Solution-Generating Techniques

Almost every closed-form metric in the exact-solutions catalogue was found the same way: not by attacking Einstein's ten coupled nonlinear field equations head-on, but by first spending symmetry to buy simplicity. The decisive fact is that a spacetime with two commuting Killing vectors — a timelike $\xi=\partial_t$ (stationarity) and a spacelike $\eta=\partial_\phi$ with closed orbits (axisymmetry) — has enough symmetry that its vacuum equations descend to a field theory on a two-dimensional quotient, the Weyl half-plane spanned by $\rho$ and $z$ . In 1968 F. J. Ernst showed that on that quotient the entire vacuum system collapses into a single complex nonlinear equation for one potential $\mathcal{E}=f+i\omega'$ , packaging the gravitational norm and the frame-dragging twist into one object. The payoff is not just compression. The Ernst equation belongs to the rare class of completely integrable nonlinear PDEs — the family of KdV, sine-Gordon, and the nonlinear Schrödinger equation. Integrability means the equation carries hidden scaffolding: a Lax pair whose compatibility reproduces the field equations, an infinite-dimensional symmetry (the Geroch group, an affine $SL(2,\mathbb{R})$ action), Bäcklund transformations, and a full inverse-scattering method. Belinski and Zakharov turned that last piece into a machine that dresses a simple seed metric with solitons — and out drops the Kerr black hole as a two-soliton solution on flat space. A black hole, in this language, is a soliton; its spin is the imaginary part of a pole. This article develops the reduction, the equation, and the generating transformations, and marks where integrability ends and numerical relativity takes over.

Key takeaways

Two commuting Killing vectors reduce Einstein's vacuum equations to the single complex Ernst equation $(\operatorname{Re}\mathcal{E})\nabla^2\mathcal{E}=\nabla\mathcal{E}\cdot\nabla\mathcal{E}$ for one potential $\mathcal{E}=f+i\omega'$ on a 2D half-plane.
The reduced system is a $SL(2,\mathbb{R})/SO(2)$ sigma model and completely integrable: Lax pair, infinite-dimensional Geroch group, Bäcklund transformations, inverse scattering.
New exact solutions are generated algebraically from known ones — Ehlers and Harrison transformations, HKX and Belinski–Zakharov soliton dressing — not by solving PDEs afresh.
Kerr is the simplest nontrivial Ernst solution, $\mathcal{E}=(px+iqy-1)/(px+iqy+1)$ with $p^2+q^2=1$ , and equivalently a two-soliton Belinski–Zakharov solution on Minkowski; Kerr–Newman follows by an $SU(2,1)$ charging transformation.
The techniques work only in highly symmetric geometries — stationary black holes, cylindrical and colliding plane waves, Gowdy cosmologies — and fail for generic 3+1 dynamics like binary mergers, which demand numerical evolution.
Key idea: two commuting Killing vectors collapse Einstein's vacuum equations to the integrable Ernst equation for a single complex potential, and integrability (Lax pair, Geroch group, Bäcklund and inverse-scattering) then lets one generate new exact metrics from old ones — with solitons corresponding to black holes.

The core insight

Two commuting Killing vectors collapse Einstein's vacuum equations to the integrable Ernst equation for a single complex potential; the integrable structure (Lax pair, Geroch group, Bäcklund and inverse-scattering transformations) then generates new exact solutions algebraically from known ones, with solitons corresponding to black holes.

Where this lives today

The most consequential modern role of these techniques is as the exact backbone against which numerical relativity is checked. Codes that evolve binary mergers — the LIGO–Virgo–KAGRA waveform pipelines depend on them — are first validated by evolving a single Kerr black hole and confirming it stays Kerr, with the exact Ernst/soliton solution as ground truth. Event Horizon Telescope images of M87 and Sgr A are interpreted by fitting Kerr shadows, and tests of the "Kerr hypothesis" measure the multipole moments $M_\ell+iS_\ell$ whose generating structure comes straight from the Ernst formalism. The same reduction underwrites the black-hole uniqueness theorems: Robinson's vacuum proof and the Mazur–Bunting electrovac proofs are, at heart, statements that the Ernst equation with black-hole boundary conditions has a unique solution. And the method leapt dimensions: Pomeransky's higher-dimensional Belinski–Zakharov construction generates the five-dimensional black rings of Emparan and Reall — "a black hole is a soliton" beyond four dimensions.

