Dear experts,
I have a simple code that fits two mass peaks simultaneously. The setup is pretty simple, RooGaussian + RooChebychev, where the mean and sigma of the gaussian are shared between the two mass regions (RooCategory). I have experienced issues in reproducing the fit results with different ROOT versions.
- The fit result looks good in ROOT 6.30/02 (I also tested it in 6.26/11)
- The fit result is totally off in ROOT 6.34/02 (I also tested it in 6.32/10 with the same outcome)
Since ROOT 6.30/02 and 6.34/02 are the versions provided in the LCG environments 105 and 107 respectively, I believe this issue is related to the one recently reported by @nicolo.trevisani.
These plots show the result in ROOT 6.30/02 and older versions. The two plots correspond to the two mass regions in the simultaneous fit:
While here is what I get with ROOT 6.34/02:
RooFit log in ROOT 6.30/02:
[#1] INFO:Eval -- RooRealVar::setRange(dimu_OS1) new range named 'fit_nll_signal_mumu_data' created with bounds [0.98,1.08]
[#1] INFO:Fitting -- RooAbsOptTestStatistic::ctor(nll_signal_mumu_data) constructing test statistic for sub-range named fit_range_signal
[#1] INFO:Minimization -- RooAbsMinimizerFcn::setOptimizeConst: activating const optimization
Minuit2Minimizer: Minimize with max-calls 1000 convergence for edm < 1 strategy 1
Info in <Minuit2>: MnSeedGenerator Computing seed using NumericalGradient calculator
Info in <Minuit2>: MnSeedGenerator Initial state: FCN = -6540.139974 Edm = -336.8412911 NCalls = 9
Info in <Minuit2>: NegativeG2LineSearch Doing a NegativeG2LineSearch since one of the G2 component is negative
Info in <Minuit2>: MnSeedGenerator Negative G2 found - new state:
Minimum value : -6759.661969
Edm : 1.708732621
Internal parameters: [ 0 -0.3905360998]
Internal gradient : [ -9.246451549 -18.78991127]
Internal covariance matrix:
[[ 0.078044331 0]
[ 0 0.00045993433]]]
Info in <Minuit2>: MnSeedGenerator Initial state
Minimum value : -6759.661969
Edm : 1.708732621
Internal parameters: [ 0 -0.3905360998]
Internal gradient : [ -9.246451549 -18.78991127]
Internal covariance matrix:
[[ 0.078044331 0]
[ 0 0.00045993433]]]
Info in <Minuit2>: VariableMetricBuilder Start iterating until Edm is < 0.001 with call limit = 1000
Info in <Minuit2>: VariableMetricBuilder 0 - FCN = -6759.661969 Edm = 1.708732621 NCalls = 26
Info in <Minuit2>: VariableMetricBuilder 1 - FCN = -6761.250973 Edm = 0.07328111069 NCalls = 32
Info in <Minuit2>: VariableMetricBuilder 2 - FCN = -6761.337042 Edm = 1.380802967e-05 NCalls = 38
Info in <Minuit2>: VariableMetricBuilder After Hessian
Info in <Minuit2>: VariableMetricBuilder 3 - FCN = -6761.337042 Edm = 1.619328576e-05 NCalls = 48
Minuit2Minimizer : Valid minimum - status = 0
FVAL = -6761.33704188590127
Edm = 1.61932857624699604e-05
Nfcn = 48
M_mumu = 1.01973 +/- 0.000397116 (limited)
width = 0.0188872 +/- 0.00032099 (limited)
Info in <Minuit2>: Minuit2Minimizer::Hesse Using max-calls 1000
Info in <Minuit2>: Minuit2Minimizer::Hesse Hesse is valid - matrix is accurate
[#1] INFO:Minimization -- RooAbsMinimizerFcn::setOptimizeConst: deactivating const optimization
[#1] INFO:Eval -- RooRealVar::setRange(dimu_OS1) new range named 'fit_nll_simul_model_datatofit' created with bounds [0.9,1.16]
RooAbsTestStatistic::initSimMode: creating slave calculator #0 for state looseBDT (7828 dataset entries)
[#1] INFO:Fitting -- RooAbsOptTestStatistic::ctor(looseBDT) constructing test statistic for sub-range named fit_range
RooAbsTestStatistic::initSimMode: creating slave calculator #1 for state tightBDT (3530 dataset entries)
[#1] INFO:Fitting -- RooAbsOptTestStatistic::ctor(tightBDT) constructing test statistic for sub-range named fit_range
[#1] INFO:Fitting -- RooAbsTestStatistic::initSimMode: created 2 slave calculators.
