Regularity Classes and Inextendibility

Whether a spacetime is extendible depends on how regular the larger metric must be. A metric may cross a boundary continuously while its connection, curvature, or Einstein equation fails there. Consequently, “inextendible” is not a complete claim until it names a class such as $C^0$, locally Lipschitz, $C^2$, smooth, or analytic—and states whether matter fields and field equations must extend as well.

Required background. Geodesics and Jacobi Fields supplies the regularity used by curve equations; Curvature supplies the derivative hierarchy. Helpful background. Lorentzian Metrics and Causal Cones supplies nondegeneracy and causal structure; Causality and Global Structure orients determinism; Exact Solutions of Einstein’s Equations supplies horizon extensions.

The hierarchy of extension classes

A $C^k$ extension of $(M,g)$ is a proper isometric embedding into a larger manifold carrying a Lorentzian metric $g'$ of class $C^k$. If $k=0$, metric components are continuous in smooth charts and the signature remains Lorentzian. At $C^1$, first derivatives exist continuously; at $C^2$, the classical Riemann tensor is continuous. Smooth means $C^k$ for all finite $k$, while analytic is stronger still.

Because every smoother extension is also a less regular one,

$$\text{a }C^\infty\text{ extension} \Longrightarrow \text{a }C^2\text{ extension} \Longrightarrow \text{a }C^1\text{ extension} \Longrightarrow \text{a }C^0\text{ extension}.$$

Taking contrapositives gives the inextendibility chain in the other direction: $C^0$-inextendibility is the strongest of these conclusions. A proof of $C^2$-inextendibility does not rule out a continuous extension.

Other classes fill the gaps. A locally Lipschitz metric has essentially bounded first weak derivatives, giving a connection almost everywhere and sufficiently controlled causal cones for many results. A $C^{1,1}$ metric has locally Lipschitz first derivatives and bounded curvature almost everywhere. Sobolev classes can make the Einstein equation meaningful weakly without supplying classical curvature at each point. The exact threshold is tied to the question.

Which geometry survives each threshold

At $C^0$, lengths of locally Lipschitz causal curves and the light-cone relation remain meaningful. Classical Christoffel symbols need not exist, so the geodesic equation as an ordinary differential equation is unavailable in its usual form. One can study maximizing causal curves variationally, but uniqueness and standard exponential-map arguments can fail.

At $C^1$, Christoffel symbols are continuous. Geodesics exist, but the usual Lipschitz condition ensuring uniqueness need not hold. At $C^{1,1}$, the connection is locally Lipschitz and classical geodesic existence and uniqueness are recovered; curvature exists almost everywhere and is locally bounded. At $C^2$, the classical curvature tensor is continuous, making pointwise curvature convergence conditions and standard Jacobi equations available.

These thresholds explain why a singularity theorem proved for $C^2$ metrics does not automatically apply to a merely continuous extension. The theorem concludes an incomplete geodesic inside the original smooth region. Whether a lower-regularity metric can continue causal curves beyond its endpoint is a separate extension problem. Steinbauer surveys both the classical architecture and carefully formulated low-regularity replacements Steinbauer 2023, §§2.3 and 4, pp. 8–9, 21–40.

Curvature blowup and extension obstructions

Suppose a scalar polynomial curvature invariant diverges along a curve approaching a boundary. If a $C^2$ extension existed and the curve approached an interior point of it with controlled frame, the continuous extended curvature would remain bounded in a compact neighborhood. The divergence therefore obstructs such an extension. It does not necessarily obstruct a $C^0$ extension, because a continuous metric has no classical curvature to continue.

Parallelly propagated curvature blowup can be stronger than scalar tests. A frame component of the Riemann tensor may diverge while all scalar polynomial invariants remain finite. Under suitable convergence of the curve and frame, this still obstructs a sufficiently regular extension. But one must state the frame and topology: divergent coordinate components can be artifacts.

Conversely, bounded curvature does not guarantee extendibility. Topology, causal structure, holonomy, or failure of the metric to approach a nondegenerate limit can obstruct an extension. The flat punctured plane is extendible; a flat cone with incompatible apex angle may not admit a smooth regular filling in the same category. Curvature provides useful obstructions, not a complete if-and-only-if test.