The physical picture

Start with the symmetry and follow it. Two commuting Killing vectors, $\xi=\partial_t$ and $\eta=\partial_\phi$ , mean the metric does not depend on $t$ or $\phi$ ; those coordinates can be "integrated out." What is left depends only on the remaining two, which in vacuum can always be brought to Weyl canonical form $(\rho,z)$ — a flat half-plane with $\rho\ge0$ . Ten coupled nonlinear equations for a four-dimensional metric become a small system for a couple of scalar fields there, and Ernst's stroke was to show the couple is really one complex field. The first figure is this collapse.

Figure

The two-Killing-vector reduction. A stationary axisymmetric spacetime with commuting ξ = ∂t and η = ∂φ is quotiented to the Weyl half-plane ρ ≥ 0, on which the ten coupled vacuum equations R_ab = 0 become one complex Ernst equation for the single potential E = f + iω′

On that half-plane the two real fields are the norm potential $f=-\xi\cdot\xi$ (roughly the redshift $\xi$ measures) and the twist potential $\omega'$ (encoding frame dragging — the failure of $\xi$ to be hypersurface-orthogonal). They are not independent scalars on a flat sheet; they are coordinates on a curved target, the hyperbolic half-plane $SL(2,\mathbb{R})/SO(2)$ , and the field equations make the map from $(\rho,z)$ into it harmonic. That is precisely a nonlinear sigma model, and 2D sigma models with this symmetric-space target are the archetype of integrability — the same reason the principal chiral model is solvable. Integrability is therefore not a gift of gravity; it is a property the two-Killing-vector reduction shares with a family of solvable field theories. Once the system is integrable, solutions come in orbits: from flat space you can reach Schwarzschild, Kerr, Kerr–NUT and the Tomimatsu–Sato tower by symmetry transformations (Ehlers, Harrison) or by dressing a seed with Belinski–Zakharov solitons. A single real pole builds a Schwarzschild-like horizon; a complex-conjugate pair adds rotation and gives Kerr; several poles give multi-black-hole strings held apart by conical struts. The second figure maps this generating web, with concrete Kerr parameters attached.

Figure

The solution-generating web. Symmetry and soliton transformations move you between exact vacuum and electrovac metrics, starting from flat space. For Kerr with spin a/M = 0.6 the Ernst parameters are p = 0.8, q = 0.6 with p² + q² = 1, rod half-length κ = √(M² − a²) = 0.8M, and outer horizon r₊ = M + κ = 1.8M

The core equations

The reduction begins with the most general stationary, axisymmetric, orthogonally-transitive vacuum metric in Weyl–Lewis–Papapetrou form. Two of the four functions ( $f$ , $\omega$ ) carry the physics; $\gamma$ is fixed afterward by a line integral, and $\rho,z$ are flat canonical coordinates.

\mathrm{d}s^2=-f(\mathrm{d}t-\omega\,\mathrm{d}\phi)^2+f^{-1}\big[e^{2\gamma}(\mathrm{d}\rho^2+\mathrm{d}z^2)+\rho^2\,\mathrm{d}\phi^2\big] \qquad (1)

where: $f$ — $=-\xi_a\xi^a$ , the norm (gravitational) potential of the timelike Killing vector; $\omega$ — off-diagonal frame-dragging potential, tied to the twist; $\rho,z$ — Weyl canonical coordinates on the 2D quotient; $\gamma$ — conformal factor, obtained by quadrature once $f,\omega$ are known. Reading it: the whole metric is fixed by two scalar fields on the half-plane, and a black hole horizon is a finite rod on the axis $\rho=0$ . Ernst's equation is what those fields obey once the twist $\omega$ is traded for a scalar potential $\omega'$ (legal in vacuum) and the pair complexified as $\mathcal{E}=f+i\omega'$ . Ten coupled equations become one.

