Harmonic and Generalized Harmonic Gauge

Einstein's field equations $G_{\mu\nu}=8\pi T_{\mu\nu}$ hide a wave equation, and the disguise is coordinate freedom. General relativity is diffeomorphism invariant: the four functions $x^\mu$ that label events carry no physics, so before a coordinate condition is fixed the equations belong to no definite PDE type. Write the Ricci tensor in metric components and its second-derivative skeleton is $-\tfrac12 g^{\alpha\beta}\partial_\alpha\partial_\beta g_{\mu\nu}$ — a clean wave operator — plus gauge terms $g_{\sigma(\mu}\partial_{\nu)}\Gamma^\sigma$ that mix the components and destroy hyperbolicity. Those terms are pure grid wobble; kill them and what remains is genuine geometry obeying a wave equation. Harmonic (de Donder) gauge kills them in the most natural way imaginable: demand that each coordinate function solve the wave equation, $\Box x^\mu=0$ . Because $\nabla_\alpha\nabla_\beta x^\mu=-\Gamma^\mu_{\alpha\beta}$ , this is exactly the vanishing of the contracted Christoffel symbols $\Gamma^\mu\equiv g^{\alpha\beta}\Gamma^\mu_{\alpha\beta}=0$ , and it turns the ten Einstein equations into ten coupled quasilinear wave equations with the hyperbolic symbol $g^{\alpha\beta}\partial_\alpha\partial_\beta$ . This is the form in which Yvonne Choquet-Bruhat proved, in 1952, that the Cauchy problem of general relativity is locally well-posed — the theorem that makes "evolving a spacetime" meaningful. Generalized harmonic gauge (GHG) is the flexible descendant. Promote the zero on the right to a freely prescribed source, $\Box x^\mu=-H^\mu(x,g)$ , so that $\Gamma^\mu=H^\mu$ . As long as $H^\mu$ never touches derivatives of the metric, the wave skeleton survives untouched while the practitioner gains a steering wheel for the slicing and shift. Add constraint damping — feedback that drives violations to zero like $e^{-\gamma_0 t}$ — and the marginally stable system becomes robustly stable. That combination is what Frans Pretorius used in 2005 to compute the first complete binary-black-hole inspiral, merger and ringdown, and it remains a pillar of numerical relativity today.

Key takeaways

Harmonic (wave) coordinates impose $\Box x^\mu=0$ , equivalent to $\Gamma^\mu\equiv g^{\alpha\beta}\Gamma^\mu_{\alpha\beta}=0$ , which strips the gauge-dependent second derivatives out of the Ricci tensor and leaves a pure wave operator on every metric component.
Einstein's equations then read as ten coupled quasilinear wave equations with the hyperbolic principal symbol $g^{\alpha\beta}\xi_\alpha\xi_\beta$ , whose characteristics are the physical null cones.
This is the form in which general relativity's Cauchy problem was first proved locally well-posed — Choquet-Bruhat 1952 — with the constraints propagated rather than re-solved at each step.
Generalized harmonic gauge replaces $0$ by a prescribable source $H^\mu(x,g)$ : $\Box x^\mu=-H^\mu$ , i.e. $\Gamma^\mu=H^\mu$ , keeping symmetric hyperbolicity while giving dynamical control of lapse and shift.
Constraint damping (Gundlach et al.; Pretorius) adds terms proportional to $C^\mu=H^\mu-\Gamma^\mu$ that force numerical violations to decay like $e^{-\gamma_0 t}$ — the switch that made stable black-hole mergers computable.
GHG underlies the SpEC / SXS pseudospectral waveforms used by LIGO–Virgo–KAGRA and sits alongside BSSN as one of the two workhorse formulations of numerical relativity.
Key idea: choosing coordinates that are themselves solutions of the wave equation — $\Box x^\mu=-H^\mu$ — cancels the gauge-dependent second-derivative terms in the Ricci tensor, leaving $g^{\alpha\beta}\partial_\alpha\partial_\beta g_{\mu\nu}$ as the principal part: ten coupled quasilinear wave equations that are symmetric hyperbolic and hence demonstrably well-posed.

