# Landau-gaussian convoluted fit looks weird

Dear experts,
When I try to make a langau fit to my histograms, for most of them, the fit works fine but for some of them, I get some strange behaviour which I cannot understand. Please can someone help?

Please have a look at the attached pdf which contains the figures and the fit output.

Kind regards,
Ruina

rootforumlangaufit.pdf (398.8 KB)

It looks like the â€śGSigmaâ€ť can get too big. Try to define this variable as â€ślimitedâ€ť and set the upper limit to something like 10.

Thanks, Iâ€™ll work on that.

Meanwhile, do you know if there is an efficient way to give reasonable parameter values when one is fitting many distributions together? Meaning, say, I need to make these langau fits for energy distributions of ~1k electronic chips, for all of which the distribution is expected to be similar, meaning they peak at the nearly the same position etc. I guess it is possible that the start params for one distribution may not be ideal for the nextâ€¦? If so, how can I solve this?

You could â€śestimateâ€ť them (well, maybe not all of them) from:
TH1::GetMean
TH1::GetMaximumBin + TH1::GetBinCenter
TH1::GetStdDev
TH1::Integral
TH1::GetMaximum
TH1::GetMaximumBin + TH1::GetBinContent

Yes, I had tried them in the past but somehow, they werenâ€™t giving good results in some cases. Iâ€™ll try them againâ€¦
Thank you very much for your prompt help.

Update:
I set it to 10 and it works fine for 97 out of a 100 cases.
For 3, it is at the upper limit.

So, try to play with these 3 and e.g. set the limit to 15 (or maybe you need to set limits to another variables, too).

Yes, I already tried 15 and it works! Thanks!

However, for the the other histogram (the blue one), I had 6 of them with this warning
`Warning in <Fit>: Abnormal termination of minimization.`
and `STATUS:FAILED`
So I tried changing the start value of MPV to 35. instead of 40. and that resulted in all 6 of them getting fixed but a new one (which was fine before) getting bad . Changing start value to 34. or 36. does not help (infact they make worse the fits for others).

Failed fit examplesâ€¦ How do I solve this?
Example 1

``````1141 hist1 std dev 26.762
1142 hist1 mean 47.0601
1143 hist1 integral 794
1144  FCN=243.635 FROM MIGRAD    STATUS=FAILED        566 CALLS         567 TOTAL
1145                      EDM=0.523113    STRATEGY= 1  ERROR MATRIX UNCERTAINTY  10.5 per cent
1146   EXT PARAMETER                APPROXIMATE        STEP         FIRST
1147   NO.   NAME      VALUE            ERROR          SIZE      DERIVATIVE
1148    1  Width        1.27874e-02   2.39655e-04   0.00000e+00  -1.51826e+02
1149    2  MP           1.30452e+01   6.92787e-04   0.00000e+00  -1.08202e+06
1150    3  Area         1.33779e+04   2.59424e+03   0.00000e+00   2.15953e+00
1151    4  GSigma       9.99316e+00   3.68144e-04  -0.00000e+00   1.71252e+02
``````

Example 2

``````25603 hist1 std dev 24.9358
25604 hist1 mean 51.7143
25605 hist1 integral 14599
25606  FCN=293.288 FROM MIGRAD    STATUS=FAILED        246 CALLS         247 TOTAL
25607                      EDM=10738.8    STRATEGY= 1  ERROR MATRIX UNCERTAINTY 100.0 per cent
25608   EXT PARAMETER                APPROXIMATE        STEP         FIRST
25609   NO.   NAME      VALUE            ERROR          SIZE      DERIVATIVE
25610    1  Width        3.16788e+00   6.53764e+00   0.00000e+00  -2.99869e+02
25611    2  MP           3.87556e+01   6.08231e+04  -0.00000e+00   2.87698e+04
25612    3  Area         1.48061e+04   5.61181e+05  -0.00000e+00   1.87737e+02
25613    4  GSigma       9.40323e+00   7.00875e+00  -0.00000e+00   1.30095e+02
``````

The errors are quite large and I donâ€™t know why the step size is zero for all the parameters!

Hi,

Finding good initial values is the most tricky part in minimization / fitting: it can determine success or failure and can change the result. I donâ€™t think we have any generic recommendations here; this depends too much on the actual problem / distribution. So indeed your best option is likely to take these distributions and use the fit panel (right click the histogram in a pad) to come up with good initial parameter values for a landau distribution.

Axel.

1 Like

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.