Cheeger–Naber Minkowski Dimension Bound

• The paper establishes a quantitative estimate on singular sets using ε-regularity and geometric packing, leading to sharp Minkowski bounds.
• It employs quantitative stratification and cone-splitting to define singular strata and precisely control the shrinkage of tubular neighborhoods.
• The approach refines classical Hausdorff dimension estimates and has key applications in analyzing area-minimizing hypersurfaces and harmonic map regularity.

The Cheeger–Naber Minkowski Dimension Bound provides a quantitative estimate of the size of singular sets arising in a broad class of geometric and variational settings. This estimate not only bounds the Hausdorff dimension—a classical measure of set size—but yields sharp upper bounds on the Minkowski (box-counting) dimension and the rate at which tubular neighborhoods of the singular set shrink as their radius decreases. This quantitative control is achieved through a combination of ε-regularity theorems, quantitative stratification, and geometric packing arguments, and has become central to regularity theory in geometric analysis.

1. Setting and Main Statement

Consider a sequence of complete Riemannian $n$-manifolds $(M_i^n, g_i)$ satisfying two key uniform hypotheses:

• Volume non-collapsing: For some $v > 0$ and all $x\in M_i$, $r \le 1$,

$\Vol_{g_i}(B_{g_i}(x,r))\ge v r^n.$

• Integral curvature bounds: For exponents $p > n/2$ and $1 \le q \le n/2$,

$\int_{M_i}|\Ric|^p \le \lambda, \qquad \int_{M_i}|\Rm|^q \le \Lambda,$

for fixed $\lambda, \Lambda > 0$.

Passing to a pointed Gromov–Hausdorff limit

$(M_i^n, g_i)$0

the limit $(M_i^n, g_i)$1 decomposes into a regular (smooth) part $(M_i^n, g_i)$2 and a singular set $(M_i^n, g_i)$3. The singular set $(M_i^n, g_i)$4 consists of points admitting no $(M_i^n, g_i)$5-regular neighborhoods, equivalently, points where no small ball is Gromov–Hausdorff close to a Euclidean ball.

The Cheeger–Naber Minkowski dimension bound asserts:

$(M_i^n, g_i)$6

Here, $(M_i^n, g_i)$7 denotes the (upper) Minkowski dimension—defined via asymptotics of the covering number by $(M_i^n, g_i)$8-balls as $(M_i^n, g_i)$9. For integer $v > 0$0, this matches the best-known Hausdorff dimension estimate, and for non-integral $v > 0$1 the same bound holds up to rounding (Qian, 2023).

2. Quantitative Stratification and ε-Regularity

The approach relies on the construction of quantitative singular strata. For each $v > 0$2, and for small parameters $v > 0$3, define

$v > 0$4

Points in $v > 0$5 fail to exhibit $v > 0$6-Euclidean symmetry at every scale larger than $v > 0$7.

The ε-regularity theorem states that if, on a ball $v > 0$8,

$v > 0$9

then the $x\in M_i$0 harmonic radius $x\in M_i$1. This result allows one to localize the singular set as the set where this regularity fails, and to stratify it according to how far the failure is from being geometric or algebraic in nature (Qian, 2023).

3. Packing Estimates and Minkowski Dimension

Central to the quantitative nature of the Cheeger–Naber bound is the derivation of sharp tube-packing estimates. By combining ε-regularity and a quantitative cone-splitting principle, one proves that for the $x\in M_i$2-th stratum,

$x\in M_i$3

For $x\in M_i$4, this yields

$x\in M_i$5

for the covering number $x\in M_i$6. Allowing $x\in M_i$7, one concludes

$x\in M_i$8

and the corresponding Minkowski content is finite. These bounds are effective, quantifying the rate at which neighborhoods of the singular set shrink in volume, and thereby refining the purely measure-theoretic Hausdorff dimension control (Qian, 2023, Focardi et al., 2014, Vedovato, 2019).

4. Technical Ingredients: Cone-Splitting and Neck Decomposition

Multiple geometric and analytic tools underpin the Cheeger–Naber approach:

• Cone-splitting lemma: If a tangent cone at a point admits two almost-splitting directions, the space must split off a line, leading to a stratification by the dimension of symmetries.
• Neck decomposition: By controlling the number of bad scales at a point, one shows the structure between those scales is a “neck,” diffeomorphic to $x\in M_i$9.
• Quantitative stratification and packing: Control of the number and arrangement of points with bad scales—through quantitative stratification and cone-splitting—yields a covering by small balls whose number satisfies a sharp power law determined by the dimension defect.
• Volume estimates using $r \le 1$0 control: Coarea and slicing arguments enable $r \le 1$1 control over the generalized second fundamental form, which integrates into packing and covering estimates (Qian, 2023, Vedovato, 2019).