Where this lives today

Generalized harmonic gauge is running on supercomputers right now. The Spectral Einstein Code (SpEC) and its successor SpECTRE, built by the SXS collaboration, evolve binary black holes in a first-order symmetric-hyperbolic reduction of GHG (Lindblom et al. 2006) with damped-harmonic sources and constraint damping. The resulting SXS catalog holds thousands of high-accuracy simulations — the templates and surrogate models that LIGO–Virgo–KAGRA match against real data, including GW150914, the first detection — a merger of black holes of about $36$ and $29\,M_\odot$ that radiated roughly $3\,M_\odot c^2$ of energy. On the analytic side, harmonic gauge remains the native language of mathematical relativity: Lindblad and Rodnianski's proof of the global nonlinear stability of Minkowski space works directly in wave coordinates. The open frontier is gauge design under extreme conditions — through common-horizon formation, for eccentric and highly spinning binaries, and for the long inspirals that next-generation detectors (LISA, Einstein Telescope, Cosmic Explorer) will demand.

The physical picture

Each metric component $g_{\mu\nu}$ carries two kinds of information at once: real gravitational radiation and the arbitrary jitter of the coordinate grid drawn on top of it. Written blindly, the Einstein equations do not distinguish them — the grid jitter enters through $\partial_\nu\Gamma^\sigma$ terms that couple the components and wreck the clean second-derivative structure. Harmonic gauge nails the grid down self-consistently: let the coordinate lines themselves ripple like free fields obeying $\Box x^\mu=0$ . Because the d'Alembertian inherits the light cones of the physical metric, the coordinates flow along the geometry's characteristics, information propagates exactly on the null cones, and hyperbolicity appears automatically.

Figure

One gauge choice turns Einstein's equations into wave equations: the naive Ricci tensor mixes gauge wobble ∂Γ into its second derivatives, but imposing Box x^μ = −Γ^μ = 0 strips those terms and leaves the pure wave operator g^αβ ∂_α∂_β, whose characteristics are the null cones g^αβ ξ_α ξ_β = 0 of the physical metric

Generalized harmonic gauge recognizes that free coordinates are not sacred. Prescribing $H^\mu$ is like fitting the grid with a programmable forcing term: you can steer the slicing to avoid a forming singularity or stop coordinates piling up near a horizon, all while leaving the second-derivative skeleton exactly as it was — provided $H^\mu$ never depends on derivatives of the metric, which would put new second derivatives into the principal part and shatter hyperbolicity. Constraint damping is the last ingredient, pure feedback control: the constraints are conserved in the continuum but drift under discretization, and a restoring term proportional to the violation pushes the solution back onto the constraint surface the way friction returns a pendulum to rest. Without it, roundoff seeds grow and crash the run; with it, they decay. The second figure is that switch.

Figure

Constraint damping is the switch that made black-hole mergers computable: a tiny seeded violation ‖C‖ grows exponentially and crashes an undamped run, but the Gundlach damping term drives it as e^(−γ₀ t) back to roundoff — the mechanism behind Pretorius 2005, the first full inspiral, merger and ringdown

The core equations

The subject grows from one covariant statement: harmonic coordinates are wave-equation solutions. Treating each $x^\mu$ as a scalar field, $\nabla_\alpha\nabla_\beta x^\mu=-\Gamma^\mu_{\alpha\beta}$ , so contracting with the inverse metric turns "the coordinate is harmonic" into "the contracted Christoffel vanishes".

\Box x^\mu = g^{\alpha\beta}\nabla_\alpha\nabla_\beta x^\mu = -\Gamma^\mu \qquad (1)

where: $x^\mu$ — the four coordinate functions, each treated as a scalar field; $\Box$ — covariant d'Alembertian $g^{\alpha\beta}\nabla_\alpha\nabla_\beta$ ; $\Gamma^\mu\equiv g^{\alpha\beta}\Gamma^\mu_{\alpha\beta}$ — the contracted Christoffel symbol. Reading: $\Box x^\mu=0$ is identical to $\Gamma^\mu=0$ , a local algebraic condition on the connection. The vanishing-connection form has an equivalent numerical relativists prefer — a plain coordinate divergence of a single density, needing no Christoffel bookkeeping.