Worked example: continuous metric and divergent curvature

A two-dimensional example makes the derivative hierarchy explicit. On $t\gt0$, take

$$ds^2=-dt^2+a(t)^2dx^2, \qquad a(t)=1+\left(\frac{t}{T}\right)^{3/2}, \qquad T\gt0.$$

The curves $x=\text{constant}$ are unit-speed timelike geodesics and reach $t=0$ in finite proper time. In the site’s curvature convention,

$$R=\frac{2\ddot a}{a} =\frac{3}{2T^2} \frac{(t/T)^{-1/2}}{1+(t/T)^{3/2}},$$

so $R$ diverges there. Nevertheless, setting $a(t)=1+(\lvert t\rvert/T)^{3/2}$ for all $t\in\mathbb R$ gives a nondegenerate $C^1$ Lorentzian extension: $a$ and $\dot a$ are continuous at zero, and $a(0)=1$. Its second derivative diverges. The supplied extension therefore continues the metric and cones while failing to continue classical curvature. A $C^2$ extension attaching an endpoint to the same approach would contradict the scalar divergence.

This is a geometric example, with no claim that the extension solves a prescribed Einstein–matter system. It shows concretely why a curvature singularity and continuous metric extendibility can coexist. It does not establish uniqueness of geodesics for arbitrary $C^1$ metrics or of any weak field evolution.

Weak null singularities and determinism

Black-hole interiors motivate the distinction sharply. At a Cauchy horizon, metric components may remain continuous while transverse derivatives or curvature grow without bound. Such a weak null singularity can be $C^0$-extendible but fail to admit a $C^2$ extension. An observer might cross a continuous metric boundary along a causal curve, yet the classical Einstein equation and deterministic evolution of differentiable data may not continue in the usual sense.

This is why formulations of strong cosmic censorship differ. A $C^0$ version asks whether generic maximal globally hyperbolic developments have any continuous Lorentzian extension. A $C^2$ version asks whether classical curvature can continue. A Christodoulou-type formulation can demand sufficient connection regularity for the Einstein equations as weak equations. Evidence for or against one version is not a verdict on the others.

Matter regularity must be matched as well. A continuous metric extension accompanied by a distributional stress tensor may not belong to the original Einstein–matter model. Conversely, a weak solution framework may legitimately admit such fields. The category includes the equations, function spaces, gauge, and notion of solution—not merely a superscript on $g$.

Analyticity is not physical maximality

Analytic extension has special rigidity: analytic functions are determined by local series where continuation exists. But generic smooth initial data are not analytic, and a smooth spacetime can fail to possess an analytic atlas compatible with a requested continuation. “Maximal analytic extension” is therefore a useful classification for exact solutions, not the universal endpoint of general-relativistic evolution.

Kruskal’s extension of the Schwarzschild exterior is analytic and removes the static-coordinate horizon. The same fact can be expressed smoothly, and the physical distinction between eternal and collapse spacetimes is not decided by analyticity. At the central boundary, curvature blowup obstructs $C^2$ extension; stronger statements about $C^0$ require separate arguments.

Exercise

Assume a spacetime has a proper $C^0$ extension but no $C^2$ extension. Which of the following statements follow: (a) it is smoothly inextendible; (b) it is $C^0$-inextendible; (c) classical curvature must remain bounded at the boundary; (d) causal cones can extend continuously?

Solution

Statement (a) follows: a smooth extension would also be $C^2$, contradicting the assumption. Statement (b) is false because a $C^0$ extension is given. Statement (c) does not follow; indeed curvature divergence is one possible obstruction to $C^2$ extension, and classical curvature need not even be defined on the continuous side. Statement (d) holds for the supplied nondegenerate continuous Lorentzian metric: its cone field extends continuously, although classical geodesic uniqueness and Einstein evolution need not.

Curvature Blowup and Measures of Singularity Strength next separates the different curvature and tidal criteria that an extension boundary may satisfy.

References

• Isenberg, James, “On Strong Cosmic Censorship”, Surveys in Differential Geometry 20, §§1–4, pp. 17–36, 2015.
• Steinbauer, Roland, “The Singularity Theorems of General Relativity and Their Low Regularity Extensions”, Jahresbericht der Deutschen Mathematiker-Vereinigung 125, §§2.3 and 4, pp. 8–9 and 21–40, 2023.