Hi,

I think we have still included a maximum step size.
Which I guess would be used to try and ensure that an integrator does 
not just completely step over an impulse.

There is nothing to suggest that the tabulation step size is a multiple 
of the maximum integrator step size, and therefore redundant.  Also they 
will always be misaligned (requiring interpolation) unless the minimum 
step size is so small that an adaptive scheme never reduces it further.

Hopefully your maximum step size and your tabulation step size are much 
larger than the step size sometimes used by the adaptive integrator. 
Here forcing results to come back at a fixed spacing forces you not to 
capture this signal properly in your graphs.  This seems really 
unfortunate, you may have correctly integrated a complicated action 
potential and then you draw a graph which completely misses it!  This 
seems really bad.  Isn't it much better is to concentrate your line 
segments in your graph in exactly the same places where the adaptive 
step size system found your solution to be complicated?

If an integrator produces "different" results depending on the 
tabulation step size then it sounds as if it hasn't converged and so the 
adaptive step size isn't working, maybe your maximum step size needs to 
be further reduced for the problems you point out about missing the 
important part?  If your tabulation is a multiple of your maximum step 
size as you suggested then I can't understand how this can happen.
(I do realise that there are probably a number of models that are 
dependent on not being converged.)

And if your integration solution is affected by the tabulated result 
spacing what if you don't want tabulated results?  You just want the end 
integration, then you have to tell your integrator to generate some 
special tabulation just to get the same final state from your model?

After thinking about where the interpolation should be performed, I 
expect that most integrators will end up interpolating somehow to 
generate values at the exact grid you specified, I do agree that the 
integrator may be able to give a far more accurate interpolation, as 
presumably it has a far finer set of values than the ones returned as 
results, however requiring a grid, specified in the model metadata, 
doesn't seem the best way to specify this.

I didn't even bring this up yesterday as I was ignored last time I 
brought it up, but it really seems wrong to me to be forcing your 
integrator to give results on a grid in every case, and even more 
concerning to me if as you suggest, the performance of your model is 
dependent on this grid.
However I am not a user or intended user of these ODE integrators.  My 
experience only comes from the ODE integrator that I maintain in CMGUI 
which is for tracking streamlines and streaklines.
So if I am outvoted I'll just take my concerns back into my hole.

Shane.
_______________________________________________
cellml-discussion mailing list
[email protected]
http://www.cellml.org/mailman/listinfo/cellml-discussion

Reply via email to