(\operatorname{Re}\mathcal{E})\,\nabla^2\mathcal{E}=\nabla\mathcal{E}\cdot\nabla\mathcal{E},\qquad \mathcal{E}=f+i\,\omega' \qquad (2)

where: $\mathcal{E}$ — complex Ernst potential, $\operatorname{Re}\mathcal{E}=f$ , $\operatorname{Im}\mathcal{E}=\omega'$ ; $\nabla,\nabla^2$ — flat gradient and Laplacian in cylindrical $(\rho,z,\phi)$ space. Reading it: a single, Möbius-covariant, nonlinear — and crucially integrable — equation. Everything downstream either writes its solutions or is a symmetry moving between them. Split into real and imaginary parts, equation (2) is a harmonic-map (sigma-model) system with hyperbolic target — the form exposing the symmetry group.

f\,\nabla^2 f=(\nabla f)^2-(\nabla\omega')^2,\qquad \nabla\cdot\!\big(f^{-2}\nabla\omega'\big)=0 \qquad (3)

where: $f,\omega'$ — coordinates on the target $SL(2,\mathbb{R})/SO(2)\cong\mathbb{H}$ with metric $(\mathrm{d}f^2+\mathrm{d}\omega'^2)/f^2$ . Reading it: the second equation conserves the twist (angular-momentum) potential; the isometries of $\mathbb{H}$ are the Ehlers transformations, seed of the full Geroch group. The simplest nontrivial solution is Kerr. In prolate spheroidal coordinates $(x,y)$ the Ernst potential is a Möbius function of a linear complex coordinate, with $p,q$ (tuned by $p^2+q^2=1$ ) carrying mass and spin.

\mathcal{E}=\frac{\zeta-1}{\zeta+1},\qquad \zeta=p\,x+i\,q\,y,\qquad p^2+q^2=1 \qquad (4)

where: $x\in[1,\infty)$ — radial-like prolate coordinate; $y\in[-1,1]$ — angular-like; $p=\sqrt{M^2-a^2}/M$ — mass parameter; $q=a/M$ — rotation parameter. Reading it: $q=0$ makes $\mathcal{E}$ real (Schwarzschild); $q\ne0$ generates the angular-momentum twist, and $p\to0$ ( $a\to M$ ) is extremal Kerr. Integrability is made explicit by a Lax pair — a linear system for a matrix $\Psi$ with a spacetime-varying spectral parameter $\lambda$ , whose compatibility is the Ernst equation.

\partial_\zeta\Psi=\frac{A}{\lambda-w}\,\Psi,\qquad \partial_{\bar\zeta}\Psi=\frac{B}{\lambda+\bar w}\,\Psi \qquad (5)

where: $\Psi(\lambda)$ — generating matrix; $\lambda$ — variable spectral parameter, tied to $\rho,z$ through $w=z+i\rho$ ; $A,B$ — matrices built from the seed potential. Reading it: the zero-curvature condition $\partial_{\bar\zeta}\partial_\zeta\Psi=\partial_\zeta\partial_{\bar\zeta}\Psi$ holds for all $\lambda$ iff the field equations hold — exactly what "integrable" means, and what opens the inverse-scattering method. The Belinski–Zakharov inverse-scattering method dresses a seed by adding poles in $\lambda$ ; each pole trajectory $\mu_k$ solves a quadratic tied to an axis point $z_k$ , with residues fixed by the seed.

\chi=\mathbb{1}+\sum_{k}\frac{R_k}{\lambda-\mu_k},\qquad \mu_k^{2}+2(z-z_k)\,\mu_k-\rho^{2}=0 \qquad (6)

where: $\chi$ — dressing matrix acting on the seed as $g_{\rm new}=\chi\,g_{\rm seed}\,\chi^{T}$ (up to a determinant); $\mu_k$ — pole trajectory (soliton position); $R_k$ — residue matrices; $z_k$ — axis anchor of soliton $k$ . Reading it: one real pole is a Schwarzschild-like mass; a complex-conjugate pair is Kerr; $n$ poles give $n$ -object configurations. The same machinery survives coupling to Maxwell: the electrovac Ernst pair adds a complex potential $\Phi$ , and the symmetry group enlarges from $SL(2,\mathbb{R})$ to $SU(2,1)$ .

(\operatorname{Re}\mathcal{E}-|\Phi|^{2})\,\nabla^{2}\mathcal{E}=(\nabla\mathcal{E}+2\bar\Phi\,\nabla\Phi)\cdot\nabla\mathcal{E} \qquad (7)

where: $\mathcal{E}$ — electrovac potential, $\mathcal{E}=f-|\Phi|^{2}+i\omega'$ ; $\Phi$ — electromagnetic Ernst potential, $\Phi=A_t+i\tilde A$ . Reading it: $\Phi$ obeys a companion equation with the same operator, and a Harrison ( $SU(2,1)$ ) transformation charges the Kerr seed into Kerr–Newman.

