Fit personal function and chi square getting

ROOT
1

Hi everyone, I’ve to fit this plot

it’s a power law

dN/dE = E^(-2)

I want the fit and the chi square and I want to write the results of the fit and chi square on the plot (in the statistical box)

I wrote the macro fitspettro.c but as you can see it doesn’t work because:

  1. The line of the fit isn’t right
  2. I don’t have the fit and chi square values in the box on the plot
  3. On the root panel the values of the fit aren’t right (as I wrote it’s a power law E^-2, so [0] must be 2);
  4. The box where there is written “Fit” is so big and on the plot, how can I change the size and move it on the right?

Thanks
fitspettro.c (1.51 KB)

2

Try: TF1 *fitspettro = new TF1("fitspettro", "[0] * TMath::Power(x, [1])", 1, 5e3); fitspettro->SetParameters(1.0e6, -2.0); or: TF1 *fitspettro = new TF1("fitspettro", "TMath::Power([0] * x, [1])", 1, 5e3); fitspettro->SetParameters(1.0e-3, -2.0);

3

Hi pepe thanks! I modified the macro using your code (See attachment), now how you can see the red line of the fit is perfect,

but:

  1. Why did you set 2 parameters SetParameters(1.0e-3, -2.0); and not just -2 if it is a power lin dN/dE=E^-2? I think the 2 parameters are the reason because I have 2 parameters on the result of the fit too
  2. I still can’t have the chi square values
  3. I still can’t have the fit and chi square values in the issue box of the plot

Thanks for helping me
fitspettro.c (1.53 KB)

4

Right in the beginning of your macro, add: gStyle->SetOptFit();
You can also try (note below “SetParameter”, not “SetParameters”): TF1 *fitspettro = new TF1("fitspettro", "[0] / x / x", 1, 5e3); fitspettro->SetParameter(0, 1.0e6);

5

Hi pepe and thanks, I updated the code as you wrote, now I have the values in the box but they are so big, because the p0 value is 7,47E5 (and it must be 2) and the chi square/ndf is 2595/853…

6

Someone to fix the problem please? In the attachment the actual macro
fitspettro.c (1.64 KB)

7

Hi,

I guess ROOT in this situation is just a tool, which is working quite well. You now have a way to carry out your fit, right (the minimisation converged)? The value of the reduced chi2 is a consequence of the error on the data and your fit model, ultimately on the physics problem you are trying to solve.

8

Looking at the graphical results of your fits, I would try two fit options (for both types of your “fitspettro” functions):
hene->Fit(“fitspettro”, “L”); // Use Loglikelihood method (default is chisquare method)
and:
hene->Fit(“fitspettro”, “W”); // Set all weights to 1 for non empty bins; ignore error bars
and then choose the one that goes better through the data points.

9

[quote=“dpiparo”]Hi,

I guess ROOT in this situation is just a tool, which is working quite well. You now have a way to carry out your fit, right (the minimisation converged)? The value of the reduced chi2 is a consequence of the error on the data and your fit model, ultimately on the physics problem you are trying to solve.[/quote]

Hi dpiparo, the set of data is simulated by CORSIKA. As sure you know, cosmic ray spectrum is a power law

dN/dE (proportional to) E^-gamma
where gamma depends by the range energy.
The author of simulations told me that simulations uses a power low E^-2, this is the reason because I’m fitting using a power low E^-2

[quote=“Pepe Le Pew”]Looking at the graphical results of your fits, I would try two fit options (for both types of your “fitspettro” functions):
hene->Fit(“fitspettro”, “L”); // Use Loglikelihood method (default is chisquare method)
and:
hene->Fit(“fitspettro”, “W”); // Set all weights to 1 for non empty bins; ignore error bars
and then choose the one that goes better through the data points.[/quote]

Hi pepe, for the case

TF1 *fitspettro = new TF1("fitspettro", "[0] / x / x", 1, 5e3); fitspettro->SetParameter(0, 1.0e6);

this is

hene->Fit("fitspettro", "L");


and this is

hene->Fit("fitspettro", "W");

for the case

TF1 *fitspettro = new TF1("fitspettro", "TMath::Power([0] * x, [1])", 1, 5e3); fitspettro->SetParameters(1.0e-3, -2.0);
this is

hene->Fit("fitspettro", "L");

and this is

hene->Fit("fitspettro", "W");

as you can see the hene->Fit("fitspettro", "L"); doesn’t work fine. The case hene->Fit("fitspettro", "W"); draws good lines, but there is still the problem of the parameter value…

10

Try to limit the fit range (e.g. from 1 to 300 GeV, so that the the average bin contents > 10):
hene->Fit(“fitspettro”, “”, “”, 1, 300);
hene->Fit(“fitspettro”, “W”, “”, 1, 300);
hene->Fit(“fitspettro”, “L”, “”, 1, 300);

11

Hi pepe, I tried to fit limiting the energy range, but unfortunatly I’ve the same problem…

12

I was thinking that using the codes

[quote=“Pepe Le Pew”]Try: TF1 *fitspettro = new TF1("fitspettro", "[0] * TMath::Power(x, [1])", 1, 5e3); fitspettro->SetParameters(1.0e6, -2.0); or: TF1 *fitspettro = new TF1("fitspettro", "TMath::Power([0] * x, [1])", 1, 5e3); fitspettro->SetParameters(1.0e-3, -2.0);[/quote]
the values of the fit are good, because the power law of comsic rays is

dN/dE= c E^-gamma

maybe the author of the simulations told me that the simulations are a power law E^-2 just meaning
dN/dE proportional to E^-2 so I understood dN/dE=E^-2…but looking the plot it must be c=1e6 (As correctly pepe wrote to set parameters)…I’m sorry!