Baryons from Mesons: A Machine Learning Perspective

Yarin Gal, Vishnu Jejjala, Damian Kaloni Mayorga Pena, Challenger MishraarXiv 2020 · arXiv

Abstract

Predicting the masses of baryons—three-quark particles like protons—from first principles in Quantum Chromodynamics (QCD) is notoriously difficult due to the non-perturbative nature of the strong force at low energies. This research explores a novel machine learning approach based on Edward Witten’s 1979 hypothesis that baryons can be viewed as solitons emerging from mesonic fields in the large-$N$ limit. By training models on meson data, the authors demonstrate that Gaussian Processes and Neural Networks can accurately predict baryon masses, with Gaussian Processes achieving 96.6% accuracy and outperforming traditional physics-based models.

Introduction

In subatomic physics, the interactions of quarks and gluons are governed by QCD. While high-energy interactions are calculable, low-energy "hadronization"—where quarks bind into particles—remains a challenge. Traditional methods like Lattice QCD or Constituent Quark Models (CQM) are computationally expensive or simplified. This study leverages machine learning to bridge the gap between mesons (quark-antiquark pairs) and baryons (three-quark states), testing the empirical validity of the meson-baryon theoretical link.

Methodology

The researchers utilized two distinct ML architectures: a single-layer feed-forward Neural Network (NN) and a Gaussian Process (GP). Particles were encoded as 13-dimensional vectors representing valence quark counts and global quantum numbers (Isospin, Angular Momentum, Parity).

The Gaussian Process was selected for its ability to handle small datasets (196 mesons) and its inherent Bayesian nature. Unlike the NN, the GP provides a measure of uncertainty alongside its predictions, which is vital for interpreting results in a sparse data environment.

Gaussian Process regression illustrating the fit to data points (blue crosses) and the associated uncertainty (green shaded region).

Results and Discussion

The GP model achieved a remarkable 96.6% accuracy (3.4% mean error) on a test set of 43 baryons that were entirely withheld during training. In contrast, the standard Constituent Quark Model (CQM) yielded an 8.6% error rate. The Neural Network also performed well (90.3% accuracy) and notably learned subtle physical nuances, such as the mass difference between the proton and neutron.

The success of these models supports the "soliton" interpretation of baryons. Furthermore, the GP's ability to provide variance estimates allowed researchers to evaluate the reliability of predictions for "exotic" candidates like tetraquarks and pentaquarks.

Conclusion

This work demonstrates that machine learning can capture complex, non-linear relationships in hadronic physics that traditional linear models fail to describe. While challenges remain—such as the unique physics of the pion and the inclusion of glueballs—the integration of theoretical physics constraints with probabilistic ML models like Gaussian Processes offers a promising path forward for mapping the subatomic spectrum.

Summary updated

Abstract

Quantum chromodynamics (QCD) is the theory of the strong interaction. The fundamental particles of QCD, quarks and gluons, carry colour charge and form colourless bound states at low energies. The hadronic bound states of primary interest to us are the mesons and the baryons. From knowledge of the meson spectrum, we use neural networks and Gaussian processes to predict the masses of baryons with 90.3% and 96.6% accuracy, respectively. These results compare favourably to the constituent quark model. We as well predict the masses of pentaquarks and other exotic hadrons.
BibTeX
@misc{gal2020baryons,
  author = {Yarin Gal and Vishnu Jejjala and Damian Kaloni Mayorga Pena and Challenger Mishra},
  title = {{Baryons from Mesons: A Machine Learning Perspective}},
  year = {2020},
  eprint = {2003.10445},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2003.10445},
}
Figure