Derivation AFrom the vacuum field equations to the Ernst equation

The goal is to show, without hand-waving, how the ten components of $R_{ab}=0$ for a stationary axisymmetric spacetime funnel down to the single complex equation (2). The plan is to introduce the twist as a one-form, prove it is closed in vacuum so it comes from a scalar potential, write the two real equations for the norm and twist potentials, and complexify them into one — seeing why the conformal factor $\gamma$ is free labor once $\mathcal{E}$ is known. Take $\xi=\partial_t$ Killing, define $f=-\xi_a\xi^a$ and the twist one-form $\omega_a=\epsilon_{abcd}\xi^b\nabla^c\xi^d$ .

Step 1. The curl of the twist obeys $\nabla_{[a}\omega_{b]}=\epsilon_{abcd}\xi^c R^d{}_e\xi^e$ , whose right side vanishes when $R_{ab}=0$ ; hence $\mathrm{d}\omega=0$ and there is a scalar $\omega'$ with $\omega_a=\nabla_a\omega'$ . Why: this replaces the metric's off-diagonal function by a genuine scalar potential, trading a gauge-laden $\omega$ for an honest field.

Step 2. Projecting $R_{ab}=0$ along and across the Killing orbit, on the flat 3-metric $\mathrm{d}\rho^2+\mathrm{d}z^2+\rho^2\mathrm{d}\phi^2$ , gives the norm equation $f\nabla^2 f=(\nabla f)^2-(\nabla\omega')^2$ . Why: this fixes the redshift function $f$ and already shows the twist feeding back on the norm — rotation gravitates.

Step 3. The remaining content is the conservation law $\nabla\cdot(f^{-2}\nabla\omega')=0$ , equivalently $f\nabla^2\omega'=2\,\nabla f\cdot\nabla\omega'$ . Why: this is the angular-momentum equation, linear in $\omega'$ at fixed $f$ .

Step 4. Set $\mathcal{E}=f+i\omega'$ . Multiplying the norm equation by $f$ and adding $i$ times the twist equation assembles $f\nabla^2 f-(\nabla f)^2+(\nabla\omega')^2+i[f\nabla^2\omega'-2\nabla f\cdot\nabla\omega']=0$ , exactly $(\operatorname{Re}\mathcal{E})\nabla^2\mathcal{E}-\nabla\mathcal{E}\cdot\nabla\mathcal{E}=0$ . Why: the messy real system is the real and imaginary parts of one complex equation — Ernst's actual discovery, and what makes the Möbius symmetry visible.

Step 5. With $\mathcal{E}$ solved, the conformal factor follows from $\partial_\zeta\gamma=2\rho\,\partial_\zeta\mathcal{E}\,\partial_\zeta\bar{\mathcal{E}}/(\mathcal{E}+\bar{\mathcal{E}})^2$ , whose integrability condition is the Ernst equation itself, so $\gamma$ is a line integral, not a new PDE. Why: this is why "solve for $\mathcal{E}$ " means "solve the whole problem" — the last metric function costs only a quadrature.

Go deeper · Derivation B · 1 minThe Ehlers transformation that turns Schwarzschild into Taub–NUT

Derivation BThe Ehlers transformation that turns Schwarzschild into Taub–NUT

The goal is to exhibit a concrete, finite element of the Geroch group at work: a real Möbius map on $\mathcal{E}$ that leaves the Ernst equation invariant yet changes the physics, manufacturing a gravomagnetic (NUT) charge out of a purely static seed. The plan is to state the invariance, apply it to the Schwarzschild potential, and read the new asymptotics. This is the gravitational analogue of a duality rotation, the elementary building block from which the whole infinite-dimensional symmetry is generated.

Step 1. The Ernst equation is invariant under $\mathcal{E}\mapsto(\alpha\mathcal{E}+i\beta)/(i\gamma\mathcal{E}+\delta)$ with $\alpha\delta-\beta\gamma=1$ — the $SL(2,\mathbb{R})$ isometry group of the hyperbolic target. Why: the equation is a harmonic map into $\mathbb{H}$ , so any isometry of $\mathbb{H}$ maps solutions to solutions for free.