[#1] INFO:Minimization -- RooAbsMinimizerFcn::setOptimizeConst: activating const optimization
[#1] INFO:Minimization -- The following expressions will be evaluated in cache-and-track mode: (signal_mumu,bmodel_ctrl)
[#1] INFO:Minimization -- The following expressions will be evaluated in cache-and-track mode: (signal_mumu,bmodel)
Minuit2Minimizer: Minimize with max-calls 4000 convergence for edm < 1 strategy 1
Info in <Minuit2>: MnSeedGenerator Computing seed using NumericalGradient calculator
Info in <Minuit2>: MnSeedGenerator Initial state: FCN = -95360.13639 Edm = -1154.076539 NCalls = 33
Info in <Minuit2>: NegativeG2LineSearch Doing a NegativeG2LineSearch since one of the G2 component is negative
Info in <Minuit2>: MnSeedGenerator Negative G2 found - new state:
Minimum value : -97081.5748
Edm : 758.9011992
Internal parameters: [ 0.375208442 0 0.6435011088 0.6435011088 -0.6376432924 0 0 -0.3928146854]
Internal gradient : [ 25.31044218 417.4800536 9.957564737 320.6898948 -3.140052728 162.39212 588.010494 794.3649351]
Internal covariance matrix:
[[ 0.18647605 0 0 0 0 0 0 0]
[ 0 0.0019809103 0 0 0 0 0 0]
[ 0 0 0.0014842965 0 0 0 0 0]
[ 0 0 0 0.0090653991 0 0 0 0]
[ 0 0 0 0 0.00066566041 0 0 0]
[ 0 0 0 0 0 0.0041729091 0 0]
[ 0 0 0 0 0 0 0.00095230311 0]
[ 0 0 0 0 0 0 0 0.0019003106]]]
Info in <Minuit2>: MnSeedGenerator Initial state
Minimum value : -97081.5748
Edm : 758.9011992
Internal parameters: [ 0.375208442 0 0.6435011088 0.6435011088 -0.6376432924 0 0 -0.3928146854]
Internal gradient : [ 25.31044218 417.4800536 9.957564737 320.6898948 -3.140052728 162.39212 588.010494 794.3649351]
Internal covariance matrix:
[[ 0.18647605 0 0 0 0 0 0 0]
[ 0 0.0019809103 0 0 0 0 0 0]
[ 0 0 0.0014842965 0 0 0 0 0]
[ 0 0 0 0.0090653991 0 0 0 0]
[ 0 0 0 0 0.00066566041 0 0 0]
[ 0 0 0 0 0 0.0041729091 0 0]
[ 0 0 0 0 0 0 0.00095230311 0]
[ 0 0 0 0 0 0 0 0.0019003106]]]
Info in <Minuit2>: VariableMetricBuilder Start iterating until Edm is < 0.001 with call limit = 4000
Info in <Minuit2>: VariableMetricBuilder 0 - FCN = -97081.5748 Edm = 758.9011992 NCalls = 62
Info in <Minuit2>: VariableMetricBuilder 1 - FCN = -97425.06857 Edm = 141.2002291 NCalls = 86
Info in <Minuit2>: VariableMetricBuilder 2 - FCN = -97490.89187 Edm = 64.54300277 NCalls = 104
Info in <Minuit2>: VariableMetricBuilder 3 - FCN = -97518.81138 Edm = 11.94799931 NCalls = 122
Info in <Minuit2>: VariableMetricBuilder 4 - FCN = -97523.53194 Edm = 3.395830549 NCalls = 140
Info in <Minuit2>: VariableMetricBuilder 5 - FCN = -97525.0164 Edm = 0.4153871578 NCalls = 158
Info in <Minuit2>: VariableMetricBuilder 6 - FCN = -97525.4484 Edm = 0.02006770115 NCalls = 176
Info in <Minuit2>: VariableMetricBuilder 7 - FCN = -97525.4665 Edm = 0.0002343436795 NCalls = 194
Info in <Minuit2>: VariableMetricBuilder After Hessian
Info in <Minuit2>: VariableMetricBuilder 8 - FCN = -97525.4665 Edm = 0.0001879779896 NCalls = 261
Minuit2Minimizer : Valid minimum - status = 0
FVAL = -97525.4665035848157
Edm = 0.000187977989554814859
Nfcn = 261
M_mumu = 1.01878 +/- 0.000282703 (limited)
Nb = 1570.53 +/- 46.7777 (limited)
