Initial-Value (Cauchy) Problem in General Relativity

The Cauchy problem is not a piece of mathematical hygiene tucked away in a textbook appendix — it is the operating system of computational gravity. Every gravitational waveform in the LIGO–Virgo–KAGRA template banks, beginning with GW150914 in 2015, is produced by a numerical-relativity code that follows the initial-value recipe to the letter: lay constraint-satisfying data (Bowen–York or puncture extrinsic curvature solving the momentum constraint, conformal thin-sandwich or Brill–Lindquist data solving the Hamiltonian constraint) on a first Cauchy slice, then march it forward through a foliation using the generalized-harmonic or BSSN reformulation, with the lapse and shift chosen adaptively to dodge the singularities. Well-posedness is what guarantees those simulations converge to a single answer as the grid is refined; constraint propagation is why "constraint violation" is the health monitor every such code watches. At the theory frontier the same framework defines the open questions: strong cosmic censorship — whether the Cauchy horizons inside realistic rotating and charged black holes are destroyed by generic perturbations, so that determinism survives — was recently sharpened and contested for near-extremal holes in de Sitter space, and the Christodoulou–Klainerman proof of the nonlinear stability of Minkowski space is a global Cauchy-problem theorem. Even quantum field theory in curved spacetime leans on it: canonical evolution of quantum data requires a globally hyperbolic background, which is exactly why anti-de Sitter holography must supply boundary conditions to render its bulk dynamics deterministic.

Prediction in general relativity is a slice-by-slice affair. Lay down one spacelike surface $\Sigma_0$ — the relativistic replacement for "the state of the universe now" — and equip it with two tensor fields: the intrinsic metric $h_{ij}$ (the shape of space) and the extrinsic curvature $K_{ij}$ (how that space is bending inside spacetime, the geometric stand-in for "velocity"). The field equations then push this data onto the next slice, and the next, sweeping out a spacetime. But two of Einstein's structures police the process: the constraints forbid you from choosing $(h_{ij},K_{ij})$ freely, and the light cones — themselves part of the metric being solved for — bound how far the data reach. The first figure is that machine in motion.

Two clauses of the definition do all the work, and the first figure hides a piece of arithmetic that the second makes explicit. Of the ten independent components of the symmetric metric $g_{\mu\nu}$, four correspond to nothing physical at all — they are the freedom to relabel spacetime with four coordinate functions. Four more are pinned down by the constraint equations, which are not extra dynamics but conditions the slice must already obey. That leaves precisely two genuine degrees of freedom per point: the two transverse-traceless polarizations $h_+$ and $h_\times$ that a passing gravitational wave stretches space with. The same split runs through the field equations themselves — four carry no second time derivative and merely constrain, six actually evolve.

Well-posedness is a separate demand layered on top of this counting. It is Hadamard's trio: a solution must exist, be unique, and depend continuously on the data, so that a small change in $(h_{ij},K_{ij})$ produces only a bounded change in the development over any finite time. The wave equation has this; the Laplace equation, run as an initial-value problem, catastrophically does not. Einstein's equations sit ambiguously between until a gauge is chosen — and the whole art of the subject is choosing a gauge that lands them on the wave-equation side of the line. Causality is the last ingredient: because the light cones belong to the very metric being solved for, the region a slice controls, its domain of dependence $D^+(\Sigma)$, is carved out by disturbances travelling at or below the speed of light, and a spacetime whose causal structure is fully governed by one such slice is called globally hyperbolic.

Everything begins with the field equations read as a system of partial differential equations for the metric. In geometrized units $G=c=1$ they are ten coupled, quasilinear, second-order equations.

where: $G_{\mu\nu}$ — Einstein tensor; $R_{\mu\nu},R$ — Ricci tensor and scalar of $g_{\mu\nu}$; $T_{\mu\nu}$ — stress-energy tensor; $g_{\mu\nu}$ — spacetime metric. Reading: ten equations for ten metric components, but the next identity shows they cannot all be evolution equations. The reason the count of genuine evolution equations is six, not ten, is a geometric identity obeyed by the left-hand side no matter what the matter does. It is the gravitational analogue of charge conservation.

where: $\nabla_\mu$ — Levi-Civita covariant derivative of $g$. Reading: the $\nu=0$ component forces $G^{0\mu}$ to contain at most first time derivatives of $g$, so the four equations $G^{0\mu}=8\pi T^{0\mu}$ cannot evolve data — they constrain it on the slice. Projecting the doubly-normal part of the field equations onto a spacelike slice gives the first constraint: a relation among the slice's own curvature and its bending, equal to the local energy density.

where: ${}^{(3)}\!R$ — Ricci scalar of the spatial metric $h_{ij}$; $K_{ij}$ — extrinsic curvature of $\Sigma_t$; $K=h^{ij}K_{ij}$ — mean curvature (trace); $\rho=n_\mu n_\nu T^{\mu\nu}$ — energy density seen by the normal observer $n^\mu$. Reading: no time derivative of $K_{ij}$ appears, so this is a condition on the data, the Hamiltonian constraint — the cousin of $\nabla\!\cdot\!\vec E=\rho$. The mixed normal-tangential projection gives the second constraint, tying the divergence of the extrinsic curvature to the momentum flowing through the slice.

