Gauge Conditions in Numerical Relativity

A solution of Einstein's equations is a geometry, not a coordinate chart — yet a computer can only ever integrate numbers on a grid, and those numbers are coordinate-dependent. This is the tension at the heart of numerical relativity. In the Cauchy (3+1) approach one slices spacetime into spacelike sheets $\Sigma_t$ and marches the induced metric $\gamma_{ij}$ and extrinsic curvature $K_{ij}$ forward in a coordinate time $t$ . But $t$ itself, and the spatial labels carried from one slice to the next, are pure gauge. The lapse function $\alpha$ sets how much proper time separates neighbouring slices along their normal; the shift vector $\beta^i$ sets how the spatial coordinates are dragged sideways as the slice is laid down. Choosing $\alpha$ and $\beta^i$ is the act of drawing a coordinate system onto the spacetime as it is built, and the Einstein evolution equations leave these four functions completely free. That freedom is a loaded gun. Point it carelessly — geodesic slicing, $\alpha=1$ , $\beta^i=0$ — and the coordinate observers fall straight into curvature, the slices fold into caustics, and the code crashes within a light-crossing time, long before any physical singularity. The craft of numerical relativity is to prescribe gauge conditions that keep coordinates civilized: that avoid singularities, resist grid stretching, respond to the geometry, and asymptote to familiar coordinates far away so that masses, spins, and waveforms can be read off. The modern answer is not to fix $\alpha$ and $\beta^i$ by hand but to give them their own hyperbolic evolution equations — the 1+log slicing and the Gamma-driver shift — that together form the moving-puncture gauge, the single ingredient that finally cracked the binary-black-hole problem in 2005.

Key takeaways

In the 3+1 split the lapse $\alpha$ and shift $\beta^i$ are not dynamical fields but gauge functions: they encode the freedom to slice and thread spacetime with coordinates, and Einstein's equations never fix them.
Geodesic slicing ( $\alpha=1$ , $\beta^i=0$ ) fails catastrophically — normal observers focus and $\partial_t K\ge\tfrac13 K^2$ drives a coordinate caustic in finite time (Schwarzschild crashes at $\tau=\pi M$ ).
Maximal slicing $K=0$ avoids singularities beautifully but costs a global elliptic solve every step; hyperbolic "live" gauges that evolve $\alpha,\beta^i$ are cheaper and parallelize.
The workhorse slicing is 1+log, $\partial_t\alpha=-2\alpha K$ , the $f=2/\alpha$ member of the Bona–Massó family; its gauge mode travels at coordinate speed $\sqrt{2\alpha}$ .
The Gamma-driver evolves the shift to chase the BSSN connection functions $\tilde\Gamma^i$ , killing slice-stretching and admitting a stationary trumpet end-state that lets a puncture translate across a fixed grid.
Gauge choice is tied to the formulation and to well-posedness: the combined metric-plus-gauge system must be strongly hyperbolic, which is why the $3/4$ and $f=2/\alpha$ constants are not arbitrary.
Key idea: lapse $\alpha$ and shift $\beta^i$ are non-dynamical gauge functions; rather than fixing them by hand one evolves them with hyperbolic "live" equations — 1+log slicing for the lapse and the Gamma-driver for the shift — so the coordinates respond to the geometry, avoid singularities, and resist grid distortion, enabling moving-puncture black-hole evolutions.

The core insight

Lapse $\alpha$ and shift $\beta^i$ are non-dynamical gauge functions; rather than fixing them by hand one evolves them with hyperbolic "live" equations — 1+log slicing for the lapse and the Gamma-driver for the shift — so that the coordinates respond to the geometry, avoid singularities, and resist grid distortion, enabling moving-puncture black-hole evolutions.

