Function convolution with parameters

Svyatoslav · April 20, 2021, 1:01pm

Hello!

I have a problem with function convolution. I used Breit-Wigner function for fitting TH1:

TF1 *core3 = new TF1("core3","[0]*TMath::BreitWigner(x,[1],[2])", inf, sup);

But now i need describe TH1 with convolution by BW and Gausn. I tried:

TF1Convolution *core3_conv = new TF1Convolution("[0]*TMath::BreitWigner(x,[1],[2])", "gausn(3)", inf, sup);
TF1 *core3 = new TF1("core3", *core3_conv, inf, sup, core3_conv->GetNpar());

and

TF1 *core3 = new TF1("core3", "NSUM(*[0]*TMath::BreitWigner(x,[1],[2]), gausn(3))", inf, sup);

but thay don’t work.

What is the correct way to convolve functions with parameters for fitting?

Please read tips for efficient and successful posting and posting code

ROOT Version: 6.22/02
Platform: Debian GNU/Linux
Compiler: Not Provided

moneta · April 21, 2021, 8:35pm

Hi,
I see you create 2 TF1 objects called `“core3”. The first one results from the convolution, while the second is instead a simple addition.
If you are interested in the convolution, you should use the first one .It looks fine to me at first sight, if it does not work, please post your code reproducing the problem, in order to be able to investigate it

Best

Lorenzo

Svyatoslav · April 22, 2021, 1:33pm

Hi,
no, my first TF1 object has name core3_conv. When I run my program I give many many mistakes Error in <TFormula::Eval>: Formula is invalid and not ready to execute and then:

moneta · April 22, 2021, 1:45pm

Hi,
can you please post your complete code you are using ?

Svyatoslav · April 22, 2021, 2:57pm

Yes, this is my code:
script.C (8.5 KB)

moneta · April 22, 2021, 3:35pm

Hi,
I would need your input file to run.
I see however an error in using the NSUM operator. If you want to add a BreitWigner with a normalized gaussian, you should do as following:

TF1 *core3 = new TF1("core3", "NSUM(TMath::BreitWigner(x,[0],[1]), gausn)", inf, sup);

Note you do not have to use gausn(3) and you need to drop the coefficient parameter that will be added automatically by NSUM.

If instead you want to do a convolution, using the same functions, you do:

TF1 *core3 = new TF1("core3", "CONV(TMath::BreitWigner(x,[0],[1]), gausn)", inf, sup);

Lorenzo

Dharshan · April 23, 2021, 8:29am

Hii…I having some difficulties in converting csv to root format.can anyone help me in this issue.

Svyatoslav · April 23, 2021, 8:44am

I changed NSUM on CONV, but it didn’t solve a problem. Input file is too big, and forum doesn’t allow to download it . File with code, whitch i run, above.

moneta · April 23, 2021, 12:31pm

HI,
Did you change in your code above NSUM or CONV as I have suggested you in my previous post ?
If yes and you have an error, please post at least the error and the code creating the TF1 object

Lorenzo

Svyatoslav · April 23, 2021, 1:54pm

Hi,
list of run:
out.txt (264.2 KB)

In last version:

TF1 *core3 = new TF1("core3", /*core3_conv*/"CONV([0]*TMath::BreitWigner(x,[1],[2]), gausn)", inf, sup);

moneta · April 23, 2021, 2:38pm

Hi,
The problem you cannot use core3 as a formula function and use its name as input in a subsequent formula. So if you are doing this:

TF1 *total3 = new TF1("tot3","core3*0.001 + bg3", inf, sup);

and core3 is defined as above, it is not allowed !
What you can do is using a lambda:

TF1 *core3 = new TF1("core3", /*core3_conv*/"CONV([0]*TMath::BreitWigner(x,[1],[2]), gausn)");
TF1 *bg3 = new TF1("bg3","pol1", inf, sup);
int npar3 = core3->GetNpar();
// note when defining a lambda for a TF1 , one needs to use [=] , i.e. capture by value
TF1 *total3 = new TF1("tot3",[=](double*x, double *p){ return core3->EvalPar(x,p)*0.001+bg3->EvalPar(x,p+npar3);}, inf, sup, npar3+bg3->GetNpar());

Lorenzo

This should work

Svyatoslav · April 26, 2021, 12:09pm

There are fewer errors, but there is no construction:
out.txt (2.7 KB)

moneta · April 28, 2021, 10:33am

Hi,
It looks like you get some Nan values in fitting. Either the function is ill defined or the parameters get some un-physical values, where the function return NaN.
Can you please share your latest script and the input files needed to run, so I can understand the problem ?
Cheers

Lorenzo

Svyatoslav · April 28, 2021, 1:25pm

Yes, of course. I only entered your code.

script.C (7.3 KB)

couet · April 28, 2021, 2:08pm

root [0] 
Processing script.C...
In file included from input_line_9:1:
/Users/couet/Downloads/script.C:1:10: fatal error: 'fun.h' file not found
#include "fun.h"
         ^~~~~~~

Svyatoslav · April 28, 2021, 2:49pm

Yes, sorry, I forgot that I recently transferred some functions and variables to the header file fun.h
fun.h (2.1 KB)

Svyatoslav · April 28, 2021, 2:50pm

I also put some of the functions and variables into a separate file. See above.

couet · April 28, 2021, 3:10pm

root [0] 
Processing script.C...
Error in <TFile::TFile>: file /home/pogodin/Data_Files/Bes/umcinc.root does not exist

Svyatoslav · April 30, 2021, 5:42pm

File with data:
https://drive.google.com/drive/folders/18KxKRD3UoXZ7aoZxJXb-k4yIovfi-YY1?usp=sharing

In file fun.h str1 is massiv of ways to file. Chenge him or str in script.C.

couet · May 3, 2021, 9:13am

  Mphi->Fit("gaus", "", "", 0.97, 1.07);//1.020 - 0.02, 1.020 + 0.02);
  Mphi->Draw();

Works…
But

  Mphi->Fit("tot3", "", "", 0.97, 1.07);//1.020 - 0.02, 1.020 + 0.02);
  Mphi->Draw();

Crashes… tot3 seems “wrong” somehow.