Null infinity, the BMS group, and the Bondi-Sachs mass-loss formula

Intuition Beginner

Throw a stone into a still pond and ripples spread outward, carrying energy away from the splash. Long after the stone has sunk, a cork floating far out on the water bobs once as a ripple passes, then settles. The cork never touched the stone; all it ever felt was the wave that travelled across the whole pond to reach it.

Gravitational waves are the spacetime version of those ripples. A pair of orbiting stars, or two black holes spiralling together, shakes the geometry around them, and the shaking races outward at the speed of light. To study what finally arrives, we want to stand infinitely far away and watch the waves wash past. But "infinitely far away" is an awkward place to stand. The clever trick is to rescale distances so that the faraway edge of spacetime is pulled in to a finite boundary we can actually draw and study.

That ideal far edge, reached by following light rays outward forever, has a name: null infinity. The waves that reach it carry a precise record of the radiation, called the news. And there is a running tally of how much mass-energy the system still has. Every burst of news drains a little of that mass away, and it never flows back.

Visual Beginner

Alt text: At the centre a vibrating source emits a family of outgoing rings, the gravitational ripples. The rings travel rightward and upward along light-ray directions and pile up against a diagonal boundary line, the rescaled edge of spacetime called future null infinity, drawn at a finite distance because all the faraway geometry has been shrunk inward. A small dial beside the source is labelled Bondi mass; as each ring detaches and heads for the boundary, the dial's needle ticks down a notch and never rises. The picture shows two ideas at once: the distant edge where radiation is read off, and the steady one-way drain of the source's mass as waves leave.

Worked example Beginner

Picture a single short burst of gravitational waves leaving a binary system, like one clap of a pair of hands. Before the clap, the system has some mass-energy; call the reading on the Bondi dial $10$ units. The burst carries away energy, and once it has fully departed the dial reads $9.7$ units. The missing $0.3$ units did not vanish: it rode out with the wave toward the distant edge.

Now follow that burst outward along the light rays it travels on. As it climbs away, its strength fades in a very orderly way — the strongest, most wave-like part falls off slowest, like the brightest part of a ringing bell that you can still hear from across a field. By the time it reaches the rescaled edge, the burst has thinned to a clean signal that an observer parked at the edge records as a single packet of news.

What this tells us: the news at the edge and the drop in the dial are two faces of one event. Read the news and you know exactly how much the dial fell. No news means the dial holds steady; loud news means a big drop. And crucially the dial only ever falls, because energy that has left on an outgoing wave cannot come back to a system that is shrinking into its own past.

Check your understanding Beginner

Formal definition Intermediate+

Let $(M,g)$ be a smooth spacetime with signature $(-,+,+,+)$. Following Penrose, we encode "asymptotically flat at large null separations" by attaching a conformal boundary. An unphysical (or conformal) spacetime $(\hat{M},\hat{g})$ with $\hat{g}={\Omega }^{2}g$ is a conformal completion of $(M,g)$ if: $M$ is the interior of $\hat{M}$, the boundary-defining function $\Omega \ge 0$ is smooth on $\hat{M}$ with $\Omega =0$ on $\partial M=:\mathcal{I}$ and $d\Omega \ne 0$ there, and $\hat{g}$ extends smoothly across $\mathcal{I}$. The boundary $$ \mathscr I = {, p \in \hat M : \Omega(p) = 0,\ d\Omega|_p \neq 0 ,} $$ is null infinity ("scri"), split into the future part ${\mathcal{I}}^{+}$ reached by outgoing null geodesics and the past part ${\mathcal{I}}^{-}$ reached by incoming ones [03.09.17 contrasts the spatial-infinity completion used for the ADM mass]. The condition $d\Omega \ne 0$ makes $\mathcal{I}$ a genuine hypersurface; the physical metric is recovered as $g={\Omega }^{-2}\hat{g}$, blowing up at the boundary exactly so that the finite $\hat{g}$-distance to $\mathcal{I}$ is infinite $g$-distance.

