Federer's Curvature Measures: Characterization

• Federer's curvature measures are finitely additive set functions that localize intrinsic volumes via tube formulas and invariant differential forms.
• They are rigorously defined through axioms like translation, rotation invariance, and continuity, ensuring a unique linear combination representation.
• Extensions to non-smooth and pseudo-Riemannian settings enable robust computational methods and stability under Hausdorff perturbations.

A curvature measure, in the sense of Federer, is a finitely additive set function that localizes global geometric invariants such as intrinsic volumes or valuations, yielding a measure-theoretic assignment quantifying geometric content around subsets of a convex body or, more generally, sets of positive reach or subanalytic sets. Federer's fundamental insight was to relate these local measures to the coefficients in tube-volume (Steiner-type) expansions of parallel (offset) sets, and to construct them via normal cycles and special differential forms. Modern advancements include precise axiomatic characterizations—especially in the context of translation invariance, rotation invariance, and functoriality—as well as extensions to pseudo-Riemannian and singular geometric settings.

1. Classical Definition and Construction

Federer's curvature measures $C_k(K,\cdot)$, for $k = 0,\dots, n$, assign to each compact convex body $K \subset \mathbb{R}^n$ a finite Borel measure on $\mathbb{R}^n$, capturing the "density" of order-$k$ curvature concentrated near given regions. Formally, for $K$ with $C^2$ boundary and Borel set $U \subset \mathbb{R}^n$,

$C_k(K,U) = \frac{1}{\Vol S^{n-k-1}} \int_{N(K) \cap \pi^{-1}(U)} \kappa_k,$

where $N(K)$ is the unit normal bundle, $\pi$ the canonical projection, and $\kappa_k$ an explicit $n-1$ form alternating over base and fiber coordinates (Schuhmacher et al., 20 Jan 2026, Saienko, 2019). The corresponding total measures coincide with the intrinsic volumes: $V_k(K) = C_k(K, \mathbb{R}^n).$ The tube formula asserts that for $t > 0$ sufficiently small,

$$\operatorname{Vol}\left( \{x : d(x, K) < t \} \right) = \sum_{i=0}^n \omega_{n-i}\, t^{n-i} V_i(K),$$

with $C_k(K, \cdot)$ as the local coefficients (0812.1390). For $C^2$ sets of positive reach, these measures are well-defined and robust under Hausdorff-perturbations.

2. Axiomatic and Representation-Theoretic Characterization

The definitive axiomatic characterization is based on translation invariance, locality, continuity, and, in the most restrictive setting, $\operatorname{SO}(n)$-invariance. Let $\operatorname{Curv}(\mathbb{R}^n)$ denote the space of all translation-invariant curvature measures on compact convex bodies:

• Locality: If $K\cap U = L\cap U$ for open $U$, then $\Phi(K, B \cap U) = \Phi(L, B \cap U)$ for all Borel $B$.
• Translation invariance: $\Phi(K + x, B + x) = \Phi(K, B)$ for all $x \in \mathbb{R}^n$.
• Continuity: If $K_i \rightarrow K$ in the Hausdorff metric, then $\Phi(K_i) \rightharpoonup \Phi(K)$ weakly.

The main structure theorem states [(Schuhmacher et al., 20 Jan 2026), Theorem 6.2]: Every $\operatorname{SO}(n)$-invariant element $\Phi \in \operatorname{Curv}(\mathbb{R}^n)$ is a unique linear combination of the classical Federer curvature measures:

$\Phi = \sum_{k=0}^n a_k C_k.$

No positivity or monotonicity assumptions are needed—$\operatorname{SO}(n)$-invariance alone suffices. This is the precise analogue of Hadwiger's theorem for valuations.

This rigidity finds further justification in the fine structure of $\operatorname{Curv}(\mathbb{R}^n)$ as a $\mathrm{GL}(n)$-module. Each graded component by degree and parity has composition length at most 2, and the subspace of measures with zero total mass is irreducible in low degrees (Schuhmacher et al., 20 Jan 2026). The classical measures are thus characterized as the only possible $\operatorname{SO}(n)$-invariant elements.

