Navigating the string landscape with machine learning techniques

Thesis

This thesis explores the possibility of using heuristic search algorithms from data science, namely genetic algorithms and reinforcement learning, to navigate the string landscape to uncover phenomenologically interesting constructions. Specifically, we apply these algorithms to construct holomorphic slope-stable vector bundles over Calabi-Yau three-folds. These vector bundles lead to the particle spectrum of the minimally supersymmetric standard model (plus uncharged moduli), via compactifications of the E8 × E8 heterotic string. We explore two types of vector bundles: sums of line bundles, which have been extensively explored in existing literature and thus serve as a benchmark for the effectiveness of the algorithms, and monad bundles, where only one quasi-realistic model was previously known. For both environments, these search algorithms were able to discover many models, while exploring as little as 10−19 of the total space. As an example of these methods in a simpler context, we also explored Froggatt-Nielsen models of quark masses.

Authors: Harvey, TR

• DOI: 10.5287/ora-y5oeq5qv7
• Type of award: DPhil
• Level of award: Doctoral
• Awarding institution: University of Oxford
• Language: English
• Keywords: quarks; string theory; standard model; genetic algorithms; machine learning; heterotic string; reinforcement learning; Calabi-Yau
• Subjects: String models