Step 2. Take the Schwarzschild seed $\mathcal{E}=(x-1)/(x+1)$ , which is real, so $\omega'=\operatorname{Im}\mathcal{E}=0$ — pure mass, no twist. Why: a static seed is the cleanest start, and any imaginary part appearing later is manufactured entirely by the symmetry.

Step 3. Apply the Ehlers element $\mathcal{E}\mapsto\mathcal{E}/(1+ic\,\mathcal{E})$ , giving $\mathcal{E}'=\dfrac{(x-1)/(x+1)}{1+ic\,(x-1)/(x+1)}$ , now with nonzero imaginary part everywhere. Why: a twist has switched on without touching the mass — the solution has moved along its Geroch orbit to a different spacetime.

Step 4. In the asymptotics, $\operatorname{Im}\mathcal{E}'$ behaves like a gravomagnetic monopole with NUT parameter $\ell\propto c$ ; the result is the Taub–NUT metric, mass unchanged. Why: a "magnetic" mass with no Newtonian analogue, produced purely by a symmetry — solution-generating is physically creative, not cosmetic.

Go deeper · Derivation C · 1 minReading the mass and spin from the Kerr Ernst potential

Derivation CReading the mass and spin from the Kerr Ernst potential

The goal is to confirm that the compact potential (4) really is Kerr, and to extract $M$ , $a$ , $J$ so the abstract $p,q$ acquire meaning. The plan is to fix the prolate scale $\kappa$ , verify the potential solves the Ernst equation, then expand $f$ and $\omega$ at large radius to read off the Komar mass and angular momentum — closing the loop between "simplest nontrivial Ernst solution" and "the rotating black hole of astrophysics." Use $\mathcal{E}=(\zeta-1)/(\zeta+1)$ with $\zeta=px+iqy$ , $p^2+q^2=1$ .

Step 1. The prolate coordinates are $\rho=\kappa\sqrt{(x^2-1)(1-y^2)}$ and $z=\kappa x y$ , so $\kappa$ sets the half-length of the source rod, from $z=-\kappa$ to $z=+\kappa$ . Why: a black hole horizon is a finite rod on $\rho=0$ , its length the geometric fingerprint of $M$ and $a$ together.

Step 2. Because $\zeta=px+iqy$ solves the flat axisymmetric Laplace equation and $\mathcal{E}$ is a Möbius function of it, direct substitution confirms $\mathcal{E}$ satisfies the Ernst equation (2). Why: the shortcut integrability provides — a linear seed, wrapped in a Möbius map, solves the full nonlinear equation.

Step 3. Expanding the norm at large $r$ gives $f\to1-2M/r+\cdots$ with $M=\kappa/p$ , the Komar mass. Why: the $1/r$ falloff of the redshift function defines mass and pins $\kappa=pM=\sqrt{M^2-a^2}$ .

Step 4. The frame-dragging potential expands as $\omega\to-2J\sin^2\theta/r+\cdots$ with $J=Ma$ , $a=qM$ ; the constraint $p^2+q^2=1$ is exactly $\kappa^2=M^2-a^2$ , and $p\to0$ is extremal $a\to M$ where the rod shrinks to a point. Why: this fixes the spin and recovers the Boyer–Lindquist relations $J=Ma$ , $r_\pm=M\pm\sqrt{M^2-a^2}$ , extremal degeneration inside the same formula.

Limiting cases

The reach of these methods is set entirely by how much symmetry the spacetime has; the table marks where integrability holds and where it fails.

Case	Result	Meaning
Two commuting Killing vectors	$[\xi,\eta]=0$	The 2D reduction and Ernst equation exist. Stationary-axisymmetric, cylindrical, plane-wave, Gowdy geometries qualify; generic ones do not.
Vacuum matter	$R_{ab}=0$	Pure Ernst equation (2), group $SL(2,\mathbb{R})$ (Geroch). Kerr, Kerr–NUT, Tomimatsu–Sato, double-Kerr live here.
Electrovac matter	$R_{ab}=2\,T^{\rm EM}_{ab}$	Integrability survives for the Ernst pair $(\mathcal{E},\Phi)$ ; group enlarges to $SU(2,1)$ . Kerr–Newman is charged Kerr.
Orthogonal transitivity	$\xi_{[a}\eta_b\nabla_c\xi_{d]}=0$	Off-orbit 2-planes must be integrable — automatic in vacuum/electrovac (circularity), broken by generic matter such as $\phi$ -currents.
Fewer than two symmetries	$N_{\rm Killing}<2$	No Lax pair, no Geroch group, no solitons. Inspirals and asymmetric collapse must be evolved numerically (ADM/BSSN, generalized harmonic, excision/punctures).

