AI and the World's Hard Problems · Physics · arXiv · 2026-10-21
Learning the open string spectrum with neural networks
The tree-level open superstring has an integer mass-squared spectrum on linear Regge trajectories, and this work asks how far consistency conditions—locality, analyticity, crossing, unitarity, and entanglement minimization in 2→2 scattering—fix it. Using a bootstrap based on the local two-channel crossing symmetric dispersion relation, a machine learning algorithm learns the mass-squared spectrum and partial-wave coefficients simultaneously, going beyond earlier work in which the spectrum was fixed. A three-channel version yields a bound on the leading Wilson coefficients, with the Virasoro-Shapiro amplitude located in the allowed region of that space.