[CIG-LONG] Fwd: CIG-LONG Digest, Vol 47, Issue 2

Louis Moresi louis.moresi at monash.edu
Wed Oct 20 16:35:41 PDT 2010

It can't be a legacy thing, because some form of this condition was
originally in underworld, though there was an assumption that diffusivity
was unlikely to change which has been strangely difficult to knock out of
the code.

One would generally use the local diffusivity and mesh size to give the
limiting condition rather than the global extremes.



 louis.moresi at sci.monash.edu.au
GoogleProfile <http://www.google.com/profiles/louis.moresi>
Mobile phone: +61 4 2850 1907


2010/10/21 Laetitia Le Pourhiet <laetitia.le_pourhiet at upmc.fr>

> Hi all,
> I was intrigued by all this problems with thermal diffusion...
> I don't understand why this limitation for the time step is not included in
> gale, I guess it is historical since diffusion is negligible for convection
> problem and gale is based on underworld but as Charmaine showed, it is not
> negligible for post-rift simulation.
> Anyway
> I found that  FiniteElementContext.c contains a parameter we can all use
> for the moment to limit the time step without using the very dangerous dt
> parameter...
>  context->maxTimeStepSize = Dictionary_GetDouble_WithDefault(
> self->dictionary, (Dictionary_Entry_Key)"maxTimeStepSize", 0.0 );
> So providing that function is really used in Gale, it means that everybody
> has to compute by hand its diffusion timestep i.e. :
> dt_diff = sqrt(min_length_of_a cell)/max_diffusivity_in_your_model...
> and  add
> <param name="maxTimeStepSize">dt_diff</param> in your xml ....
> to their xml
> I think we should all include that in our simulation because for a 3km mesh
> and a diffusivity of 1e-6 it makes a timestep of 300ka which is about or a
> little bit less than what I get with dtfactor=0.5 for a 1cm/year crustal
> extension problem.
> hope this help
> Laetitia
> Charmaine Thomas wrote:
>> Hi Walter,
>> I did a series of experiments to get Gale to do pure thermal conductivity
>> problems without having a non-zero velocity boundary applied to the
>> right/left walls, ie velocity=0.
>> Firstly I tried turning off the Stokes flow and the uzawa condition, and
>> although everything ran very smoothly and quickly, it took ridiculously big
>> timesteps, despite having an explicitly set 'dt' (I assumed later that
>> Stokes flow has to be turned on for this parameter to kick in?). This
>> predictably resulted in very high temperatures. I had to do this experiment
>> because I was dealing with non-newtonian rheologies, and needed to  model a
>> time of quiescence in my crust.
>> Finally I found the best option was to leave the Stokes/uzawa stuff turned
>> on and to apply either a very low strain-rate, or set the right/left
>> velocities as zero, but also have a right and left wall stress boundary
>> condition. This second option runs smoothly even with non-newtonian
>> rheologies, and takes more reasonably sized timesteps - I could also try
>> changing the dtfactor to speed things up. So does this method sound more
>> reasonable? Is there a better way?
>> Cheers,
>> Charmaine Thomas
>> School of Geosciences | University of Sydney
>> On Wed, Oct 20, 2010 at 10:21 AM, Walter Landry <walter at geodynamics.org<mailto:
>> walter at geodynamics.org>> wrote:
>>    <Guillaume.Duclaux at csiro.au> wrote:
>>    > Indeed.
>>    >
>>    > But, shouldn't it be possible to solve purely thermal problem with
>>    > Gale?  (let's pretend the thermal expansion is null is Nicolas'
>>    > problem).  ie a sill at a temperature of 1000 K has intruded a mass
>>    > of rock at constant temperature (600 K) and I want to simulate the
>>    > thermal evolution of the system as I change the thickness of the
>>    > dyke or the radiogenic heat production of one or the other material.
>>    It is possible to do pure thermal conductivity problems with Gale.
>>    You have to turn off all of the Stokes flow stuff, but it does seem to
>>    work.
>>    > To ensure the solver timestepping is not missing the temperature
>>    > perturbation timescale, how should the time be scaled?
>>    > I guess viscosity doesn't matter if the problem is purely thermal,
>>    > but as soon as the thermal expansion is on, some body forces act too
>>    > creating some 'slow' displacement.
>>    For this particular case, the displacement is so slow that it can be
>>    neglected.  If you still want to solve the Stokes flow, then you can
>>    set the timestep explicitly with 'dt' (see Appendix A.1.4).  Gale
>>    should probably take the thermal diffusivity into account when
>>    deciding upon a timestep, but it does not do that now.
>>    Cheers,
>>    Walter Landry
>>    walter at geodynamics.org <mailto:walter at geodynamics.org>
>>    _______________________________________________
>>    CIG-LONG mailing list
>>    CIG-LONG at geodynamics.org <mailto:CIG-LONG at geodynamics.org>
>>    http://geodynamics.org/cgi-bin/mailman/listinfo/cig-long
>> ------------------------------------------------------------------------
>> _______________________________________________
>> CIG-LONG mailing list
>> CIG-LONG at geodynamics.org
>> http://geodynamics.org/cgi-bin/mailman/listinfo/cig-long
> _______________________________________________
> CIG-LONG mailing list
> CIG-LONG at geodynamics.org
> http://geodynamics.org/cgi-bin/mailman/listinfo/cig-long

louis.moresi at monash.edu
GoogleProfile <http://www.google.com/profiles/louis.moresi> | GoogleCalendar
Geodynamics Software <http://www.underworldproject.org/>
Mobile phone: +61 4 2850 1907 ☎
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://geodynamics.org/pipermail/cig-long/attachments/20101021/ed43e138/attachment.htm 

More information about the CIG-LONG mailing list