\Gamma^\mu \equiv g^{\alpha\beta}\Gamma^\mu_{\alpha\beta} = 0 \quad\Longleftrightarrow\quad \partial_\nu\!\left(\sqrt{-g}\,g^{\mu\nu}\right)=0 \qquad (2)

where: $\Gamma^\mu_{\alpha\beta}$ — Christoffel symbols of the second kind; $g$ — determinant of $g_{\mu\nu}$ (negative for Lorentzian signature); $\sqrt{-g}\,g^{\mu\nu}$ — the densitized inverse metric (the "Gothic" metric $\mathfrak{g}^{\mu\nu}$ ). Reading: via $\Gamma^\mu=-(\sqrt{-g})^{-1}\partial_\nu(\sqrt{-g}\,g^{\mu\nu})$ , the harmonic condition is a conservation law for $\mathfrak{g}^{\mu\nu}$ . The payoff appears when this is fed into the Ricci tensor, which, organized by derivative order, splits into a scalar wave operator on $g_{\mu\nu}$ , a gauge piece from $\Gamma^\sigma$ , and lower-order quadratic terms.

R_{\mu\nu} = -\tfrac12 g^{\alpha\beta}\partial_\alpha\partial_\beta g_{\mu\nu} + g_{\sigma(\mu}\partial_{\nu)}\Gamma^\sigma + Q_{\mu\nu}(g,\partial g) \qquad (3)

where: $R_{\mu\nu}$ — Ricci tensor; $g_{\sigma(\mu}\partial_{\nu)}\Gamma^\sigma$ — the gauge term, the only obstruction to hyperbolicity; $Q_{\mu\nu}$ — terms quadratic in first derivatives of the metric; $(\mu\nu)$ — symmetrization. Reading: everything that spoils the wave structure sits in the one term proportional to $\Gamma^\sigma$ , so $\Gamma^\sigma=0$ removes it. Impose harmonic gauge and insert the trace-reversed Einstein equation: the ten equations become ten quasilinear wave equations — the Choquet-Bruhat "reduced" system.

g^{\alpha\beta}\partial_\alpha\partial_\beta\, g_{\mu\nu} = -2\,Q_{\mu\nu}(g,\partial g) + 16\pi\!\left(T_{\mu\nu}-\tfrac12 g_{\mu\nu}T\right) \qquad (4)

where: $g^{\alpha\beta}\partial_\alpha\partial_\beta$ — the scalar wave operator, now the principal part of every component; $T_{\mu\nu}$ — stress-energy; $T=g^{\mu\nu}T_{\mu\nu}$ — its trace. Reading: in vacuum the right side is just $-2Q_{\mu\nu}$ , and the metric propagates on its own null cones like ten coupled scalar fields. Generalized harmonic gauge is a one-symbol edit: replace the zero on the right of (1) by a prescribed source.

\Box x^\mu = -H^\mu(x,g)\quad\Longleftrightarrow\quad \Gamma^\mu = H^\mu, \qquad g^{\alpha\beta}\partial_\alpha\partial_\beta\, g_{\mu\nu} = -2\nabla_{(\mu}H_{\nu)} - 2Q_{\mu\nu} + 16\pi\!\left(T_{\mu\nu}-\tfrac12 g_{\mu\nu}T\right) \qquad (5)

where: $H^\mu(x,g)$ — the prescribed gauge source, depending on $x^\mu$ and $g_{\mu\nu}$ but never on $\partial g$ ; $H_\nu=g_{\nu\mu}H^\mu$ — its lowered form. Reading: $H^\mu$ enters only through $\nabla_{(\mu}H_{\nu)}$ , leaving the symbol $g^{\alpha\beta}\xi_\alpha\xi_\beta$ intact — steerable but still hyperbolic. Finally, the object that guards the scheme: the gauge constraint $C^\mu$ , which vanishes on physical solutions and obeys its own homogeneous wave equation.

C^\mu \equiv H^\mu - \Gamma^\mu = 0,\qquad \Box C_\nu + R_\nu{}^\mu C_\mu = 0 \qquad (6)

where: $C^\mu$ — the gauge (harmonic) constraint; $R_\nu{}^\mu$ — mixed Ricci tensor in the homogeneous propagation. Reading: nothing sources $C$ but $C$ itself, so $C^\mu=0$ and $\partial_t C^\mu=0$ initially force $C^\mu\equiv0$ forever — the constraints are propagated, not re-solved.