Nb_ctrl = 6548.31 +/- 90.2702 (limited)
Ns = 1958.77 +/- 50.7567 (limited)
Ns_ctrl = 1278.53 +/- 53.8043 (limited)
a0 = -0.362474 +/- 0.0423579 (limited)
b0 = -0.286239 +/- 0.0210441 (limited)
width = 0.0117274 +/- 0.000270008 (limited)
Info in <Minuit2>: Minuit2Minimizer::Hesse Using max-calls 4000
Info in <Minuit2>: Minuit2Minimizer::Hesse Hesse is valid - matrix is accurate
[#1] INFO:Fitting -- RooAbsPdf::fitTo(simul_model) Calculating sum-of-weights-squared correction matrix for covariance matrix
Info in <Minuit2>: Minuit2Minimizer::Hesse Using max-calls 4000
Info in <Minuit2>: Minuit2Minimizer::Hesse Hesse is valid - matrix is accurate
[#1] INFO:Minimization -- RooAbsMinimizerFcn::setOptimizeConst: deactivating const optimization
RooFitResult: minimized FCN value: -97525.5, estimated distance to minimum: 0.000187445
covariance matrix quality: Full, accurate covariance matrix
Status : MINIMIZE=0 HESSE=0 HESSE=0
Floating Parameter InitialValue FinalValue +/- Error GblCorr.
-------------------- ------------ -------------------------- --------
M_mumu 1.0197e+00 1.0188e+00 +/- 2.84e-04 <none>
Nb 1.7650e+03 1.5705e+03 +/- 4.68e+01 <none>
Nb_ctrl 6.2624e+03 6.5483e+03 +/- 9.04e+01 <none>
Ns 2.8240e+03 1.9588e+03 +/- 5.08e+01 <none>
Ns_ctrl 6.2624e+03 1.2785e+03 +/- 5.38e+01 <none>
a0 0.0000e+00 -3.6247e-01 +/- 4.24e-02 <none>
b0 0.0000e+00 -2.8624e-01 +/- 2.10e-02 <none>
width 1.8887e-02 1.1727e-02 +/- 2.70e-04 <none>
RooFit log in ROOT 6.34/02:
[#1] INFO:Eval -- RooRealVar::setRange(dimu_OS1) new range named 'fit_nll_signal_mumu_data' created with bounds [0.98,1.08]
[#1] INFO:Fitting -- RooAbsPdf::fitTo(signal_mumu_over_signal_mumu_Int[dimu_OS1|fit_range_signal]) fixing normalization set for coefficient determination to observables in data
[#1] INFO:Fitting -- using CPU computation library compiled with -mavx2
[#1] INFO:Fitting -- RooAddition::defaultErrorLevel(nll_signal_mumu_over_signal_mumu_Int[dimu_OS1|fit_range_signal]_data) Summation contains a RooNLLVar, using its error level
[#1] INFO:Minimization -- RooAbsMinimizerFcn::setOptimizeConst: activating const optimization
Minuit2Minimizer: Minimize with max-calls 1000 convergence for edm < 1 strategy 1
Info in <Minuit2>: MnSeedGenerator Computing seed using NumericalGradient calculator
Info in <Minuit2>: MnSeedGenerator Initial state: FCN = -6540.139974 Edm = 138022.4473 NCalls = 9
Info in <Minuit2>: NegativeG2LineSearch Doing a NegativeG2LineSearch since one of the G2 component is negative
Info in <Minuit2>: MnSeedGenerator Negative G2 found - new state:
Minimum value : -6759.661969
Edm : 1.708732621
Internal parameters: [ 0 -0.3905360998]
Internal gradient : [ -9.246451549 -18.78991127]
Internal covariance matrix:
[[ 0.078044331 0]
[ 0 0.00045993433]]]
Info in <Minuit2>: MnSeedGenerator Initial state
Minimum value : -6759.661969
Edm : 1.708732621
Internal parameters: [ 0 -0.3905360998]
Internal gradient : [ -9.246451549 -18.78991127]
Internal covariance matrix:
[[ 0.078044331 0]
[ 0 0.00045993433]]]
Info in <Minuit2>: VariableMetricBuilder Start iterating until Edm is < 0.001 with call limit = 1000