where: $D_j$ — covariant derivative intrinsic to $\Sigma_t$; $S^i=-h^i{}_\mu n_\nu T^{\mu\nu}$ — momentum density seen by $n^\mu$; $K^{ij}$ — extrinsic curvature with indices raised by $h^{ij}$. Reading: three equations, again with no $\partial_t K_{ij}$ — the momentum constraint, the cousin of $\nabla\!\cdot\!\vec B=0$; together with (3) they are the four constraints. Half of the true dynamics is kinematic: the extrinsic curvature is, up to lapse and shift, the time derivative of the spatial metric. This equation defines $K_{ij}$ and evolves $h_{ij}$.

where: $\alpha$ — lapse (proper time per unit coordinate time for the normal observer); $\beta^i$ — shift (coordinate drift of spatial points); $\partial_t$ — derivative along $t^\mu=\alpha n^\mu+\beta^\mu$. Reading: $\alpha$ and $\beta^i$ are freely choosable gauge — they thread coordinates through spacetime and never appear in a constraint. The other half is the genuine second-order dynamics, the tangential-tangential projection of the field equations, which supplies $\partial_t K_{ij}$ and closes the system.

where: ${}^{(3)}\!R_{ij}$ — Ricci tensor of $h_{ij}$; $S_{ij}=h_i{}^\mu h_j{}^\nu T_{\mu\nu}$ — spatial stress, $S=h^{ij}S_{ij}$. Reading: equations (5)–(6) are the ADM system — six evolution equations for the phase-space pair $(h_{ij},K_{ij})$, driven by curvature and gauge. The ADM system as written is not yet hyperbolic. The decisive move is to fix the gauge by demanding the coordinate functions be harmonic, which surgically removes the ill-behaved part of the Ricci tensor.

where: $\Box_g$ — covariant d'Alembertian; $\Gamma^\mu_{\nu\rho}$ — Christoffel symbols of $g$; $x^\mu$ — coordinates imposed to satisfy the wave equation. Reading: the condition $\Gamma^\mu\equiv g^{\nu\rho}\Gamma^\mu_{\nu\rho}=0$ is what turns Einstein's equations into wave equations. In that gauge the principal (highest-derivative) part of the Ricci tensor collapses to the scalar wave operator acting on each component, and the vacuum equations become a diagonal quasilinear wave system.

where: $Q_{\mu\nu}$ — terms quadratic in first derivatives of the metric (lower order). Reading: the principal symbol is $-\tfrac12 g^{\rho\sigma}\xi_\rho\xi_\sigma$ — exactly the wave operator — so classical hyperbolic existence theory applies to Sobolev data with $s>5/2$. Causal determinism is quantified by the domain of dependence: the region whose entire physics is fixed by data on the slice.

where: $\Sigma$ — achronal initial-data surface; $D^+(\Sigma)$ — its future domain of dependence, bounded by the Cauchy horizon $H^+(\Sigma)$. Reading: because light cones are set by the metric being solved for, the reach of the data is dynamical, not fixed in advance. Finally, solving the constraints on the slice is an elliptic problem, and the conformal method is its workhorse: rescale the spatial metric by a factor $\psi$ and the Hamiltonian constraint becomes a single scalar equation for $\psi$.

where: $\psi$ — conformal factor; $\hat h_{ij}$ — chosen background metric with Laplacian $\bar\nabla^2$ and scalar $\hat R$; $\hat A^{ij}$ — conformal trace-free part of $K^{ij}$. Reading: this is the Lichnerowicz–York equation; in the time-symmetric conformally-flat limit ($K_{ij}=0,\hat R=0$) it degenerates to the flat Laplace equation $\nabla^2\psi=0$.

The Cauchy framework specializes and degenerates in ways that each teach a lesson, from the manifestly well-posed linear limit to the outright failure of the raw ADM system.

1 · Conformally-flat black-hole initial data (fully worked)

Construct time-symmetric initial data for one, then two, black holes on a single slice. Show the Hamiltonian constraint reduces to a flat Laplace equation, identify the solutions, and read off the ADM mass. Use geometrized units $G=c=1$.

Step 1. Choose the data to be time-symmetric, $K_{ij}=0$. Then $K=0$ and every term of the momentum constraint (4) vanishes identically — the momentum constraint is satisfied for free. The only surviving field equation on the slice is the Hamiltonian constraint (3), which with $K_{ij}=0$ and no matter reads simply ${}^{(3)}\!R=0$.

Step 2. Adopt the conformal ansatz $h_{ij}=\psi^4\,\delta_{ij}$ with a flat background, so $\hat R=0$ and $\hat A_{ij}=0$. Equation (10) then collapses to the flat Laplace equation $\nabla^2\psi=0$. A nonlinear constraint has become the most standard linear elliptic equation in physics — and linearity means solutions superpose.