Derivation AHarmonic gauge makes the Cauchy problem well-posed

The goal is to show that imposing $\Gamma^\mu=0$ converts the Einstein equations from a system of no fixed type into a symmetric-hyperbolic set of ten quasilinear wave equations whose characteristic surfaces are the null cones of $g_{\mu\nu}$ — the property that lets standard existence-and-uniqueness theorems apply and makes "give data on a slice and evolve" well defined. We expand the Ricci tensor, drop the gauge term, insert the trace-reversed field equation, read off the principal symbol, and confirm the reduced solution solves the full equations.

Step 1. Expand the Ricci tensor in coordinate components and regroup by derivative order, giving $R_{\mu\nu}=-\tfrac12 g^{\alpha\beta}\partial_\alpha\partial_\beta g_{\mu\nu}+g_{\sigma(\mu}\partial_{\nu)}\Gamma^\sigma+Q_{\mu\nu}$ , with $Q_{\mu\nu}$ quadratic in $\partial g$ . Why: every gauge-dependent second derivative is isolated into the single middle term proportional to $\Gamma^\sigma$ , so the obstruction to hyperbolicity has one address.

Step 2. Impose $\Gamma^\sigma=0$ . The middle term vanishes identically and $R_{\mu\nu}=-\tfrac12 g^{\alpha\beta}\partial_\alpha\partial_\beta g_{\mu\nu}+Q_{\mu\nu}$ . Why: the highest-derivative operator acting on each $g_{\mu\nu}$ is now the plain scalar wave operator $g^{\alpha\beta}\partial_\alpha\partial_\beta$ , identical on all ten components.

Step 3. Insert the trace-reversed Einstein equation $R_{\mu\nu}=8\pi(T_{\mu\nu}-\tfrac12 g_{\mu\nu}T)$ to obtain the reduced system $g^{\alpha\beta}\partial_\alpha\partial_\beta g_{\mu\nu}=-2Q_{\mu\nu}+16\pi(T_{\mu\nu}-\tfrac12 g_{\mu\nu}T)$ . Why: the right side now holds only $g$ and $\partial g$ , so the system is quasilinear — strictly linear in the second derivatives that fix its type.

Step 4. Read off the principal symbol $P(\xi)=g^{\alpha\beta}\xi_\alpha\xi_\beta$ . For Lorentzian signature this is a real, hyperbolic symbol whose zero set $g^{\alpha\beta}\xi_\alpha\xi_\beta=0$ is the null cone of the metric. Why: this hyperbolicity is exactly the hypothesis of the existence-and-uniqueness theorems for quasilinear wave equations, so well-posedness follows at once.

Step 5. Recover full Einstein from reduced plus constraint: since $R^{\rm red}_{\mu\nu}=R_{\mu\nu}+\nabla_{(\mu}C_{\nu)}$ and $C^\mu$ propagates as zero, a reduced solution whose constraint and its time derivative vanish initially solves the true field equations for all time. Why: the evolved geometry is a genuine Einstein spacetime, not a look-alike. *Verified: the Ricci split and the trace-reversed reduced system match MTW section 35.4 and Wald section 10.2.*

Go deeper · Derivation B · 1 minConstraint propagation and the damping that stabilizes mergers

Derivation BConstraint propagation and the damping that stabilizes mergers

The goal is to explain why free GHG evolution can ignore the constraints during the run and still produce a physical spacetime, and why an added damping term is nonetheless essential in floating-point practice. We define $C^\mu=H^\mu-\Gamma^\mu$ , use the contracted Bianchi identity to show it obeys a source-free wave equation so zero initial data stays zero in the continuum, and add the Gundlach–Calabrese–Hinder–Martín-García term that converts marginal stability into exponential decay of the seeded violations that would otherwise blow the simulation apart.

Step 1. Compare reduced and full: substituting $\Gamma^\mu\to H^\mu$ changes the true Ricci tensor by a symmetrized gradient of the constraint, $R^{\rm red}_{\mu\nu}=R_{\mu\nu}+\nabla_{(\mu}C_{\nu)}$ . Why: this measures how far an off-constraint solution sits from a true Einstein solution — a gap linear in $C^\mu$ , hence controllable.

Step 2. Apply the contracted Bianchi identity $\nabla^\mu(R_{\mu\nu}-\tfrac12 g_{\mu\nu}R)=0$ to a solution of the reduced equations. Why: Bianchi holds for every geometry, so on the reduced solution it turns the divergence of the constraint gap into an equation for $C^\mu$ itself.

