Trapped Surfaces, Topology

of Black Holes, and the

Positive Mass Theorem

Lan-Hsuan Huang and Dan A. Lee

1. Introduction precisely? Do black holes really exist in nature? What do
Albert Einstein’s theory of general relativity, first published black holes have to do with mathematics?
in 1915, successfully unified special relativity with grav- These questions and others have been pondered by gen-
ity and led to many predictions that have since been veri- erations of researchers. The 2020 Nobel Prize in Physics
fied, marking one of the greatest triumphs of 20th century was awarded to mathematical physicist Roger Penrose for
physics. Perhaps one of the most sensational features of his “discovery that black hole formation is a robust pre-
the theory is the concept of a black hole—a region from diction of the general theory of relativity,” and to astro-
which even light cannot escape. But what is a black hole, physicists Reinhard Genzel and Andrea Ghez for their “dis-
covery of a supermassive compact object at the centre of
Lan-Hsuan Huang is a professor of mathematics at the University of Connecti- our galaxy.” Although the very first nontrivial solution to
cut. Part of the work is supported by NSF CAREER DMS-1452477 and grant Einstein’s equations ever discovered—the Schwarzschild
DMS-2005588. Her email address is [email protected].
spacetime—describes a black hole, many physicists, in-
Dan A. Lee is an associate professor of mathematics at Queens College and the
cluding Einstein, believed that black hole regions might
CUNY Graduate Center. His email address is [email protected].
only be present in highly symmetric solutions that were
Communicated by Notices Associate Editor Daniela De Silva.
not realistic enough to describe nature. Penrose’s sem-
For permission to reprint this article, please contact:
inal 1965 paper [Pen65] implied that the singular be-
[email protected].
havior associated with black holes persists even without
DOI: https://doi.org/10.1090/noti2453

536 NOTICES OF THE AMERICAN MATHEMATICAL SOCIETY VOLUME 69, NUMBER 4

 Figure 3. This image shows the elliptical galaxy Messier 87.
The right top inset gives a close-up of two shockwaves
Figure 1. This computer-simulated image shows a created by a jet emanating from the galaxy’s supermassive
supermassive black hole at the center of a galaxy. The black black hole. The right bottom inset shows an image of that
region represents a snapshot of the event horizon of the black black hole produced by the Event Horizon Telescope [Eve19].
hole. Light from background stars is stretched and distorted
due to the strong gravity of the black hole.
pictures of the black hole (or rather, the “shadow” created
symmetry, which was enough to convince many that black by it) at the heart of Messier 87, an elliptical galaxy 55 mil-
hole formation was a real physical phenomenon. But per- lion light-years from Earth. See Figure 3.
haps just as importantly, Penrose revolutionized the study In this article we provide an exposition of Pen-
of general relativity by introducing global differential geo- rose’s groundbreaking concepts of trapped surfaces and
metric and topological methods, in contrast to the more marginally outer trapped surfaces (MOTS), and some of
calculation-based approaches of the past. their applications. Specifically, we will discuss Stephen
Also starting in the 1960s, the study of quasars even- Hawking’s result on the topology of black holes and some
tually led astrophysicists to hypothesize that there were recent developments on the positive mass theorem that go
black holes at the centers of most galaxies, including our back to the work of Richard Schoen and Shing-Tung Yau.
own galaxy, the Milky Way. It is difficult to directly ob- The study of MOTS gives an effective way to understand
serve something that cannot emit light, but starting in the some properties of black holes and turns out to have many
1990s, separate teams led by Genzel and Ghez began mak- analogies with the study of minimal surfaces. The theory
ing detailed observations of the movements of stars near of minimal surfaces is a mathematically rich topic with a
the center of the Milky Way, and after decades of collect- long history that goes back to Lagrange’s work in the 18th
ing increasingly accurate data, we can now be confident century. Minimal surfaces also have many applications to
that those movements are consistent with the existence of such diverse fields as architecture, biology, and engineer-
a black hole whose mass is 4.3 million times that of our ing, in addition to general relativity. See Figure 4. Explor-
Sun [EG96, GKMB98]. See Figure 2. ing the intriguing connections between MOTS and mini-
Even more recently, in 2019, the Event Horizon mal surfaces has led to fruitful developments in both gen-
Telescope—an international collaboration linking radio eral relativity and differential geometry.
telescopes across the globe—treated us to spectacular
2. Trapped Surfaces
In this article our basic setting is a 4-dimensional space-
time, representing one time dimension and three spatial
dimensions, and since we would like to describe an (ef-
fectively) isolated gravitational system, such as a galaxy,
we assume that this spacetime is asymptotic to the triv-
ial flat spacetime, which is usually called the Minkowski
spacetime, near “infinity.” A rigorous general mathemati-
cal definition of a black hole is quite technical, but roughly
Figure 2. The W. M. Keck Observatory is a two-telescope speaking, when we refer to the black hole region of a space-
astronomical observatory near the summit of Mauna Kea in time, we mean a region that has the property that light rays
Hawaii. Starting in the 1990s, Ghez and her team used these
emanating from the black hole region can never reach its
telescopes and adaptive optics systems to track multiple stars
orbiting the center of our galaxy. complement, while from every point of the complement,
one should be able to “escape to infinity” by following a

APRIL 2022 NOTICES OF THE AMERICAN MATHEMATICAL SOCIETY 537

 Figure 5. Typically, the outgoing “light shell” has increasing
area form as it exits the surface, as shown for the surface Σ1 .
Meanwhile, for the trapped surface Σ2 , the area form is
decreasing.