Step 3. The elementary regular-at-infinity solution with one point source is $\psi=1+\dfrac{M}{2r}$. This is exactly the isotropic form of the Schwarzschild slice, with an apparent horizon at isotropic radius $r=M/2$. Reading the falloff $\psi\to 1+M/2r$ against the ADM mass integral $M_{\text{ADM}}=-\tfrac{1}{2\pi}\oint_\infty\partial_i\psi\,dS^i$ gives $\oint\partial_i\psi\,dS^i=-2\pi M$, hence $M_{\text{ADM}}=M$.

Step 4. Because $\nabla^2$ is linear, two punctures superpose exactly: $\psi=1+\dfrac{m_1}{2r_1}+\dfrac{m_2}{2r_2}$, the Brill–Lindquist solution, with $r_1,r_2$ the flat distances to the two centres. The ADM mass is the sum of the falloff coefficients, $M_{\text{ADM}}=m_1+m_2$. Taking $m_1=m_2=0.5$ gives $M_{\text{ADM}}=\mathbf{1.0}$; scaling to $m_1=m_2=15\,M_\odot$ gives $M_{\text{ADM}}=30\,M_\odot$ — a stand-in for the roughly $36+29\,M_\odot$ progenitors of GW150914. This superposed puncture data is literally the first slice of a binary-black-hole simulation.

2 · Counting the degrees of freedom in the linear limit

Take the linearized theory $g_{\mu\nu}=\eta_{\mu\nu}+h_{\mu\nu}$, $|h|\ll1$. Show the Cauchy problem is manifestly well-posed and that exactly two physical degrees of freedom survive, travelling at the speed of light.

Step 1. Form the trace-reversed perturbation $\bar h_{\mu\nu}=h_{\mu\nu}-\tfrac12\eta_{\mu\nu}h$ and impose the Lorenz gauge $\partial^\mu\bar h_{\mu\nu}=0$ — four gauge conditions. The linearized field equations become $\Box\bar h_{\mu\nu}=-16\pi T_{\mu\nu}$, and in vacuum $\Box\bar h_{\mu\nu}=0$: a flat, diagonal wave equation, the textbook well-posed hyperbolic system.

Step 2. Count what is left. The symmetric $h_{\mu\nu}$ has 10 components; the four Lorenz conditions remove 4; residual gauge transformations $\xi^\mu$ with $\Box\xi^\mu=0$ remove 4 more. That leaves $10-4-4=\mathbf{2}$ independent components — the transverse-traceless polarizations $h_+$ and $h_\times$, the same 2 the full nonlinear count gives.

Step 3. Insert a plane wave $\bar h_{\mu\nu}\propto\exp(ik_\alpha x^\alpha)$ into $\Box\bar h_{\mu\nu}=0$. The wave operator gives $k_\alpha k^\alpha=0$, i.e. $\omega^2=|\vec k|^2$, so $\omega/|\vec k|=1$: gravitational waves propagate at exactly the speed of light $c$. The linear limit thus displays, in miniature, the entire logic of the Cauchy problem — constraints (the Lorenz conditions), gauge (residual $\xi^\mu$), two physical polarizations, and hyperbolic evolution at the causal speed.

If a Cauchy surface is where the data live, the theorem that those data determine a whole universe is the reason the subject exists — and that theorem carries one name above all others.

Texts. Misner, Thorne & Wheeler, Gravitation (1973), Ch. 21 · Wald, General Relativity (1984), Ch. 10 · Carroll, Spacetime and Geometry (2004), Ch. 4 and App. · Baumgarte & Shapiro, Numerical Relativity (2010), Ch. 2–3 · Ringström, The Cauchy Problem in General Relativity (2009) · Choquet-Bruhat, General Relativity and the Einstein Equations (2009) · Hartle, Gravity (2003), Ch. 21 · Schutz, A First Course in General Relativity (2nd ed., 2009). All equations use geometrized units $G=c=1$. Local well-posedness in harmonic gauge holds for Sobolev data with $s>5/2$; Klainerman–Rodnianski later lowered the threshold toward the $L^2$-curvature bound. The time-symmetric conformally-flat limit gives $\nabla^2\psi=0$; the isotropic Schwarzschild solution $\psi=1+M/2r$ has $M_{\text{ADM}}=M$ and an apparent horizon at $r=M/2$; the Brill–Lindquist two-puncture solution gives $M_{\text{ADM}}=m_1+m_2$. The identity $M_{\text{ADM}}=k$ for $\psi=1+k/2r$ follows directly from the ADM mass integral $M_{\text{ADM}}=-\tfrac{1}{2\pi}\oint_\infty\partial_i\psi\,dS^i$.

Prerequisites. ADM 3+1 Decomposition of Spacetime · Gauss–Codazzi–Mainardi Equations · Hamiltonian and Momentum Constraints. Leads to. Generalized Harmonic Gauge · BSSN Formulation · Conformal Method for Constraint Solving · Numerical Relativity and Binary Black Hole Mergers.

Ask the tutor: Want to walk through how harmonic gauge collapses the Ricci tensor to a wave operator, or how puncture data actually solve the constraints that seed a black-hole merger?