Riemannian Penrose Inequality

• Riemannian Penrose Inequality is a sharp geometric inequality that relates the ADM mass to the area of the outermost minimal surface in time-symmetric, asymptotically flat Riemannian manifolds.
• It is rigorously established using techniques such as Huisken–Ilmanen’s weak inverse mean curvature flow and Bray’s conformal metric evolution, achieving equality for Schwarzschild exteriors.
• Recent advances extend the inequality to higher dimensions, charged settings, and manifolds with noncompact boundaries, offering deeper insights into cosmic censorship and rigidity.

The Riemannian Penrose inequality is the time-symmetric Penrose inequality. In its classical three-dimensional form, for an asymptotically flat Riemannian manifold with nonnegative scalar curvature and outermost minimal surface $S_m$, it asserts

$M_{ADM}\ge \sqrt{\frac{|S_m|}{16\pi}}.$

In higher dimension, the corresponding formula is

$$m_{\mathrm{ADM}}(M,g)\ge \frac12\left(\frac{\mathcal H_g^{n-1}(\Sigma)}{n\omega_n}\right)^{\frac{n-2}{n-1}},$$

with equality exactly for the Riemannian Schwarzschild exteriors. The inequality relates total mass to horizon area, is motivated by Penrose’s cosmic censorship heuristic, and forms the completed time-symmetric part of the Penrose program in mathematical relativity (0906.5566, Bi et al., 1 May 2026).

1. Geometric formulation

In the time-symmetric setting, an initial data set is an asymptotically flat Riemannian manifold $(\Sigma,\gamma)$ with vanishing second fundamental form $A_{ij}=0$. Under the weak or dominant energy condition this gives

$R(\gamma)=16\pi \rho \ge 0,$

so the problem becomes purely Riemannian. The relevant horizon is the outermost minimal surface $S_m$, which in this setting is the boundary of the trapped region. The classical inequality is

$M_{ADM} \geq \sqrt{\frac{|S_m|}{16\pi}},$

while the higher-dimensional analogue is

$$M_{ADM} \geq \frac{1}{2}\left(\frac{|S_0|}{\omega_{n-1}}\right)^{\frac{n-2}{n-1}}$$

in the Bray–Lee formulation, and

$$m_{\mathrm{ADM}}(M,g)\ge \frac12\left(\frac{\mathcal H_g^{n-1}(\Sigma)}{n\omega_n}\right)^{\frac{n-2}{n-1}}$$

in the all-dimensional formulation with possibly singular outer-minimizing minimal boundary (0906.5566, Bi et al., 1 May 2026).

The inequality is sharp on the Schwarzschild family. In dimension $M_{ADM}\ge \sqrt{\frac{|S_m|}{16\pi}}.$0, the Riemannian Schwarzschild metric is

$M_{ADM}\ge \sqrt{\frac{|S_m|}{16\pi}}.$1

on

$M_{ADM}\ge \sqrt{\frac{|S_m|}{16\pi}}.$2

and its horizon has area

$M_{ADM}\ge \sqrt{\frac{|S_m|}{16\pi}}.$3

Solving for $M_{ADM}\ge \sqrt{\frac{|S_m|}{16\pi}}.$4 gives exactly the right-hand side of the Penrose inequality, so equality is realized by Schwarzschild and only by Schwarzschild (Bi et al., 1 May 2026).

A basic structural distinction concerns connected versus disconnected horizons. Huisken–Ilmanen’s inverse-mean-curvature-flow theorem gives, for components $M_{ADM}\ge \sqrt{\frac{|S_m|}{16\pi}}.$5 of the outermost minimal surface,

$M_{ADM}\ge \sqrt{\frac{|S_m|}{16\pi}}.$6

whereas Bray’s theorem gives the full inequality in terms of the total area of the outermost horizon, with no connectedness assumption (0906.5566).

