Study of the \(K^{+}K^{-}\) S-wave amplitude near threshold in \(D^{+}\rightarrow K^{-}K^{+}K^{+}\)

Sebastian Ordoñez*, Alberto Reis, Diego Milanés

jsordonezs@unal.edu.co

Universidad Nacional de Colombia

Charm WG meeting

February 22 , 2023

Outline

  • Introduction
  • Analysis strategy
  • Data samples
    • Data
    • Monte Carlo
  • Selection
    • Pre-selection
    • MC reweighting
    • MVA
    • Figures of merit
  • \(K^{-}K^{+}K^{+}\) Invariant mass fit
    • Dalitz plot final sample
  • Outlook 

Introduction

S. Ordoñez (UNAL-COL)

Study of the \(K^{-}K^{+}\)  S-wave amplitude near treshhold in \(D^{+}\rightarrow K^{-}K^{+}K^{+}\)

February 22, 2023

Analysis strategy

Study of the \(K^{-}K^{+}\)  S-wave amplitude near treshhold in \(D^{+}\rightarrow K^{-}K^{+}K^{+}\)

S. Ordoñez (UNAL-COL)

February 22, 2023

Perform a study of the \(K^{-}K^{+} S\)-wave amplitude in the \(D^{+}\rightarrow K^{-}K^{+}K^{+}\) decay channel using the data set of Run 2.

Methodology

  • Data
    • Sample used for the CPV search in  \(D^{+}\rightarrow K^{-}K^{+}K^{+}\) is not adequate for DP analysis \(\rightarrow\) New selection to be made.
    • Only a small region of the phase space will be used \(\rightarrow\) Dalitz folded
      • Given that there is no interference between the two \(K^{-}K^{+}\) channels in the region of interest, it is possible to use the Dalitz plot folded.
  • Simulations
    • Large Monte Carlo (MC) samples available
    • Need to match data and MC distributions \(\rightarrow\) GB-Reweighting

This analysis follows a similar strategy to that used in the analysis of \(D_{(s)}^{+}\rightarrow \pi^{-}\pi^{+}\pi^{+}\)

Data samples

Study of the \(K^{-}K^{+}\)  S-wave amplitude near treshhold in \(D^{+}\rightarrow K^{-}K^{+}K^{+}\)

S. Ordoñez (UNAL-COL)

  • Data collected in \(pp\) collisions at \(\sqrt{s} = 13\) TeV by LHCb in the years 2016-2018.
  • The candidates are required to pass any of the L0 hardware trigger lines.
  • The data set comes from HLT2 Turbo line: Hlt2CharmHadDpToKmKpKp
  • Statistics of the data samples after trigger level cuts:

Data sample

The total data sample consists of appx. 30M candidates in the whole phase space.

In the region \(s_{low}^{DTF} < 1.18\) GeV\(^{2}\)  the number of candidates is reduced to appx. 14M.

February 22, 2023

Study of the \(K^{-}K^{+}\)  S-wave amplitude near treshhold in \(D^{+}\rightarrow K^{-}K^{+}K^{+}\)

S. Ordoñez (UNAL-COL)

Data sample

February 22, 2023

\(s_{12}\) [GeV\(^{2}\)]

\(s_{12}\) [GeV\(^{2}\)]

\(s_{13}\) [GeV\(^{2}\)]

\(\phi(1020)\) 

Cloned tracks

Cloned tracks

  • Notation: \(s_{12} = s_{K^{-}K^{+}} = m_{K^{-}K^{+}}^{2}\) and \(s_{13} = s_{K^{-}K^{+}} = m_{K^{-}K^{+}}^{2}\)

\(\phi(1020)\) 

\(\phi(1020)\) 

Events

Study of the \(K^{-}K^{+}\)  S-wave amplitude near treshhold in \(D^{+}\rightarrow K^{-}K^{+}K^{+}\)

S. Ordoñez (UNAL-COL)

Data sample

February 22, 2023

\(m(K^{-}K^{+}K^{+})\) [MeV]

\(s_{low}^{DTF}\) [GeV\(^{2}\)]

\(s_{high}^{DTF}\) [GeV\(^{2}\)]

Cloned tracks

\(\phi(1020)\) 

Events

Data samples

Study of the \(K^{-}K^{+}\)  S-wave amplitude near treshhold in \(D^{+}\rightarrow K^{-}K^{+}K^{+}\)

S. Ordoñez (UNAL-COL)

Monte Carlo

February 22, 2023

  • Large Monte Carlo (MC) samples are available for each year and polarity (here only used a sub-sample).
  • MC data is used for the extraction of the PDF parameters employed in fits.
  • MC (reweighted) is used as signal proxy in the MVA selection and the calculation of the efficiency map.