Step 3. The result is the homogeneous wave equation $\Box C_\nu+R_\nu{}^\mu C_\mu=0$ . With $C^\mu=0$ and $\partial_t C^\mu=0$ initially, uniqueness gives $C^\mu\equiv0$ . Why: this is the continuum guarantee — one evolves the unconstrained reduced system and the constraints ride along for free.

Step 4. Add the Gundlach damping term $\gamma_0\,(t_{(\mu}C_{\nu)}-\tfrac12 g_{\mu\nu}\,t^\sigma C_\sigma)$ , with $t^\mu$ the timelike unit normal and $\gamma_0>0$ the damping rate; it vanishes when $C^\mu=0$ . Why: the physical solution is untouched, yet off-surface dynamics are reshaped to fight discretization drift.

Step 5. For suitable energy norms the modified system gives $\partial_t\!\int C^\mu C_\mu\,\mathrm{d}V\le-2\gamma_0\!\int C^\mu C_\mu\,\mathrm{d}V+\dots$ , so seeded violations decay like $e^{-\gamma_0 t}$ rather than growing. Why: this exponential decay is the whole ball game — clean ringdown versus blow-up in a fraction of an orbit. *Verified: reproduces the constraint-damping structure of Gundlach, Calabrese, Hinder and Martín-García (2005), the scheme Pretorius (2005) used.*

Go deeper · Derivation C · 1 minLinearized harmonic gauge is the Lorenz gauge of gravitational waves

Derivation CLinearized harmonic gauge is the Lorenz gauge of gravitational waves

The goal is to connect the full machinery to the weak-field picture by showing that, to first order in a perturbation $h_{\mu\nu}$ about flat space, the harmonic condition $\Gamma^\mu=0$ becomes the Lorenz gauge $\partial_\nu\bar h^{\mu\nu}=0$ and the reduced equations become the flat-space wave equation for the trace-reversed perturbation — so gravitational radiation at $c$ is the linear shadow of the structure that governs binary black holes. We write $g_{\mu\nu}=\eta_{\mu\nu}+h_{\mu\nu}$ , define the trace reverse, linearize, and read off the wave equation.

Step 1. Define the trace-reversed perturbation $\bar h_{\mu\nu}=h_{\mu\nu}-\tfrac12\eta_{\mu\nu}h$ with $h=\eta^{\alpha\beta}h_{\alpha\beta}$ . Why: trace-reversal diagonalizes the linearized Einstein tensor, collapsing it to a single d'Alembertian on $\bar h_{\mu\nu}$ .

Step 2. Linearize the contracted Christoffel: to first order $\Gamma^\mu\approx\partial_\nu\bar h^{\mu\nu}$ , so $\Gamma^\mu=0$ is exactly the Lorenz gauge $\partial_\nu\bar h^{\mu\nu}=0$ . Why: "Lorenz gauge" is literally the linearized harmonic condition, not merely an analogue.

Step 3. In Lorenz gauge the linearized Einstein tensor is $-\tfrac12\Box_\eta\bar h_{\mu\nu}$ , so the field equations read $\Box_\eta\bar h_{\mu\nu}=-16\pi T_{\mu\nu}$ , and in vacuum $\Box_\eta\bar h_{\mu\nu}=0$ . Why: this radiation at $c$ is the $H^\mu=0$ weak-field corner of the same reduced system Pretorius evolves in full. *Verified: standard linearized-gravity result (Carroll 7.1, MTW ch. 35).*

Limiting cases

Generalized harmonic gauge collapses to familiar structures in the appropriate regimes.

