📝 SummarXiv

Abstract

Identifying the principles that determine neural population activity is paramount in the field of neuroscience. We propose the Principle of Isomorphism (PIso): population activity preserves the essential mathematical structures of the tasks it supports. Using grid cells as a model system, we show that the neural metric task is characterized by a flat Riemannian manifold, while path integration is characterized by an Abelian Lie group. We prove that each task independently constrains population activity to a toroidal topology. We further show that these perspectives are unified naturally in Euclidean space, where commutativity and flatness are intrinsically compatible and can be extended to related systems including head-direction cells and 3D grid cells. To examine how toroidal topology maps onto single-cell firing patterns, we develop a minimal network architecture that explicitly constrains population activity to toroidal manifolds. Our model robustly generates hexagonal firing fields and reveals systematic relationships between network parameters and grid spacings. Crucially, we demonstrate that conformal isometry, a commonly proposed hypothesis, alone is insufficient for hexagonal field formation. Our findings establish a direct link between computational tasks and the hexagonal-toroidal organization of grid cells, thereby providing a general framework for understanding population activity in neural systems and designing task-informed architectures in machine learning.