Electromagnetic tails and the Kerr–Newman Penrose inequality

Proof. Let X=(x^1,x^2,x^3), T=(-x^2,x^1,0), and w=(x^1)^2+(x^2)^2, with \flat denoting Euclidean metric dual in the following formulas. The primitives and tensor have the Cartesian expressions

These are smooth on r\ge1, with no denominator involving distance to the axis. They also show rotation invariance and, since X\cdot T=0, trace-freeness. The exact flux forms are closed, so \dd(\iota_{\widetilde E}\dd V_\delta)= (\operatorname{div}_\delta\widetilde E)\dd V_\delta=0, and likewise for \widetilde B.

For the momentum identity we compute off the axis and then use smoothness. Since \dd V_\delta=r^2\sin\theta\,\dd r\wedge\dd\theta\wedge\dd\phi, the spherical coordinate components are

In particular an orthonormal polar component is r times the displayed polar component. The covector \dd V_\delta(\widetilde E,\widetilde B,\cdot) has only an azimuthal component, namely

To check all three tensor-divergence components, put a=\sigma_{r\phi}=-s^2c^2\sin^4\theta/r^2 and use the coordinate identity

The last term vanishes: the metric is diagonal in these coordinates, whereas \sigma^{jk} has only r\phi and \phi r entries. Thus

This proves (2.2). The last expression also proves the stated support property; for r\ge2L both fields are radial and parallel.

Finally, for j\ge1 the Cartesian derivatives of c obey

Indeed, all such derivatives are supported where r is comparable to L, and differentiating \chi(r/L) gives this bound. The coefficients of the two uncut primitive one-forms in (2.4) are smooth homogeneous functions of degree -1; exterior differentiation therefore gives the field bounds in (2.3). The uncut Cartesian tensor coefficients are smooth homogeneous functions of degree -3, which gives the tensor bounds by the same product rule. Smooth cutoff jets give the estimates also at r=L and r=2L.□

Proof. Since \dd V_g=u^6\dd V_\delta, the flux forms are unchanged:

Their closedness proves the two Maxwell divergence constraints. The cross-product convention gives, for any vector Z,

Here is the tensor calculation in Cartesian coordinates. Put b=2\log u, so g=e^{2b}\delta and K=e^{-b}\sigma. The connection difference is

with indices on the right raised by \delta. Symmetry and trace-freeness give

Consequently the derivative of e^{-b} cancels the second connection contraction, yielding

Together with Lemma 2.1, this is exactly the momentum constraint.

For completeness, substituting the same D into

gives derivative terms -b_{ij}-\delta_{ij}\Delta_\delta b and quadratic terms b_i b_j-\delta_{ij}|\nabla_\delta b|^2. Contracting therefore gives

The norm identities are

Thus the Hamiltonian constraint R_g-|K|_g^2=2(|\mathcal E|_g^2+|\mathcal B|_g^2) becomes (2.9), with the coefficients in (2.10). Those explicit expressions follow from (2.5) and |\sigma|_\delta^2=2a^2/(r^2\sin^2\theta). Their smoothness, nonnegativity, and derivative bounds follow either from their definitions as squared norms or directly from Lemma 2.1 and the product rule.□

Proof. The bound on H gives \|H\|_{L^1}\le4\pi a/L. The Newton kernel and its first derivatives are locally integrable. Localizing around a point and separating a small ball about its kernel singularity proves that the potential and its first derivatives are continuous; the singular part of the gradient integral is bounded by a constant times the ball’s radius. The fundamental-solution identity -\Delta(4\pi|x|)^{-1}=\delta_0, obtained by integration by parts around the pole, therefore gives the distributional equation by Fubini’s theorem.

For |x|\le L/2, source separation allows differentiation under the integral to every order and gives

For r=|x|\ge L/2, split the integral at |x-y|=r/2. In the inner part |y|\ge r/2, and for j=0,1 its absolute value is at most

The outer part is bounded by Cr^{-1-j}\|H\|_{L^1}\le CaL^{-1}r^{-1-j}. This proves (3.5).

Write I(x)=x/r^2. The Kelvin identity in dimension three is

Its right side vanishes on \Omega because |I(x)|\le1<L. Along each ray x=r\omega, direct differentiation gives

which proves the Robin condition including its angular dependence. Positivity follows from the two positive kernels. Combining (3.5) in the two radial regions gives the first two bounds in (3.6). For the last bound both potential arguments are separated from the source. More explicitly, put n(t)=N_H(t\omega); then

The separated estimates for j=0,1,2, together with the corresponding bound for \partial_r^2N_H(r\omega), give the claim.

