What Asymptotic Flatness Must Specify

Saying that a metric “approaches Minkowski space” is not yet a definition of asymptotic flatness. One must say which infinity is approached, in what geometry or coordinates, with what topology and differentiability, how rapidly every relevant field and derivative decays, which transformations preserve the boundary conditions, and whether the limiting boundary is complete. Those choices determine whether energy, angular momentum, radiation, or scattering data are even well defined. This page gives a specification checklist for four-dimensional general relativity; it does not assert that one checklist is optimal for every initial-value, radiative, or global problem.

Required background. The Riemann Curvature Tensor supplies the curvature and conformal-curvature language used to distinguish genuine decay from a coordinate effect.

Helpful background. Pseudo-Riemannian Metrics and Signature fixes the Lorentzian conventions; Affine and Variational Geodesics explains geodesic endpoints and completeness; Causality and Global Structure provides conformal boundaries and domains of dependence; Formulations of General Relativity supplies Cauchy and characteristic data; and Exact Solutions of Einstein’s Equations provides benchmark solutions.

Three inequivalent approaches to infinity

An isolated system has several limiting regimes, and a condition natural in one need not control another.

• Spatial infinity is approached along spacelike directions, for example by taking $r\to\infty$ on an asymptotically Euclidean Cauchy hypersurface. ADM energy and momentum live here. A standard end has topology $\mathbb R^3\setminus B$, but multiple ends, asymptotically locally Euclidean quotients, or other topologies must be declared rather than hidden in the word “flat.”
• Future or past null infinity, denoted $\mathscr I^+$ or $\mathscr I^-$, is approached along outgoing or incoming null directions with retarded or advanced time held finite. Bondi news, radiative flux, and the Bondi–Metzner–Sachs symmetry group live here. In the usual four-dimensional isolated-system framework, each component has topology $S^2\times\mathbb R$.
• Timelike infinity, $i^+$ or $i^-$, describes late or early limits along timelike trajectories. Massive fields and scattering states can carry information there even when the geometry near $\mathscr I$ is controlled.

The same radial symbol does not identify these limits. In a conformal completion with physical metric $\widehat g_{ab}$ and unphysical metric

$$g_{ab}=\Omega^2\widehat g_{ab},$$

one normally has $\Omega\sim r^{-1}$ when approaching null infinity but $\Omega\sim r^{-2}$ when approaching the point $i^0$ along spacelike directions. Correspondingly, $\nabla_a\Omega$ is nonzero on a regular $\mathscr I$, whereas at $i^0$ one imposes $\Omega=0$, $\nabla_a\Omega=0$, and a normalized nondegenerate Hessian. These different orders are explicit in the conformal definitions of null and spatial infinity in Ashtekar and Khera 2024, §§2.1 and 2.3, pp. 4–12.

Thus a statement such as $g_{\mu\nu}=\eta_{\mu\nu}+O(r^{-1})$ is ambiguous until the limiting direction is fixed. It may describe a spatial end in asymptotically Cartesian coordinates, but it is not by itself a definition of a smooth null boundary and says nothing about $i^+$.

Coordinate falloff and conformal regularity

Two common languages encode complementary information.

On a spacelike slice, let $u^a$ be the future-directed unit normal, with $u^a u_a=-1$. We use

$$K_{ij}=-h_i{}^a h_j{}^b\nabla_a u_b =-\frac12\mathcal L_u h_{ij}.$$

In a coordinate formulation on a spatial end, one might require induced data

$$h_{ij}=\delta_{ij}+O_2(r^{-1}), \qquad K_{ij}=O_1(r^{-2}).$$

Here $f=O_k(r^{-p})$ means $\partial^\alpha f=O(r^{-p-\lvert\alpha\rvert})$ for every multi-index $\lvert\alpha\rvert\leq k$ in the stated asymptotically Cartesian chart. The subscript matters: curvature and surface charges contain derivatives, so a zeroth-order estimate need not control them. One must also specify lapse, shift, matter decay, constraint falloff, and the class of allowed coordinate changes. The familiar leading falloffs alone can leave logarithmic divergences in the symplectic form or ambiguities in Lorentz generators; parity or related boundary conditions cure particular problems but select different asymptotic symmetry content Henneaux and Troessaert 2019, §§III–VI, pp. 5–24.

In a conformal formulation, the boundary is attached geometrically. The data include a manifold-with-boundary $M$, a defining function $\Omega$, the regularity class of $g_{ab}$, the topology and causal character of the boundary, and field-equation decay. This makes null infinity and its generators visible without choosing a Bondi radial coordinate. It does not eliminate choices: $\Omega$ has conformal freedom $\Omega' = \omega\Omega$, and finite differentiability, polyhomogeneous logarithms, or direction-dependent limits at $i^0$ lead to genuinely different categories.

Neither language automatically implies the other at the advertised strength. Coordinate falloffs may not produce a smooth conformal boundary. Conversely, a conformal definition near one boundary component need not provide the precise weighted Sobolev estimates needed for a Cauchy evolution theorem.

A complete asymptotic specification

A usable definition should answer all of the following questions.