Figure 4. Left: The roof of the Denver International Airport is a

surface feels a bit like saying that even though the light
tensioned fabric structure that employs double-curved rays are “directed outward,” they are still “moving inward,”
minimal surfaces. Right: Triply periodic minimal surfaces are which vaguely captures the idea of light not being able
used in industry as cost-effective functional components to to “escape.” Because of this heuristic and Penrose’s singu-
produce high-complexity customized structures via 3D larity theorem, physicists often associate trapped surfaces
printing technology.
with the existence of black holes, and in fact, under cer-
tain global hypotheses, one can even prove that trapped
light ray. The boundary of the black hole region is called surfaces must lie inside the black hole region. Because of
the event horizon. Given a spatial 3-dimensional “snapshot this relationship, the (weakly) trapped region of 𝑀, which
in time” 𝑀 in our spacetime, we would like to understand we define to be the region of 𝑀 enclosed by either trapped
which points in 𝑀 lie inside the black hole, but unfortu- surfaces or MOTS in 𝑀, can be thought of as a simpler
nately, since the black hole region can only be technically stand-in for the intersection of 𝑀 with the black hole re-
defined in terms of global properties of the spacetime, it gion, and the apparent horizon of 𝑀, which we define to be
is impossible to tell whether or not a given point in 𝑀 lies the boundary of the trapped region of 𝑀, can be thought
inside the black hole without complete knowledge of the of as a simpler stand-in for the intersection of 𝑀 with the
long-term spacetime future of 𝑀. event horizon. The advantage of these concepts is that
Penrose’s concept of a trapped surface offers us an acces- the trapped region and apparent horizon are entirely deter-
sible way to understand the location of a black hole with- mined by data along the spatial 3-dimensional “snapshot
out knowing its long-term future: Given a 2-surface Σ in in time” 𝑀.
spacetime, imagine shooting a light ray from each point of
Σ, and then define Σ𝑡 to be the surface obtained by follow- 3. Spacetime Geometry
ing these light rays for parameter-time 𝑡, so that Σ𝑡 can be Since we want to consider 4-dimensional spacetimes but
thought of as a “shell of light” emanating from Σ. (Note without a uniquely determined “time coordinate,” the
that we should not think of this 𝑡 as actual “time” since natural setting of general relativity is a 4-manifold. A
light does not experience passage of time.) We typically Lorentzian metric 𝐠 on a 4-manifold 𝑁 defines an inner
expect that if we shoot these light rays inward, the area of product with signature (−, +, +, +) on each tangent space
Σ𝑡 will decrease in 𝑡, and that if we shoot them outward, of 𝑁, smoothly depending on the base point. This means
the area of Σ𝑡 will increase, as is the case for a standard that at each point of 𝑁, there is an orthogonal basis of
sphere in the Minkowski spacetime. However, in the pres- tangent vectors {𝑒 0 , 𝑒 1 , 𝑒 2 , 𝑒 3 } such that 𝐠(𝑒 0 , 𝑒 0 ) = −1 and
ence of strong gravity, it is possible for the family of outgo- 𝐠(𝑒 𝑖 , 𝑒 𝑖 ) = +1 for 𝑖 = 1, 2, 3. So if a tangent vector 𝑣 is
ing light shells to have decreasing area forms at each point given by (𝑣0 , 𝑣1 , 𝑣2 , 𝑣3 ) when written in this basis, then
of Σ, and in this case we say that Σ is an (outer) trapped 𝑔(𝑣, 𝑣) = −(𝑣0 )2 + (𝑣1 )2 + (𝑣2 )2 + (𝑣3 )2 . If one thinks of
surface. See Figure 5. Meanwhile, a marginally outer trapped Riemannian geometry as being locally modeled on the Eu-
surface (or MOTS) refers to the borderline case in which the clidean metric 𝑑𝑠2 = 𝑑𝑥2 + 𝑑𝑦2 + 𝑑𝑧2 , one can analogously
area forms at each point are unchanging to first-order in 𝑡. think of Lorentzian geometry as being locally modeled on
Penrose’s famed singularity theorem states that under the Minkowski metric 𝑑𝑠2 = −𝑑𝑡2 + 𝑑𝑥2 + 𝑑𝑦2 + 𝑑𝑧2 . Spe-
certain physically reasonable assumptions, the existence cial relativity is essentially the physics of Minkowski geom-
of a closed trapped surface implies that there is a light ray etry, so the reason why a Lorentzian manifold is the natu-
emanating from the trapped surface that eventually runs ral setting for general relativity is that the theory should
into a singularity. Intuitively, the behavior of a trapped be locally modeled on special relativity. Moreover, we

538 NOTICES OF THE AMERICAN MATHEMATICAL SOCIETY VOLUME 69, NUMBER 4

Figure 6. At the tangent space of each point, one can define
the future and past light cones. Two future null vectors 𝑒 0 + 𝑒 1
and 𝑒 0 − 𝑒 1 are given. The yellow curve is null as its tangent
vector is null at each point. The submanifold at the
bottom-right is spacelike because its tangent vectors are
spacelike at each point.