Where this lives today

Gauge conditions are the running gear of gravitational-wave astronomy. Every numerical-relativity waveform in the LIGO–Virgo–KAGRA template banks came from a code that had to make exactly these choices, and the two dominant strategies map onto the two great code families. The moving-puncture codes — the BSSN and CCZ4 evolutions descended from the 2006 breakthrough — run 1+log slicing and the Gamma-driver shift essentially verbatim, letting black holes drift across a fixed Cartesian mesh as regularized punctures with no excision boundary. The spectral codes, above all the Spectral Einstein Code (SpEC) and its successor SpECTRE, instead use the generalized-harmonic formulation with damped-harmonic gauge sources and surgical excision of the interiors, delivering the long, high-accuracy inspirals that calibrate the effective-one-body and phenomenological models. The choice is not cosmetic: the decades-long failure to simulate a merger was in large part a gauge failure, with well-matched formulations crashing at merger because their coordinate conditions went ill-posed under the extreme fields. The same conditions govern the frontier now — neutron-star mergers where the shift must survive a hypermassive remnant and its possible collapse, high-spin binaries that stress the trumpet end-state, and the sub-percent waveform accuracy that Einstein Telescope, Cosmic Explorer, and LISA will demand. Before any of these runs the gauge is vetted on the "Apples-with-Apples" test beds — the propagating gauge wave and the random-noise robust-stability test — built precisely to expose an ill-posed gauge–formulation pairing before it ruins a production run.

The physical picture

Build a four-dimensional spacetime the way a computer must: one spacelike slice at a time, stacking cards. The lapse $\alpha$ tells you how much proper time elapses when you step straight up off the current card along its unit normal $n$ ; the shift $\beta^i$ tells you how far you slide the new card sideways before laying it down. Neither changes the deck — the geometry is fixed by the initial data and the field equations — but together they decide how the gridlines drawn on one card connect to those on the next. The coordinate-time direction is $\partial_t=\alpha n+\beta^i\partial_i$ : a piece straight up (the lapse) plus a piece tangent to the slice (the shift). The first figure makes this the whole story.

Figure

The 3+1 foliation of spacetime: the lapse α advances proper time α dt along the unit normal n from slice Σₜ to Σₜ₊dₜ, while the shift βⁱ drags the spatial coordinates sideways by βⁱ dt — both are freely choosable gauge functions, and 1+log slicing evolves the lapse by ∂ₜα = −2αK so it collapses toward zero wherever the mean curvature K blows up

Singularity avoidance is nothing more than the lapse learning to collapse. Where curvature blows up, a good slicing drives $\alpha\to0$ , so proper time freezes and the coordinate observers never actually reach the singularity even as their physical infall continues in the true geometry. The 1+log condition does this automatically: $\partial_t\alpha=-2\alpha K$ , and the mean curvature $K$ grows without bound near a forming singularity, so the lapse is dragged down exactly where it must be. The shift plays the complementary role of damage control on the spatial grid. Near a black hole the infalling normal observers would pile coordinate lines toward the horizon and stretch the mesh until it tears; the Gamma-driver pushes the spatial coordinates outward to fight this "slice-stretching," holding the connection functions $\tilde\Gamma^i$ — a proxy for grid distortion — from running away. Together the two conditions conspire to turn the black hole's interior into a stationary trumpet: a geometry with a nonzero limiting lapse at a finite areal radius and an infinitely long throat, so the singularity is never on the grid and the whole structure can translate rigidly as the hole orbits. The second figure is the quantitative signature of that collapse.

Figure

The collapse of the lapse: for the Schwarzschild trumpet the singularity-avoiding lapse α(R) climbs from 0 at the limiting throat radius R₀ = 3M/2 toward 1 far away, passing α ≈ 0.33 near R = 2M — geodesic slicing instead holds α = 1 everywhere and crashes into the singularity at proper time τ = πM ≈ 15.5 μs per solar mass

The core equations

The gauge lives in the line element. Writing the 4-metric in 3+1 form makes $\alpha$ and $\beta^i$ appear explicitly as the ingredients that connect one slice to the next, alongside the dynamical spatial metric $\gamma_{ij}$ .

