📝 SummarXiv

Abstract

The classical fiber product in algebraic geometry provides a powerful tool for studying loci where two morphisms to a base scheme, $\phi: X \to S$ and $\psi: Y \to S$, coincide exactly. This condition of strict equality, however, is insufficient for describing the geometric structure of semantic spaces in modern large language models, whose foundational architecture is the Transformer. The token spaces of these models are fundamentally arithmetic, and recent work has revealed complex geometric singularities, challenging the classical manifold hypothesis. This paper develops a new framework to study and quantify the nature of approximate alignment between morphisms in the context of arithmetic geometry, using the tools of \'etale homotopy theory. We introduce the central object of our work, the \'etale mismatch torsor, a sheaf of torsors over the product scheme $X \times_S Y$. The structure of this sheaf serves as a rich, intrinsic, and purely algebraic measure of the global relationship between the two morphisms. Our main theorem provides a complete classification of these structures, establishing a bijection between their isomorphism classes and the first \'etale cohomology group $H^1_{\text{\'et}}(X \times_S Y, \underline{\pi_1^{\text{\'et}}(S)})$. This result yields a discrete, algebraic invariant that replaces the non-intrinsic, metric-based notions of approximation first explored in applied contexts. We apply this framework to the concrete case of generalized Howe curves over finite fields to illustrate its connection to classical problems. This work introduces new homotopical invariants to arithmetic geometry, opening new avenues for the structural analysis of complex discrete systems.