2. Penrose heuristic and the time-symmetric reduction

The physical origin of the inequality is Penrose’s argument from gravitational collapse and cosmic censorship. In rough form, one expects

$M_{ADM}\ge \sqrt{\frac{|S_m|}{16\pi}}.$7

because a trapped surface indicates collapse, weak cosmic censorship suggests that the singularity is hidden behind an event horizon, Hawking’s area theorem makes the event horizon area nondecreasing, and the final stationary black hole should satisfy the Kerr or Schwarzschild area-mass relation. In the time-symmetric case, apparent-horizon geometry simplifies drastically: null expansions reduce to the mean curvature, so marginally trapped surfaces become minimal surfaces. This converts the spacetime inequality into a statement about asymptotically flat Riemannian manifolds with nonnegative scalar curvature (0906.5566).

This reduction explains both the strength and the limitations of the Riemannian theorem. It is strong because it gives a sharp lower bound purely from scalar curvature and minimal-surface geometry. It is limited because it uses $M_{ADM}\ge \sqrt{\frac{|S_m|}{16\pi}}.$8; the fully general Penrose inequality for non-time-symmetric initial data remains open. Mars’s survey identifies Bray–Khuri’s generalized-Jang proposal as a major route toward the general case, but presents it explicitly as a proposal rather than a completed proof (0906.5566).

3. Proof methods

The first successful proof strategy was Huisken–Ilmanen’s weak inverse mean curvature flow. For a smooth flow, the Geroch mass is monotone when $M_{ADM}\ge \sqrt{\frac{|S_m|}{16\pi}}.$9, and a connected minimal initial surface has initial Geroch mass

$$m_{\mathrm{ADM}}(M,g)\ge \frac12\left(\frac{\mathcal H_g^{n-1}(\Sigma)}{n\omega_n}\right)^{\frac{n-2}{n-1}},$$0

The analytic obstacle is that smooth IMCF develops singularities. Huisken–Ilmanen therefore introduced a weak level-set formulation based on

$$m_{\mathrm{ADM}}(M,g)\ge \frac12\left(\frac{\mathcal H_g^{n-1}(\Sigma)}{n\omega_n}\right)^{\frac{n-2}{n-1}},$$1

and a variational functional

$$m_{\mathrm{ADM}}(M,g)\ge \frac12\left(\frac{\mathcal H_g^{n-1}(\Sigma)}{n\omega_n}\right)^{\frac{n-2}{n-1}},$$2

This yields the connected-horizon theorem and the componentwise bound quoted above (0906.5566).

Bray’s proof is conceptually different. Instead of moving surfaces in a fixed metric, it evolves the metric conformally,

$$m_{\mathrm{ADM}}(M,g)\ge \frac12\left(\frac{\mathcal H_g^{n-1}(\Sigma)}{n\omega_n}\right)^{\frac{n-2}{n-1}},$$3

with a harmonic velocity field chosen so that the horizon area stays constant and the ADM mass is nonincreasing. The flow is designed to converge to Schwarzschild, so the initial manifold inherits the Schwarzschild lower bound. This method resolves the disconnected-horizon case and gives the full Riemannian Penrose inequality in dimension three; Bray–Lee extend the same architecture to dimensions $$m_{\mathrm{ADM}}(M,g)\ge \frac12\left(\frac{\mathcal H_g^{n-1}(\Sigma)}{n\omega_n}\right)^{\frac{n-2}{n-1}},$$4 (0906.5566).