Definition (asymptotic simplicity). A spacetime is asymptotically simple if it admits such a completion in which every inextendible null geodesic of $M$ has past and future endpoints on $\mathcal{I}$, and asymptotically flat if in addition the Ricci tensor vanishes in a neighbourhood of $\mathcal{I}$ (a vacuum exterior). The defining computation is the next theorem: for an asymptotically flat vacuum spacetime, $\mathcal{I}$ is a null hypersurface, and $\hat{g}(\nabla \Omega ,\nabla \Omega )=0$ on $\mathcal{I}$. The Penrose-diagram pictures of 13.06.01 and the conformal-cone invariance of 13.02.01 are the qualitative shadow of this statement; here it becomes a metric fact.

Bondi-Sachs coordinates. Near ${\mathcal{I}}^{+}$ one introduces retarded coordinates $(u,r,x^A)$, with $u$ a retarded time labelling the outgoing null cones $\{u=\text{const}\}$, $r$ an areal radius along the generators, and $x^A=(\theta ,\varphi )$ coordinates on the cut $S^2$. The physical metric admits the asymptotic expansion (Bondi-Sachs form) $$ g = -\Big(1 - \frac{2 m_B(u, x^A)}{r}\Big),du^2 - 2,du,dr + r^2, h_{AB},dx^A dx^B + \cdots, $$ where $h_{AB}$ is the round 2-sphere metric and the dots are subleading shear, angular-momentum, and gauge terms. The function $m_B(u,x^A)$ is the Bondi mass aspect.

Definition (the news function). Let $(\ell ,n,m,\bar{m})$ be a Newman-Penrose null tetrad adapted to the outgoing congruence at ${\mathcal{I}}^{+}$, with $\sigma$ the complex shear of that congruence [03.02.18 supplies the NP formalism and the peeling falloff ${\Psi }_n\sim {\Psi }_n^{(5-n)}/r^{5-n}$]. The leading asymptotic shear ${\sigma }^0(u,x^A)$ is the $O(1/r^2)$ coefficient of $\sigma$ in the $1/r$ expansion. The Bondi news function is its retarded-time derivative, $$ N(u, x^A) ;=; \dot\sigma^0(u, x^A), \qquad \dot{} = \frac{\partial}{\partial u}. $$ The news is the genuine radiative datum at ${\mathcal{I}}^{+}$: it is conformally weighted of spin $2$ on the cut, and it is gauge-invariant under the supertranslations defined below, whereas ${\sigma }^0$ alone is not.

Definition (the BMS group). The asymptotic symmetry group is the group of conformal diffeomorphisms of $({\mathcal{I}}^{+},q_{AB},{\ell }^a)$ preserving the universal structure of ${\mathcal{I}}^{+}$ — the degenerate metric $q_{AB}$ (the round $S^2$, defined up to conformal rescaling) together with the null generator direction ${\ell }^a$ along $u$. This is the Bondi-Metzner-Sachs group $$ \mathrm{BMS} ;=; \mathrm{SO}^+(3,1) \ltimes \mathcal S, \qquad \mathcal S \cong C^\infty(S^2), $$ the semidirect product of the (restricted) Lorentz group, acting as the conformal group of the celestial $S^2$, with the abelian group of supertranslations $u\mapsto u+f(x^A)$ for an arbitrary smooth $f$ on the sphere. The ordinary translations sit inside $\mathcal{S}$ as the four lowest spherical harmonics ($\ell =0,1$), but there is no canonical Poincaré subgroup: the normal subgroup of translations is the unique four-dimensional one only after the supertranslation freedom is fixed.

Key theorem with proof Intermediate+

Theorem (Bondi-Sachs mass-loss). Let $(M,g)$ be an asymptotically flat vacuum spacetime with future null infinity ${\mathcal{I}}^{+}$ and a Bondi cut foliation by retarded time $u$. Define the Bondi mass at retarded time $u$ as the integral of the mass aspect over the cut, $$ m_B(u) ;=; \frac{1}{4\pi}\oint_{S^2} m_B(u, x^A), dS . $$ Then $m_B(u)$ is non-increasing, and its rate of change is $$ \frac{d m_B}{du} ;=; -\frac{1}{4\pi}\oint_{S^2} |N(u, x^A)|^2 , dS ;\le; 0, $$ where $N={\dot{\sigma }}^0$ is the news function. Equality holds at a given $u$ iff the news vanishes everywhere on that cut: a radiating isolated system strictly loses Bondi mass, and a non-radiating one conserves it.