5. Comparison with Hausdorff Dimension Bounds

Cheeger’s earlier work established that

$r \le 1$2

in the sense of Hausdorff dimension (Qian, 2023). The Cheeger–Naber Minkowski bound is strictly stronger: the Minkowski dimension controls not just the measure of the singular set, but the growth rate of neighborhood volumes. This refinement is essential for applications that require uniform quantitative control over neighborhoods of singularities, rather than mere negligibility in Hausdorff measure (Focardi et al., 2014).

6. Applications and Extensions

Diffeomorphism Finiteness and Bubble-Tree Decomposition

In the critical case $r \le 1$3, the singular set is a finite set of isolated points. By constructing a “bubble-tree” decomposition, where "body" regions have uniformly bounded regularity radius and "neck" regions are modeled on tubular neighborhoods of spherical quotients, the finiteness of diffeomorphism types follows. The bounded height of the tree, coupled with the Cheeger–Gromov finiteness theorem for regular regions, ensures only finitely many diffeomorphism classes occur (Qian, 2023).

Area-Minimizing Hypersurfaces and Beyond

For area-minimizing hypersurfaces of dimension $r \le 1$4 in $r \le 1$5-manifolds, the Cheeger–Naber Minkowski bound specifies that the singular set has dimension at most $r \le 1$6, extending beyond the Hausdorff setting. This result is crucial in regularity theory for geometric variational problems and the analysis of mean curvature flow and $r \le 1$7-bubbles in general relativity (Brendle et al., 9 Apr 2026).

Other Variational Problems

The Cheeger–Naber argument and quantitative stratification have direct extensions to:

• $r \le 1$8-energy minimizing maps (yielding singular sets of Minkowski dimension $r \le 1$9) (Vedovato, 2019).
• Dirichlet-minimizing $\Vol_{g_i}(B_{g_i}(x,r))\ge v r^n.$0-valued functions (Focardi et al., 2014).
• Stationary harmonic maps, varifolds with $\Vol_{g_i}(B_{g_i}(x,r))\ge v r^n.$1 mean curvature, and almost-minimizers of perimeter in codimension one (Focardi et al., 2014).

These generalizations rely on verifying quantitative stratification and local ε-regularity, then applying abstract packing principles and discrete/continuous Reifenberg theorems.

7. Summary Table of Main Quantitative Dimension Bounds

Geometric Object	Upper Minkowski Bound	Reference
GH limits with $\Vol_{g_i}(B_{g_i}(x,r))\ge v r^n.$2 curvature bounds	$\Vol_{g_i}(B_{g_i}(x,r))\ge v r^n.$3	(Qian, 2023)
Area-minimizing hypersurfaces in $\Vol_{g_i}(B_{g_i}(x,r))\ge v r^n.$4-manifolds	$\Vol_{g_i}(B_{g_i}(x,r))\ge v r^n.$5 (codimension-one)	(Brendle et al., 9 Apr 2026)
$\Vol_{g_i}(B_{g_i}(x,r))\ge v r^n.$6-energy minimizing maps, domain dim $\Vol_{g_i}(B_{g_i}(x,r))\ge v r^n.$7	$\Vol_{g_i}(B_{g_i}(x,r))\ge v r^n.$8	(Vedovato, 2019)
Dir-minimizing $\Vol_{g_i}(B_{g_i}(x,r))\ge v r^n.$9-valued functions	$p > n/2$0	(Focardi et al., 2014)
Mean curvature flow (Brakke flows)	$p > n/2$1-stratum: $p > n/2$2	(Cheeger et al., 2012)

• "Singular sets on spaces with integral curvature bounds and diffeomorphism finiteness for manifolds" (Qian, 2023)
• "Quantitative regularity for $p > n/2$3-minimizing maps through a Reifenberg Theorem" (Vedovato, 2019)
• "Quantitative Stratification and the Regularity of Mean Curvature Flow" (Cheeger et al., 2012)
• "A dimension descent scheme for the positive mass theorem in high dimensions" (Brendle et al., 9 Apr 2026)
• "Improved estimate of the singular set of Dir-minimizing Q-valued functions via an abstract regularity result" (Focardi et al., 2014)