Several later works revisit the connected-horizon theorem with new techniques. A nonlinear-potential-theoretic proof replaces weak IMCF by the level sets of the $$m_{\mathrm{ADM}}(M,g)\ge \frac12\left(\frac{\mathcal H_g^{n-1}(\Sigma)}{n\omega_n}\right)^{\frac{n-2}{n-1}},$$5-capacitary potential of the horizon and proves a new monotonicity formula along those level sets for $$m_{\mathrm{ADM}}(M,g)\ge \frac12\left(\frac{\mathcal H_g^{n-1}(\Sigma)}{n\omega_n}\right)^{\frac{n-2}{n-1}},$$6; letting $$m_{\mathrm{ADM}}(M,g)\ge \frac12\left(\frac{\mathcal H_g^{n-1}(\Sigma)}{n\omega_n}\right)^{\frac{n-2}{n-1}},$$7 recovers

$$m_{\mathrm{ADM}}(M,g)\ge \frac12\left(\frac{\mathcal H_g^{n-1}(\Sigma)}{n\omega_n}\right)^{\frac{n-2}{n-1}},$$8

but the proof does not establish a full equality characterization (Agostiniani et al., 2022). A different refinement proves the connected-horizon theorem under the optimal asymptotic decay assumption $$m_{\mathrm{ADM}}(M,g)\ge \frac12\left(\frac{\mathcal H_g^{n-1}(\Sigma)}{n\omega_n}\right)^{\frac{n-2}{n-1}},$$9, $(\Sigma,\gamma)$0, and establishes

$(\Sigma,\gamma)$1

relating the ADM mass to Huisken’s isoperimetric mass through a comparison of Hawking mass, potential-theoretic Hawking mass, and isoperimetric mass (Benatti et al., 2022).

4. Equality, rigidity, and stability

In the smooth theorem, equality characterizes the spatial Schwarzschild manifold outside its horizon (0906.5566). Recent work sharpens this rigidity in singular and near-equality settings.

For asymptotically flat manifolds with corners, equality in the Penrose inequality forces much more than the distributional scalar-curvature condition $(\Sigma,\gamma)$2. Shi–Wang–Yu prove that equality implies

$(\Sigma,\gamma)$3

that the interior region is static with vanishing scalar curvature, and, under an additional geometric assumption on distance-minimizing geodesics to the horizon, that the interior region is Schwarzschild (Shi et al., 2017). Lu–Miao strengthen the singular rigidity picture: a suitable metric with corners attaining the optimal Penrose value is necessarily smooth across the corner in the specified coordinates, and the glued manifold is Schwarzschild; the same argument yields rigidity for isometric hypersurfaces with the same mean curvature in Schwarzschild (Lu et al., 2020).

Near equality also has a stability theory. In dimension three, if a sequence of asymptotically flat manifolds with nonnegative scalar curvature has outermost minimal boundary area $(\Sigma,\gamma)$4 and

$(\Sigma,\gamma)$5

then, after discarding negligible domains and allowing a small perturbation of the boundary area, the exterior regions converge to the corresponding Schwarzschild manifold in pointed measured Gromov–Hausdorff topology. The theorem is formulated with a major boundary component whose area tends to $(\Sigma,\gamma)$6 and a minor boundary part whose area tends to zero, reflecting the fact that thin necks can obstruct stronger global convergence statements (Dong, 2024).

5. Higher-dimensional, boundary, and geometric analogues

The dimensional gap in the classical theory has now been closed. In arbitrary dimension $(\Sigma,\gamma)$7, the Riemannian Penrose inequality holds for smooth complete asymptotically flat manifolds with nonnegative scalar curvature and compact outer-minimizing minimal boundary, even when the boundary has a singular set of Hausdorff dimension at most $(\Sigma,\gamma)$8. The proof extends Bray’s conformal flow through singular outer-minimizing enclosures and establishes rigidity to the Riemannian Schwarzschild exteriors (Bi et al., 1 May 2026).

There are also Penrose-type theorems for asymptotically flat manifolds with non-compact boundary. In the three-dimensional half-space setting, with asymptotic end modeled on $(\Sigma,\gamma)$9, nonnegative scalar curvature, nonnegative boundary mean curvature, and a connected free boundary minimal horizon $A_{ij}=0$0, Koerber proves

$A_{ij}=0$1

with equality exactly for one-half of the spatial Schwarzschild half-space. The proof develops a weak free-boundary inverse mean curvature flow and a monotone free-boundary Hawking mass (Koerber, 2019). A higher-dimensional boundary analogue uses a doubling procedure for asymptotically flat half-spaces and yields

$A_{ij}=0$2

for $A_{ij}=0$3, with rigidity to Schwarzschild half-space (Eichmair et al., 2023).

