Time steps/stages in Numerics field

[Michael Hurd]

I understand that the adaptive time step control can be used to resolve convergence failures. What I'm struggling with is the criteria for selecting the number of steps and Maximum stages. I've read some discussion within the help sections of WUFI but am having difficulty understanding how to select the number of steps and found nothing regarding the selection of stages. Is there some additional guidance you can give us on the use of this function? Also, the help section seems to imply that if you have hourly weather data you should not select a higher number of steps than the 1 hour, I believe that all the weather data for the cities in the United states is hourly, so if I select more steps than 1 am I using this function incorrectly?

[Manfred Kehrer]

Hello Michael,

the time steps and ATSC steps are not the same. To be more precise the ATSC steps are the amount of sub steps of one time step.

For Instance: Time step=0.5 h, ATSC steps = 5 means, if no convergence is achieved in any time step WUFI will recalculate this time step by making 5 times a 0.5/5 time step.

Which combination of ATSC Steps,Stages work the best depends on the specific calculation case so just play around with these numbers.

The only thing I can say is the higher the numbers the longer a calculation can take, but will achieve convergence more likely, too.

[Thomas]

I understand that the adaptive time step control can be used to resolve convergence failures. What I'm struggling with is the criteria for selecting the number of steps and Maximum stages. I've read some discussion within the help sections of WUFI but am having difficulty understanding how to select the number of steps and found nothing regarding the selection of stages. Is there some additional guidance you can give us on the use of this function?

I'm afraid there are no easy a-priori criteria. In general, you simply increase the numbers until it works.

The number of subdivision "Steps" increases the numerical work and thus the required computation time. If you tell WUFI to subdivide a problematic time step into, say, 8 smaller steps, then WUFI will have to perform roughly 8 times the computational work for such a time step. If one of these substeps is still not successful and WUFI has to proceed to the next stage, subdividing this substep again, there is another 8-fold increase in computation time. So this parameter should not be set much higher than actually needed (but you'll have to find out by trial and error how much is needed).

You can set the "Maximum Stages" to the highest available number (currently "10") without any penalty. If the Adaptive Time Step Control doesn't need the number of subdivision stages you allowed, it simply doesn't use all of them.

If numerical problems need to be addressed by the Adaptive Time Step Control (because adapting the numerical grid didn't solve the problems), I would suggest to simply enable the time step control, using the default parameters for 'Steps' and 'Max. Stages', and to repeat the calculation. If this still doesn't help, you can try increasing the parameters until (hopefully) the problems disappear. It is difficult to tell which one of the two adjustable parameters can solve the problems more efficiently. This seems to depend on the kind of numerical problems.

Leaving the Adaptive Time Step Control enabled all the time doesn't hurt. If no numerical problems occur, the time step control is not needed, and the computation time is not affected. However, we feel that silently resolving numerical problems might often be misleading or even counterproductive, and the user should in general be made aware of the presence of problems. This will give the user a chance to try more direct solutions first, for example revising the numerical grid, or checking material data for possible inconsistencies.

For the same reason, the "Max. Stages" has not simply been set to a fixed number (which would not incur a loss of computational performance, as explained above). If severe problems require the user to fiddle around a bit more, he/she may be made more acutely aware of the presence of problems, assessing the results more critically.

Also, the help section seems to imply that if you have hourly weather data you should not select a higher number of steps than the 1 hour, I believe that all the weather data for the cities in the United states is hourly, so if I select more steps than 1 am I using this function incorrectly?

If you have hourly weather data, then in general there is simply no point in using finer time steps. You are always allowed to set the default time step (in the dialog 'Calculation Period/Profiles') to something smaller (say, 0.5 hours or 0.1 hours or 0.001 hours...) but the calculation results will be basically the same, and they will come at the price of higher computation time.

See the topic 'WUFI's Performance and Limitations' in the Appendix of the online help for a discussion of the effect of different time step widths.

So under normal circumstances, using a finer time step is not wrong, it's just not needed. If numerical problems occur, however, finer time steps (and the correspondingly smaller thermal and hygric movements to be simulated during a time step) may be able to overcome these problems.

Kind regards,
Thomas

[Michael Hurd]

As always, thanks for the help guys, I really appreciate your quick responses!