\mathrm{d}s^2=-\alpha^2\,\mathrm{d}t^2+\gamma_{ij}\,(\mathrm{d}x^i+\beta^i\,\mathrm{d}t)(\mathrm{d}x^j+\beta^j\,\mathrm{d}t) \qquad (1)

where: $\alpha$ — lapse, proper time advanced per unit coordinate time along the slice normal; $\beta^i$ — shift, tangential drag of the spatial coordinates between slices; $\gamma_{ij}$ — induced spatial 3-metric on $\Sigma_t$ (dynamical); $x^i,t$ — spatial and time coordinates whose threading $\alpha,\beta^i$ define. Read it as: the true separation of two events is a proper-time part ( $\alpha\,\mathrm{d}t$ ) minus a spatial part measured after undoing the coordinate slide $\beta^i\,\mathrm{d}t$ . The dynamics enter through the metric's own evolution, and this is where lapse and shift make their presence felt. The rate of change of $\gamma_{ij}$ is set by the extrinsic curvature (the "velocity" of the slice) scaled by the lapse, plus a Lie-drag along the shift.

\partial_t\gamma_{ij}=-2\alpha K_{ij}+D_i\beta_j+D_j\beta_i \qquad (2)

where: $K_{ij}$ — extrinsic curvature of the slice embedded in spacetime; $D_i$ — covariant derivative compatible with $\gamma_{ij}$ . Setting $\alpha=1,\beta^i=0$ recovers geodesic slicing; the shift terms are precisely the freedom to reshuffle spatial labels without touching the geometry. The first genuine gauge condition historically was maximal slicing, defined by demanding the slices have vanishing mean curvature, $K=\gamma^{ij}K_{ij}=0$ , for all time. Preserving that demand turns into an elliptic equation for the lapse.

D^iD_i\alpha=\alpha\,\big(K_{ij}K^{ij}+4\pi(\rho+S)\big) \qquad (3)

where: $\rho$ — energy density measured by the normal observer; $S$ — trace of the spatial stress tensor $S_{ij}$ ; the operator $D^iD_i$ is the covariant Laplacian on the slice. Because the right-hand side is non-negative, the maximum principle forces $\alpha$ to dip toward zero wherever curvature is large — the "collapse of the lapse." To escape the cost of a global elliptic solve every step, one replaces the condition with a hyperbolic evolution. The Bona–Massó family is the general single-function slicing that keeps the coupled system well-posed.

\partial_t\alpha=-\alpha^2 f(\alpha)\,K,\qquad v_g=\alpha\sqrt{f(\alpha)} \qquad (4)

where: $f(\alpha)>0$ — the free function selecting the member; $v_g$ — the coordinate speed of the resulting gauge mode, found by linearizing about flat space; $K$ — mean curvature, large near a forming singularity. Harmonic slicing is $f=1$ ( $v_g=\alpha$ , luminal); the production choice $f=2/\alpha$ is 1+log, with $v_g=\sqrt{2\alpha}$ . Written in the form actually integrated in a code — and made comoving with the shifted coordinates by an advection term — the 1+log slicing is a cheap, local, parallelizable update.

\partial_t\alpha=\beta^i\partial_i\alpha-2\alpha K \qquad (5)

where: $\beta^i\partial_i\alpha$ — advection so that the condition is dragged along with the moving coordinates; $-2\alpha K$ — the $f=2/\alpha$ Bona–Massó driving term. This is the single most-used slicing condition in the moving-puncture era. The shift needs its own live equation. The Gamma-driver evolves $\beta^i$ to chase the BSSN conformal connection functions $\tilde\Gamma^i$ , which measure how badly the spatial grid is being distorted, with an auxiliary field and a damping term that make it a well-behaved second-order hyperbolic system.

\partial_t\beta^i=\tfrac34 B^i,\qquad \partial_t B^i=\partial_t\tilde\Gamma^i-\eta\,B^i \qquad (6)

where: $\tilde\Gamma^i=-\partial_j\tilde\gamma^{ij}$ — BSSN conformal connection functions, a proxy for grid distortion; $B^i$ — auxiliary field making the driver second-order hyperbolic; $\eta$ — damping parameter with dimension of inverse mass, typically $\eta\sim 1\text{–}2/M$ , controlling puncture drift; $\tfrac34$ — the coefficient tuned so the shift's gauge speed matches the slicing's, preserving strong hyperbolicity. A wholly different but equally valid strategy hides the gauge inside the coordinates themselves. The generalized harmonic formulation promotes the four coordinates $x^\mu$ to fields obeying wave equations with prescribed source functions.

