Singular Set of Nondifferentiability Points

• Singular Set of Nondifferentiability Points is the collection of domain points where functions or currents lack differentiability, often showing fractal and measure-zero properties.
• The analysis employs tools like blow-up methods, frequency functions, and rectifiability theory to classify the geometric and variational behavior of these sets.
• Applications span minimal surface theory, convex variational problems, and PDEs, offering insights into regularity, stratification, and rigidity of solutions.

A singular set of nondifferentiability points refers to the subset of the domain of a function (or the support of an object, such as a current or a surface), where the function fails to be differentiable. In analysis, geometry, and PDEs, the structure, size, and geometric properties of these sets are central both for classification and for understanding regularity phenomena, fractal behaviors, and rigidity properties of solutions to variational and geometric problems.

1. Foundational Definitions and General Notions

For a function $f: \mathbb{R}^n \to \mathbb{R}$ (or more generally, a map between manifolds or a section of some bundle), the singular set $S_f$ is defined as

$$S_f := \{ x \in \mathbb{R}^n : f \text{ is not differentiable at } x \}.$$

For geometric objects such as area-minimizing integral currents $T$, the singular set $\operatorname{Sing}(T)$ is the set of points in the support of $T$ where the current is not, locally, a smooth embedded submanifold with integer multiplicity (Lellis et al., 2023).

Singular sets often exhibit intricate structure: they are typically "small" in measure-theoretic terms (measure zero), but can be "large" in topological or fractal sense (positive or even full Hausdorff dimension), and their rectifiability is a central aspect of their study.

2. Singular Sets in Geometric Measure Theory

Area-minimizing currents—generalizations of minimal surfaces—admit a stratified structure of their singular set. Consider an $m$-dimensional area-minimizing integral current $T$ in a manifold $\Sigma$. Write $\operatorname{Reg}(T)$ for the regular set and $\operatorname{Sing}(T)$ for the set of singular points. If every tangent cone at $p \in \operatorname{Sing}(T)$ is a flat multiplicity-$Q$ plane, $p$ is called a flat singular point.

Key results for the detailed structure of singular sets include:

• The singular set $\operatorname{Sing}(T)$ has Hausdorff dimension at most $m-2$.
• For flat singular points with singularity degree $I(p) > 1$, the set $F_{Q,>1}(T)$ is $(m-2)$-rectifiable; for those with $I(p) = 1$, $F_{Q,1}(T)$, the $(m-2)$-dimensional Hausdorff measure is zero.
• Thus, $\operatorname{Sing}(T)$ is $(m-2)$-rectifiable and satisfies $\dim_H \operatorname{Sing}(T) = m-2$ (Lellis et al., 2023).

The central analytical tools are frequency function monotonicity, blow-up analysis to Dir-minimizing multivalued maps, and rectifiable-Reifenberg theory for controlling the geometry using $\beta$-numbers.

3. Singular Sets for Lipschitz and Continuous Functions

For real-valued Lipschitz functions $f: \mathbb{R}^n \to \mathbb{R}$, Rademacher's theorem asserts almost-everywhere differentiability and hence $S_f$ has Lebesgue measure zero. However, classification of possible singular sets for generic Lipschitz or continuous functions yields:

• Lipschitz functions: A closed purely unrectifiable set (one that intersects every $C^1$ curve in a set of zero length) can be realized as the singular set of some Lipschitz function (Dymond et al., 2019). Analytic sets covered by countably many closed purely unrectifiable sets are precisely those where a typical Lipschitz function is "severely non-differentiable" everywhere on the set.
• Typical continuous functions: By Banach-Mazur category methods, the exceptional set where a typical $f \in C([0,1])$ is not maximally nondifferentiable can be any residual family of $F_\sigma$ sets (most commonly Lebesgue null, Hausdorff dimension $\leq s$, etc.), but not countable or nowhere dense sets (Preiss et al., 2012).

