Function Spaces for Einstein Evolution

An Einstein evolution theorem is incomplete until it specifies the topology in which data are controlled. Local Sobolev spaces govern derivatives and nonlinear products; weighted spaces encode asymptotic decay; Hölder spaces serve elliptic and geometric regularity; energy spaces follow propagation. No single norm automatically controls all four, and asymptotic charges require more than local regularity.

Required background. Analysis for Mathematical Relativity supplies Sobolev embedding and weak convergence, The Einstein Equation identifies the nonlinear system, PDEs and Green Functions for Gravitation supplies energy spaces, and Tensor Fields and Natural Operations supplies coordinate-independent tensor norms.

Helpful background. Causality and Global Structure supplies the geometric meaning of a Cauchy foliation.

Sobolev regularity for Cauchy data

On a three-dimensional slice $\Sigma$, the local model is

$$\|u\|_{H^s}^2 =\int_{\mathbb R^3}(1+|\xi|^2)^s|\widehat u(\xi)|^2\,d^3\xi,$$

where $\widehat u(\xi)=(2\pi)^{-3/2}\int e^{-ix\cdot\xi}u(x)\,d^3x$. Fix a reference length when adding $1$ to $|\xi|^2$; equivalently, use dimensionless chart coordinates. On a manifold the coordinate definitions are patched by a partition of unity, or integer-order norms are defined by covariant derivatives relative to a smooth reference metric. For metric initial data one classically takes

$$h_{ij}\in H^s_{\mathrm{loc}}, \qquad K_{ij}\in H^{s-1}_{\mathrm{loc}}, \qquad s>\frac52.$$

The one-derivative offset reflects that $K_{ij}$ is a normal derivative of the metric. In three dimensions, $H^s$ is an algebra for $s\gt3/2$, and $H^s\hookrightarrow C^1$ for $s\gt5/2$. These facts control products such as $g^{-1}\partial g\partial g$ and make the coefficients of the reduced wave operator continuously differentiable. Ringström develops Sobolev spaces and embedding in Ringström 2009, Chs. 5–6, pp. 33–56.

The threshold is formulation- and theorem-dependent. Modern low-regularity results can exploit null structure, curvature estimates, or geometric parametrices below the classical coordinate threshold. It is therefore inaccurate to call $s>5/2$ the sharp regularity of Einstein’s equation. It is the transparent threshold for the standard quasilinear metric argument used here.

Endpoint inequalities are another source of errors. The embeddings above are strict: $H^{5/2}(\mathbb R^3)$ does not generally embed in $C^1$, and $H^{3/2}$ is not generally an algebra. A proof that silently replaces $>$ with $\ge$ may lose precisely the coefficient or product bound needed for uniqueness.

Decay and weighted spaces

Local Sobolev regularity says nothing about spatial infinity. On an asymptotically Euclidean end, take a smooth reference metric $\bar h$ that is Euclidean sufficiently far out and write $\langle r\rangle=(1+r^2)^{1/2}$ in units of a fixed reference length. For an integer $m\ge0$, one explicit convention is

$$\|u\|_{H^m_\delta}^2 =\sum_{j=0}^{m} \int_\Sigma \langle r\rangle^{2(\delta+j)} |\bar D^j u|_{\bar h}^2\,\langle r\rangle^{-3}d\mu_{\bar h}.$$

Fractional orders require an interpolation or equivalent fractional-derivative definition; a sum over integer derivatives alone does not define a fractional $H^s$ norm. Weight conventions differ between authors, so a theorem must define its norm rather than quote only the symbol $H^s_\delta$.

To read this convention, test $u\sim r^{-p}$, smoothly cut off near the origin. Every derivative term has large-radius radial integrand proportional to $r^{2(\delta-p)-1}dr$. It is integrable precisely when $p\gt\delta$; $p=\delta$ is logarithmically divergent. Thus positive $\delta$ asks for decay stronger than $r^{-\delta}$ in this convention. A metric is measured through $h-\bar h$, not through the nondecaying Euclidean background itself.

For asymptotically flat data, one controls $h_{ij}-\delta_{ij}$ and $K_{ij}$ in compatible weighted spaces, often with parity or additional expansion conditions when a particular Hamiltonian charge is needed. Finite weighted norm alone does not guarantee that ADM energy, momentum, angular momentum, and center of mass all exist. Each boundary integral has its own decay and cancellation requirements. The constraint literature’s asymptotic classes are reviewed in Bartnik and Isenberg 2004, §§4.1–4.6, pp. 14–28.

Hölder and energy control

Hölder spaces $C^{k,\alpha}$ measure pointwise regularity and are natural for Schauder estimates in the elliptic constraint equations. Sobolev spaces are usually better suited to hyperbolic energy estimates because derivatives are controlled in $L^2$ and commute systematically with the equation. One commonly proves a solution in Sobolev spaces and then uses persistence of regularity: smoother data produce a smoother solution without changing the lifespan dictated by a lower controlling norm.

For a field $u$ on a foliation $\Sigma_t$, an integer-order energy, $m\ge1$, has the form

$$E_m(t) \simeq \sum_{|I|\le m-1} \int_{\Sigma_t} \left(|\nabla_nD^Iu|^2+|DD^Iu|_h^2+|D^Iu|^2\right)d\mu_h.$$

Uniform equivalence of this energy to a Sobolev norm requires lapse, induced metric, and inverse metric bounds. A continuation argument therefore controls geometry as well as field derivatives. If the slice degenerates or the injectivity radius collapses, coordinate Sobolev norms may cease to represent the same geometric regularity even while an algebraic expression remains finite.

The energy construction and its dependence on a uniformly positive time matrix are described in Friedrich and Rendall 2000, §3.1. This is why a function-space norm and a nondegeneracy condition belong together in the local theorem.

The correct choice of space follows the claim:

• local existence needs enough derivatives and product control;
• asymptotic flatness needs decay weights;
• elliptic constraint solving may need Sobolev or Hölder Fredholm theory;
• stability needs energies adapted to the background and expected decay;
• compactness needs a declared weak or strong topology.

These distinctions prevent a frequent category error: convergence weakly in $H^1$ does not imply convergence of quadratic curvature terms, and bounded local energy does not imply the existence of an asymptotic charge.

Check your understanding

Why does the standard metric argument use $s>5/2$ rather than merely the algebra threshold $s>3/2$?

Solution

The differentiated metric lies in $H^{s-1}$, so products of first metric derivatives are controlled by the algebra property at $s-1\gt3/2$, not merely by an algebra property for the undifferentiated metric. In addition, the first derivatives of the principal coefficients enter the energy estimate. Both requirements are met by $s\gt5/2$, which also gives $H^s\hookrightarrow C^1$. More refined arguments can lower the threshold by exploiting structure rather than these direct embeddings.

Continue with Geometric Uniqueness Modulo Diffeomorphism to see why analytic uniqueness must be translated back into a geometric statement.

References

• Bartnik, Robert, and Jim Isenberg. “The Constraint Equations.” In The Einstein Equations and the Large Scale Behavior of Gravitational Fields, 1–38. Birkhäuser, 2004. arXiv:gr-qc/0405092.
• Friedrich, Helmut, and Alan D. Rendall. “The Cauchy Problem for the Einstein Equations.” In Einstein’s Field Equations and Their Physical Implications, Lecture Notes in Physics 540, 127–224. Springer, 2000. arXiv:gr-qc/0002074.
• Ringström, Hans. The Cauchy Problem in General Relativity. ESI Lectures in Mathematics and Physics. European Mathematical Society, 2009. Publisher record.