Proof. Work in Bondi-Sachs coordinates and use the Newman-Penrose hierarchy at ${\mathcal{I}}^{+}$. The peeling theorem 03.02.18 gives the falloff ${\Psi }_n\sim {\Psi }_n^0/r^{5-n}$ along the outgoing generators, isolating the coefficients ${\Psi }_2^0,{\Psi }_3^0,{\Psi }_4^0,{\sigma }^0$ as the asymptotic data. Two of the asymptotic Bianchi/NP equations at the boundary are the relevant ones.

Step 1: the mass aspect is the real part of ${\Psi }_2^0$. The Bondi mass aspect is fixed by the leading coefficient of the "Coulombic" Weyl scalar through $$ m_B(u, x^A) ;=; -,\mathrm{Re},\Big(\Psi_2^0 + \sigma^0,\dot{\bar\sigma}^{,0}\Big), $$ the second term being the shear correction that makes $m_B$ the correctly supertranslation-covariant quantity. Integrating over the cut and using that the angular ($ð$-derivative) pieces are total divergences on $S^2$ leaves $$ m_B(u) = -\frac{1}{4\pi}\oint_{S^2}\mathrm{Re},\Psi_2^0, dS . $$

Step 2: the asymptotic Bianchi identity for ${\dot{\Psi }}_2^0$. The Newman-Penrose contracted Bianchi identity, expanded to the leading order at ${\mathcal{I}}^{+}$, reads $$ \dot\Psi_2^0 ;=; \eth^2 \dot{\bar\sigma}^{,0} ;+; \sigma^0,\ddot{\bar\sigma}^{,0} ;-; \Psi_3^0\text{-divergence}, $$ where $ð$ is the spin-weighted derivative on the cut and the last term integrates to zero over the closed $S^2$. Taking the real part and integrating over the sphere, the ${ð}^2$ term is a total angular derivative and drops out, leaving the energy flux entirely in the shear terms.

Step 3: assemble the flux. Differentiate the Bondi mass and substitute Steps 1-2: $$ \dot m_B = -\frac{1}{4\pi}\oint_{S^2}\mathrm{Re},\dot\Psi_2^0, dS = -\frac{1}{4\pi}\oint_{S^2}\mathrm{Re}\big(\sigma^0\ddot{\bar\sigma}^{,0} + |\dot\sigma^0|^2 - \tfrac{d}{du}(\sigma^0\dot{\bar\sigma}^{,0})\big), dS . $$ The exact-derivative piece $\frac{d}{du}({\sigma }^0{\dot{\bar{\sigma }}}^0)$ is absorbed into the shear correction of $m_B$ from Step 1 (this is exactly why that correction is there), and what remains is the manifestly non-positive $$ \frac{d m_B}{du} = -\frac{1}{4\pi}\oint_{S^2} |\dot\sigma^0|^2, dS = -\frac{1}{4\pi}\oint_{S^2} |N|^2, dS \le 0 . $$ The integrand is a non-negative real number on the cut, so the bound is automatic, with equality iff $N={\dot{\sigma }}^0\equiv 0$ on the cut. $■$

The formula realises a physical slogan as a theorem: gravitational radiation is the news, and a radiating isolated system loses Bondi mass monotonically, at a rate set by the integrated square of the news. The quantity $-{\dot{m}}_B$ is the total gravitational-wave luminosity, and in the weak-field limit it reduces to the linearized-GR luminosity built from the third time-derivative of the quadrupole moment 13.07.01.