For uniqueness of the homogeneous problem, let w be harmonic on \Omega, C^1 up to S, with w=O(r^{-1}) and w_r+\tfrac12w=0 on S. Extend it to 0<r<1 by r^{-1}w(I(x)). Values and tangential derivatives match at S, and the interior radial derivative there is -w-w_r=w_r. Thus the extension \widehat w is distributionally harmonic across S and hence harmonic on \R^3\setminus\{0\}. It is bounded near zero and is O(r^{-1}) at infinity. For sufficiently small \varepsilon and large R, the maximum principle applied to \pm\widehat w on \varepsilon<|x|<R gives |\widehat w(x)|\le C\varepsilon/|x|+C/R, with one constant C: the harmonic right side bounds the absolute value on both boundary spheres. Letting \varepsilon\downarrow0 and R\to\infty proves uniqueness.□

Proof. Fix a point x_0. Choose a smooth compactly supported cutoff \zeta equal to one near x_0, and keep this cutoff fixed when differentiating with respect to x. Writing G(x)=(4\pi|x|)^{-1}, for |\alpha|=j+1 choose \alpha=\beta+e_i, |\beta|=j. Distributional integration by parts gives, near x_0,

There are no boundary terms since \zeta H has compact support. The first integral is continuous by local integrability of \partial_iG and the same small-ball argument as above; the second is smooth near x_0 by source separation and integrability at infinity. The formula, also applied at lower orders, proves C^{j+1} regularity.

For the estimate at |x_0|=r\ge L/2, take \zeta supported in B(x_0,r/2) and equal to one on B(x_0,r/4), with |D^i\zeta|\le C_i r^{-i}. On its support every Leibniz term in D^\beta(\zeta H) is bounded by C_j a_jr^{-4-j}. The first integral in (3.7) at x_0 is therefore bounded by

The second is bounded by

These are precisely the claimed bounds at order j+1; the same argument gives the lower orders. On |x|\le L/2, differentiation of the separated integral already gives every derivative estimate.□

Proof of Proposition 3.1. The coefficient estimates from Lemmas 2.1 and 2.2 imply, for every j\ge0,

Let \mathcal C=\{v\in C_b(\Omega):v\ge0\}, a complete metric space in the sup norm. For v\in\mathcal C, put

The source is extended by zero to \R^3; this extension is continuous because it vanishes on r<L. Since u_0+v\ge1, the mean-value theorem for the negative powers gives

By Lemma 3.2, T_s maps the whole cone \mathcal C into itself, with

Choose L_0>4 so large that the second constant is at most 1/2 and the first, for s=1, is at most one. The contraction theorem now gives an actual fixed point v\in\mathcal C, unique in that cone. The operator commutes with rotation about the prescribed axis, so its unique fixed point is axisymmetric. It also gives v=0 when s=0. Lemma 3.2 yields (3.2), 1\le u_0\le u\le3, the Robin condition, and the scalar equation distributionally. The potential bounds give v=O(r^{-1}), hence u\to1.

For completeness we bootstrap both smoothness and the full decay. The fixed point is C^1 by Lemma 3.2; near S it is already smooth by separation of both potential arguments from the source. Suppose derivatives of u-1 through order j have the uniform bounds C_i r^{-1-i}. The case j=0 follows from (3.5) and u_0=1+r^{-1}. Because u\ge1, the chain and product rules with (3.8) give

The zero extension is C^j since the source vanishes near S. Lemma 3.3 gains one derivative for N_v with the stated bounds. For the reflected term, the separated estimates and the homogeneity of I(x) give at every order k

In the region 1\le r\le L/2, the ordinary term is bounded by C_ks^2L^{-2-k}; in the other region it is bounded by C_ks^2(r^{-2-k}+L^{-1}r^{-1-k}). Each bound is at most C_ks^2L^{-1}r^{-1-k} in its region. This proves the next induction step and (3.3) at every order, with constants independent of L,s. In particular u is smooth and solves (2.9) pointwise. If another solution has v\ge0 and v=O(r^{-1}), subtracting \mathcal RH_v leaves a decaying homogeneous harmonic Robin solution. Lemma 3.2 makes that difference zero; cone uniqueness then identifies the solutions.

Finally, fix L\ge L_0. The coefficients have the form p_s=s^4p_1, \ell_s=s^2\ell_1. Thus uniformly for v\in\mathcal C,

Using the contraction constant 1/2 in the fixed point equations gives

Combining this with the pointwise source Lipschitz bound yields

The right side is integrable with both weights 1 and |y|^{-1}. It proves continuity of the two source integrals, and (3.10) gives their quantitative bounds (3.4). Uniform convergence also gives continuity of the restriction u_s|_S, which will imply continuity of the horizon area. No finite first moment or multipole expansion has entered the argument.□

Proof. The domain is connected and oriented by the standard orientation of \R^3; its boundary is a connected two-sphere. Outside each sufficiently large ball it is a single product end. Smoothness up to S and across the rotation axis follows from Lemma 2.1 and Proposition 3.1; the constraints and maximality follow from Lemma 2.2. The seeds are invariant under the standard rotations, and uniqueness of the fixed point makes u invariant as well. Thus these rotations preserve g,K,\mathcal E,\mathcal B, with generator \eta=\partial_\phi of period 2\pi. The action is effective because a rotation fixing every point of \Omega has angle 0 modulo 2\pi.