Common trapIntegrability is a property of the reduction, not of general relativity itself

It is tempting, after seeing Kerr fall out as a two-soliton solution, to imagine Einstein's equations are "an integrable system." They are not. Integrability is a feature of the two-commuting-Killing-vector reduction — drop below two symmetries and the Lax pair, the Geroch group, and the entire soliton construction vanish; generic 3+1 evolution is thoroughly non-integrable. A binary black hole merger has generically no continuous symmetry at all, so no Bäcklund or inverse-scattering trick will ever write its metric in closed form. A second trap: the imaginary part of $\mathcal{E}$ is the twist potential $\omega'$ , not a metric component — reading $\operatorname{Im}\mathcal{E}$ as an off-diagonal metric entry is a common and costly mistake.

Worked examples

1 · Kerr parameters and horizons from a measured spin (fully worked)

An astrophysical black hole has dimensionless spin $a_\ast\equiv a/M=0.6$ . Work out its Ernst parameters $p,q$ , source-rod length $\kappa$ , and horizon radii in units $G=c=1$ with $M$ symbolic, and check $p^2+q^2=1$ .

Step 1. The rotation parameter is read straight off: $q=a/M=\mathbf{0.6}$ . The mass parameter is $p=\sqrt{M^2-a^2}/M=\sqrt{1-a_\ast^2}=\sqrt{0.64}=\mathbf{0.8}$ . These are the two numbers that go into the Ernst potential.

Step 2. Check the constraint of equation (4): $p^2+q^2=0.64+0.36=\mathbf{1}$ , as required. This is the identity $\kappa^2=M^2-a^2$ in disguise, since $p^2=(M^2-a^2)/M^2$ and $q^2=a^2/M^2$ .

Step 3. The prolate rod scale is $\kappa=pM=\sqrt{M^2-a^2}=0.8M$ , so the horizon rod runs from $z=-0.8M$ to $z=+0.8M$ , total length $2\kappa=\mathbf{1.6\,M}$ ; the extremal case $a\to M$ shrinks it to a point.

Step 4. The horizons sit at $r_\pm=M\pm\kappa=M\pm0.8M$ : outer $r_+=\mathbf{1.8\,M}$ , inner $r_-=\mathbf{0.2\,M}$ . The angular momentum is $J=Ma=0.6\,M^2$ , and the potential is $\mathcal{E}=(0.8x+0.6\,iy-1)/(0.8x+0.6\,iy+1)$ . So $a_\ast=0.6$ maps to $p=0.8$ , $q=0.6$ , $\kappa=0.8M$ , $r_+=1.8M$ , $r_-=0.2M$ .

2 · Where a Belinski–Zakharov soliton pole sits

A soliton is anchored at $z_k=0$ on a flat seed. Using the pole equation in (6), locate its two pole trajectories $\mu$ at $(\rho,z)=(3,4)$ (units of $M$ ), and show what happens as the point approaches the axis.

Step 1. With $z_k=0$ the pole equation is $\mu^2+2z\,\mu-\rho^2=0$ , i.e. $\mu^2+8\mu-9=0$ . Solving, $\mu=\dfrac{-8\pm\sqrt{64+36}}{2}=\dfrac{-8\pm10}{2}$ , giving $\mu=\mathbf{+1}$ and $\mu=\mathbf{-9}$ . A soliton contributes two pole branches — the two roots of one quadratic.

Step 2. Sanity-check by Vieta: $\mu_1\mu_2=(1)(-9)=-9=-\rho^2$ and $\mu_1+\mu_2=-8=-2z$ , matching the coefficients exactly. The two poles are not independent; they are locked together by the seed geometry.

Step 3. On the axis $\rho\to0$ the equation degenerates to $\mu(\mu+2z)=0$ , so $\mu\to0$ or $\mu\to-2z$ ; the trajectory reaching $\mu=0$ is where the pole touches the axis — an endpoint of a horizon rod. A real pole at $(3,4)$ sits at $\mu=1$ or $\mu=-9$ ; on axis the poles pin to rod endpoints, which is how BZ solitons build horizons.