Bridge. This result builds toward the global picture of an isolated gravitating system as an object whose total energy is read at a boundary rather than a point: the foundational reason the Bondi mass is well defined is that conformal completion turns "infinitely far away" into a finite null hypersurface where flux integrals converge. The peeling falloff that appears again in the proof is exactly the same $1/r$ expansion proved in 03.02.18; this is exactly the asymptotic data that the news repackages. Putting these together, the central insight is that the supertranslation freedom of the BMS group is dual to the freedom in choosing the cut at which mass is evaluated, so the news — which is supertranslation-invariant — is the only honest radiative observable, and it generalises the flat-space notion of outgoing energy flux to the full nonlinear theory.

Exercises Intermediate+

Advanced results Master

Geroch's intrinsic characterisation of $\mathcal{I}$. Rather than carry $\Omega$ around, Geroch isolated the universal structure that every asymptotically flat ${\mathcal{I}}^{+}$ shares: a smooth 3-manifold $\mathbb{R}\times S^2$ carrying a degenerate metric $q_{ab}$ of signature $(0,+,+)$ whose null direction is the generator ${\ell }^a$, with $q_{ab}$ and ${\ell }^a$ defined only up to the conformal rescaling $q_{ab}\mapsto {\omega }^2{q}_{ab}$, ${\ell }^a\mapsto {\omega }^{-1}{\ell }^a$ inherited from the freedom in $\Omega$ ^{[Geroch 1977]}. The BMS group is then defined as the automorphisms of this universal structure, which is why it is spacetime-independent: any two asymptotically flat spacetimes share the same $\mathcal{I}$-symmetry group. This is the foundational reason the supertranslations are unavoidable: the conformal rescaling freedom $\omega (x^A)$ on the sphere is exactly what generates the $u\mapsto u+f$ shifts.

The news as curvature: $N$ and ${\Psi }_4^0$. The peeling hierarchy gives ${\Psi }_4\sim {\Psi }_4^0/r$ as the dominant "purely radiative" Weyl component, and the asymptotic NP equations identify ${\Psi }_4^0=-{\ddot{\bar{\sigma }}}^0=-\dot{\bar{N}}$. Thus the news is one time-integral away from the leading radiative curvature: $N$ is the strain-rate at ${\mathcal{I}}^{+}$ and ${\Psi }_4^0$ its time derivative. In the gravitational-wave detector language of 13.07.01, ${\sigma }^0$ is the waveform $h$, the news $N={\dot{\sigma }}^0$ is $\dot{h}$, and ${\Psi }_4^0\sim \ddot{h}$ is what an interferometer's curvature coupling responds to. The mass-loss integrand $|N{|}^2=|\dot{h}{|}^2$ is exactly the energy-flux density of the linearized theory, now derived nonlinearly.

Memory and the supertranslation field. A burst of news permanently changes ${\sigma }^0$ between early and late cuts, $\Delta {\sigma }^0=\int Ndu$, and this jump is a pure supertranslation ${ð}^{2}f$ precisely when the burst is "ordinary," shifting the relative clock readings of inertial detectors — the gravitational-wave memory effect. Memory is therefore the statement that the early and late vacua at ${\mathcal{I}}^{+}$ are related by a BMS supertranslation rather than being identical, which ties the asymptotic symmetry group to a physically measurable permanent displacement. This is the bridge from the kinematics of BMS to an observable, and it generalises the flat notion of a soft graviton to the nonlinear boundary data.

Positivity and the spatial-to-null contrast. The Bondi mass is non-negative for physically reasonable matter (the Bondi-mass positivity theorem, proved by Ludvigsen-Vickers, Horowitz-Perry, and Schoen-Yau by adapting the spinorial Witten argument), exactly paralleling the ADM positivity of 03.09.17 but now at retarded rather than spatial infinity. The two masses bracket the radiated energy: the ADM mass is the total energy on a Cauchy slice reaching spatial infinity $i^0$, the Bondi mass is the energy remaining after radiation up to retarded time $u$, and the difference $M_{\mathrm{ADM}}-m_B(u)={\int }_{-\infty }^u\frac{1}{4\pi }\oint |N{|}^{2}dSd{u}^{\prime }$ is the total energy radiated to null infinity up to that time, a manifestly non-negative quantity.