Item	What must be fixed	What can fail if it is omitted
End and topology	Spatial, null, or timelike regime; number of ends; topology of cuts or generators	Charges may refer to different boundaries, and angular modes may be misidentified
Limiting structure	An asymptotic chart, conformal completion, hyperboloidal blowup, or weighted function space	“Small” perturbations can change under an unrestricted coordinate transformation
Rate and derivatives	Orders for metric, inverse metric, connection, curvature, matter, and enough derivatives	Curvature, symplectic forms, or surface integrals may not converge
Regularity type	Smooth, finite $C^k$, Sobolev, conormal, or polyhomogeneous	Full peeling and Taylor expansions may be claimed without their hypotheses
Equations and matter	Vacuum near infinity or an explicit decay law for $T_{ab}$ and constraints	The boundary can cease to be null, or matter flux can alter balance laws
Gauge and symmetry	Transformations preserving all conditions and transformations regarded as pure gauge	Physical large symmetries can be discarded, or coordinate artifacts counted as charges
Parity and corner data	Antipodal/parity conditions and compatibility where spatial and null infinity meet	Angular momentum or matching across $i^0$ can remain ambiguous
Completeness	Completeness of $\mathscr I$ generators, endpoints of physical null geodesics, or Cauchy completeness—stated separately	A truncated boundary can masquerade as an isolated system

This list is intentionally more detailed than a single decay equation. Penrose’s asymptotic-simplicity framework, for example, includes endpoints on the conformal boundary for every inextendible physical null geodesic; a merely local completion near $\mathscr I$ does not. The definition and its relation to asymptotic flatness are stated in Frauendiener 2004, §2.3, pp. 15–18. Likewise, completeness of the vector field generating $\mathscr I^+$ is an additional condition in a frequently used null-infinity definition, not a consequence of the topology $S^2\times\mathbb R$ Ashtekar and Khera 2024, §2.1, pp. 4–6.

What a definition licenses

The desired observable should determine the boundary conditions, not the other way around.

For ADM four-momentum, one needs a controlled spatial end and a transformation class under which the surface integrals transform properly. For Bondi mass loss, one needs sufficient regularity at null infinity, a conformal frame or an invariant replacement, field equations, and integrable news. For an $S$-matrix-style conservation law linking $\mathscr I^-$ to $\mathscr I^+$, one additionally needs matching through spatial infinity and assumptions excluding uncontrolled leakage or singular behavior at the corner. For classical peeling, one needs more differentiability than is required merely to define leading radiative curvature; polyhomogeneous null infinity separates those statements.

Minkowski spacetime satisfies all standard smooth versions, but it is too special to test their distinctions. Schwarzschild and Kerr have acceptable asymptotically flat exterior regions and spatial charges, yet their black-hole interiors prevent full asymptotic simplicity. Radiating solutions may possess a useful $\mathscr I$ with weaker regularity than a smooth Taylor expansion. Asymptotically de Sitter and anti-de Sitter spacetimes can approach highly symmetric metrics, but their conformal boundaries have different causal character and are not asymptotically flat in this sense.

The phrase “asymptotically flat” should therefore be followed, at least implicitly, by a contract: flat where, in which category, to what order, modulo which transformations, and complete in which sense? The next pages state separate contracts for null infinity and spatial infinity.

Common pitfalls

A component estimate is not geometric by itself. A coordinate transformation can create or remove $1/r$ terms while preserving the same spacetime. Restrict the allowed asymptotic charts or replace component statements with a conformal or weighted geometric definition.

Curvature decay does not fix global topology. A locally flat quotient or a manifold with several asymptotically Euclidean ends can have small curvature at each end. State the topology and count each connected end separately.

Completeness has several meanings. Completeness of the generators of $\mathscr I^+$, endpoint completeness of all physical null geodesics, and geodesic completeness of the physical spacetime are not interchangeable. Name the curve family, metric, parameter, and direction of completeness.

Exercises

Two papers both write $g_{\mu\nu}=\eta_{\mu\nu}+O(r^{-1})$. Paper A means that only the ten component differences are bounded by $C/r$ in one unspecified chart. Paper B fixes an end diffeomorphic to $\mathbb R^3\setminus B$, an asymptotically Cartesian chart class, $O_2(r^{-1})$ decay for the induced metric, $O_1(r^{-2})$ decay for extrinsic curvature, vacuum constraints outside a compact set, and Regge–Teitelboim parity. Which paper has stated enough to investigate ADM and Lorentz charges, and what does neither paper establish?

Solution

Paper B has supplied the ingredients needed to begin an ADM Hamiltonian analysis: topology, derivative-controlled falloffs, equations, allowed chart behavior, and parity data. Paper A has not controlled derivatives or coordinate freedom, so even convergence and invariance of a surface integral are undecided. Neither statement establishes a smooth or complete $\mathscr I^+$, Bondi news, peeling, endpoints for all null geodesics, or matching between $\mathscr I^-$ and $\mathscr I^+$; those are null or global conditions not implied by spatial $1/r$ falloff.

References

• Ashtekar, Abhay, and Neev Khera. “Unified Treatment of Null and Spatial Infinity III: Asymptotically Minkowski Space-times.” Journal of High Energy Physics 2024, no. 2 (2024): 210. DOI.
• Frauendiener, Jörg. “Conformal Infinity.” Living Reviews in Relativity 7 (2004): 1. DOI.
• Henneaux, Marc, and Cédric Troessaert. “The Asymptotic Structure of Gravity at Spatial Infinity in Four Spacetime Dimensions.” arXiv:1904.04495 [hep-th] (2019). arXiv.