Case	Result	Meaning
Linearized / weak field	$\Box_\eta\bar h_{\mu\nu}=-16\pi T_{\mu\nu},\ \partial^\mu\bar h_{\mu\nu}=0$	Harmonic gauge becomes the Lorenz gauge; the reduced system becomes the flat-space wave equation for radiation at $c$ .
Vacuum, pure harmonic	$g^{\alpha\beta}\partial_\alpha\partial_\beta\, g_{\mu\nu}=-2\,Q_{\mu\nu}(g,\partial g)$	With $H^\mu=0$ the principal part is the scalar wave operator on each component; $Q_{\mu\nu}$ is quadratic in $\partial g$ . The Choquet-Bruhat form.
GHG with sources	$g^{\alpha\beta}\partial_\alpha\partial_\beta\, g_{\mu\nu}=-2\nabla_{(\mu}H_{\nu)}-2Q_{\mu\nu}+16\pi(T_{\mu\nu}-\tfrac12 g_{\mu\nu}T)$	Free choice of $H^\mu(x,g)$ steers slicing and shift; $H^\mu=0$ recovers the harmonic case. Hyperbolic iff $H^\mu$ is independent of $\partial g$ .
Constraint-damped	$C^\mu=H^\mu-\Gamma^\mu,\ \Box C^\mu\simeq(\text{lower order})\,C^\nu$	Adding $\gamma_0(t^{(\mu}C^{\nu)}-\tfrac12 g^{\mu\nu}t_\sigma C^\sigma)$ , $\gamma_0>0$ , makes $C^\mu$ decay like $e^{-\gamma_0 t}$ ; undamped GHG as $\gamma_0\to0$ .

Common trapA gauge source can depend on the metric — but never on its derivatives

The single most consequential rule in the subject is that $H^\mu$ may depend on the coordinates $x^\mu$ and on the metric $g_{\mu\nu}$ , but never on derivatives of the metric. The reason is structural: $H^\mu$ enters through $\nabla_{(\mu}H_{\nu)}$ , so a $\partial g$ dependence would put $\partial\partial g$ back into that term, re-contaminating the principal symbol $g^{\alpha\beta}\xi_\alpha\xi_\beta$ and potentially destroying its real, diagonalizable, hyperbolic character — voiding the well-posedness the gauge secures. A softer pitfall: the harmonic condition is only local. Free harmonic coordinates generically develop caustics as they focus, which is why global black-hole evolutions cannot use $H^\mu=0$ and instead need dynamical gauge sources (damped harmonic gauge) or black-hole excision.

Worked examples

1 · The harmonic radial coordinate of a Schwarzschild black hole (fully worked)

Find the radial coordinate in which the Schwarzschild solution is harmonic, and see where the horizon lands. Work in geometrized units $G=c=1$ , so mass $M$ has units of length.

Step 1. Set up the harmonic condition on the static Schwarzschild background $ds^2=-(1-2M/r)\,dt^2+(1-2M/r)^{-1}dr^2+r^2 d\Omega^2$ , with $\sqrt{-g}=r^2\sin\theta$ and $g^{rr}=1-2M/r$ . A Cartesian-like coordinate $x^i=R(r)\,n^i$ ( $n^i$ the unit radial vector, an $\ell=1$ pattern) has its harmonicity set by a single radial equation for $R(r)$ .

Step 2. For an $\ell=1$ function the d'Alembertian gives $\dfrac{d}{dr}\!\left[(r^2-2Mr)\,R'\right]-2R=0$ , the $-2R$ being the angular ( $\ell(\ell+1)=2$ ) contribution — the exact condition $\Box x^i=0$ , i.e. $\Gamma^i=0$ .

Step 3. Try $R=r-M$ , so $R'=1$ : then $(r^2-2Mr)$ has derivative $2r-2M$ , and $2r-2M-2(r-M)=0$ identically. The harmonic radial coordinate is $r_H=r-M$ — the classical Fock result — with $t$ harmonic by staticity.

Step 4. Locate the horizon. The event horizon at $r=2M$ maps to $r_H=2M-M=\mathbf{M}$ : the harmonic chart slides every radius inward by one mass, so the sphere $r_H=M$ is the horizon and $r_H<M$ is unreachable. The explicit result $r_H=r-M$ shows harmonic coordinates form a genuine regular-down-to-the-horizon chart, not an abstract existence statement.

2 · Counting the physical waves inside ten wave equations

The reduced system evolves all ten components $g_{\mu\nu}$ as wave equations. If each obeys a wave equation, why does gravity carry only two radiative degrees of freedom?

Step 1. A symmetric $4\times4$ metric has $\tfrac{4\cdot5}{2}=\mathbf{10}$ independent components, and the reduced GHG system propagates every one hyperbolically.

Step 2. The four gauge conditions $\Gamma^\mu=H^\mu$ (the four functions $x^\mu$ one is free to redefine) subtract $\mathbf{4}$ , leaving $10-4=6$ .