will often implicitly assume that 𝐠 is asymptotic to the Figure 7. The Schwarzschild metric in the ingoing
Minkowski metric in some sense. Eddington-Finkelstein coordinates (one space dimension
For any tangent vector 𝑣 of 𝑁, we call it timelike if suppressed). Any light ray that starts at a point where 𝑟 < 2𝑚
𝐠(𝑣, 𝑣) < 0, null if 𝐠(𝑣, 𝑣) = 0, and spacelike if 𝐠(𝑣, 𝑣) > 0. will crash into the 𝑟 = 0 singularity. The surfaces Σ𝑣,𝑟 defined
by constant 𝑣, 𝑟 coordinates are trapped surfaces whenever
So for example, 𝑒 0 is timelike, 𝑒 1 is spacelike, and 𝑒 0 + 𝑒 1 is 𝑟 < 2𝑚, and they are MOTS when 𝑟 = 2𝑚.
null. The null vectors form a double cone which separates
the timelike vectors from the spacelike ones. A spacetime describes the distribution of energy in the spacetime.2 One
is a Lorentzian 4-manifold (𝑁, 𝐠) equipped with a globally can view these equations as a complicated nonlinear sys-
defined unit timelike vector field, which we may select as tem of partial differential equations on 𝐠, with the stress-
our 𝑒 0 at every point. This choice allows us to further dis- energy tensor as a source term. An initial data set (𝑀, 𝑔, 𝑘)
tinguish between future null or timelike vectors, which lie may be regarded as Cauchy data for this system of partial
on or above the upper half of the null cone, and past null differential equations, as explicated by the fundamental
or timelike vectors, which lie on or below the lower half of work of Yvonne Choquet-Bruhat on well-posedness of the
the null cone. Einstein equations.
A submanifold of 𝑁 is called spacelike if all of its tan- As alluded to earlier, the Schwarzschild metrics were the
gent vectors are spacelike, or equivalently, if 𝐠 induces a first nontrivial solutions to be discovered with source term
Riemannian metric on it. In particular, we define a space- equal to zero. In one particular choice of coordinates, the
like slice of 𝑁 to be a 3-dimensional spacelike hypersurface Schwarzschild metric of mass 𝑚 can be written as
𝑀, which is what we earlier referred to as a “snapshot in
time.” The induced Riemannian metric 𝑔 and the second 2𝑚
𝐠𝑚 = − (1 − ) 𝑑𝑣2 + 2𝑑𝑣𝑑𝑟 + 𝑟2 𝑑Ω2 ,
fundamental form1 𝑘 of a spacelike slice 𝑀 can largely cap- 𝑟
ture the spacetime geometry along 𝑀, and we will refer to which is a smooth Lorentzian metric on ℝ × (0, ∞) × 𝑆 2 ,
(𝑀, 𝑔, 𝑘) as an initial data set. A curve is called null (or time- where 𝑣 ∈ ℝ, 𝑟 ∈ (0, ∞), and 𝑑Ω2 is the standard Rie-
like) if its tangent vector is null (or timelike) at each point. mannian metric on the sphere 𝑆 2 . In this spacetime, one
See Figure 6. As in Riemannian geometry, a Lorentzian can show that a light ray emanating from within the re-
metric 𝐠 gives us a concept of “straight lines,” which we gion 𝑟 ≤ 2𝑚 can never enter the region 𝑟 > 2𝑚, while
call geodesics. The path of a light ray traces out a future any point in the region 𝑟 > 2𝑚 can be connected to “infin-
null geodesic in the spacetime, while a massive test parti- ity” by a light ray. Or in other words, the region 𝑟 ≤ 2𝑚
cle will trace out a future timelike geodesic. A test particle is a black hole region, with its boundary 𝑟 = 2𝑚 as the
that traces out a spacelike geodesic would travel faster than event horizon. It is also a fact that as 𝑟 approaches zero, the
the speed of light, and thus it is unphysical. metric becomes singular there in the sense that the curva-
Einstein’s equations demand that certain “curvatures” ture blows up, and moreover, any light ray (or massive test
of 𝐠 must be equal to the stress-energy tensor, which
2Specfically, the curvatures referred to here are the Einstein tensor of 𝐠, defined
1
1The second fundamental form of 𝑀 ⊂ (𝑁, 𝐠) is defined to be the tangential to be 𝐆 = 𝐑𝐢𝐜 − 𝐑𝐠 where 𝐑𝐢𝐜 and 𝐑 are the Ricci and scalar curvatures of
2
part of ∇𝐠 𝐧 where 𝐧 is the future unit normal of 𝑀. 𝐠, respectively.

