Steven G. Johnson <stevenj....@...> writes:

> 
> No, that is fundamentally not how these algorithms work...
> 
> 
> ...I wouldn't call the fractional steps a "waste" then---they are what  
> allow the algorithm to find an optimum (or at least a local optimum)  
> without exponential complexity...
> 
> 
> Steven
> 


Steven,

Thanks so much for your assistance!  I greatly appreciate it.

I was under the impression that having these fractional steps sacrificed speed
for the sake of accuracy.  This was obviously an ill-founded assumption and I
apologize. 

As you know, I'm only concerned with finding an optimum on the scale of whole
numbers.  However, I am not as concerned about having to round or truncate my
doubles to integers as I am with the speed of the algorithm.  

My assumption now is that the only way to increase the speed or rather, decrease
the precision of the algorithm is to set xtol_abs1 to 1.0 (in my case) or ftol
to reflect some acceptable %error.

If there is any other way to speed up the algorithm I would love to hear it. 
Otherwise, in the absence of a response I will just assume the way I described
above is the only way.

Thanks again for your help!

-Adam Openshaw





_______________________________________________
NLopt-discuss mailing list
[email protected]
http://ab-initio.mit.edu/cgi-bin/mailman/listinfo/nlopt-discuss

Reply via email to