Minuit, Fumili & cie clarification?

ROOT
1

Hi,

I would like to make some likelihood tests with Minuit and Fumili, but things remain unclear to me : are there fondamental differences between Minuit (resp. Fumili) and Minuit2 (resp. Fumili2) ? I mean, is there any reason for which I would want to use the first instead of the latter, or are they kept only for backward compatibility ?

I’ve also read that Fumili was much faster than Minuit, is it also less accurate ? How to chose one or the other, just by trial and error ?

Thanks for any clarification.

2

[quote]I would like to make some likelihood tests with Minuit and Fumili, but things remain unclear to me : are there fondamental differences between Minuit (resp. Fumili) and Minuit2 (resp. Fumili2) ? I mean, is there any reason for which I would want to use the first instead of the latter, or are they kept only for backward compatibility ?
[/quote]
We would like to get more feedback from people using Minuit2,Fumili2.
We have no plans to remove Minuit/Fumili in the short/medium term, too many applications
using these packages directly.

[quote]I’ve also read that Fumili was much faster than Minuit, is it also less accurate ? How to chose one or the other, just by trial and error ?
[/quote]
I would say Fumili is faster than Minuit if you have an easy case close to the solution,
but Minuit is probably safer. Again it would be nice to get more input.

Rene

3

hmm,

I hear only silence …

Have a look at this

root.cern.ch/viewcvs?rev=23610&r … t&view=rev

Still silent ?

4

Sorry for that, I have put these tests aside for the moment. Actually I plan to make a few tests against NAG libraries that we are currently using, so I’ll try to think about keeping you informed. Don’t expect any feedback before a few weeks though…

Thanks for the link and the precisions.