Synthesis. The asymptotic-structure programme stitches together several threads of the corpus into one nonlinear statement. It is the foundational reason that an isolated gravitating system has a well-defined, monotonically decreasing energy: conformal completion turns the unreachable far zone into a finite null boundary, and on that boundary the peeling expansion of 03.02.18 supplies exactly the data — the mass aspect ${\Psi }_2^0$ and the shear ${\sigma }^0$ — that the mass-loss law consumes. This is exactly the nonlinear completion of the linearized luminosity of 13.07.01: the linearized quadrupole flux is the weak-field shadow of $\oint |N{|}^2$. The construction is dual to the ADM story of 03.09.17 — the central insight being that energy can be read at spatial infinity (ADM, a single number on a Cauchy slice) or at null infinity (Bondi, a function of retarded time), and putting these together their difference is the radiated energy. And the whole edifice generalises the flat-space Poincaré group to the infinite-dimensional BMS group, whose supertranslation enhancement is not a defect but the carrier of gravitational-wave memory, so that the symmetry group, the radiative data, and the energy balance are three views of the single object ${\mathcal{I}}^{+}$.

Full proof set Master

Proposition 1 ($\mathcal{I}$ is null for an asymptotically flat vacuum spacetime). Let $\hat{g}={\Omega }^{2}g$ be a conformal completion with $\mathrm{Ric}(g)=0$ near $\mathcal{I}$. Then $\hat{g}(\nabla \Omega ,\nabla \Omega )=0$ on $\mathcal{I}$, i.e. $\mathcal{I}$ is a null hypersurface.

Proof. The Ricci tensors of conformally related metrics in dimension $4$ are related by $$ R_{ab} = \hat R_{ab} + 2\Omega^{-1}\hat\nabla_a\hat\nabla_b\Omega + \hat g_{ab}\Big(\Omega^{-1}\hat\Box\Omega - 3\Omega^{-2}\hat g^{cd}\hat\nabla_c\Omega,\hat\nabla_d\Omega\Big). $$ Impose $R_{ab}=0$ and multiply through by ${\Omega }^2$. Every term then carries a non-negative power of $\Omega$ except the last, so taking the limit $\Omega \to 0$ on $\mathcal{I}$ and keeping the coefficient of ${\Omega }^0$ in the trace-free and trace parts forces $$ \hat g^{cd},\hat\nabla_c\Omega,\hat\nabla_d\Omega ;=; 0 \quad\text{on } \mathscr I . $$ Since ${\hat{\nabla }}_a\Omega =d\Omega \ne 0$ on $\mathcal{I}$ by hypothesis, the normal to $\mathcal{I}$ is a non-zero null vector, so $\mathcal{I}$ is a null hypersurface. $■$

Proposition 2 (the news is supertranslation-invariant). Under a BMS supertranslation $u\mapsto u-f(x^A)$ with $f\in C^{\infty }(S^2)$, the asymptotic shear transforms as ${\sigma }^0\mapsto {\sigma }^0-{ð}^{2}f$, and the news $N={\dot{\sigma }}^0$ is unchanged.

Proof. A supertranslation relabels the retarded-time cuts by an angle-dependent constant shift, $u^{\prime }=u-f(x^A)$, with $f$ independent of $u$. The spin-2 shear acquires the inhomogeneous term $-{ð}^{2}f$ from the change of the null cuts (the second $ð$-derivative because ${\sigma }^0$ has spin weight $2$). Differentiating in retarded time, $N^{\prime }={\partial }_{u^{\prime }}({\sigma }^0-{ð}^{2}f)$; since ${\partial }_{u^{\prime }}={\partial }_u$ to leading order and ${\partial }_{u}f=0$, the inhomogeneous term drops and $N^{\prime }={\dot{\sigma }}^0=N$. Hence the news is a genuine, supertranslation-invariant field on ${\mathcal{I}}^{+}$, unlike ${\sigma }^0$ itself. $■$

Proposition 3 (the radiated energy is non-negative and bracketed by ADM and Bondi). For an asymptotically flat vacuum spacetime, the energy radiated between retarded times $u_1<u_2$ equals $m_B(u_1)-m_B(u_2)=\frac{1}{4\pi }{\int }_{u_1}^{u_2}{\oint }_{S^2}|N{|}^{2}dSdu\ge 0$, and in the limit $u_1\to -\infty$ the initial Bondi mass equals the ADM mass.

