Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

CGAN-Based Framework for Meson Mass and Width Prediction

S. Rostami, M. Malekhosseini, M. Rahavi Ezabadi, K. Azizi

TL;DR

This work tackles the challenge of undetermined meson masses and decay widths by applying a conditional GAN to generate physically consistent synthetic data conditioned on meson properties. By augmenting a limited dataset and training multiple CGANs with bagging, the approach yields mass and width predictions that align more closely with experimental values and lattice/QCD-sum-rule results, while providing quantified uncertainties. The results span ordinary, exotic, and fully-heavy tetraquark states, demonstrating improved accuracy over previous DNN-based methods and suggesting a useful data-driven complement to traditional hadron spectroscopy. The study also outlines a roadmap for extending CGAN-based data augmentation to broader hadronic systems and potential applications in fast detector-level simulations and hadronization modeling.

Abstract

Mesons play a crucial role in understanding the strong interaction in the framework of quantum chromodynamics (QCD). However, the mass and decay width of several ordinary and exotic mesons remain experimentally undetermined. In this work, we propose a novel application of advanced machine learning techniques to deal with this challenge. Due to the limited available meson datasets, traditional data-driven methods are norm To overcome this, we employ a Conditional Generative Adversarial Network (CGAN) to generate synthetic meson data based on known physical parameters. This not only augments the dataset but also retain the underlying physics of the original mesons data. With the extended dataset, we train multiple copies of CGAN and apply a bagging technique to predict uncertainties, improving the robustness and reliability of the predictions. As our findings indicate, the CGAN models are capable of well describing meson properties and their structure relations, offering a potent novel instrument for hadron spectroscopy. This calculation opens a promising future for data-driven hadron physics studies.

CGAN-Based Framework for Meson Mass and Width Prediction

TL;DR

This work tackles the challenge of undetermined meson masses and decay widths by applying a conditional GAN to generate physically consistent synthetic data conditioned on meson properties. By augmenting a limited dataset and training multiple CGANs with bagging, the approach yields mass and width predictions that align more closely with experimental values and lattice/QCD-sum-rule results, while providing quantified uncertainties. The results span ordinary, exotic, and fully-heavy tetraquark states, demonstrating improved accuracy over previous DNN-based methods and suggesting a useful data-driven complement to traditional hadron spectroscopy. The study also outlines a roadmap for extending CGAN-based data augmentation to broader hadronic systems and potential applications in fast detector-level simulations and hadronization modeling.

Abstract

Mesons play a crucial role in understanding the strong interaction in the framework of quantum chromodynamics (QCD). However, the mass and decay width of several ordinary and exotic mesons remain experimentally undetermined. In this work, we propose a novel application of advanced machine learning techniques to deal with this challenge. Due to the limited available meson datasets, traditional data-driven methods are norm To overcome this, we employ a Conditional Generative Adversarial Network (CGAN) to generate synthetic meson data based on known physical parameters. This not only augments the dataset but also retain the underlying physics of the original mesons data. With the extended dataset, we train multiple copies of CGAN and apply a bagging technique to predict uncertainties, improving the robustness and reliability of the predictions. As our findings indicate, the CGAN models are capable of well describing meson properties and their structure relations, offering a potent novel instrument for hadron spectroscopy. This calculation opens a promising future for data-driven hadron physics studies.

Paper Structure

This paper contains 10 sections, 5 equations, 2 figures, 6 tables.

Table of Contents

  1. Introduction
  2. Generative Adversarial Networks
  3. Data augmentation and preprocessing
  4. Discussion
  5. Ordinary or exotic mesons: an ongoing debate
  6. Light mesons
  7. Exotic mesons
  8. Fully heavy tetraquarks
  9. Meson width
  10. Summary and conclusions

Figures (2)

  • Figure 1: Illustration of the bagging technique applied to the CGAN model. The method involves training $N$ independent CGAN models on different subsets of the training data, which are created through random sampling (bootstrapping).
  • Figure 2: Side-by-side comparison of heatmaps for the original and augmented meson datasets, demonstrating how the augmented data preserves the patterns and distributions of the original dataset.