[CIG-SHORT] SCEC Banchmarks

[Brad Aagaard]

Huihui,

In your previous email you mentioned changing the time step, but you 
didn't mention if you changed the scales for the nondimensionalization 
or any of the other parameters. This is likely affecting the results.

If you run the simulation at 100m resolution at time step of 0.01s, the 
simulation can resolve shear waves (assuming 10 cells per wavelength) 
with periods down to about 0.3s. If you run the simulation at 50m 
resolution with a time step of 0.005s, the minimum wave period decreases 
to 0.15s. In dynamic simulations, we use the shear wave speed, the mass 
density, and minimum wave period to nondimensionalize the problem. This 
means, that as you change the spatial resolution, you should be changing 
the corresponding nondimensionalization parameters as well, in this case 
the minimum period. However, changing the time step alone does not 
change the minimum period, so the scales should remain unchanged.

However, there can be other side effects of changing just the time step. 
For example, with explicit time stepping the normalized viscosity to 
reduce high frequency oscillations is scaled by the time step (see 
Section 4.2.2 of the PyLith Manual). In most cases, we do not change the 
normalized viscosity as we change the spatial resolution. However, 
changing the time step alone will alter the artificial damping. Thus, 
you will need to adjust the normalized viscosity accordingly if you 
change the time step to maintain the exact same problem. I don't think 
you will need to change the zero sliding tolerance.

Regards,
Brad


On 12/03/2014 08:49 AM, Brad Aagaard wrote:
> Huihui,
>
> To assess convergence, you should be looking at both the time step and
> discretization size and whatever fields you are interested in. The
> solution is generally more sensitive to discretization size than time
> step, so I would first assess convergence by varying the discretization
> size.
>
> It is also important to note that not all of the SCEC benchmarks are
> well formulated so that the problem descriptions are independent of time
> step and discretization size. Additionally, the ruptures in some cases
> (including TPV205) abruptly hit the edges of the fault rather than
> gradually dying out. As a result, it takes a long time for the slip rate
> to decrease to zero.
>
> Regards,
> Brad
>
>
>
> On 12/03/2014 01:30 AM, 翁辉辉 wrote:
>>
>>
>> Dear Brad,
>>
>>
>> I am trying to run some SCEC Dynamic Rupture Benchmarks cases. I use
>> different time_steps(0.01s,0.05s,0.001s,0.0005s,all are smaller than
>> the stable time_step) to run a same
>> model(tpv205-2d,tri3_100m_gradient.cfg). I find that the coseismic
>> slip of the model with different time_steps are different(show as the
>> attached figure).All these cases are run by pylith 2-0-0. And the
>> rupture velocity seems to decrease as the time_step decreases. How to
>> choose the appropriate time_step?
>>
>> ps. that model is in the CIG SVN Repository
>> https://geodynamics.org/svn/cig/short/3D/PyLith/benchmarks
>>
>> regard,
>>
>> Huihui
>>
>