Allard's interior $\varepsilon$-Regularity Theorem in Alexandrov spaces
Marcos Agnoletto, Julio C. Correa Hoyos, Márcio Fabiano da Silva, Stefano Nardulli
TL;DR
This work develops an intrinsic theory for $m$-dimensional varifolds in Alexandrov spaces with double-sided curvature bounds, culminating in Allard's Interior $\varepsilon$-Regularity Theorem in a nonsmooth ambient setting. By building an intrinsic varifold framework (Definitions, First Variation, Monotonicity, Constancy, and Tangent/Rectifiability) and then deriving Caccioppoli inequalities, density estimates, and affine-approximation results, the paper achieves a Euclidean-like regularity theory that crucially avoids Nash embeddings. The results provide explicit constants depending only on dimension, curvature, and injectivity radius, enabling a density-dominated approach to represent varifolds locally as $\mathcal{C}^{1,\alpha}$ graphs over intrinsic submanifolds. The methodology leverages harmonic submanifolds and harmonic coordinates to manage curvature terms and delivers a robust regularity toolkit for nonsmooth metric-measure spaces with controlled geometry.
Abstract
In this paper, we prove Allard's Interior $\varepsilon$-Regularity Theorem for $m$-dimensional varifolds with generalized mean curvature in $L^p_{loc}$, for $p \in \mathbb{R}$ such that $p>m$, in Alexandrov spaces of dimension $n$ with double-sided bounded intrinsic sectional curvature. We first give an intrinsic proof of this theorem in the case of varifolds in Riemannian manifolds of dimension $n$ whose metric tensor is at least of class $\mathcal{C}^2$, without using Nash's Isometric Embedding Theorem. This approach provides explicitly computable constants that depend only on $n$, $m$, the injectivity radius and bounds on the sectional curvature, which is essential for proving our main theorem, as we establish it through a density argument in the topological space of Riemannian manifolds with positive lower bounds on the injectivity radius and double-sided bounds on sectional curvature, equipped with the $\mathcal{C}^{1,α}$ topology, for every $α\in ]0,1[$ (in fact, it is enough with the $W^{2,q}$ topology for some suitable $q$ large enough).