Configurations used to simulate the data per year

Selection

Pre-selection

Study of the \(K^{-}K^{+}\)  S-wave amplitude near treshhold in \(D^{+}\rightarrow K^{-}K^{+}K^{+}\)

S. Ordoñez (UNAL-COL)

  • On top of the trigger level cuts we also apply  offline pre-selection cuts.
  • The total number of candidates in the 2016-2018 samples after the pre-selection is appx. 4.1M
  • After pre-selection 29.5% of the candidates in the region \(s_{low}^{DTF} < 1.18\) GeV\(^2\) are kept.
  • From the pre-selected sample an estimated 12.9% of the events correspond to potential signal events.

February 22, 2023

Selection

Pre-selection

Study of the \(K^{-}K^{+}\)  S-wave amplitude near treshhold in \(D^{+}\rightarrow K^{-}K^{+}K^{+}\)

February 22, 2023

S. Ordoñez (UNAL-COL)

  • Comparison between the \(KKK\) mass distribution before and after the pre-selection for the region \(s_{low}^{DTF} < 1.18\) GeV\(^2\).

\(m(K^{-}K^{+}K^{+})\) [MeV]

\(m(K^{-}K^{+}K^{+})\) [MeV]

Events

Events

Selection

Study of the \(K^{-}K^{+}\)  S-wave amplitude near treshhold in \(D^{+}\rightarrow K^{-}K^{+}K^{+}\)

S. Ordoñez (UNAL-COL)

February 22, 2023

  • \(KKK\) invariant mass distribution and Dalitz plots for one of the pre-selected subsamples

\(m(K^{-}K^{+}K^{+})\) [MeV]

\(s_{12}\) [GeV\(^{2}\)]

\(s_{13}\) [GeV\(^{2}\)]

\(s_{low}^{DTF}\) [GeV\(^{2}\)]

\(s_{high}^{DTF}\) [GeV\(^{2}\)]

2016-Down

Pre-selection

Events

Folding 

Study of the \(K^{-}K^{+}\)  S-wave amplitude near treshhold in \(D^{+}\rightarrow K^{-}K^{+}K^{+}\)

February 22, 2023

S. Ordoñez (UNAL-COL)

  • Previous to the MVA it is necessary to prepare the MC samples that will be used as signal proxy.
  • The data samples for each year and polarity are randomly divided into two reproducible sub-samples:

Monte Carlo reweighting

After the folding, it is ensured that each fold has the same number of events coming from each sub-sample. 

Fold 1

Fold 2

sWeights

Study of the \(K^{-}K^{+}\)  S-wave amplitude near treshhold in \(D^{+}\rightarrow K^{-}K^{+}K^{+}\)

S. Ordoñez (UNAL-COL)

  • To obtain signal weights via sPlot technique which are required for the reweight procedure, a fit to the invariant \(K^{-}K^{+}K^{+}\) mass is performed for each sub-sample.

February 22, 2023

Monte Carlo reweighting

\(m(K^{-}K^{+}K^{+})\) [MeV]

\(m(K^{-}K^{+}K^{+})\) [MeV]

\(m(K^{-}K^{+}K^{+})\) [MeV]

\(D^{+}\) candidate mass vs background weight

\(D^{+}\) candidate mass vs signal weight

Invariant mass \(K^{-}K^{+}K^{+}\) fit

Results for the sub-sample 2016-MagDown (Fold 1)

Dynamics correction

Study of the \(K^{-}K^{+}\)  S-wave amplitude near treshhold in \(D^{+}\rightarrow K^{-}K^{+}K^{+}\)

S. Ordoñez (UNAL-COL)

  • The MC sample is generated as phase space, therefore the dynamics is emulated by applying a resonant structure weight \(w_{data}\).
  • A correction in the number of SPDHits is applied as a weight, \(w_{nSPDHits}\).

February 22, 2023

Monte Carlo weighting

Reweight procedure

Study of the \(K^{-}K^{+}\)  S-wave amplitude near treshhold in \(D^{+}\rightarrow K^{-}K^{+}K^{+}\)

S. Ordoñez (UNAL-COL)

February 22, 2023

  • The \(w_{nSPDHits}\) and \(w_{data}\) are insufficient to match data and MC, a third weight is required.
  • Inputs for the MVA reweighting are: MC sample weighted (PID, SPDhits correction, phase space dynamics) and the data sample with background subtracted from sPlot.