Proof. Integrate the mass-loss formula of the Key Theorem from $u_1$ to $u_2$: $$ m_B(u_2) - m_B(u_1) = \int_{u_1}^{u_2}\dot m_B, du = -\frac{1}{4\pi}\int_{u_1}^{u_2}\oint_{S^2}|N|^2, dS, du . $$ The right side is the negative of an integral of a pointwise non-negative quantity, so $m_B(u_2)\le m_B(u_1)$ and the radiated energy $m_B(u_1)-m_B(u_2)$ is non-negative. For the limit, the cut at $u\to -\infty$ along ${\mathcal{I}}^{+}$ approaches spatial infinity $i^0$; the standard matching of the Bondi and ADM expansions there (Ashtekar-Hansen, Wald §11.2 ^{[Wald 1984]}) identifies ${\lim }_{u\to -\infty }{m}_B(u)=M_{\mathrm{ADM}}$, the energy on a Cauchy slice. Hence $M_{\mathrm{ADM}}-m_B(u)=\frac{1}{4\pi }{\int }_{-\infty }^u\oint |N{|}^{2}dSd{u}^{\prime }\ge 0$ is the total energy radiated to null infinity up to retarded time $u$. $■$

The Bondi-mass positivity (that $m_B(u)\ge 0$ itself, not merely its monotonicity) is the deeper statement, proved spinorially in the manner of 03.09.17; we state it in the Advanced results and refer to Ludvigsen-Vickers and Horowitz-Perry ^{[Penrose-Rindler Vol. 2]} for the Nester-Witten 2-form argument at ${\mathcal{I}}^{+}$.

Connections Master

• Peeling, the Newman-Penrose scalars, and the Petrov classification 03.02.18. That unit proves the peeling falloff ${\Psi }_n\sim {\Psi }_n^{(5-n)}/r^{5-n}$ and defines the Weyl scalars on a null tetrad; this unit consumes exactly that asymptotic data, reading the Bondi mass off ${\Psi }_2^0$, the news off the shear ${\sigma }^0$, and the radiative curvature off ${\Psi }_4^0=-\dot{N}$. Peeling is the premise; the Bondi-Sachs construction is the conclusion built on it.

• Linearized GR and gravitational waves 13.07.01. The mass-loss law is the full nonlinear parent of the linearized quadrupole luminosity established there: ${\sigma }^0$ is the waveform $h$, the news $N=\dot{h}$, and $|N{|}^2$ integrated over the sphere is the energy-flux density whose weak-field limit is the quadrupole formula. The history paragraph of that unit names Bondi and the news; this unit supplies the definitions and the theorem it was pointing at.

• ADM mass and positive-mass 03.09.17. That unit gives the energy at spatial infinity via the Witten spinor argument; this unit gives the energy at null infinity. The two are bracketed: $M_{\mathrm{ADM}}$ is the total on a Cauchy slice, $m_B(u)$ is the residual after radiation up to retarded time $u$, and their difference is the radiated energy. The same Nester-Witten spinor technology proves positivity in both settings.

• Penrose diagrams and conformal infinity 13.06.01 and conformal structure on manifolds 13.02.01. Those units draw ${\mathcal{I}}^{\pm }$ as the null boundary of the Carter-Penrose diagram and establish the conformal invariance of the null cone that makes the rescaling $\hat{g}={\Omega }^{2}g$ admissible. This unit upgrades that picture to a theorem-level structure: the same ${\mathcal{I}}^{+}$ becomes the carrier of the BMS group, the news, and the mass-loss law.