Physics of the portrait

Behind the whole apparatus is one deceptively simple move — packaging two real gravitational potentials as a single complex number — which reorganized the exact-solutions field around integrable systems.

Origins · 1 minFrederick J. Ernst and how this was discovered

Frederick J. Ernst

1933–2023 · complex-potential reformulation of stationary axisymmetric gravity

Frederick J. Ernst was an American mathematical physicist who, in two 1968 papers in the Physical Review, changed how the general-relativity community thought about exact solutions. Trained at Princeton and working for much of his career at the Illinois Institute of Technology, Ernst asked a narrow-looking question — what is the cleanest form of the stationary axisymmetric vacuum equations? — and found an answer with enormous reach. His insight was to combine the norm potential $f$ and twist potential $\omega'$ into a single complex field $\mathcal{E}=f+i\omega'$ , whereupon the coupled real system collapses to the one complex equation that now bears his name, and Kerr and Tomimatsu–Sato become transparent Möbius functions of a linear seed. The deeper reward, harvested by Geroch, Kinnersley, Belinski, and Zakharov through the 1970s, was that the Ernst equation is completely integrable, carrying a Lax pair, an infinite-dimensional symmetry group, Bäcklund transformations, and inverse scattering. Ernst extended the framework to electrovacuum, coupling $\mathcal{E}$ to an electromagnetic potential $\Phi$ and enlarging the symmetry to $SU(2,1)$ , which put Kerr–Newman within algebraic reach. He remained active in the exact-solutions community until his death in 2023.

History

1917

— Weyl writes the static axisymmetric vacuum metric in canonical ρ, z coordinates

1963

— Kerr finds the rotating black hole metric by algebraic Petrov type-D methods

1968

— Ernst recasts stationary axisymmetric vacuum as one complex equation for E

1972

— Geroch conjectures the infinite-dimensional symmetry group of the reduced equations

1978

— Belinski and Zakharov give the inverse-scattering soliton construction, Kerr as a two-soliton

Going further

Texts. Misner, Thorne & Wheeler — Gravitation (1973), §33 · Wald — General Relativity (1984), Chs. 7, 12 · Stephani, Kramer, MacCallum, Hoenselaers & Herlt — Exact Solutions of Einstein's Field Equations (2nd ed., 2003), Chs. 19–20, 34 · Belinski & Verdaguer — Gravitational Solitons (2001) · Griffiths — Colliding Plane Waves in General Relativity (1991).

Provenance. The reduction and Ernst equation are standard; primary sources are Ernst, Phys. Rev. 167, 1175 and 168, 1415 (1968), Geroch, J. Math. Phys. 12, 918 (1971) and 13, 394 (1972), and Belinski & Zakharov, Sov. Phys. JETP 48, 985 (1978). A verifier should confirm: (1) the Weyl–Lewis–Papapetrou reduction against Stephani et al. §19; (2) the Ernst form (2) and Kerr potential $\mathcal{E}=(px+iqy-1)/(px+iqy+1)$ with $\kappa^2=M^2-a^2$ , $J=Ma$ , $r_\pm=M\pm\sqrt{M^2-a^2}$ against §20; (3) the arithmetic $a_\ast=0.6\Rightarrow p=0.8,q=0.6,\kappa=0.8M,r_+=1.8M,r_-=0.2M$ ; (4) the BZ pole $\mu^2+2(z-z_k)\mu-\rho^2=0$ giving $\mu=1,-9$ at $(\rho,z)=(3,4)$ , $z_k=0$ . All numbers are textbook Kerr values.

Prerequisites. Killing Vectors and Spacetime Symmetry · Kerr Metric · ADM 3+1 Decomposition of Spacetime · Initial-Value (Cauchy) Problem in General Relativity. Leads to. Numerical Relativity and Binary Black Hole Mergers · Black Hole Excision and Puncture Methods · Black Hole Uniqueness Theorems.

Ask the tutor: Ask me to walk through Kerr as a two-soliton Belinski–Zakharov solution, or to run an Ehlers transformation on a seed of your choice and read off the new charges.

Why package two real potentials as one complex number instead of just solving for f and ω′?