Several related inequalities are analogues rather than literal extensions of the classical Riemannian theorem. For conformally flat asymptotically flat manifolds, one has all-dimensional Penrose-type lower bounds in terms of Euclidean boundary area rather than intrinsic horizon area, obtained by combining the conformal scalar-curvature formula with the Euclidean Minkowski inequality; these estimates can be stronger than the usual total-area bound when many boundary components are present, but they are not sharp in the Schwarzschild case and do not require outermost minimal surfaces (Jauregui, 2011). In weighted asymptotically flat geometry, a conformal relation

$A_{ij}=0$4

reduces the weighted Penrose inequality to the classical theorem, giving

$A_{ij}=0$5

under $A_{ij}=0$6, with equality characterized by $A_{ij}=0$7-Schwarzschild metrics (McCormick, 27 Jan 2026). In a different direction, an extrinsic analogue for asymptotically flat support surfaces $A_{ij}=0$8 proves

$A_{ij}=0$9

with equality exactly for the half-catenoid; this is presented as the resolution of Huisken’s conjecture in extrinsic geometry (Eichmair et al., 2024).

6. Charged versions and remaining frontiers

The charged Riemannian Penrose inequality refines the Schwarzschild bound by replacing the model solution with Reissner–Nordström. In the time-symmetric Einstein–Maxwell setting, the natural charged formula is

$R(\gamma)=16\pi \rho \ge 0,$0

where $R(\gamma)=16\pi \rho \ge 0,$1. For connected outermost horizons, Khuri, Weinstein, and Yamada isolate a structural asymmetry in the equivalent pair

$R(\gamma)=16\pi \rho \ge 0,$2

They show that the lower bound follows directly from Gibbons’ area-charge inequality

$R(\gamma)=16\pi \rho \ge 0,$3

together with the positive mass theorem with charge, whereas the upper bound is the branch suggested by Penrose’s cosmic-censorship heuristic (Khuri et al., 2013).

For multiple black holes in the time-symmetric Einstein–Maxwell case, the physically expected upper bound

$R(\gamma)=16\pi \rho \ge 0,$4

has been proved via a charged version of Bray’s conformal flow. Under the additional hypothesis $R(\gamma)=16\pi \rho \ge 0,$5, this becomes the charged Penrose inequality

$R(\gamma)=16\pi \rho \ge 0,$6

with equality only for Reissner–Nordström. These results are specifically formulated for strongly asymptotically flat data with no charged matter outside the horizon and allow disconnected outermost horizons (Khuri et al., 2013, Khuri et al., 2014).

Allowing charge density outside the horizon changes the picture. If the exterior charge densities are compactly supported, the upper bound still holds, but if the charges extend to infinity there are spherically symmetric counterexamples (Khuri et al., 2014). Jang’s inverse-mean-curvature-flow argument can also be adapted when the exterior charge density has a favorable sign: with suitable sign assumptions one obtains a charged Penrose inequality with either horizon charge or total charge, and with the opposite sign one gets counterexamples. This identifies the sign condition as a genuine substitute for charge conservation when $R(\gamma)=16\pi \rho \ge 0,$7 (McCormick, 2019).

The broader Penrose program remains incomplete outside the Riemannian setting. Mars’s survey presents the Riemannian theorem as resolved, but the full non-time-symmetric Penrose inequality, including the expected treatment of apparent horizons and extrinsic curvature, remains open. In that sense, the Riemannian Penrose inequality is both a completed theorem and a model problem: it provides the sharp scalar-curvature case against which charged, boundary, weighted, and fully dynamical extensions continue to be measured (0906.5566).