Historical & philosophical context Master

The problem of defining gravitational radiation invariantly — distinguishing a genuine wave from a coordinate artefact, the issue that had stalled the subject since Einstein's own vacillation over whether waves were physical — was resolved in a remarkable burst between 1960 and 1965. Hermann Bondi, with M. G. J. van der Burg and A. W. K. Metzner, introduced retarded Bondi coordinates and the mass-loss formula, proving that an axisymmetric radiating system loses mass at a rate fixed by the square of the news function ^{[Bondi 1962]}. Rainer Sachs, removing the axisymmetry assumption, discovered that the asymptotic symmetry group is not the Poincaré group one would expect but its infinite-dimensional enlargement by supertranslations — the result that so surprised the community that the group now bears the names Bondi, Metzner, and Sachs ^{[Sachs 1962]}.

Roger Penrose then reorganised the entire subject geometrically. His 1965 paper recast asymptotic flatness as the existence of a smooth conformal boundary $\mathcal{I}$, replacing delicate $1/r$ expansions with the clean statement that the unphysical metric $\hat{g}={\Omega }^{2}g$ extends across $\Omega =0$; the peeling theorem, the null character of $\mathcal{I}$, and the conformal invariance of zero-rest-mass fields all became corollaries of this single move ^{[Penrose 1965]}. The mature synthesis is Chapter 9 of Penrose and Rindler's Spinors and Space-Time, Vol. 2, which develops the spinor formulation of $\mathcal{I}$, the BMS group, the news, and the Bondi-Sachs mass within the two-component spinor calculus ^{[Penrose-Rindler Vol. 2]}; Robert Geroch's 1977 lectures distilled the spacetime-independent universal structure of $\mathcal{I}$ from which the BMS group follows intrinsically ^{[Geroch 1977]}. Philosophically the episode is a paradigm of how the right boundary — infinitely far in physical distance, finite in the conformal completion — converts an ill-posed question about energy in general relativity into a sharp theorem, and how a symmetry group thought to be Poincaré turns out to be infinite-dimensional, a discovery whose implications for soft theorems and memory are still being mined.

Bibliography Master

@article{Penrose1965Asymptotic,
  author  = {Penrose, Roger},
  title   = {Zero rest-mass fields including gravitation: asymptotic behaviour},
  journal = {Proceedings of the Royal Society of London A},
  volume  = {284},
  pages   = {159--203},
  year    = {1965}
}

@article{BMS1962,
  author  = {Bondi, Hermann and van der Burg, M. G. J. and Metzner, A. W. K.},
  title   = {Gravitational waves in general relativity. VII. Waves from axi-symmetric isolated systems},
  journal = {Proceedings of the Royal Society of London A},
  volume  = {269},
  pages   = {21--52},
  year    = {1962}
}

@article{Sachs1962,
  author  = {Sachs, Rainer K.},
  title   = {Gravitational waves in general relativity. VIII. Waves in asymptotically flat space-time},
  journal = {Proceedings of the Royal Society of London A},
  volume  = {270},
  pages   = {103--126},
  year    = {1962}
}

@book{PenroseRindler1986,
  author    = {Penrose, Roger and Rindler, Wolfgang},
  title     = {Spinors and Space-Time, Volume 2: Spinor and Twistor Methods in Space-Time Geometry},
  publisher = {Cambridge University Press},
  year      = {1986},
  note      = {Chapter 9: conformal infinity, the BMS group, the news, peeling}
}

@incollection{Geroch1977,
  author    = {Geroch, Robert},
  title     = {Asymptotic structure of space-time},
  booktitle = {Asymptotic Structure of Space-Time},
  editor    = {Esposito, F. P. and Witten, L.},
  publisher = {Plenum Press},
  pages     = {1--105},
  year      = {1977}
}

@book{Wald1984,
  author    = {Wald, Robert M.},
  title     = {General Relativity},
  publisher = {University of Chicago Press},
  year      = {1984},
  note      = {§11.1--11.2: asymptotic flatness at null infinity, the Bondi mass}
}

@book{Stewart1991,
  author    = {Stewart, John},
  title     = {Advanced General Relativity},
  publisher = {Cambridge University Press},
  year      = {1991},
  note      = {Ch. 3: asymptopia, the Newman-Penrose formalism at scri}
}