APRIL 2022 NOTICES OF THE AMERICAN MATHEMATICAL SOCIETY 539

 at each 𝑝 ∈ Σ. Note that the sign of 𝐻 will depend on
choice of 𝜈. See Figure 8.
One can then show that for any outward normal
variation with first-order variation 𝑋 = 𝑒ᵆ 𝜈, we have
𝜕| ᵆ
| 𝑑𝜎𝑡 = 𝑒 𝐻 𝑑𝜎. From this we see that if 𝐻 > 0 on
𝜕𝑡 𝑡=0
Σ, then all small outward normal variations will increase
area, and if 𝐻 < 0 on Σ, then all small outward normal vari-
ations will decrease area. A minimal surface is defined to be
a surface with 𝐻 = 0 everywhere. It is called this because
any surface whose area minimizes area among all small
outward and inward normal variations must have 𝐻 = 0.
(Note that it is a bit of a misnomer since a minimal surface
Figure 8. Parallel surfaces of distance 𝑡 away from Σ in the need not minimize area among variations.)
direction 𝜈 are shown. Positive mean curvature of Σ is Now consider a spacelike 2-surface Σ in a 4-dimensional
characterized by the property that the area of Σ𝑡 increases in 𝑡
for small 𝑡.
spacetime (𝑁, 𝐠). Now, instead of having a 1-dimensional
space of normal vectors at each point of Σ, we have a 2-
dimensional space of normal vectors. However, our space-
particle) starting at a point where 𝑟 < 2𝑚 will inevitably
time geometry picks out exactly two future null directions,
crash into this singularity. See Figure 7.
which correspond to two normal directions from which
4. Variations of Surface Area light rays can originate. Specifically, if 𝜈 is a spacelike unit
vector orthogonal to Σ and 𝑒 0 is a future timelike unit vec-
Earlier we described trapped surfaces in terms of decreas-
tor orthogonal to both Σ and 𝜈, then those two future null
ing area forms of “light shells.” Here we will make that
directions are given by 𝑒 0 + 𝜈 and 𝑒 0 − 𝜈. Suppose that ℓ is
precise, but first we discuss the simpler concept of varying
a future null vector field defined along Σ which is normal
the area form of a surface Σ in a Riemannian 3-manifold
to Σ at each point. Note that we cannot demand a “unit
(𝑀, 𝑔). Suppose 𝜈 is a unit normal vector on Σ, which
length” normalization since ℓ has “length” zero. Multi-
we will regard as the “outward” direction (regardless of
plying ℓ by a positive function yields another future null
whether Σ separates 𝑀 into an “outside” and “inside”),
normal vector field defined along Σ, but modulo this sort
and let Φ𝑡 be a family of diffeomorphisms on 𝑀 with Φ0
of rescaling, there can only be two choices of future null
equal to the identity. Then Σ𝑡 ∶= Φ𝑡 (Σ) defines a fam-
normal, and we may designate one of them as “outward.”
ily of surfaces, sometimes called variations of Σ, whose
Given a choice of (future) outward null normal ℓ for Σ, we
first-order variation vector field 𝑋 along Σ is defined by
𝜕 use it in the following construction:
𝑋(𝑝) ∶= || Φ𝑡 (𝑝) for each 𝑝 ∈ Σ. Let us consider out- Define a family Σ𝑡 = Φ𝑡 (Σ) in 𝑁 by defining Φ𝑡 (𝑝) =
𝜕𝑡 𝑡=0
ward normal variations, which are those for which 𝑋 = 𝑒ᵆ 𝜈 exp𝑝 (𝑡ℓ), where exp𝑝 is the exponential map at 𝑝. This is
along Σ, for some smooth function 𝑢. the family of “light shells” referred to in Section 2, and it
One simple variation of Σ is the family of parallel sur- is analogous to the family of parallel surfaces in the Rie-
faces, obtained by taking Φ𝑡 (𝑝) = exp𝑝 (𝑡𝜈) for small 𝑡, at mannian setting. Now we define the (outward) null expan-
each 𝑝 ∈ Σ, where exp𝑝 is the exponential map at 𝑝. This sion analogously to how we defined the mean curvature: It
just means that Φ𝑡 (𝑝) is obtained by starting at 𝑝, and then is the unique function 𝜃 on Σ with the property that
moving along the geodesic pointing in the 𝜈 direction for
𝜕|
𝑡 units of arclength. These are called parallel surfaces be- | 𝑑𝜎 = 𝜃 𝑑𝜎
𝜕𝑡 |𝑡=0 𝑡
cause Σ𝑡 will be exactly a (signed) distance 𝑡 away from Σ
in the 𝜈 direction (for small enough 𝑡). For this family, at each 𝑝 ∈ Σ. Since 𝜃 depends on the exact choice of ℓ,
the first-order variation 𝑋 is just 𝜈, and this family gives us and there is no natural choice of scaling for ℓ, it turns out
one way to define the mean curvature 𝐻: It measures the lo- that only the sign of 𝜃 is a physical or geometric property
cal increase (or decrease) in area as we move through the of Σ. We can now define Σ to be an (outer) trapped surface if
family of parallel surfaces. Explicitly, if 𝑑𝜎 denotes the in- 𝜃 < 0, an (outer) untrapped surface if 𝜃 > 0, or a marginally
duced area form on Σ and 𝑑𝜎𝑡 denotes the induced area outer trapped surface (or MOTS) if 𝜃 = 0.
form on Σ𝑡 (and then pulled back to Σ via Φ𝑡 ), then the Now suppose that the 2-surface Σ lies in a designated
mean curvature 𝐻 is defined to be the unique function on spacelike slice 𝑀 of the spacetime (𝑁, 𝐠). In this case, a
Σ such that choice of outward normal 𝜈 to Σ in 𝑀 gives us a choice of
𝜕| outward null normal ℓ on Σ by taking ℓ = 𝑒 0 + 𝜈 as above,
| 𝑑𝜎 = 𝐻 𝑑𝜎,
𝜕𝑡 |𝑡=0 𝑡 where 𝑒 0 is the future timelike unit normal to 𝑀. In the

540 NOTICES OF THE AMERICAN MATHEMATICAL SOCIETY VOLUME 69, NUMBER 4

special case where 𝑀 is totally geodesic in 𝑁, 𝜃 is equal to The dominant energy condition is a physically realis-
𝐻, and hence a MOTS is just a minimal surface. Because tic assumption on the stress-energy tensor4, which we will
of this, MOTS can be thought of as generalizations of min- come back to later. This result was generalized by Gre-
imal surfaces, and therefore any general facts or heuristics gory Galloway and Schoen, who showed that higher di-
about MOTS arising from physics automatically translate mensional analogues of apparent horizons must be topo-
into statements about minimal surfaces. Conversely, some logically Yamabe positive [GS06].
parts of the highly developed theory of minimal surfaces
can be used to attack questions concerning MOTS. For ex- 6. The Locally Outermost Property
ample, the existence theory for minimal surfaces using bar- Schoen and Yau were the first to notice that minimal sur-
riers was adapted by Lars Andersson and Jan Metzger and faces could be used to study scalar curvature. The scalar
by Michael Eichmair to prove a corresponding existence curvature 𝑅𝑀 of a Riemannian metric (𝑀, 𝑔) is a scalar func-
theorem for MOTS, and they were also able to prove that tion on 𝑀 which is defined to be the full trace of the Rie-
an apparent horizon must be a smooth MOTS [AEM11]. mann curvature tensor of 𝑔, and as its name suggests, it
is the simplest scalar function that can be computed from
5. Topology of Black Holes 𝑔 that is invariant under change of coordinates. In three
Going back to the example of the Schwarzschild spacetime, dimensions, it has the property that for small 𝑟 > 0, the
each choice of (𝑣, 𝑟) defines a 2-sphere Σ𝑣,𝑟 . In this case, volume of the geodesic ball of radius 𝑟 around 𝑝 ∈ 𝑀 is
4 2𝜋
there is a natural choice for which of the two future null given by 𝜋𝑟3 − 𝑅𝑀 (𝑝)𝑟5 +𝑂(𝑟7 ), so 𝑅𝑀 (𝑝) measures the
3 45
directions is “outward” and we can choose it as our ℓ, and deviation of volumes of small balls around 𝑝 from their
then we can compute the null expansion 𝜃 of Σ𝑣,𝑟 to see Euclidean comparison balls. A similar formula holds in
that Σ𝑣,𝑟 is trapped when 𝑟 < 2𝑚, untrapped when 𝑟 > 2𝑚, other dimensions, and in particular, for a 2-surface Σ, 𝑅Σ
and is a MOTS when 𝑟 = 2𝑚. See Figure 7. In fact, one can is just twice the more familiar Gauss curvature 𝐾Σ .
show that for a spacelike slice 𝑀 of Schwarzschild, the ap- While Hawking exploited the relationship between an
parent horizon in 𝑀 is actually equal to the intersection of outermost MOTS Σ, the dominant energy condition, and
𝑀 with the event horizon. More generally, this is true for the Gauss curvature of Σ, Schoen and Yau similarly ex-
any slice 𝑀 of a “stationary” spacetime.3 Using this fact, ploited the relationship between an area minimizing sur-
Hawking was able to show, under reasonable physical hy- face Σ, scalar curvature of the ambient Riemannian space,
potheses, that any cross-section (i.e. a spacelike 2-surface) and the Gauss curvature of Σ. To be more precise, sup-
of an event horizon in a stationary spacetime must be a pose we have a closed orientable surface Σ in a Riemann-
topological sphere [Haw72]. Or in simpler terms, the sur- ian 3-manifold with nonnegative scalar curvature. Schoen
face of a black hole must be a topological sphere. Based on and Yau observed that if Σ minimizes area compared to all
a suggestion by Gary Gibbons, Hawking was able to gener- small variations of Σ, then Σ must be topologically either
alize his argument to show that even without the station- a sphere or a torus. By a refined argument, Mingliang Cai
ary hypothesis (which is very strong), apparent horizons and Galloway showed that if Σ is strictly area-minimizing,
must be topological spheres [Haw73]. We will explain this then Σ must be a topological sphere. We will explain a
result below. similar result that is a special case of Hawking’s theorem.
Let 𝑀 be a spacelike slice of a spacetime. Given a surface Given a surface Σ with a choice of “outward” normal
Σ in 𝑀 with a choice of “outward” normal 𝜈, we say that Σ 𝜈, we will say that Σ is a locally outermost minimal surface if
is a locally outermost MOTS in 𝑀 if it is a MOTS and there it is a minimal surface and there are no arbitrarily small
do not exist arbitrarily small outward normal variations of outward normal variations of Σ with 𝐻 ≤ 0.
Σ in 𝑀 with 𝜃 ≤ 0. An apparent horizon in 𝑀 must be a
locally outermost MOTS in 𝑀, and this is the relevant prop- Theorem 2. Any orientable locally outermost closed minimal
erty used in Hawking’s proof, which is a beautiful combi- surface Σ in a Riemannian 3-manifold (𝑀, 𝑔) with nonnegative
nation of calculus of variations and the Gauss-Bonnet the- scalar curvature must be a topological sphere.
orem.

