Title:

Improving Accuracy of Scattering Predictions with Machine Learning

The Scattering Matrix (S-Matrix) is a fundamental object in Quantum Field Theory that contains the probability amplitudes of particles scattering off each other. In the 1970s, D. Atkinson solved for families of solutions to the equations that govern this scattering with phase-shift ambiguities. Research on this topic stopped for several decades, until 2023, when a paper by Dersey, Schwartz, and Zhiboedov was published where they reproduced and improved upon Atkinson’s results using machine learning. Last year, we modified this program and were able to further improve the results, lowering a parameter called sinμ, which governs if the scattering angle is unique, to 1.447 from 1.67 by Dersey et al. and 2.15 by Atkinson, and lowering the loss in solving our solutions. This year, we continued to try to improve the program using Kolmorgorov-Arnold Networks (KANs), and were able to improve loss in regions with a single-phase solution. Using KANs also allowed us to use better optimizers for solving and lowered the time it takes to reach solutions.

Name

Comment*

Comments are viewable only by submitter

Captcha*