[aspect-devel] Progress in writing the mantle convection code Aspect

[Rene Gassmoeller]

As I remember, the last time we changed the default stabilization 
parameters was a year ago, when we unified the chemical field and 
temperature field handling. At that time I tried to find parameters that 
where as low as possible to prevent unnecessary diffusion in benchmark 
cases like Rayleigh-Taylor and Stokes sphere. With this, we wanted to 
show that chemical fields can be advected with very little diffusion.
The resulting overshooting in the first time steps (5-10% in the 
beginning for a quite sharp interface sounds like the values I got 
there) seemed to decay over time, so I was fine with that.
It is always possible to increase beta (or Cr if the overshooting grows 
over time) at the cost of a bit higher numerical diffusion to reduce the 
overshooting, so if you are fine with that, feel free to test it. The 
best choice of beta and Cr is always a bit dependent on the model 
parameters, but of course we can discuss, whether we should increase 
them by default.

Best
Rene


On 10/17/2013 05:01 AM, Magali Billen wrote:
> We are modeling sinking of the oceanic lithosphere into the mantle (subduction). The lithosphere is not just a thermal
> boundary layer, but also has compositional layering: the basaltic crust, a harzburgitic layer (mostly olivine) beneath that
> and then normal mantle. Each of these compositions undergoes phase changes at different depths depending on its composition.
> Also each layer has different rheologic properties (harzburgite is dryer than mantle so stronger, basalt has a lower viscosity, etc...).
> So, this is what we'd like to be able model in Aspect. We've been doing this in CitcomS using tracer particles for each field but
> it requires a lot of particles and is computationally expensive.
>
> It would great to see how the stabilization parameter affected this.
> -Magali
>
> On Oct 16, 2013, at 4:42 PM, Timo Heister wrote:
>
>>> We're quantifying them for an AGU poster so we'll have more information then.  If I recall correctly, they are on the order of 5-10% of the gradient (so for a gradient of 10^4, the overshoot would be on the order of 10^2 to 10^3), versus up to 20-30% for CitcomS.  Finer resolution will make them more localized but generally worse in magnitude.
>> If you can set up simple to run examples, I am interested in seeing
>> them. You can likely remove overshoots by increasing the stabilization
>> parameter beta. What is your compositional field supposed to
>> represent?
>>
>> -- 
>> Timo Heister
>> http://www.math.clemson.edu/~heister/
> _______________________________________________
> Aspect-devel mailing list
> Aspect-devel at geodynamics.org
> http://geodynamics.org/cgi-bin/mailman/listinfo/aspect-devel