Step 3. The Hamiltonian and momentum constraints — four relations the data must satisfy and that then propagate — subtract another $\mathbf{4}$ , leaving $6-4=\mathbf{2}$ .

Step 4. Interpret. The two survivors are exactly the transverse-traceless polarizations $h_+$ and $h_\times$ of a gravitational wave. The ten wave equations are honest, but eight of their solutions are pure gauge or constrained away; only two carry physical energy — the two independent stretch-and-squeeze patterns LIGO records.

Physics of the portrait

Harmonic coordinates predate general relativity's initial-value problem, but one theorem turned them into its foundation.

Origins · 1 minYvonne Choquet-Bruhat and how this was discovered

Yvonne Choquet-Bruhat

b. 1923, French mathematician · proved general relativity's Cauchy problem well-posed in harmonic gauge

The harmonic coordinate condition drifted through relativity for decades as a computational convenience — Einstein used a harmonic-type condition to derive gravitational waves in 1916, Théophile de Donder wrote it cleanly in 1921, and Vladimir Fock made harmonic coordinates physically preferred. But nobody had shown that Einstein's equations, given data on a spacelike slice, actually determine a spacetime — that the theory has an initial-value formulation at all. Yvonne Choquet-Bruhat closed that gap in 1952: working in harmonic gauge, she recognized the reduced equations as quasilinear wave equations, proved local existence and uniqueness, and showed the constraints propagate, so a genuine Einstein spacetime emerges from the data — the theorem that makes numerical relativity conceivable. In 1969 she went further, proving with Robert Geroch that every set of initial data has a unique maximal globally hyperbolic development — the definitive statement of determinism in general relativity. Her argument is still the skeleton of every modern well-posedness result, from the nonlinear stability of Minkowski space to the generalized-harmonic evolutions that produced the first merger waveforms half a century later.

History

1916

— Einstein derives gravitational waves using coordinates obeying a harmonic-type condition

1921

— de Donder introduces the harmonic coordinate condition in clean covariant form

1952

— Choquet-Bruhat proves local existence and uniqueness for the vacuum Cauchy problem in harmonic gauge

1985

— Friedrich generalizes to prescribed gauge source functions and proves symmetric hyperbolicity

2005

— Pretorius evolves the first full binary-black-hole inspiral, merger and ringdown in generalized harmonic gauge with constraint damping

Going further

Texts. Misner, Thorne and Wheeler, Gravitation (1973), section 35.4 · Wald, General Relativity (1984), section 10.2 · Carroll, Spacetime and Geometry (2004), section 7.1 · Baumgarte and Shapiro, Numerical Relativity (2010), chapter 11 · Y. Choquet-Bruhat, Acta Math. 88 (1952) 141 · H. Friedrich, Commun. Math. Phys. 100 (1985) 525 · F. Pretorius, Phys. Rev. Lett. 95 (2005) 121101 · C. Gundlach, J. M. Martín-García, G. Calabrese and I. Hinder, Class. Quantum Grav. 22 (2005) 3767 · L. Lindblom et al., Class. Quantum Grav. 23 (2006) S447.

Provenance. Equations cross-checked against MTW section 35.4, Wald section 10.2, Carroll section 7.1 and the numerical-relativity literature (Pretorius 2005; Gundlach et al. 2005; Lindblom et al. 2006). A verifier should confirm: the Ricci split and $\Box x^\mu=-\Gamma^\mu$ from $\nabla_\alpha\nabla_\beta x^\mu=-\Gamma^\mu_{\alpha\beta}$ ; the densitized identity $\Gamma^\mu=-(\sqrt{-g})^{-1}\partial_\nu(\sqrt{-g}\,g^{\mu\nu})$ ; the Schwarzschild harmonic coordinate $r_H=r-M$ in $\tfrac{d}{dr}[(r^2-2Mr)R']-2R=0$ (checked symbolically here, residual exactly zero), horizon $r=2M\mapsto r_H=M$ ; and that the linearized harmonic condition equals Lorenz gauge, yielding $\Box_\eta\bar h_{\mu\nu}=-16\pi T_{\mu\nu}$ . Historical attributions (de Donder 1921, Choquet-Bruhat 1952, Friedrich 1985, Pretorius 2005) are standard.

