2-D Gaussian fit to a data file

ROOT
1

Hi,

I plot a data file using TGraph2D(), the file has three columns and I can plot this file without problem.

TGraph2D *g = new TGraph2D("datafile.dat");
 g->Draw("surf1z");

I want to do a 2-D gaussian fit to this graph to obtain constant, mean and sigma values.

Is there any way to make it??

Thanks,

Manuel

2

Create a double gaussian like

TF2 *f2 = new TF2("f2","[0]*TMath::Gaus(x,[1],[2])*TMath::Gaus(y,[3],[4])",0,10,0,10); f2-<SetParameters(1,3,1,4,1); g->Fit(f2);
Rene

3

Hi Rene,

I did as you said, with the following results:

EXT PARAMETER APPROXIMATE STEP FIRST NO. NAME VALUE ERROR SIZE DERIVATIVE 1 p0 7.82702e+01 4.65184e+05 3.24934e+00 1.24906e-10 2 p1 -4.24749e+00 1.08480e+03 2.28853e-01 5.14651e-08 3 p2 9.99192e-01 1.08488e+02 1.47939e-01 2.86447e-07 4 p3 7.23658e+00 3.88612e+03 2.87589e-01 -4.53026e-09 5 p4 2.41924e+00 6.49284e+04 4.96411e-01 -3.64138e-09

So, in this table, p0 corresponds to the amplitude (in this case, 2-D, it is the peak height) of approximately 78 and p1 is the mean value in the x-axis, it is a negative value. I do not understood what happens, because the mean value in the x-axis is approximately 17 and why the amplitude value is 78 if the peak height in the graph is about 58??

I attach the graph.

Thanks for your attention

Manuel

Graph_iter.png696×474 9.93 KB

4

Hi,

to have a correct fit you need to provide some errors for your data points, for example 0 for x and y and sqrt(z) for z if z represents counts). Then you can fill and then fit a TGraph2DErrors. Without errors all the points have the same weights and the function will be pushed towards the tails

Best Regards

Lorenzo

5

Hi,
I’m also having problems with drawing 2D gaussians (nothing to do with fitting). The amplitude drawn is most of the time completely different from the one set.
For example
TF2 *f2t = new TF2(“f2t”,"[0]*TMath::Gaus(x,[1],[2])*TMath::Gaus(y,[3],[4])",-10,10,-10,10);
f2t->SetParameters(0.9,3,.22,4,.03);

Gives me the attached figure (amplitude of .1 e-3). if I do cout<<" par0:"<< f2t->GetParameter(0);
this returns me the correct set amplitude ie: 0.9. Does anybody know what I might be doing wrong?
Many Thanks,
Sebastien

6

Hi,

your gaussian has a very small width in the y direction. You should decrease the range in x and y. For example plot between -10 and + 10 sigma in x and y and you will get a right plot. Or you can increase the precision of th e plot by using more points. For example do
f2t->SetNpy(1000)

but it will take a longer time to draw

Best Regards

Lorenzo

7

Hi,
Yes thanks it works fine with SetNpy()…
Best,
Sebastien

8

Hello I trying to create a 2D Gaussian where x and y are dependent on each other with the variable “rho”.

TwoDGaus=ROOT.TF2("TwoDGaus","[0]*TMath::Exp(-1/(2(1-TMath::Power([1],2))*(TMath::Power((x-[2]),2))/TMath::Power([3],2)+TMath::Power((y-[4]),2))/TMath::Power([5],2)-(2*[1]*(x-[2])*(y-[4]))/([3]*[5]))",0,10,0,10)

But I am getting a error messages: has 0 parameters instead of 2.
I was hoping to save my self some headache, So I was wondering if there is a pre-difinde function for this in ROOT? I am unable to find one in the TMath library. Dose any one have knowledge of this. I am using Pyroot if that changes anything.

Thanks,
-F

9
10

Note that there is a ROOT function called

which is the bivariate normal distribution that you want to use.

(Since March 16):
root.cern.ch/gitweb/?p=root.git … ad7c858db9
root.cern.ch/gitweb/?p=root.git … 41364ea63c

11

Note that there is a ROOT function called

bigaus

which is the bivariate normal distribution that you want to use.

(Dont know why the previous post is corrupted)