Theorem 1 (Hawking). Any orientable locally outermost Proof. Let Σ𝑡 be an outward normal variation of Σ in 𝑀,
closed MOTS in an initial data set satisfying the dominant en- with first-order variation 𝑋 = 𝑒ᵆ 𝜈, and let 𝐻𝑡 denote the
ergy condition must be a topological sphere. mean curvature of Σ𝑡 (pulled back to Σ). One can always

4Explicitly, we say that an initial data set (𝑀, 𝑔, 𝑘) satisfies the dominant en-
ergy condition if the Einstein tensor satisfies 𝐆(𝑒 0 , 𝑣) ≥ 0 where 𝑒 0 is the future
3A spacetime is called stationary if it admits a global Killing vector field that is normal to 𝑀 and 𝑣 is any future null or timelike vector. The condition can also
asymptotically timelike. be expressed purely in terms of (𝑔, 𝑘).

APRIL 2022 NOTICES OF THE AMERICAN MATHEMATICAL SOCIETY 541

Figure 9. Lawson’s minimal surfaces of genus 2, 3, 4, 5, and 6 in 𝑆3 , stereographically projected to ℝ3 in two different ways.

find an outward normal variation such that 𝜕𝐻𝑡

Since Σ𝑡 is a torus and is constant over Σ𝑡 , integrating
𝜕𝑡
𝜕𝐻𝑡 | 𝜕𝐻
| = 𝜆𝑒ᵆ (1) both sides of the above inequality over Σ𝑡 shows that 𝑡 ≤
𝜕𝑡 |𝑡=0 𝜕𝑡
0 for all 𝑡 ≥ 0, which contradicts the locally outermost
for some constant 𝜆. (More precisely, 𝑒ᵆ is chosen to be the assumption. □
principal eigenfunction of the linearized mean curvature
operator.) The locally outermost assumption on Σ implies Without the locally outermost assumption, a minimal
that 𝜆 ≥ 0, because otherwise, we would have 𝐻𝑡 < 0 for surface can have higher genus. For example, the stan-
small 𝑡 > 0. dard 3-sphere admits closed minimal surfaces of arbitrary
Routine geometric computations, including use of the genus. See Figure 9. Over the last decade Fernando Mar-
Gauss equation, show that ques and André Neves have advanced the study of min-
𝜕𝐻𝑡 | imal surfaces using min-max methods, and building on
| their work, Haozhao Li and Xin Zhou proved that a generic
𝜕𝑡 |𝑡=0
1 1 (2) closed Riemannian 3-manifold with positive Ricci curva-
= 𝑒ᵆ (−ΔΣ 𝑢 − |∇𝑢|2 + 𝐾Σ − 𝑅𝑀 − |𝐴|2 ) ture admits closed minimal surfaces with arbitrarily high
2 2
≤ 𝑒ᵆ (−ΔΣ 𝑢 + 𝐾Σ ) , genus, and Antoine Song proved Yau’s 1982 conjecture
that all closed Riemannian 3-manifolds admit infinitely
where ΔΣ denotes the Laplace-Beltrami operator on Σ, 𝐴 many closed minimal surfaces.
is the second fundamental form of Σ in 𝑀, and we used The proof of Theorem 1 is conceptually similar to that
the assumption that 𝑅𝑀 ≥ 0. Combining this with (1), we of Theorem 2. In that case, we start with an orientable
have locally outermost closed MOTS Σ in an initial data set
𝜆 ≤ −ΔΣ 𝑢 + 𝐾Σ , (𝑀, 𝑔, 𝑘). We still look at outward normal variations of Σ
and integrating this over Σ and using the Gauss-Bonnet in 𝑀, but instead of (2), we obtain
Theorem, it follows that 𝜕𝜃𝑡 |
| ≤ 𝑒ᵆ (−ΔΣ 𝑢 + 𝐾Σ + divΣ 𝑊) ,
0 ≤ 𝜆 ⋅ (Area Σ) ≤ 2𝜋𝜒(Σ). 𝜕𝑡 |𝑡=0

