24360 elements:
> ------------------------------------
>     Function Name    Average      Std Dev      Minimum   (id)    Maximum  (id)
>   -----------------  ---------    ---------    --------------    -------------
>     Condition No.    1.118e+00    3.692e-01    1.000e+00 (2846)    1.231e+01 (24023)
>
> In that case, I have no option but to refine the mesh throughout which will give me a huge number of cells. Then I won't be able to run it on my machine, for sure.
>
>
> Can I use Pylith's refine mesh capability to help me here? Can it refine selectively in one material region?
>
>
> Thanks and regards
> Birendra
>
> --- On Mon, 11/26/12, Charles Williams <willic3 at gmail.com> wrote:
>
>
> From: Charles Williams <willic3 at gmail.com>
> Subject: Re: [CIG-SHORT] Dynamic faulting under gravity
> To: "Birendra jha" <bjha7333 at yahoo.com>
> Cc: cig-short at geodynamics.org
> Date: Monday, November 26, 2012, 3:32 AM
>
>
>
> Hi Birendra,
>
>
> What sort of maximum condition number values are you getting?  There still appear to be some very flat cells in your mesh.
>
>
> Cheers,
> Charles
>
>
>
>
>
> On 25/11/2012, at 5:53 PM, Birendra jha wrote:
>
>
> Hi all
>
> I removed gravity (and initial stress), improved the aspect ratio of my grid (condition number is much better now), still the nonlinear solver is not converging (linear solver converges in <50 iterations). SNES Function Norm decreases for first 2 iterations then increases (from 2.8E-02 to 3.08E-02 in 4 iterations).
>
> I have the attached the 3 files necessary to run the problem.
>
> Problem description: a slanted fault in a box domain. Dynamic fault. Static friction model with 0 cohesion.
> Boundary condition: roller boundary on -x, +y, -y. Dirichlet displacement on +x of -1 cm to apply compression on the fault.
>
> Please help!
>
> Thanks and regards
> Birendra
>
> --- On Thu, 11/22/12, Brad Aagaard <baagaard at usgs.gov> wrote:
>
>
> From: Brad Aagaard <baagaard at usgs.gov>
>
> Subject: Re: Dynamic faulting under gravity
>
> To: "Birendra jha" <bjha7333 at yahoo.com>
>
> Cc: cig-short at geodynamics.org
>
> Date: Thursday, November 22, 2012, 3:18 AM
>
> Birendra,
>
>
>
> You have very flat hex cells in your mesh. This results in a
>
> very poorly
>
> conditioned system. I think this prevents getting a decent
>
> guess for the
>
> slip, so the nonlinear solve fails to converge.
>
>
>
> It looks like you are trying to resolve very small scale
>
> geometric
>
> features with a coarse hexahedral mesh. I think you will
>
> need a mesh
>
> with much closer to cube-like cells to get the nonlinear
>
> solve to converge.
>
>
>
> Regards,
>
> Brad
>
>
>
>
>
> On 11/20/2012 11:32 AM, Birendra jha wrote:
>
>
> Hi,
>
>
>
>
>
> With that, the linear solver converges but nonlinear
>
> solver didn't converge in 100 iterations.
>
>
> To improve the gravity modeling, I also modified the
>
> boundary conditions such that now I have zero normal
>
> displacement on +x,-x,+y,-y,-z faces except the fault nodes
>
> on +y,-y,-z faces. With this setup, in absence of fault
>
> model in pylith, the case runs fine with almost 0
>
> displacement everywhere except on the fault nodes which are
>
> free. This is as expected. But with fault friction, it does
>
> not converge. The unconverged solution also shows large
>
> displacements on the fault nodes, I am just not sure if the
>
> numbers are accurate.
>
>
>
>
>
> The modified files are attached.
>
>
>
>
>
> Thanks
>
>
> Birendra
>
>
>
>
>
> --- On Tue, 11/20/12, Brad Aagaard <baagaard at usgs.gov>
>
> wrote:
>
>
>
>
>
>
> From: Brad Aagaard <baagaard at usgs.gov>
>
>
>
> Subject: Re: Dynamic faulting under gravity
>
>
>
> To: "Birendra jha" <bjha7333 at yahoo.com>
>
>
>
> Cc: cig-short at geodynamics.org
>
>
>
> Date: Tuesday, November 20, 2012, 10:28 PM
>
>
>
> Birendra,
>
>
>
>
>
>
>
> The linear solve did not converge. If the linear
>
> solves
>
>
>
> within the
>
>
>
> nonlinear solve doesn't converge, the nonlinear
>
> solve won't
>
>
>
> converge.
>
>
>
> You are using ASM for the preconditioner. Try using
>
> the
>
>
>
> field split
>
>
>
> preconditioner with ML on the elastic part and the
>
> custom
>
>
>
> fault
>
>
>
> preconditioner. See examples/3d/tet4/step04 for
>
> PETSc
>
>
>
> settings.
>
>
>
>
>
>
>
> Brad
>
>
>
>
>
>
>
> On 11/19/2012 09:36 PM, Birendra jha wrote:
>
>
>
>
> Dear all
>
>
>
>
>
>
>
>
>
> I have a relatively simple case of a slanted
>
> fault
>
>
>
> plane (along y axis) inside a layered box. I am
>
> using
>
>
>
> FaultCohesiveDyn with static friction model (please
>
> see the
>
>
>
> attached file). +x, -x boundaries are fixed in x,y
>
>
>
> directions and bottom is fixed in z direction. y
>
> faces are
>
>
>
> free.
>
>
>
>
> I have gravity on and I use
>
> initial_stress.spatialdb
>
>
>
> with a density of 2400 kg/m3.
>
>
>
>
>
>
>
>
>
> The nonlinear solver is not converging. Can you
>
> please
>
>
>
> help? What am I missing? There are four thin layers
>
> in the
>
>
>
> grid sandwiched between thick layers above and
>
> below. Is
>
>
>
> that the problem?
>
>
>
>
>
>
>
>
>
> I am using the Pylith 1.8.0 windows binary.
>
>
>
>
>
>
>
>
>
> Thanks and regards
>
>
>
>
> Birendra
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
> <faultedgrid3.zip>_______________________________________________
> CIG-SHORT mailing list
> CIG-SHORT at geodynamics.org
> http://geodynamics.org/cgi-bin/mailman/listinfo/cig-short
>
>
>
>
>
> Charles A. Williams
> Scientist
> GNS Science
> 1 Fairway Drive, Avalon
> PO Box 30368
> Lower Hutt  5040
> New Zealand
> ph (office): 0064-4570-4566
> fax (office): 0064-4570-4600
> C.Williams at gns.cri.nz
>
>
>
> _______________________________________________
> CIG-SHORT mailing list
> CIG-SHORT at geodynamics.org
> http://geodynamics.org/cgi-bin/mailman/listinfo/cig-short
>