My calculation was interupted at 96 % and it said that Maxit was to small..What does this mean? can i use the calculation even though it interupted it?
should i just increase maxit to solve this problem?
Best regards
calculation interrupted
Re: calculation interrupted
Please see explanation in following post
http://www.wufi-forum.com/viewtopic.php ... axit#p2100
Regards,
http://www.wufi-forum.com/viewtopic.php ... axit#p2100
Regards,
Re: calculation interrupted
Okay i have tried increasing my maxit and I have experienced the following:veitner wrote:Please see explanation in following post
http://www.wufi-forum.com/viewtopic.php ... axit#p2100
Regards,
- Calculation interrupted - maxit to small (ended at 97 %)
- Calculation interrupted - maxit to small (ended at 100 %)
What does this percentage mean? I understand that it is a measure of how far the calculations but fx when it interrupts at 97 % - is it then missing calculations for 3 % of the model geometry ?
Best regards
Re: calculation interrupted
I'm sorry but I do know knothing about this percentage. Where did you find that?Calculation interrupted - maxit to small (ended at 97 %)
I only know
- maxit_msg.png (622 Bytes) Viewed 11634 times
Re: calculation interrupted
Oh i can see my post could be misunderstood - its not that it displays "(ended at 97 %)" im talking about the large blue line at the bottom of the screen. Sorry for the misconception.veitner wrote:I'm sorry but I do know knothing about this percentage. Where did you find that?Calculation interrupted - maxit to small (ended at 97 %)
I only know .
Best regards
Re: calculation interrupted
I see - the blue bar is an overall progress indicator e.g. 97% means current_time_step/total_time_steps=0.97
And *** interrupted - maxit too small *** means the convergence criterion was not reached within maxit iterations - which does not necessarily mean you got a "wrong solution" but there is a chance to get a slightly better solution for that specific timestep by increasing maxit. Sometimes one can get some kind of oscillation: current temperature field results in temperature dependent moisture transport coefficients leading to a moisture field yielding to different moisture dependent heat transport coefficients resulting in different temperature field and so on - without finding a "converged solution". WUFI assumes a solution of both temperatur and r.h. as converged as soon as two subsequent solutions of temperature and r.h. do not differ more than "convergence crit." (maximum norm, residual). The independent solutions of both temperature and r.h. are exact for each single inner iteration - but the coupled solution has not underrun the user definable convergence criterion perhaps because of huge nonlinearity.
Following chart (hypothetical values) might illustrate the iterative process and meaning of the convergence criterion.
iterative Naeherungsloesung: iterative approximative solution
unbekannte Loesung: unknown solution (the exact or analytical solution)
Unterschied zum vorhergehenden Loesungsschritt: difference to last solution (residual)
Konvergenzkriterium: convergence criterion
Abweichung: difference between "converged solution" and "exact solution"
The left axis is for the red and the black line
The right axis for the blue and dark green line
The convergence criterion has been reached at about iteration 800 (light green line) and at this point the difference between the exact and iterative solution is about 0.0056 degree celsius.
But remember: the exact solution is unknown - therefor WUFI
And *** interrupted - maxit too small *** means the convergence criterion was not reached within maxit iterations - which does not necessarily mean you got a "wrong solution" but there is a chance to get a slightly better solution for that specific timestep by increasing maxit. Sometimes one can get some kind of oscillation: current temperature field results in temperature dependent moisture transport coefficients leading to a moisture field yielding to different moisture dependent heat transport coefficients resulting in different temperature field and so on - without finding a "converged solution". WUFI assumes a solution of both temperatur and r.h. as converged as soon as two subsequent solutions of temperature and r.h. do not differ more than "convergence crit." (maximum norm, residual). The independent solutions of both temperature and r.h. are exact for each single inner iteration - but the coupled solution has not underrun the user definable convergence criterion perhaps because of huge nonlinearity.
Following chart (hypothetical values) might illustrate the iterative process and meaning of the convergence criterion.
- konvergenz.png (12.32 KiB) Viewed 11631 times
unbekannte Loesung: unknown solution (the exact or analytical solution)
Unterschied zum vorhergehenden Loesungsschritt: difference to last solution (residual)
Konvergenzkriterium: convergence criterion
Abweichung: difference between "converged solution" and "exact solution"
The left axis is for the red and the black line
The right axis for the blue and dark green line
The convergence criterion has been reached at about iteration 800 (light green line) and at this point the difference between the exact and iterative solution is about 0.0056 degree celsius.
But remember: the exact solution is unknown - therefor WUFI
Re: calculation interrupted
Here comes another chart illustrating the residuals comparing two different approaches of source term linearization.
This shows the real course of the residuals during the iterative process for a single time step.
You can create a similiar chart by setting "Print residuals at iteration modulo" (computational settings/enhanced) to 1. Then redo the calculation and open the "result analyzer" (shows up without logarithmic axis scaling). Do not forget to reset the "print residuals modulo" before saving your data because a value of 1 leads to a really huge "iter.txt" file.
- iter_small.png (4.18 KiB) Viewed 11631 times
You can create a similiar chart by setting "Print residuals at iteration modulo" (computational settings/enhanced) to 1. Then redo the calculation and open the "result analyzer" (shows up without logarithmic axis scaling). Do not forget to reset the "print residuals modulo" before saving your data because a value of 1 leads to a really huge "iter.txt" file.
Re: calculation interrupted
Tanks, makes a lot mre sense now 
Best Regards
Best Regards
Re: calculation interrupted
But just to clarify one last thing: If it interrupts the calculations at 97 % saying that maxit is to small. Does it mean that 97 % of the time interval have converged?
Best Regards
Best Regards
Re: calculation interrupted
Hmhh - it does not interrupt the calculation at 97%
It does not reach the convergence criterion at timestep n/8760=0.97->n=0.97*8760=8498 (assuming you intend to simulate 1 year=8760 timesteps of 1 h each).
After that it continues with timestep 8499 using the "unconverged" solution as a starting point. But unconverged does not mean bad - it just could be better and has not reached the user configurable convergence criterion.
It does not reach the convergence criterion at timestep n/8760=0.97->n=0.97*8760=8498 (assuming you intend to simulate 1 year=8760 timesteps of 1 h each).
After that it continues with timestep 8499 using the "unconverged" solution as a starting point. But unconverged does not mean bad - it just could be better and has not reached the user configurable convergence criterion.