Info in <Minuit2>: VariableMetricBuilder 0 - FCN = -6759.661969 Edm = 1.708732621 NCalls = 26
Info in <Minuit2>: VariableMetricBuilder 1 - FCN = -6761.250973 Edm = 0.07328111069 NCalls = 32
Info in <Minuit2>: VariableMetricBuilder 2 - FCN = -6761.337042 Edm = 1.380802967e-05 NCalls = 38
Info in <Minuit2>: VariableMetricBuilder After Hessian
Info in <Minuit2>: VariableMetricBuilder 3 - FCN = -6761.337042 Edm = 1.619328585e-05 NCalls = 48
Minuit2Minimizer : Valid minimum - status = 0
FVAL = -6761.33704188590127
Edm = 1.61932858511379351e-05
Nfcn = 48
M_mumu = 1.01973 +/- 0.000397116 (limited)
width = 0.0188872 +/- 0.00032099 (limited)
Info in <Minuit2>: Minuit2Minimizer::Hesse Using max-calls 1000
Info in <Minuit2>: Minuit2Minimizer::Hesse Hesse is valid - matrix is accurate
[#1] INFO:Minimization -- RooAbsMinimizerFcn::setOptimizeConst: deactivating const optimization
[#1] INFO:Eval -- RooRealVar::setRange(dimu_OS1) new range named 'fit_nll_simul_model_datatofit' created with bounds [0.9,1.16]
[#1] INFO:Fitting -- RooAbsPdf::fitTo(simul_model) fixing normalization set for coefficient determination to observables in data
[#1] INFO:Fitting -- RooAddition::defaultErrorLevel(nll_simul_model_datatofit) Summation contains a RooNLLVar, using its error level
[#1] INFO:Minimization -- RooAbsMinimizerFcn::setOptimizeConst: activating const optimization
Minuit2Minimizer: Minimize with max-calls 4000 convergence for edm < 1 strategy 1
Info in <Minuit2>: MnSeedGenerator Computing seed using NumericalGradient calculator
[#0] ERROR:InputArguments -- RooArgSet::checkForDup: ERROR argument with name _tightBDT_signal_mumu_over_signal_mumu_Int[dimu_OS1]_Int[] is already in this set
[#0] ERROR:InputArguments -- RooArgSet::checkForDup: ERROR argument with name _tightBDT_bmodel_over_bmodel_Int[dimu_OS1]_Int[] is already in this set
[#0] ERROR:InputArguments -- RooArgSet::checkForDup: ERROR argument with name _looseBDT_signal_mumu_over_signal_mumu_Int[dimu_OS1]_Int[] is already in this set
[#0] ERROR:InputArguments -- RooArgSet::checkForDup: ERROR argument with name _looseBDT_bmodel_ctrl_over_bmodel_ctrl_Int[dimu_OS1]_Int[] is already in this set
Info in <Minuit2>: MnSeedGenerator Initial state: FCN = -66667.33607 Edm = 4947424.615 NCalls = 33
Info in <Minuit2>: NegativeG2LineSearch Doing a NegativeG2LineSearch since one of the G2 component is negative
Info in <Minuit2>: MnSeedGenerator Negative G2 found - new state:
Minimum value : -70794.20641
Edm : 2382.596038
Internal parameters: [ -1.553566819 0 0.6435011088 -4.679872678 -1.504705178 0 0 1.559260277]
Internal gradient : [ 0.3836497869 -1697.017873 -3099.523339 15.23508383 9.686819045 135.1214134 726.4935634 -10.20517518]
Internal covariance matrix:
[[ 0.089816107 0 0 0 0 0 0 0]
[ 0 0.0015378866 0 0 0 0 0 0]
[ 0 0 0.00046715083 0 0 0 0 0]
[ 0 0 0 0.0042688608 0 0 0 0]
[ 0 0 0 0 0.014028471 0 0 0]
[ 0 0 0 0 0 0.0068249823 0 0]
[ 0 0 0 0 0 0 0.00092151164 0]
[ 0 0 0 0 0 0 0 0.002260788]]]
Info in <Minuit2>: MnSeedGenerator Initial state
Minimum value : -70794.20641
Edm : 2382.596038
