Back to publications

Precision String Phenomenology

Paper Details

Published: 2024/07/18

DOI: 10.48550/arXiv.2407.13836

ARXIV ID: 2407.13836v1

In order to make physical predictions, we require certain geometric data on a given Calabi-Yau manifold. This is recovered by considering the system of partial differential equations which a given geometric object should satisfy, subsequently solved using a neural network ansatz. Function optimisation on general manifolds is highly nontrivial - any sensible model must obey a set of geometric constraints imposed by the topology of the manifold. The authors demonstrate, standard neural network solvers have difficulty adhering to these conditions tabula rasa, resulting in nonphysical predictions. The authors a very general ansatz, valid for any Calabi-Yau manifold, which respects these conditions by construction, and allows the modelling of tensor fields on Calabi-Yau manifold of arbitrary degree. These novel methods agree with independent numerical computations and leave the authors well-poised to generalise this program to a wider class of models.