H^\mu\equiv\Box x^\mu=-g^{\alpha\beta}\Gamma^\mu_{\alpha\beta},\qquad H^\mu=0\ (\text{harmonic}) \qquad (7)

where: $H^\mu$ — prescribed gauge source functions encoding the coordinate choice; $\Gamma^\mu_{\alpha\beta}$ — Christoffel symbols, contracted; $\Box=g^{\mu\nu}\nabla_\mu\nabla_\nu$ — covariant d'Alembertian. Setting $H^\mu=0$ gives harmonic coordinates; damped-harmonic choices add singularity-avoiding terms and are the workhorse of excision-based spectral codes.

Derivation AWhy geodesic slicing focuses the grid into a caustic

The goal is to show, from the trace of the extrinsic-curvature evolution equation, that the innocent-looking choice $\alpha=1,\ \beta^i=0$ is not merely inefficient but genuinely fatal: the mean curvature obeys a Raychaudhuri-type inequality that forces it to diverge in a finite coordinate time even in vacuum, producing a coordinate caustic where neighbouring normal geodesics cross and the grid tears — the formal reason a live, singularity-avoiding lapse is mandatory rather than a convenience.

Step 1. Start from the trace of the $K_{ij}$ evolution equation, which reads $\partial_t K=\beta^i\partial_i K-D^iD_i\alpha+\alpha(K_{ij}K^{ij}+4\pi(\rho+S))$ — the same relation whose vanishing gave maximal slicing. Why: this is the master equation for how fast the slices are converging, so any slicing choice shows its consequences here first.

Step 2. Impose geodesic slicing, $\alpha=1$ and $\beta^i=0$ . The advection term drops, the Laplacian term $D^iD_i\alpha$ vanishes because $\alpha$ is constant, and one is left with $\partial_t K=K_{ij}K^{ij}+4\pi(\rho+S)$ . Why: with unit lapse the normal observers are freely falling geodesics, so this is literally the Raychaudhuri focusing equation for the congruence that defines the coordinates.

Step 3. Bound the shear-free part with Cauchy–Schwarz: for any symmetric $K_{ij}$ one has $K_{ij}K^{ij}\ge\tfrac13(\gamma^{ij}K_{ij})^2=\tfrac13 K^2$ . Even in vacuum, where $\rho=S=0$ , this leaves $\partial_t K\ge\tfrac13 K^2$ . Why: it converts the exact evolution into a differential inequality that can be integrated without knowing the details of the geometry — the collapse is universal, not special to one spacetime.

Step 4. Integrate the inequality. If the slice is initially converging, $K_0>0$ , then $K(t)\ge K_0/(1-\tfrac13 K_0 t)$ , which diverges at a finite coordinate time $t_\ast\le 3/K_0$ . The normal geodesics have focused to a caustic; the coordinate volume element collapses and the discretized fields develop unbounded gradients. Why: this is a coordinate singularity — pure gauge pathology — arriving strictly before any curvature singularity, so it crashes the code for no physical reason at all.

Step 5. Anchor it to a known case. For a momentarily static observer released at the throat ( $R=2M$ ) of Schwarzschild, radial free fall reaches $r=0$ in proper time $\tau=\pi M$ , so geodesic slicing of Schwarzschild dies at $\tau=\pi M$ — about $15.5\ \mu\text{s}$ per solar mass. Why: it turns the abstract inequality into the concrete textbook number that motivated the whole search for singularity-avoiding gauges.

Go deeper · Derivation B · 2 minThe 1+log slicing condition and its logarithmic integral

Derivation BThe 1+log slicing condition and its logarithmic integral

The goal is to derive why the $f=2/\alpha$ member of the Bona–Massó family is called "1+log" and why it mimics maximal slicing so closely at a fraction of the cost. We take the general slicing condition, specialize $f$ , then use the trace of the metric evolution equation to relate the lapse to the spatial volume element, and integrate — exhibiting an algebraic relation $\alpha=1+\ln\gamma$ that both names the condition and makes its singularity-avoiding behaviour transparent.