Prerequisites. ADM 3+1 Decomposition of Spacetime · Initial-Value (Cauchy) Problem in General Relativity · Hamiltonian and Momentum Constraints · Gauss-Codazzi-Mainardi Equations. Leads to. Numerical Relativity and Binary Black Hole Mergers · Gauge Conditions in Numerical Relativity · BSSN Formulation · Black Hole Excision and Puncture Methods.

Ask the tutor: ask me to walk through the Ricci-tensor split term by term, derive the constraint-propagation wave equation from the Bianchi identity, or show how damped harmonic gauge chooses $H^\mu$ from the lapse and shift.

Why call it "harmonic" gauge — what is harmonic about it?

Because each coordinate function is required to be a harmonic function of the geometry it lives in: it satisfies $\Box x^\mu=0$ , exactly as an electrostatic potential satisfies $\nabla^2\phi=0$ in a source-free region. The name comes from potential theory. The equivalent labels you will meet — "de Donder gauge", "wave coordinates", and "Lorenz gauge" in the linearized limit — all name the same condition $\Gamma^\mu=0$ , and it is this harmonicity that makes the d'Alembertian the principal operator of the field equations.

What is the difference between harmonic and generalized harmonic gauge?

Harmonic gauge sets $\Box x^\mu=0$ , i.e. $\Gamma^\mu=0$ — the coordinates are free wave-equation solutions. Generalized harmonic gauge sets $\Box x^\mu=-H^\mu(x,g)$ , i.e. $\Gamma^\mu=H^\mu$ , with $H^\mu$ a source you design. The difference matters in practice: with $H^\mu=0$ the coordinates focus and form caustics as a black hole develops, wrecking the grid, whereas a well-chosen $H^\mu$ (damped harmonic gauge) keeps them smooth and the horizon well-resolved. Because $H^\mu$ enters only at first-derivative order, the wave structure and well-posedness are preserved.

If the constraints propagate automatically, why does anyone need constraint damping?

Because "automatically" is a continuum statement. In the exact theory the contracted Bianchi identity keeps constraints satisfied forever. On a computer the equations are discretized, and roundoff and truncation error seed tiny violations that the undamped system propagates — and can amplify through nonlinear feedback until the run crashes. Constraint damping adds terms proportional to $C^\mu$ that change nothing on the physical solution but drive off-constraint drift down like $e^{-\gamma_0 t}$ . It is the difference between a run that survives dozens of orbits and one that dies in a fraction of one; it was the missing ingredient before 2005.

How does GHG relate to the ADM 3+1 and BSSN formulations?

ADM slices spacetime into spatial hypersurfaces and evolves the spatial metric and extrinsic curvature, with lapse and shift as free gauge functions; but the plain ADM equations are only weakly hyperbolic and go unstable. GHG and BSSN are the two cures. BSSN keeps the 3+1 split and repairs hyperbolicity with conformal variables and an evolved contracted connection with its own gauge conditions (moving punctures). GHG instead works in four-dimensional covariant form, imposing the gauge through $H^\mu$ . In GHG the source $H^\mu$ plays the role the slicing and shift conditions play in the 3+1 languages — the same gauge freedom, one derivative order lower.

Can harmonic coordinates cover an entire black-hole spacetime?

Not globally, in general. Free harmonic coordinates are a local construction: like the characteristics they follow, they focus and cross, producing caustics where a smooth solution of $\Box x^\mu=0$ ceases to exist. The explicit Schwarzschild chart $r_H=r-M$ is regular from infinity down to the horizon $r_H=M$ but does not extend across it, and for merging black holes the free-harmonic grid degrades quickly. This is why numerical relativity uses generalized harmonic gauge with dynamical $H^\mu$ plus black-hole excision, rather than a single global chart.

Why does gravity have only two propagating polarizations if there are ten wave equations?

The ten wave equations are honest, but most of their content is not physical radiation: four of the ten metric components are pure coordinate freedom removed by $\Gamma^\mu=H^\mu$ , and four more are fixed by the Hamiltonian and momentum constraints. Ten minus four minus four leaves the two transverse-traceless polarizations $h_+$ and $h_\times$ — the only patterns that carry energy to infinity. Evolving all ten is a computational convenience, not a claim that spacetime ripples in ten independent ways.