Monte Carlo weighting

Monte Carlo matching

Reweight procedure

Study of the \(K^{-}K^{+}\)  S-wave amplitude near treshhold in \(D^{+}\rightarrow K^{-}K^{+}K^{+}\)

S. Ordoñez (UNAL-COL)

February 22, 2023

  • The Kolgomorov-Smirnov test is employed to estimate how different data and MC are.
  • The output of the reweight procedure is a gb_weight.

gb_weight

Results for the KS test before the reweighting

2016-Down (Fold 2)

Monte Carlo matching

Reweight procedure

Study of the \(K^{-}K^{+}\)  S-wave amplitude near treshhold in \(D^{+}\rightarrow K^{-}K^{+}K^{+}\)

S. Ordoñez (UNAL-COL)

February 22, 2023

  • The matching between the data and MC variables is said to be successful, confirmed by the KS test and ROC curve.

2016-Down (Fold 2)

Results for the KS test after the reweighting

Monte Carlo matching

Reweight procedure

Study of the \(K^{-}K^{+}\)  S-wave amplitude near treshhold in \(D^{+}\rightarrow K^{-}K^{+}K^{+}\)

S. Ordoñez (UNAL-COL)

February 22, 2023

  • Variables after applying the gb_weights

MVA selection

Study of the \(K^{-}K^{+}\)  S-wave amplitude near treshhold in \(D^{+}\rightarrow K^{-}K^{+}K^{+}\)

February 22, 2023

S. Ordoñez (UNAL-COL)

  • To further refine the candidate selection, a BDT is trained and applied using the TMVA package.
  • Fold 1 is used for training and the resulting BDT weights are applied to Fold 2, and vice versa.
  • A single BDT is trained with all of the sub-samples (year and polarities).

Selection

N. Background events N. Signal events
477845 802422

Fold 1

N. Background events N. Signal events
480365 800592

Fold 2

For each fold, 700k signal events are used for training and 100k for testing. Similarly, 370k bkg events are used for training, and 100k for testing.

  • Training phase:
    • MC samples are used as signal proxy, all weights are applied.
    • The data sidebands are used as background: \([1820,1840]\) MeV and \([1900,1920]\) MeV.

MVA selection

Study of the \(K^{-}K^{+}\)  S-wave amplitude near treshhold in \(D^{+}\rightarrow K^{-}K^{+}K^{+}\)

February 22, 2023

S. Ordoñez (UNAL-COL)

Selection

  • The set of variables is chosen based on their discriminant power and not having a significant efficiency variation across the Dalitz plot.

Fold 1

Training variables for BDT classifier
D_IP_CHI2
log(D_FD_CHI2)
DIRA
D_ENDVERTEX_CHI2
logIP
D_BPVTRPOINTING

MVA selection

Study of the \(K^{-}K^{+}\)  S-wave amplitude near treshhold in \(D^{+}\rightarrow K^{-}K^{+}K^{+}\)

February 22, 2023

S. Ordoñez (UNAL-COL)

Selection

  • Testing phase:
    • To perform the selection, two classifiers were tested: BDT and BDTG.
    • From the ROC curves a similar performance is observed, there is no strong motivation for a specific choice. The BDTG is used for the application.
Condition BDT BDTG
NTrees 700 700
MinNodeSize 2.5% 2.5%
MaxDepth 4 4
BoostType AdaBoost Grad
AdaBoostBeta 0.5 none
Shrinkage none 0.1
UseBaggedBoost true true
BaggedSampleFraction 0.5 0.5
SeparationType GinIndex none
nCuts 20 20

TMVA configurations for the training of BDT classifiers.

MVA selection

Study of the \(K^{-}K^{+}\)  S-wave amplitude near treshhold in \(D^{+}\rightarrow K^{-}K^{+}K^{+}\)

February 22, 2023

S. Ordoñez (UNAL-COL)

Selection

  • Performance results for the BDTG classifier.

Similar results are obtained for Fold 2.

Figures of merit

Study of the \(K^{-}K^{+}\)  S-wave amplitude near treshhold in \(D^{+}\rightarrow K^{-}K^{+}K^{+}\)

February 22, 2023

S. Ordoñez (UNAL-COL)

  • From a fit to the MC sample with the cut val_BDTG > 0, a \(\sigma_{eff} = 4.34\) MeV was estimated.
  • The signal window used will be \(2\sigma_{eff}\), which corresponds to: \([1861.41,1878.77]\) MeV

Selection

Fold 1

Fold 2

A sample with high purity is required for this analysis.

