A Neural-Network-defined Gaussian Mixture Model for particle identification applied to the LHCb fixed-target programme

Particle identification in large high-energy physics experiments typically relies on classifiers obtained by combining many experimental observables. Predicting the probability density function (pdf) of such classifiers in the multivariate space covering the relevant experimental features is usually challenging. The detailed simulation of the detector response from first principles cannot provide the reliability needed for the most precise physics measurements. Data-driven modelling is usually preferred, though sometimes limited by the available data size and different coverage of the feature space by the control channels. In this paper, we discuss a novel approach to the modelling of particle identification classifiers using machine-learning techniques. The marginal pdf of the classifiers is described with a Gaussian Mixture Model, whose parameters are predicted by Multi Layer Perceptrons trained on calibration data. As a proof of principle, the method is applied to the data acquired by the LHCb experiment in its fixed-target configuration. The model is trained on a data sample of proton-neon collisions and applied to smaller data samples of proton-helium and proton-argon collisions collected at different centre-of-mass energies. The method is shown to perform better than a detailed simulation-based approach, to be fast and suitable to be applied to a large variety of use cases.