Internal parameters: [ -1.553566819 0 0.6435011088 -4.679872678 -1.504705178 0 0 1.559260277]
Internal gradient : [ 0.3836497869 -1697.017873 -3099.523339 15.23508383 9.686819045 135.1214134 726.4935634 -10.20517518]
Internal covariance matrix:
[[ 0.089816107 0 0 0 0 0 0 0]
[ 0 0.0015378866 0 0 0 0 0 0]
[ 0 0 0.00046715083 0 0 0 0 0]
[ 0 0 0 0.0042688608 0 0 0 0]
[ 0 0 0 0 0.014028471 0 0 0]
[ 0 0 0 0 0 0.0068249823 0 0]
[ 0 0 0 0 0 0 0.00092151164 0]
[ 0 0 0 0 0 0 0 0.002260788]]]
Info in <Minuit2>: VariableMetricBuilder Start iterating until Edm is < 0.001 with call limit = 4000
Info in <Minuit2>: VariableMetricBuilder 0 - FCN = -70794.20641 Edm = 2382.596038 NCalls = 179
Info in <Minuit2>: VariableMetricBuilder 1 - FCN = -73527.13564 Edm = 8.125858153 NCalls = 202
Info in <Minuit2>: VariableMetricBuilder 2 - FCN = -73536.24614 Edm = 0.3369509244 NCalls = 220
Info in <Minuit2>: VariableMetricBuilder 3 - FCN = -73536.50858 Edm = 0.04668680801 NCalls = 238
Info in <Minuit2>: VariableMetricBuilder 4 - FCN = -73536.54625 Edm = 0.004214518765 NCalls = 256
Info in <Minuit2>: VariableMetricBuilder 5 - FCN = -73536.54991 Edm = 0.0001099576611 NCalls = 273
Info in <Minuit2>: VariableMetricBuilder After Hessian
Info in <Minuit2>: VariableMetricBuilder 6 - FCN = -73536.54991 Edm = 0.0001633640581 NCalls = 334
Minuit2Minimizer : Valid minimum - status = 0
FVAL = -73536.549905802487
Edm = 0.00016336405813592651
Nfcn = 334
M_mumu = 1.017 +/- 7.87593e-05 (limited)
Nb = 3530 +/- 0.976784 (limited)
Nb_ctrl = 7828 +/- 0.675656 (limited)
Ns = 3530 +/- 4.7914 (limited)
Ns_ctrl = 2.36629e-05 +/- 6.92393 (limited)
a0 = -0.485823 +/- 0.0439686 (limited)
b0 = -0.330843 +/- 0.0204784 (limited)
width = 0.05 +/- 2.22373e-05 (limited)
Info in <Minuit2>: Minuit2Minimizer::Hesse Using max-calls 4000
Info in <Minuit2>: Minuit2Minimizer::Hesse Hesse is valid - matrix is accurate
[#1] INFO:Fitting -- RooAbsPdf::fitTo(simul_model) Calculating sum-of-weights-squared correction matrix for covariance matrix
Info in <Minuit2>: Minuit2Minimizer::Hesse Using max-calls 4000
Info in <Minuit2>: Minuit2Minimizer::Hesse Hesse is valid - matrix is accurate
[#1] INFO:Minimization -- RooAbsMinimizerFcn::setOptimizeConst: deactivating const optimization
RooFitResult: minimized FCN value: -73536.5, estimated distance to minimum: 0.00016331
covariance matrix quality: Full, accurate covariance matrix
Status : MINIMIZE=0 HESSE=0 HESSE=0
Floating Parameter InitialValue FinalValue +/- Error GblCorr.
-------------------- ------------ -------------------------- --------
M_mumu 1.0197e+00 1.0170e+00 +/- 6.04e-08 <none>
Nb 1.7650e+03 3.5300e+03 +/- 1.13e-02 <none>
Nb_ctrl 6.2624e+03 7.8280e+03 +/- 8.30e-03 <none>
Ns 2.8240e+03 3.5300e+03 +/- 1.26e-01 <none>
Ns_ctrl 6.2624e+03 2.3663e-05 +/- 2.56e-02 <none>
a0 0.0000e+00 -4.8582e-01 +/- 4.40e-02 <none>
b0 0.0000e+00 -3.3084e-01 +/- 2.05e-02 <none>
width 1.8887e-02 5.0000e-02 +/- 6.45e-08 <none>
Can you please let me know if any change in the minimizer was done in recent ROOT/RooFit versions? I would be happy to provide a reproducer if needed.
Thank you and best,
Federica