Because the complex packaging is what makes the symmetry visible and the equation single. The pair $(f,\omega')$ are coordinates on the hyperbolic half-plane $SL(2,\mathbb{R})/SO(2)$ , and the group that maps vacuum solutions to vacuum solutions — the Ehlers/Geroch group — acts there by Möbius transformations. In the variable $\mathcal{E}=f+i\omega'$ those become literal fractional-linear maps, so a symmetry that looks opaque in $(f,\omega')$ is a one-line substitution, and generating new solutions becomes algebra rather than fresh integration. It is the same reason electromagnetism is cleaner with $\mathbf{E}+i\mathbf{B}$ than with the two real fields kept apart.

In what sense is a black hole a soliton?

In the precise sense of soliton theory. The reduced Einstein equations are an integrable system, and integrable systems build localized, stable solutions by adding poles to a scattering matrix — those poles are solitons. In the Belinski–Zakharov method each pole on a flat seed is a horizon-bearing object: one real pole gives a Schwarzschild-like mass, a complex-conjugate pair adds angular momentum to give Kerr. So "black hole = soliton" is not a loose metaphor; the Kerr metric is literally a two-soliton solution on Minkowski, its mass in the pole positions and spin in their imaginary part. Stacking more poles builds multi-black-hole configurations, which generically need conical struts to keep from collapsing together.

What exactly is the Geroch group, and is it really a symmetry?

The Geroch group is the infinite-dimensional symmetry group of the reduced vacuum equations. Its finite part is the $SL(2,\mathbb{R})$ of Ehlers transformations acting on $\mathcal{E}$ ; the full group is an affine Kac–Moody extension, realized by the Kinnersley–Chitre transformations. It is a genuine symmetry: every element maps a vacuum solution to another. Geroch conjectured it acts transitively on the space of such solutions — meaning every stationary axisymmetric vacuum spacetime is reachable from flat space by a group element — essentially established for the relevant sectors. That transitivity is the deep reason solution-generating works: the whole solution space is one orbit, and the transformations travel around it.

Do these techniques ever produce genuinely new physics, or just re-dress known metrics?

Both, and the line is subtle. Applied to a known seed, a symmetry transformation can produce a physically distinct spacetime — the Ehlers map turns Schwarzschild into Taub–NUT, adding a gravomagnetic charge with no Newtonian analogue, new physics from a symmetry. Soliton superposition produces genuinely new multi-black-hole and double-Kerr solutions whose equilibrium properties (the necessity of conical struts, the impossibility of static like-spin balance) were not obvious in advance and taught real lessons about the no-force problem. What they cannot do is escape their symmetry class: everything they generate still has two commuting Killing vectors. They are creative within the integrable sector and silent outside it.

How does the electrovac case differ, and why does the group grow to SU(2,1)?

Adding a Maxwell field introduces a second complex potential, the electromagnetic Ernst potential $\Phi=A_t+i\tilde A$ , alongside $\mathcal{E}=f-|\Phi|^2+i\omega'$ ; the two satisfy a coupled pair of Ernst equations that remain integrable. The sigma-model target enlarges from $SL(2,\mathbb{R})/SO(2)$ to $SU(2,1)/S(U(2)\times U(1))$ , so the solution-generating group grows from $SL(2,\mathbb{R})$ to $SU(2,1)$ . The extra generators are the Harrison transformations, which "charge up" a vacuum seed: applied to Kerr they produce Kerr–Newman. The whole vacuum machinery — Lax pair, Geroch group, inverse scattering — carries over, which is why charged rotating black holes are as algebraically accessible as uncharged ones.

Where precisely does this hand off to numerical relativity?

At the boundary marked by symmetry. Solution-generating techniques require two commuting Killing vectors; the instant a problem has fewer — a precessing binary, asymmetric collapse, a generic cosmology — the Lax pair and Geroch group evaporate and no closed form exists. That is the regime numerical relativity was built for: the ADM and BSSN formulations, generalized harmonic gauge, and excision or puncture methods evolve the full 3+1 equations with no symmetry assumed. The two are complementary. The analytic side supplies exact benchmarks (Kerr initial data, asymptotics, waveform cross-checks) that validate the codes; the numerical side handles the generic strong-field dynamics the analytic side provably cannot touch. Modern gravitational-wave astronomy uses both at once.