4. Regularity and Rectifiability: Stratification and Dimension

Geometric and variational problems often stratify the singular set by the symmetry of tangent cones: for minimizers or stationary points $u$ of convex functionals,

$$S^k(u) = \{ x \in \operatorname{Sing}(u) : \text{no tangent map at } x \text{ is } k\text{-symmetric} \}$$

Each stratum is $k$-rectifiable and carries locally finite $\mathcal{H}^k$ measure. Monotonicity formulas and blowup analysis (frequency, Weiss, or Almgren functionals) underpin these results (Sinaei, 2017).

For the Stefan problem (a prototype free boundary PDE), the singular set in space-time has parabolic Hausdorff dimension $\leq n-1$, and outside an $(n-2)$-dimensional set, the solution admits a full $C^\infty$ expansion at singular points (Figalli et al., 2021).

5. Explicit Singular Sets for Model Functions

Nowhere-differentiable functions (e.g., Takagi, Lebesgue's singular function, singular Cantor functions) admit explicit characterizations for their singular (or exceptional) sets:

• For Takagi's function $T(x)$, the set where $T'(x)$ is infinite has Hausdorff dimension $1$ and can be characterized exactly in terms of the binary digit gaps of $x$ (Allaart et al., 2010).
• For Lebesgue's singular function $L_a(x)$, the set where $L'_a(x)=0$ has full measure and dimension $1$; the set where $L'_a(x)=+\infty$ is also of full (fractal) dimension in parameter regimes, depending on limiting binary digit densities (Kawamura, 2010).
• Singular Cantor-type continuous functions can have their sets of positive/finiteslope differentiability prescribed to any $F_\sigma$ null set, with maximal Hausdorff dimension in every interval (Kossaczka et al., 2021).

6. Singular Sets in Convex and PDE Problems

In variational convex settings, singular sets can exhibit rigidity. For solutions $u$ of Newton's least resistance problem,

• The set of extreme points of the epigraph is contained within the closure of the singular set (points of nondifferentiability).
• The singular set $\operatorname{Sing}(u)$ uniquely determines $u$ up to translation, by a convexity-rigidity principle (Plakhov, 2021).
• The fine local geometry is by convex dihedral cones or ridges, and the singular set typically forms a closed nowhere-dense union of Lipschitz curves.

Distance functions from $C^1$ hypersurfaces $\Omega$ admit a precise geometric criterion for the density of the singular set: if $\Omega$ fails the uniform inner ball condition everywhere, the singular set of nondifferentiability can be dense or even have nonempty interior. For Hamilton–Jacobi and eikonal equations, this allows viscosity solutions with exceptionally large singular sets (Santilli, 2020).

7. Connections, Methods, and Open Questions

Modern analysis of singular sets uses:

• Rectifiability techniques: L^2 $\beta$-number control, discrete and continuous Reifenberg theorems, frequency pinching, and quantitative stratification.
• Category methods: Banach–Mazur games, permutation-invariant sets, and zero-one laws for classification of exceptional sets in function spaces.
• Explicit constructions: Iterated function systems, Cantor-type constructions, and Cauchy-integral methods (for constructing entire functions with prescribed singular/nowhere-differentiable sets).

Open questions include sharp Hausdorff dimension bounds for singular sets in geometric PDEs, extensions to lower regularity boundaries, and a finer classification of singularities in convex and variational settings.

---

Key References:

• (Lellis et al., 2023) Rectifiability and dimension of singular sets in area-minimizing currents.
• (Sinaei, 2017) Stratification and rectifiability for minimizers and stationary points of convex functionals.
• (Dymond et al., 2019) Dichotomy of analytic sets as singular sets for Lipschitz functions.
• (Kawamura, 2010, Allaart et al., 2010, Kossaczka et al., 2021) Explicit singular sets for classic nowhere-differentiable and singular functions.
• (Santilli, 2020) Singular sets for distance functions and the eikonal equation.
• (Plakhov, 2021) Rigidity and determination of convex minimizers by their singular set.
• (Preiss et al., 2012) Baire category classification of singular sets for typical continuous functions.

These results provide foundational structure theorems and explicit constructions, establishing singular (nondifferentiability) sets as central objects in analysis, geometry, and partial differential equations.