Figures of merit

Study of the \(K^{-}K^{+}\)  S-wave amplitude near treshhold in \(D^{+}\rightarrow K^{-}K^{+}K^{+}\)

February 22, 2023

S. Ordoñez (UNAL-COL)

  • A cut on val_BDTG > 0.65 gives a purity of 92%, with a signal efficiency of around 50%.

Selection

Fold 1

Fold 2

Study of the \(K^{-}K^{+}\)  S-wave amplitude near treshhold in \(D^{+}\rightarrow K^{-}K^{+}K^{+}\)

February 22, 2023

S. Ordoñez (UNAL-COL)

\(K^{-}K^{+}K^{+}\) Invariant mass fit

  • The final signal region corresponds to \(2\sigma_{eff}\) is: \([1860.38,1879.31]\) MeV
  • Within the signal region there are: 237940 signal events and  21708 background events. 

Study of the \(K^{-}K^{+}\)  S-wave amplitude near treshhold in \(D^{+}\rightarrow K^{-}K^{+}K^{+}\)

February 22, 2023

S. Ordoñez (UNAL-COL)

\(K^{-}K^{+}K^{+}\) Invariant mass fit

  • Considering events within the signal region the Dalitz plot below is obtained.
  • A clear interference between the \(S\)-wave and the \(\phi(1020)\) resonance is observed.

Dalitz plot of the selected data

\(\phi(1020)\) 

Events

Study of the \(K^{-}K^{+}\)  S-wave amplitude near treshhold in \(D^{+}\rightarrow K^{-}K^{+}K^{+}\)

February 22, 2023

S. Ordoñez (UNAL-COL)

\(K^{-}K^{+}K^{+}\) Invariant mass fit

  • Taking events in the regions \([1805,1850]\) MeV and \([1890,1935]\) MeV the Dalitz plot below is obtained.
  • The remaining background is composed by two components: combinatorial and a peaking background at the \(\phi(1020)\) mass.

Dalitz plot of the selected data

\(\phi(1020)\) 

Events

Study of the \(K^{-}K^{+}\)  S-wave amplitude near treshhold in \(D^{+}\rightarrow K^{-}K^{+}K^{+}\)

February 22, 2023

S. Ordoñez (UNAL-COL)

Outlook

  • Background study
    • Determine the fractions of the two components of the remaining background: combinatorial and \(\phi(1020)\).
    • Next: Produce the background model using a toys with the computed background fractions.
  • Efficiency
    • Next: Calculate the efficiency map using the Monte Carlo samples and the final selection

Study of the \(K^{-}K^{+}\)  S-wave amplitude near treshhold in \(D^{+}\rightarrow K^{-}K^{+}K^{+}\)

February 22, 2023

S. Ordoñez (UNAL-COL)

Thank you!

Study of the \(K^{-}K^{+}\)  S-wave amplitude near treshhold in \(D^{+}\rightarrow K^{-}K^{+}K^{+}\)

February 22, 2023

S. Ordoñez (UNAL-COL)

Back up

Dynamics correction cross check

Study of the \(K^{-}K^{+}\)  S-wave amplitude near treshhold in \(D^{+}\rightarrow K^{-}K^{+}K^{+}\)

S. Ordoñez (UNAL-COL)

  • To check if the weights were correctly calculated one can plot:

February 22, 2023

Monte Carlo weighting

MVA selection

Study of the \(K^{-}K^{+}\)  S-wave amplitude near treshhold in \(D^{+}\rightarrow K^{-}K^{+}K^{+}\)

February 22, 2023

S. Ordoñez (UNAL-COL)

Selection

MVA selection

Study of the \(K^{-}K^{+}\)  S-wave amplitude near treshhold in \(D^{+}\rightarrow K^{-}K^{+}K^{+}\)

February 22, 2023

S. Ordoñez (UNAL-COL)

Selection

  • Application phase:
    • Application of the weights obtained from Fold 1 in Fold 2

Similar results are obtained for the application in Fold 1.

Figures of merit

Study of the \(K^{-}K^{+}\)  S-wave amplitude near treshhold in \(D^{+}\rightarrow K^{-}K^{+}K^{+}\)

February 22, 2023

S. Ordoñez (UNAL-COL)

Selection

Fold 1

Fold 2

S-wave D2KKK

By Sebastian Ordoñez

S-wave D2KKK

  • 456