3. Integral Representation and Smooth Case

Federer's curvature measures admit explicit integral representations via forms on the sphere bundle $S\mathbb{R}^n = \mathbb{R}^n \times S^{n-1}$ (Saienko, 2019): $C_k(K, U) = \int_{N(K) \cap \pi^{-1}(U)} \omega_k,$ where $\omega_k$ is the unique (up to scaling) $\operatorname{SO}(n)$-invariant primitive horizontal $(k, n-1-k)$ form: $\omega_k(x, y) = \frac{1}{k! (n-k-1)! s_{n-k-1}} \sum_{\pi\in S_n} \sgn(\pi)\, y_{\pi(n)}\, x^{\pi(1)} \wedge \cdots \wedge x^{\pi(k)} \wedge y^{\pi(k+1)} \wedge \cdots \wedge y^{\pi(n-1)}.$ These forms ensure that $C_k$ recovers, on smooth domains, the integral of the $k$th elementary symmetric function of principal curvatures over $U \cap \partial K$. Under the globalization map, these curvature measures push forward to the intrinsic volumes, providing a finite-dimensional decomposition of the space of $\operatorname{SO}(n)$-invariant curvature measures and valuations (Saienko, 2019).

4. Extensions: Singular and Non-Smooth Settings

Federer's original theory was for sets of positive reach, but developments have extended curvature measures to arbitrary closed sets (Santilli, 2017). Given $A \subset \mathbb{R}^n$, one defines a normal bundle $N(A)$ and introduces the second fundamental form $Q_A(a,u)$ for $(a, u) \in N(A)$. The principal curvatures $\kappa_i(a,u)$ are then the eigenvalues of $Q_A(a,u)$.

The curvature measures $\mu_m(A,\omega)$ are expressed as

$$\mu_{m}(A,\omega) = \frac{1}{(m+1)\,\alpha(m+1)}\, \int_{\omega} \sigma_{m}\bigl(\kappa_{1},\dots,\kappa_{n-1}\bigr)\, d\mathcal{H}^{n-1}(a,u),$$

for Borel $\omega \subset N(A)$, where $\sigma_m$ is the $m$-th elementary symmetric polynomial in the principal curvatures (Santilli, 2017). This generalization relies on the rectifiability of the normal bundle and a fine analysis of the approximate differentials involved.

5. Stability and Computational Aspects

The curvature measures exhibit quantitative stability with respect to Hausdorff perturbations of the underlying set. If $K,K' \subset \mathbb{R}^n$ are close in the Hausdorff distance and have sufficiently large positive reach (or $\mu$-reach), then for any smooth compactly supported test function $f$,

$$\left| C_k(K, f) - C_k(K', f) \right| \leq C \cdot d_H(K, K')^{1/2},$$

where $C$ depends on the reach, covering numbers, and smoothness constants (0812.1390, 0706.2153). This 1/2-Hölder-type stability allows for robust approximation of curvature measures from point-cloud samples via offsets and unions of balls, leveraging explicit formulas for curvature integrals over such geometric primitives.

For sets sampled as point clouds $P$, one approximates $K$ via the union of balls $P_r$, computes curvature measures via the local combinatorics of the corresponding sphere-polyhedron, and controls error through covering and combinatorial complexity (0706.2153, 0812.1390).

6. Generalizations to the Dual Brunn–Minkowski and Manifold Contexts

In the dual Brunn–Minkowski theory, dual analogues of Federer's curvature measures $\widetilde C_q(K, \cdot)$ are constructed using the radial function $\rho_K$ and the radial Gauss map $a_K(u)$: $$\widetilde C_q(K,\omega) = \int_{a_K(\omega)} \rho_K(u)^q\, du,$$ with dual Steiner-type expansions for the volume of dual parallel bodies. Existence and uniqueness questions (dual Minkowski problems) are resolved via subspace mass inequalities that tightly control the distribution of mass in candidate Borel measures (Huang et al., 8 Feb 2025). Classical cases are recovered as limiting values of the parameter $q$.

On manifolds, Federer’s measures generalize as Lipschitz–Killing curvature measures, defined via pairs of invariant forms $(\lambda_k, \kappa_k)$ acting on conormal cycles and integrating curvature information over both interior and boundary-type parts of singular sets (Bernig et al., 2019, Bernig et al., 2020). These measures are the unique isometric-embedding-invariant assignments—that is, they satisfy the Weyl principle, are functorial, and their invariance characterizes them up to linear combination on any (pseudo-)Riemannian manifold.

7. Summary and Significance

Federer’s curvature measures are characterized by precise geometric, algebraic, and representation-theoretic properties: locality, translation invariance, continuity, and—in the strongest case—rotation invariance. These measures arise as local coefficients in tube formulas and have canonical integral-geometric constructions via normal cycles and invariant differential forms. Their uniqueness (classification) results extend to various settings: general convex bodies, sets of positive reach, singular or non-smooth sets, dual Brunn–Minkowski theory, and pseudo-Riemannian geometry (Schuhmacher et al., 20 Jan 2026, Saienko, 2019, Bernig et al., 2020, Huang et al., 8 Feb 2025, 0812.1390, Santilli, 2017). The robust stability and computability of curvature measures have established them as a central tool in geometric analysis, convex geometry, and computational geometry.