I have tested out the “Wood’s funcion in four parameters” minimization problem from the Minuit tutorial (http://seal.web.cern.ch/seal/documents/minuit/mntutorial.pdf)
When I minimize I get an error that “Edm is above max”.
This is because the estimated vertical distance to the minimum is above the max.
I am not sure how I can make this work.
I have attached the source code.
Regards
Lars Schouw
testWoods.C (2.26 KB)
I tried with the Powells example but ran into the same problem.
Attached source code.
Lars
testPowells.C (2.24 KB)
Your two tests work correctly with TMinuit.
Lorenzo will investigate the problem with minuit2
Rene
Hi Lars,
thank you for your tests.
In the case of the Powell function, I found a problem in Simplex which I will fix now in CVS. In general, I would recommend you in this minimization problems to increase the parameter error to something larger like 0.1 or 1.
For the Woods function, I think the formula in the tutorial must have an error. When I run simplex on your example I get the minimum at
FVAL = -3.9, when the mimimum of the Wood function is at Zero.
FVAL = -3.92377
Edm = 3.41739e-10
Nfcn = 501
w = 0.584371 +/- 0.282853
x = 0.335878 +/- 0.159534
y = 0.985087 +/- 0.619821
z = -1.64947e-11 +/- 1.5409e-06
Minuit2 : minimum values for -3,-1,-3,-1
par0 = 0.584371
par1 = 0.335878
par2 = 0.985087
par3 = -1.64947e-11
I will try to find out the correct formula. I think it might be the one I modified in the attached script,
Cheers,
Lorenzo
testWoods.C (2.32 KB)