[CIG-SHORT] slower convergence in PyLith 1.4 compared with PyLith 1.3

[Brad Aagaard]

Tabrez-

If the shear modulus used in the nondimensionalization and all of the BC 
are the same for PyLith 1.3 and 1.4, you should see almost identical 
values for the residual at each iteration (last few digits could be 
different). This is what we see in the examples and benchmarks.

Brad


Tabrez Ali wrote:
> Brad
> 
> PyLith 1.3 was taking <40 iterations where as 1.4 with the following 
> normalizer options takes around 1250 iterations (same solver/pc):
> 
> normalizer.length_scale = 1.0*m
> normalizer.shear_modulus = 5.0e+10*Pa
> 
> All my dimensions are in meters and the shear modulus value is close to 
> the value used for the fault material in v1.3
> 
> Tabrez
> 
> 
> On Jul 23, 2009, at 2:48 PM, Brad Aagaard wrote:
> 
>> Tabrez-
>>
>> You need to be careful with the nondimensionalization scales. If you 
>> achieved good convergence with PyLith 1.3, in PyLith 1.4 you should 
>> setup the nondimensionalization scales so that the pressure scale 
>> matches the shear modulus associated with the fault physical 
>> properties used in PyLith 1.3.
>>
>> Brad
>>
>> Tabrez Ali wrote:
>>> Brad
>>> A model of mine (containing faults) that would converge fairly 
>>> quickly in PyLith 1.3 shows extremely slow convergence with PyLith 1.4.
>>> Are there any significant changes in the input file besides whats 
>>> listed below in your mail.
>>
> 
>