Therefore Σ is either a torus or a sphere. for some quantity 𝑊, where this time the inequality fol-
To rule out the torus, suppose to the contrary that Σ is lows from the dominant energy condition rather than non-
a torus. Then 𝜒(Σ) = 0, and thus 𝜆 = 0. In this case, an in- negative scalar curvature. The rest of the proof is essentially
verse function theorem argument can be used to construct the same since the integral of the extra divergence term is
𝜕𝜃
an outward normal variation Σ𝑡 with the added property zero. However, dealing with the case where 𝑡 || = 0 re-
𝜕𝑡 𝑡=0
that each Σ𝑡 has constant mean curvature 𝐻𝑡 . Let 𝑋𝑡 = 𝑒ᵆ𝑡 𝜈𝑡 quires an additional argument (by Galloway) because the
be the first-order variation of Σ𝑡 at an arbitrary 𝑡. As in (2), 𝜕𝜃
formula for 𝑡 at a general 𝑡 has an extra term involving
we have 𝜕𝑡
𝜕𝐻𝑡 −ᵆ𝑡 𝜃𝑡 (which happens to vanish when 𝑡 = 0).
𝑒 ≤ −ΔΣ𝑡 𝑢𝑡 + 𝐾Σ𝑡 .
𝜕𝑡

542 NOTICES OF THE AMERICAN MATHEMATICAL SOCIETY VOLUME 69, NUMBER 4

7. Positive Mass Theorem
We will now discuss how a version of the topological argu-
ment from the previous section can be used to prove the
celebrated positive mass theorem. Let (𝑀, 𝑔, 𝑘) be an ini-
tial data set and assume that it is asymptotically flat, mean-
ing that in coordinates, the metric 𝑔𝑖𝑗 is asymptotic to the
Euclidean metric 𝛿 𝑖𝑗 while the second fundamental form
𝑘𝑖𝑗 is asymptotic to zero, in some precise sense that we will
not describe here. An asymptotically flat initial data set has
a well-defined total ADM energy-momentum5 (𝐸, 𝑃1 , 𝑃2 , 𝑃3 ).
Theorem 3 (Positive mass theorem). Let 𝑀 be a complete
asymptotically flat 3-dimensional initial data set satisfying the Figure 10. Under the (contradictory) assumption 𝐸 < |𝑃|, two
dominant energy condition. Then the ADM energy-momentum coordinate planes Π1 and Π2 , together with the lateral side of
a cylinder of large radius, provide barriers for the existence of
(𝐸, 𝑃1 , 𝑃2 , 𝑃3 ) satisfies 𝐸 ≥ |𝑃|. Furthermore, the equality 𝐸 =
a MOTS Σ𝑟 with prescribed boundary. A subsequential limit
|𝑃| holds if and only if (𝑀, 𝑔) can be isometrically embedded into surface of Σ𝑟 as 𝑟 → ∞ is a complete MOTS satisfying a
the Minkowski spacetime with second fundamental form 𝑘. stability property.

This is called the “positive mass theorem” because an

object with future timelike energy-momentum 𝐸 > |𝑃| a totally geodesic slice of the spacetime. In this case, the
is said to have positive mass. Explicitly, the mass is positive mass theorem reduces to a statement about Rie-
√𝐸 2 − |𝑃|2 . Note that a spacelike slice of the Schwarzschild mannian geometry, which is often called the Riemannian
spacetime with mass 𝑚 will (unsurprisingly) have ADM positive mass theorem: If (𝑀, 𝑔) is a complete asymptoti-
mass equal to 𝑚. Negative mass, which is unphysical, cally flat manifold with nonnegative scalar curvature, then
would correspond to past timelike energy momentum 𝐸 < 𝐸 > 0 unless (𝑀, 𝑔) is Euclidean. This special case was
−|𝑃|, while spacelike energy momentum |𝐸| < |𝑃|, which first proved by Schoen and Yau in 1979 using minimal sur-
would correspond to “imaginary mass,” is also unphysi- faces, and soon later they proved that 𝐸 ≥ 0 in the gen-
cal since it is associated with objects moving faster than eral case using the Jang equation [SY79, SY81]. Edward
the speed of light. Meanwhile, null energy-momentum Witten was able to prove that 𝐸 ≥ |𝑃| using a spinor ar-
|𝐸| = |𝑃| corresponds to zero mass. gument [Wit81]. This might be more accurately called a
The positive mass theorem is highly desirable for phys- “nonnegative mass theorem” since it does not handle the
ical reasons. The dominant energy condition can be de- second statement about 𝐸 = |𝑃| in Theorem 3, which we
scribed as the reasonable physical assumption that the will refer as the equality case of the positive mass theorem. In
sources for Einstein’s equations cannot travel faster than a 2015 article with Eichmair and Schoen, we gave an al-
light, as measured by any observer. The positive mass the- ternative proof of this nonnegative mass theorem by ex-
orem loosely asserts that as long as these sources cannot tending Schoen and Yau’s argument and by replacing the
travel faster than light, the entire configuration of sources, minimal surfaces by MOTS [EHLS16]. In that article we
as viewed from far away, should not behave like an ob- also tackled the more technically challenging case of 𝑛-
ject traveling faster than light, or as fast as light. An exam- dimensional slices, for 3 < 𝑛 < 8, by introducing a new
ple of a violation of the positive mass theorem could be functional that mimics the first variation of the area func-
a configuration of positive mass sources that somehow re- tional. There are recent results of Schoen and Yau and of
pels far away objects instead of attracting them. Because of Lohkamp that deal with higher dimensions. One can also
the nonlinearities of Einstein’s equations, it is highly non- weaken the hypotheses of Theorem 3 to allow 𝑀 to have
trivial to prove that such perversities cannot happen. a boundary as long as that boundary is a closed trapped
The study of the positive mass theorem has a long his- surface or MOTS [GHHP83, LLU21]. Physically, this corre-
tory. A particularly important special case is when 𝑀 is sponds to allowing for the possibility of a black hole with-
out having to assume too much about the geometry inside
5The ADM energy-momentum was formulated by physicists Arnowitt, Deser,
the black hole.
and Misner, and explicitly, the numbers (𝐸, 𝑃1 , 𝑃2 , 𝑃3 ) are defined by We outline the proof of the nonnegative mass theorem
𝜕𝑔𝑖𝑗 𝜕𝑔 𝑥𝑗 for 3-dimensional 𝑀.
16𝜋𝐸 = lim ∫ ∑( − 𝑖𝑖 ) 𝑑𝜎,
𝑟→∞ |𝑥|=𝑟
𝑖,𝑗
𝜕𝑥𝑖 𝜕𝑥𝑗 𝑟
𝑥𝑗
8𝜋𝑃𝑖 = lim ∫ ∑(𝑘𝑖𝑗 − (tr𝑘)𝑔𝑖𝑗 ) 𝑑𝜎. Outline of proof. By a subtle density theorem (see Section 6
𝑟→∞ |𝑥|=𝑟 𝑟
𝑗 of [EHLS16]), one can show that without loss of generality,
we can assume that the “strict” dominant energy condition

