Machine learning-based study of open-charm hadrons in proton-proton collisions at the Large Hadron Collider

In proton-proton and heavy-ion collisions, the study of charm hadrons plays a pivotal role in understanding the QCD medium and provides an undisputed testing ground for the theory of strong interaction, as they are mostly produced in the early stages of collisions via hard partonic interactions. The lightest open charm, <a:math xmlns:a="http://www.w3.org/1998/Math/MathML" display="inline"><a:msup><a:mi>D</a:mi><a:mn>0</a:mn></a:msup></a:math> meson (<c:math xmlns:c="http://www.w3.org/1998/Math/MathML" display="inline"><c:mrow><c:mi>c</c:mi><c:mover accent="true"><c:mrow><c:mi>u</c:mi></c:mrow><c:mrow><c:mo stretchy="false">¯</c:mo></c:mrow></c:mover></c:mrow></c:math>), can originate from two separate sources. The prompt <g:math xmlns:g="http://www.w3.org/1998/Math/MathML" display="inline"><g:msup><g:mi>D</g:mi><g:mn>0</g:mn></g:msup></g:math> originates from either direct charm production or the decay of excited open charm states, while the nonprompt stems from the decay of beauty hadrons. In this paper, using different machine learning (ML) algorithms such as XGBoost, CatBoost, and Random Forest, an attempt has been made to segregate the prompt and nonprompt production modes of the <i:math xmlns:i="http://www.w3.org/1998/Math/MathML" display="inline"><i:msup><i:mi>D</i:mi><i:mn>0</i:mn></i:msup></i:math> meson signal from its background. The ML models are trained using the invariant mass through its hadronic decay channel, i.e., <k:math xmlns:k="http://www.w3.org/1998/Math/MathML" display="inline"><k:msup><k:mi>D</k:mi><k:mn>0</k:mn></k:msup><k:mo stretchy="false">→</k:mo><k:msup><k:mi>π</k:mi><k:mo>+</k:mo></k:msup><k:msup><k:mi>K</k:mi><k:mo>−</k:mo></k:msup></k:math>, pseudoproper time, pseudoproper decay length, and distance of closest approach of <n:math xmlns:n="http://www.w3.org/1998/Math/MathML" display="inline"><n:msup><n:mi>D</n:mi><n:mn>0</n:mn></n:msup></n:math> meson, using 8 simulated <p:math xmlns:p="http://www.w3.org/1998/Math/MathML" display="inline"><p:mi>p</p:mi><p:mi>p</p:mi></p:math> collisions at <r:math xmlns:r="http://www.w3.org/1998/Math/MathML" display="inline"><r:msqrt><r:mi>s</r:mi></r:msqrt><r:mo>=</r:mo><r:mn>13</r:mn><r:mtext> </r:mtext><r:mtext> </r:mtext><r:mi>TeV</r:mi></r:math>. The ML models used in this analysis are found to retain the pseudorapidity, transverse momentum, and collision energy dependence. In addition, we report the ratio of nonprompt to prompt <t:math xmlns:t="http://www.w3.org/1998/Math/MathML" display="inline"><t:msup><t:mi>D</t:mi><t:mn>0</t:mn></t:msup></t:math> yield, the self-normalized yield of prompt and nonprompt <v:math xmlns:v="http://www.w3.org/1998/Math/MathML" display="inline"><v:msup><v:mi>D</v:mi><v:mn>0</v:mn></v:msup></v:math>, and explore the charmonium, <x:math xmlns:x="http://www.w3.org/1998/Math/MathML" display="inline"><x:mi>J</x:mi><x:mo>/</x:mo><x:mi>ψ</x:mi></x:math> to open charm, <z:math xmlns:z="http://www.w3.org/1998/Math/MathML" display="inline"><z:msup><z:mi>D</z:mi><z:mn>0</z:mn></z:msup></z:math> yield ratio as a function of transverse momenta and normalized multiplicity. The observables studied in this paper are well predicted by all the ML models compared to the simulation.