BandQ: Optimise the existing Lb2LcDs(h)(h) lines (with MVA)

The BDT is applied to select a detached \Lambda_c samples.

For the training samples:

• The signal is the simulated events for both \Lambda_b\to \Lambda_c D_{s} K^+ K^- (\pi^+ \pi^-)
• The background sample is the NoBias data in Spruce24c4 (with m(\Lambda_c) \in [2310,2340]~MeV)

Preselections are required for the training samples:

p, K, π selection
p_T	>200 MeV
P	>2.5 GeV
\chi_{IP}^2	>4
PID
p	DLL_p>0 and ISMUON==0
K	DLL_K>0
\pi	DLL_{\pi}>0

Charm hadron
p_T
D_s^-	>900 MeV
\Lambda_c^+	>1000 MeV
∑p_T
D_s^-	>1.4 GeV
\Lambda_c^+	>1.45 GeV
ρ_{PV}	>0.1 mm
DIRA	>0.995
\chi_{DOCA}^2	<12
max[DOCA]	<0.18 mm
\chi_{vtx}^2 /n_{dof}	<9
\chi_{IP}^2	>3
average \chi_{IP}^2 (p,K,π)	>8
\chi_{FD}^2	>30
\tau_{OWNPV}	>0.2 ps
m(D_s^-)	(1910, 2025) MeV/c^2
m(\Lambda_c^+)	(2230, 2340) MeV/c^2

5 varialbles are used as the features of the BDT:

• log(p_T) for \Lambda_c
• \chi^{2}_{vtx}/n_{dof} of \Lambda_c
• \chi^2_{IP} of p_{\Lambda_c}
• F.MAXDOCA for \Lambda_c
• \chi^2_{FD} for \Lambda_c

Good seperation for each variables:

And no over-training found:

Go with lhcb-datapkg/ParamFiles!129 (merged)

FYI @jfu