APRIL 2022 NOTICES OF THE AMERICAN MATHEMATICAL SOCIETY 543

holds and that the initial data (𝑔, 𝑘) has especially “nice” plane-fronted waves with parallel rays (or pp-waves for
asymptotics at spatial infinity. short), which model radiation moving at the speed of light,
The proof proceeds by contradiction. Suppose that and thus any spacelike slices naturally have 𝐸 = |𝑃| (so
𝐸 < |𝑃|. The “nice” asymptotics imply that there exist non- long as these quantities can be defined). The counterex-
intersecting coordinate planes Π1 and Π2 such that Π1 is amples come from slices of pp-waves which are asymptot-
trapped and Π2 is untrapped; namely, 𝜃 < 0 on Π1 and ically flat enough to satisfy the hypotheses of the nonneg-
𝜃 > 0 on Π2 . See Figure 10. From a PDE perspective, this ative mass theorem, but not asymptotically flat enough to
means that they provide barriers for the MOTS equation satisfy the hypotheses of the equality case of the positive
𝜃 = 0. Using these barriers, an existence theorem due to mass theorem.
Eichmair [Eic09] allows us to construct a MOTS with pre- While Penrose’s advances were only recently honored
scribed boundary that is sandwiched between the planes by the Nobel Prize in Physics, the topological and geomet-
Π1 and Π2 . By taking that prescribed boundary larger and ric methods that he introduced helped to build a long, in-
larger, we can extract a subsequential limit surface which is timate relationship between mathematics and general rel-
a complete MOTS Σ sandwiched between Π1 and Π2 , and ativity over the past several decades. We discussed ground-
we can show that Σ itself is asymptotically planar. While breaking work of Hawking, Schoen, and Yau as fine ex-
this Σ need not be a locally outermost MOTS, it still en- amples of the intriguing interactions between geometry,
joys a “MOTS stability” property that, together with the topology, analysis, and general relativity. New applica-
strict dominant energy condition, can be used to show that tions and interconnections between mathematics and gen-
∫Σ 𝐾Σ 𝑑𝜎 > 0 in a manner that is conceptually similar to eral relativity are continually being discovered in more re-
the proof of Theorem 2. The “strict” dominant energy con- cent developments, as described above. For a more exten-
dition is what forces this inequality to be strict. Finally, the sive introduction to this field of research, please see the re-
Gauss-Bonnet Theorem with boundary implies that this is cent graduate-level textbook by the second author [Lee19].
impossible for an asymptotically planar surface. □ We expect that the exchange of ideas between physics and
mathematics will continue to energize these directions of
To prove the equality case of the positive mass the-
inquiry, which we have only lightly touched upon here.
orem requires significant extra work. One must show
Penrose once stated, “We have a closed circle of consis-
that if 𝐸 = |𝑃|, then the slice 𝑀 actually sits inside the
tency here: the laws of physics produce complex systems,
Minkowski spacetime, in the sense described in the the-
and these complex systems lead to consciousness, which
orem. This actually implies that 𝐸 = |𝑃| = 0 since any
then produces mathematics, which can then encode in
slice of Minkowski has this property. The equality case
a succinct and inspiring way the very underlying laws of
was proved for all spin manifolds by work of Robert Beig
physics that gave rise to it.”
and Piotr Chruściel [BC96], and Chruściel and Daniel
Maerten. This covers all 3-manifolds since all 3-manifolds References
carry spin structures. More recently, we were able to pro- [AEM11] Lars Andersson, Michael Eichmair, and Jan Metzger,
vide a separate proof that avoids the use of a spin as- Jang’s equation and its applications to marginally trapped sur-
sumption [HL20b]. Our proof uses a variational argument faces, Complex analysis and dynamical systems IV. Part 2,
among initial data sets satisfying the dominant energy con- Contemp. Math., vol. 554, Amer. Math. Soc., Providence,
dition, which turns out to have an intriguing connection to RI, 2011, pp. 13–45, DOI 10.1090/conm/554/10958.
the question of “improving” the dominant energy condi- MR2884392
[BC96] Robert Beig and Piotr T. Chruściel, Killing vectors
tion studied by Justin Corvino and the first author [CH20].
in asymptotically flat space-times. I. Asymptotically transla-
The improvability of the dominant energy condition man- tional Killing vectors and the rigid positive energy theo-
ifestly relates to the fundamental problem of scalar curva- rem, J. Math. Phys. 37 (1996), no. 4, 1939–1961, DOI
ture deformation in differential geometry and was further 10.1063/1.531497. MR1380882
explored in our recent work [HL20a]. [CH20] Justin Corvino and Lan-Hsuan Huang, Localized de-
One curious feature of both our proof and the spinor formation for initial data sets with the dominant energy con-
proof is that in higher dimensions, the equality case of the dition, Calc. Var. Partial Differential Equations 59 (2020),
positive mass theorem seems to require a stronger defini- no. 1, Paper No. 42, 43, DOI 10.1007/s00526-019-1679-9.
tion of asymptotic flatness than the standard one needed MR4062040
for the “nonnegative mass theorem” to be true. Some- [EG96] A. Eckart and R. Genzel, Observations of stellar
proper motions near the galactic centre, Nature 383 (1996),
what surprisingly, it turns out that when the spatial dimen-
no. 6599, 415–417.
sion is greater than eight, there do exist counterexamples [Eic09] Michael Eichmair, The Plateau problem for marginally
to the expected strict inequality 𝐸 > |𝑃|. See Example outer trapped surfaces, J. Differential Geom. 83 (2009), no. 3,
7 of [HL20a]. Those examples arise from an important 551–583. MR2581357
family of exact solutions to the Einstein equations, called