The bound 1\le u\le M gives \delta\le g\le M^4\delta. The Euclidean intrinsic length metric on the closed exterior is complete: a Cauchy sequence has a Euclidean limit in \Omega, and local smooth boundary charts show convergence also in the intrinsic length metric. Comparison of curve lengths now proves completeness for g with S included. The derivative bounds in Lemma 2.1 and Proposition 3.1 and the product rule give (4.1).

The electrovacuum constraints identify

The fields vanish for r\le L and are radial, hence parallel, for r\ge2L. Consequently J_{\mathrm{con}} is supported in the compact annulus L\le r\le2L. Since \dd V_g=u^6\dd x\le M^6\dd x, both required densities are integrable. The horizon and enclosing-surface assertions are proved in Lemmas 4.2 and 4.3 below.□

Proof. The normal to S_r toward infinity is \nu_r=u^{-2}\partial_r. Since \sigma has only mixed radial–azimuthal components, K|_{TS_r\times TS_r}=0, including on the axis by smoothness. The mean curvature is therefore also the future outward expansion, and direct divergence gives

This formula permits angular dependence of u: the normal has no angular components. The Robin condition (3.1) gives W(1,\omega)=0. For r\ge2, (3.2) gives

On 1\le r\le2, the same estimates yield

Increasing L once, uniformly over s\in[0,1], we obtain

Thus the strict positivity holds even arbitrarily close to S.

Let \Sigma\subset\operatorname{int}\Omega be any compact smooth embedded enclosing surface, without a symmetry or connectedness restriction. To cover also the interpretation allowing components with interior edges, let \Sigma_{\mathrm{cl}} be the union of its components without boundary. The other components do not affect separation. Indeed, a path avoiding \Sigma_{\mathrm{cl}} can be perturbed to miss their edges and cross their interiors transversely at finitely many points. For each crossing, choose an embedded arc in that component from the crossing to an edge. A narrow tubular strip along the arc gives a detour on one side of the component, around the edge, and back on the other side, removing the crossing without introducing another. Compactness and disjointness of the components, and their location in \operatorname{int}\Omega, let every detour avoid the other components and S. Thus any path avoiding \Sigma_{\mathrm{cl}} can avoid all of \Sigma, so \Sigma_{\mathrm{cl}} still encloses S and is nonempty.

Choose p\in\Sigma_{\mathrm{cl}} where r has its global maximum R>1 on \Sigma_{\mathrm{cl}}. The increasing radial ray beyond p misses \Sigma_{\mathrm{cl}}. Starting just beyond p, the same detours past any edged components connect this side to infinity without meeting \Sigma. Hence the prescribed normal at p is \nu_\Sigma(p)=u(p)^{-2}\partial_r, and T_p\Sigma=T_pS_R. With the convention H_\Sigma=\operatorname{div}_\Sigma \nu_\Sigma, restriction of the ambient Hessian gives, at p,

For the last identity, the tangential Hessian of a defining function for a level surface equals its second fundamental form times the length of its gradient. It follows that H_\Sigma(p)\ge H_{S_R}(p)>0. Tangency also gives \tr_\Sigma K(p)=0, so \theta_+(\Sigma)(p)>0. This excludes precisely the condition \theta_+\le0 everywhere.□

Proof. Let \pi:\Omega\longrightarrow\Sph^2 be the radial projection \pi(r,\omega)=\omega, and let \dd A_{\Sph^2} denote the unit round area form. Define the smooth closed two-form

It is closed because every two-form on \Sph^2 is closed. Its comass, namely the supremum of |\beta(e_1,e_2)| over g-orthonormal pairs, is

Indeed, \beta annihilates radial vectors and its value on an oriented orthonormal angular pair is the displayed ratio. Furthermore,

by Lemma 4.2; hence ru(r,\omega)^2\ge u(1,\omega)^2. The comass estimate controls every tangent plane, so it does not require \Gamma to be a radial graph.

