THIS IS AN AUTOMATED MESSAGE, DO NOT REPLY.

The following task has a new comment added:

FS#238 - Improve terminology in Newton output / parameters
User who did this - Bernd Flemisch (bernd)

----------
Out of a private email from Jos Maas:
I looked at your terminology issue on  absolute and relative errors. I agree 
that it is currently confusing and that probably there is no standard 
terminology.  On the other hand, it must be possible to come to a more clearer 
description of what is being monitored when dealing with the tolerance 
settings. In my view, the confusion that exists now is 2-dimensional: your 
"absolute error" is not absolute but relative (normalised) just like your 
relative error; and secondly, the residuals and the "delta's" are measuring 
different things.

Firstly, I think it is important to keep a clear distinction between the error 
margins of solving the material balance equations, i.e. the residual of the 
functions on the one hand, and looking at delta's in the primary variables on 
the other hand. The delta's are often called "shifts".

The residuals drive the NR process and translate immediately into a 
mass-balance error, either over each cell/box individually or over the whole 
model. The mass-balance error can be reported as an average, possibly weighted 
and there are several choices one can make for the weighing. A mass-balance 
tolerance therefore could be offered to the user, either in absolute terms 
(kg's) or in relative terms, un-weighted or weighted.

A tolerance on shifts is less clear in how that impacts the final results in 
terms of e.g. production forecast of an oil or gas reservoir. Clearly, a 
tolerance on shifts may need to take into account scaling: a shift in pressure 
is something completely different from a shift in saturation (or temperature 
for thermal runs, assuming the DuMux would use temperature as a primary 
variable (I never used thermal in DuMux yet)). One could offer a tolerance on 
relative shifts, or even separate tolerances on separate variables, i.e. a 
maximum pressure shift, a maximum saturation shift and a maximum temperature 
shift. Different problems may require different tolerance uses.

For your information, in Shell people would focus on mass-balance errors and 
have less interest in maximum shifts. For my simulations of core analysis 
experiments, both the errors and the maximum shifts are important, but still I 
would first require a good tolerance on the mass-balance (1.E-14 or smaller, 
for your current version of "NewtonAbsTolerance") and then be interested in a 
maximum shift in saturations (1.E-10).


----------

More information can be found at the following URL:
http://www.dumux.org/flyspray/index.php?do=details&task_id=238#comment498

You are receiving this message because you have requested it from the Flyspray 
bugtracking system.  If you did not expect this message or don't want to 
receive mails in future, you can change your notification settings at the URL 
shown above.
_______________________________________________
Dumux mailing list
[email protected]
https://listserv.uni-stuttgart.de/mailman/listinfo/dumux

Reply via email to