544 NOTICES OF THE AMERICAN MATHEMATICAL SOCIETY VOLUME 69, NUMBER 4

[EHLS16] Michael Eichmair, Lan-Hsuan Huang, Dan A. Lee,
and Richard Schoen, The spacetime positive mass theo-
rem in dimensions less than eight, J. Eur. Math. Soc.
(JEMS) 18 (2016), no. 1, 83–121, DOI 10.4171/JEMS/584.
MR3438380
[Eve19] Event Horizon Telescope Collaboration, First M87
Event Horizon Telescope Results. I. The Shadow of the Super-
massive Black Hole, ApJL 875 (April 2019), no. 1, L1, avail-
able at 1906.11238.
[GS06] Gregory J. Galloway and Richard Schoen, A general-
ization of Hawking’s black hole topology theorem to higher di- Lan-Hsuan Huang Dan A. Lee
mensions, Comm. Math. Phys. 266 (2006), no. 2, 571–576,
DOI 10.1007/s00220-006-0019-z. MR2238889 Credits
[GKMB98] A. M. Ghez, B. L. Klein, M. Morris, and E. E. Beck- Opening image is courtesy of NASA’s Goddard Space Flight
lin, High Proper-Motion Stars in the Vicinity of Sagittarius A*: Center; background, ESA/Gaia/DPAC.
Evidence for a Supermassive Black Hole at the Center of Our Figure 1 is courtesy of NASA, ESA, and D. Coe, J. Anderson,
Galaxy, ApJ 509 (December 1998), no. 2, 678–686, avail- and R. van der Marel (STScI).
able at astro-ph/9807210. Figure 2 is courtesy of Left_Coast_Photographer.
[GHHP83] G. W. Gibbons, S. W. Hawking, Gary T. Horowitz, Figure 3 is courtesy of NASA/JPL-Caltech/IPAC/Event Hori-
and Malcolm J. Perry, Positive mass theorems for black zon Telescope Collaboration.
holes, Comm. Math. Phys. 88 (1983), no. 3, 295–308.
Figure 4 (left) is courtesy of arinahabich via Getty.
MR701918
Figure 4 (right) is courtesy of Julián Norato, adapted from
[Haw72] S. W. Hawking, Black holes in general relativity,
ntopology.com.
Comm. Math. Phys. 25 (1972), 152–166. MR293962
[Haw73] S. W. Hawking, The event horizon, Les Houches Sum- Figures 5–8 and 10 are courtesy of Lan-Hsuan Huang.
mer School of Theoretical Physics: Black Holes, 1973. Figure 9 is courtesy of Nick Schmitt.
[HL20a] Lan-Hsuan Huang and Dan A. Lee, Bartnik mass min- Photo of Lan-Hsuan Huang is courtesy of Damin Wu.
imizing initial data sets and improvability of the dominant en- Photo of Dan A. Lee is courtesy of Dan A. Lee.
ergy scalar, arXiv:2007.00593 [math.DG] (2020).
[HL20b] Lan-Hsuan Huang and Dan A. Lee, Equality in
the spacetime positive mass theorem, Comm. Math. Phys.
376 (2020), no. 3, 2379–2407, DOI 10.1007/s00220-019-

AMS AUTHOR
03619-w. MR4104553
[Lee19] Dan A. Lee, Geometric relativity, Graduate Studies in

RESOURCE CENTER
Mathematics, vol. 201, American Mathematical Society,
Providence, RI, 2019. MR3970261
[LLU21] Dan A. Lee, Martin Lesourd, and Ryan Unger, Density
and positive mass theorems for initial data sets with boundary, The Author Resource Center is a collection of
arXiv:2112.12017 [math.DG] (2021).
[Pen65] Roger Penrose, Gravitational collapse and space-time information and tools available to assist you to
singularities, Phys. Rev. Lett. 14 (1965), 57–59, DOI successfully write, edit, illustrate, and publish
10.1103/PhysRevLett.14.57. MR172678 your mathematical works.
[SY79] Richard Schoen and Shing Tung Yau, On the proof of
the positive mass conjecture in general relativity, Comm. Math.
Phys. 65 (1979), no. 1, 45–76. MR526976 To begin utilizing
[SY81] Richard Schoen and Shing Tung Yau, Proof of the posi- these important
tive mass theorem. II, Comm. Math. Phys. 79 (1981), no. 2,
231–260. MR612249 resources, visit:
[Wit81] Edward Witten, A new proof of the positive energy
theorem, Comm. Math. Phys. 80 (1981), no. 3, 381–402.
MR626707 www.ams.org/authors

APRIL 2022 NOTICES OF THE AMERICAN MATHEMATICAL SOCIETY 545