For completeness, take any connected smooth codimension-zero submanifold D\subset\Omega that is closed in \Omega, has interior contained in \operatorname{int}\Omega, contains the entire sufficiently distant part of the end, and has compact smooth embedded two-sided intrinsic boundary \Gamma. Choose R beyond that boundary and beyond the radius where D contains the entire end. Then

Near \Gamma the truncation leaves D unchanged, including at contacts with S; near S_R it is a regular level-set truncation in the interior of D. These boundary pieces are disjoint, so D_R has no corners. Here D_R inherits the orientation of \Omega, and its boundary is the full intrinsic boundary. In particular, any portion coinciding with S already belongs to \Gamma and contributes to this boundary integral. Stokes’ theorem with the induced boundary orientations gives

Using (4.5), including all components, we conclude

The choice D=\Omega is allowed and gives \Gamma=S, so the infimum equals A and is attained. No symmetry or finite-perimeter approximation has entered this comparison.

Finally, u(1,\omega)=2+v(1,\omega) with 0\le v(1,\omega)\le Cs^2/L^2. Since u\le M,

This proves (4.2). Supremum-norm continuity of v in s from Proposition 3.1 also proves continuity of A(s,L).□

Proof. The Cartesian decay K=O(r^{-3}), with \tau=0, makes the integrals in (1.5) of order O(r^{-1}). This proves the first part of (5.1).

The conformal weights leave both Maxwell flux forms unchanged. On every coordinate sphere, the electric flux is

The identical calculation for h gives zero magnetic flux. Equivalently, these are integrals of globally smooth exact two-forms over closed surfaces. The charge limits are therefore zero.

Substituting g_{ij}=u^4\delta_{ij} in (1.4) gives

Since u-1=O(R^{-1}) and u_r=O(R^{-2}), replacing u^3 by one has an integrated error O(R^{-1}). The source H is integrable and the ordinary Newton potential satisfies the exact flux identity

For the reflected term, smoothness of N at the origin gives

Together with u_0=1+1/r, these identities prove the existence of the limit and the exact formula (5.2). In particular no first-moment assumption on H or multipole expansion of the ordinary potential is needed. The estimates of Proposition 3.1 give N(0)=O(s^2/L^2), \int H=O(s^2/L), and all contributions in (5.2) are nonnegative. Since the linear momentum is zero, m=E_{\mathrm{ADM}}.

On r\ge2L the cutoff is one, and

Using just the electric field and u\le M,

As \int_{\Sph^2}\cos^2\theta\,\dd A=4\pi/3, this gives

The mass formula proves (5.3).

Finally \nu_r=u^{-2}\partial_r, \dd A_g=u^4\dd A_\delta, and K=u^{-2}\sigma, so all conformal weights in the angular-momentum integrand cancel. The gravitational flux at every coordinate sphere is

It is constant for r\ge2L, proving (5.4) and existence of the prescribed limit. The two nonzero angular polynomials in (5.6), multiplied by the positive factor u^{-6}, also prove the final assertion.□

Proof. Since A_B is invariant and A_B(\eta)=h, Cartan’s identity gives

Using \dd A_B=\iota_{\mathcal B}\dd V_g and evaluating on \mathcal E gives

The symmetry of K and the Killing equation imply

Combining this with \operatorname{div}_g\mathcal E=0 proves (5.8). Directly from the flux form,

Now apply (5.7).□

Proof. Write the uniform area estimate as

The lower mass bound (5.3) and A\ge64\pi imply

Choose L above the thresholds for the previous propositions, (5.12), and positivity of the quadratic coefficient on the right of (5.13). For this fixed L, the right side is positive at sufficiently small nonzero s.

The fixed point depends continuously on s in the supremum norm. Restriction to S therefore gives continuity of A, and the integrable source bounds give continuity of N(0) and \int H. The mass formula and (5.4) show that D(s,L) is continuous. Its positive value at a nonzero small parameter and its negative value at s=1 yield a root s_*\in(0,1) by the intermediate value theorem.

At this root, Q_e=Q_b=0, J\ne0, and both electromagnetic fields are nonzero. The strict branch and the equality equation give

Thus the Kerr–Newman member with these parameters would be a subextremal zero-charge member, namely Kerr with zero Maxwell tensor. Indeed the dyonic Kerr–Newman potential is linear in its electric and magnetic charges; see [CKST18, Equation (3)]. When both vanish, F=0. Every spacelike slice then has

independently of the slice or its future unit normal. This contradicts the nonzero constructed fields, and proves the asserted obstruction.□

Proof of Theorem 1.1. Lemmas 2.1 and 2.2 and Propositions 3.1 and 4.1 provide smooth maximal electrovacuum data with every topological, asymptotic, boundary, outermostness, and area condition in Section 1.1. Proposition 5.1 supplies the finite ADM and charge limits, positive mass and zero linear momentum. The physical branch is strict by (5.10). For s=1 and sufficiently large finite L, (5.12) gives the first conclusion. For the same large L, Proposition 5.3 gives the second.□