[aspect-devel] mesh refinement strategy question

Fri Jun 8 09:19:44 PDT 2018

Magali,
When setting up similar models, I have also found that an alternative to
specifying the minimum refinement function is to run one timestep with the
coarsening fraction set to zero and the refinement fraction set to a
relatively large value. Then restart for the remainder of the calculation
with both coarsening and refinement.

Max

On Thu, Jun 7, 2018 at 11:51 PM, Magali Billen <mibillen at ucdavis.edu> wrote:

> Thank you Renee and Felipe
> Both the time-dependent approach or restarting are good options for me. I
> had read that it
> could be time-dependent, but it had not registered. Also, I might first
> run this initial model as purely conductive for a few million years to
> smooth the temperature discontinuity at the top of the slab before running
> the full dynamic simulation.
>
> Also, thanks for clarifying that the refinement scheme does re-evaluate
> the initial temperature field at
> each initial refinement. I was wondering about that when I tried that, but
> I had not fully thought about
> how to control the amount of refinement with the refinement parameters.
> I’m guessing that whatever parameters I would need to get this to work for
> the first time-step would end up being in appropriate for
> the rest of the run, but its useful to think about, as it could also be
> used with the restart option, as this
> initial temperature field would only need to be created once and then
> could be re-used for various models
> exploring other things (like the rheology).
>
> And don’t worry, I’m not planning on using  Time steps between mesh
> refinement = 1, I’m just slowly
> building the model input and testing these before moving to a
> time-dependent simulation.
>
> -Magali
>
>
> On Jun 8, 2018, at 12:38 AM, Rene Gassmoeller <
> rene.gassmoeller at mailbox.org> wrote:
>
> Hi Magali,
> there are several ways to achieve this, with slightly different effects,
> so you will probably want to try a number of them so see what works best
> for you. The simplest guess is probably to make the minimum refinement
> criterion time-dependent. E.g. you can replace the following line for the
> minimum refinement function:
>
>     set Function expression = if( (YMAX-y<=LITH)||((YMAX-y<=SM)&
> &(x<=XTR+3*RC)&&(x>=XTR-0.5*RC)),9,0)
>
> by this:
>
>     set Function expression = if (t<1e5, if( (YMAX-y<=LITH)||((YMAX-y<=SM)&
> &(x<=XTR+3*RC)&&(x>=XTR-0.5*RC)),9,0), 0)
>
> which would disable the minimum refinement function after 100kyr (or any
> other time you want). I do not quite remember if you have to put 't' into
> the list of variable names of the function for this to work, or not (try
> that if it does not work out of the box).
>
> A second approach could be to run a model with an end time of 0 and write
> a checkpoint once it has finished (see Section 4.5 in the manual, and the
> parameter 'Termination criteria/Checkpoint on termination'). You can then
> restart from this checkpoint and remove the mimum refinement function for
> the restarted model (this also works at any point later in time).
>
> Lastly, the brute force approach of starting at the finest resolution and
> only coarsen should also work. I do this occasionally for testing purposes,
> but for production models it is often too expensive.
>
> Let us know if one of those approaches works for your purpose.
>
> A final remark, you mentioned "...  It does not work to just use Strategy
> = temperature, because this just refines on the temperature field defined
> on the coarsest mesh (see picture).  ...". I do not think this is what we
> do, we reapply the initial conditions before every initial refinement on
> the new mesh and then make the next initial refinement (this happens in
> lines 1593-1625 in core.cc if you use ASPECT 2.0, might be slightly
> different for other versions).
> What could be a problem though is that ASPECT's algorithm looks to refine
> a certain "percentage" of the error in a particular field. It looks like
> your initial condition has a discontinuous jump in temperature right at the
> slab edge. Thus, the algorithm might select some cells right at the edge,
> and then decide since the refined error is so much bigger than everything
> else, that is has already refined the error "enough", and just coarsen the
> rest of the cells. This is of course not what we would want in this case,
> but without changes to the underlying algorithm or the initial condition
> this could be 'technically correct'.
>
> Hope that helps a bit,
> Rene
>
>
> On 06/07/2018 08:48 AM, Magali Billen wrote:
>
> Hello,
>
> I am starting a project for subduction modeling in Aspect and I have a  question
> about how certain mesh refinement strategies work together. Specifically, I
> need to define a “proto-slab” based on temperature,
>  which is needed to start the subduction process. However, once the slab
> starts sinking its location in the model domain depends on the dynamics and
> it can move 1000s of kilometers from its starting location.
>
> So, I have first set this up using the  Strategy = minimum refinement
> function (see below for the
> full subsection definition). And, I can get a nice looking initial slab
> (see below).  However, after the initialization,
> I would like the mesh refinement to depend on things like the temperature,
> (and later composition, and strain-rate).  It does not work to just use
> Strategy = temperature, because this just refines on the temperature field
> defined on the coarsest mesh (see picture).
>
> So, what is not clear to me is whether there is a way to use one Strategy
> during the initial adaptive refinement
> and a different (set) of criteria after that point, or to combine them
> such that the minimum refinement function
> is no longer have much impact after the first step.
>
> Has anyone else had a problem like this?  Any ideas?
> Instead of thinking about refining, should I start with a really fine mesh
> everywhere and then use the
> Strategy = temperature to coarsen the mesh where refinement isn’t needed.
> It doesn’t seem like it set up
> to work that way?
>
> - Magali
> Here’s the prm file for the nice looking slab using the function to refine
> based on location:
>
>
>
> Here’s just the mesh  refinement section:
>
> subsection Mesh refinement
>   set Initial global refinement                = 4
>   set Minimum refinement level                 = 4
>   set Initial adaptive refinement              = 5
>   set Time steps between mesh refinement       = 1
>   set Strategy = minimum refinement function
>   subsection Minimum refinement function
>     set Coordinate system   = cartesian
>     set Variable names      = x,y
>     set Function constants = YMAX=1e6,LITH=1.5e5,SM=2.5e5,XTR=2.2e6,RC=4e5
>     set Function expression = if( (YMAX-y<=LITH)||((YMAX-y<=SM)&
> &(x<=XTR+3*RC)&&(x>=XTR-0.5*RC)),9,0)
>   end
> end
>
> And, here are some screen-shots
>
> Full domain
>
>
>
>
>
>
> Zoom-in on slab
>
>
>
>
> This is what happens if I just use Strategy = temperature
>
>
>
> And if I combine them  (this is just the 0th time-step), I need to set up
> some other things before
> I can run forward in time.
>
>
>
> ____________________________________________________________
> Professor of Geophysics
> Earth & Planetary Sciences Dept., UC Davis
> Davis, CA 95616
> 2129 Earth & Physical Sciences Bldg.
> Office Phone: (530) 752-4169
> http://magalibillen.faculty.ucdavis.edu
>
> Currently on Sabbatical at Munich University (LMU)
> Department of Geophysics (PST + 9 hr)
>
> Avoid implicit bias - check before you submit:
> http://www.tomforth.co.uk/genderbias/
> ___________________________________________________________
>
>
>
> _______________________________________________
> Aspect-devel mailing listAspect-devel at geodynamics.orghttp://lists.geodynamics.org/cgi-bin/mailman/listinfo/aspect-devel
>
>
> --
> Rene Gassmoellerhttps://gassmoeller.github.io/
>
> _______________________________________________
> Aspect-devel mailing list
> Aspect-devel at geodynamics.org
> http://lists.geodynamics.org/cgi-bin/mailman/listinfo/aspect-devel
>
>
> ____________________________________________________________
> Professor of Geophysics
> Earth & Planetary Sciences Dept., UC Davis
> Davis, CA 95616
> 2129 Earth & Physical Sciences Bldg.
> Office Phone: (530) 752-4169
> http://magalibillen.faculty.ucdavis.edu
>
> Currently on Sabbatical at Munich University (LMU)
> Department of Geophysics (PST + 9 hr)
>
> Avoid implicit bias - check before you submit:
> http://www.tomforth.co.uk/genderbias/
> ___________________________________________________________
>
>
> _______________________________________________
> Aspect-devel mailing list
> Aspect-devel at geodynamics.org
> http://lists.geodynamics.org/cgi-bin/mailman/listinfo/aspect-devel
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.geodynamics.org/pipermail/aspect-devel/attachments/20180608/82142ed6/attachment.html>