Step 1. Insert $f(\alpha)=2/\alpha$ into the Bona–Massó condition $\partial_t\alpha=-\alpha^2 f(\alpha)K$ . The two powers of $\alpha$ collapse to one, giving $\partial_t\alpha=-\alpha^2\cdot(2/\alpha)\,K=-2\alpha K$ in the zero-shift form. Why: this fixes which member of the one-parameter family we are studying, and it is exactly the production slicing that codes integrate.

Step 2. Relate the mean curvature to the volume element. Taking the trace of $\partial_t\gamma_{ij}=-2\alpha K_{ij}$ (with $\beta^i=0$ ) and using $\partial_t\ln\sqrt\gamma=\tfrac12\gamma^{ij}\partial_t\gamma_{ij}$ gives $\partial_t\ln\sqrt\gamma=-\alpha K$ . Why: it re-expresses the abstract quantity $K$ as the fractional rate at which a coordinate cell's proper volume is shrinking — the very thing a code cares about.

Step 3. Combine the two. Since $\partial_t\alpha=-2\alpha K$ and $-\alpha K=\partial_t\ln\sqrt\gamma$ , we have $\partial_t\alpha=2\,\partial_t\ln\sqrt\gamma=\partial_t\ln\gamma$ . The two time derivatives are proportional, so the relation integrates immediately. Why: proportional time derivatives mean the condition has a first integral — a purely algebraic tie between lapse and metric, unusual and useful.

Step 4. Integrate to $\alpha=1+\ln\gamma$ (fixing the slice-independent constant to reproduce $\alpha=1$ in flat space). The lapse equals one plus the logarithm of the determinant ratio — hence "1+log." Where matter or curvature compresses a cell, $\gamma\to0$ , the logarithm drives $\alpha$ down toward zero. Why: this is the collapse of the lapse again, now visible as an explicit formula, and it is why 1+log tracks maximal slicing so faithfully near strong fields.

Step 5. Note the practical payoff. Unlike maximal slicing's global elliptic solve (equation 3), equation (5) is a local time-evolution equation with a finite gauge speed $v_g=\sqrt{2\alpha}$ ; it is cheap, causal, and parallelizes across a grid. Why: this is the entire reason hyperbolic "live" slicings displaced the elliptic ones in production numerical relativity.

Go deeper · Derivation C · 2 minThe stationary trumpet that lets a puncture sit still

Derivation CThe stationary trumpet that lets a puncture sit still

The goal is to show that the coupled 1+log and Gamma-driver system admits a genuinely time-independent solution — the trumpet — and that its existence is exactly what allows a black hole to be represented as a puncture that translates rigidly across a fixed mesh without excision. We look for a stationary state of the full live gauge, show what each equation demands, and identify the resulting geometry and its characteristic radius for Schwarzschild.

Step 1. Demand stationarity, $\partial_t\alpha=\partial_t\beta^i=\partial_t\tilde\Gamma^i=0$ . The shift equation $\partial_t\beta^i=\tfrac34 B^i=0$ immediately forces the auxiliary field $B^i=0$ . Why: a time-independent shift can carry no residual "velocity," so the driver's internal variable must switch off first.

Step 2. Feed $B^i=0$ into $\partial_t B^i=\partial_t\tilde\Gamma^i-\eta B^i=0$ , which reduces to $\partial_t\tilde\Gamma^i=0$ . The connection functions — the proxy for grid distortion — have stopped growing; slice-stretching is defeated. Why: this is the shift doing its job, freezing the spatial grid so nothing tears even though the interior geometry is enormously strained.

Step 3. Impose stationarity on the advective 1+log slicing (equation 5): $\partial_t\alpha=\beta^i\partial_i\alpha-2\alpha K=0$ , i.e. $\beta^i\partial_i\alpha=2\alpha K$ . The inward advection by the shift exactly balances the $-2\alpha K$ collapse, permitting a nonzero stationary lapse away from the throat. Why: without the shift there would be no balance and the lapse would simply collapse to zero everywhere; the shift is what keeps a live, workable coordinate patch.

Step 4. Identify the geometry. For Schwarzschild this stationary state is a trumpet slice whose areal radius is bounded below by a limiting throat value. For the maximal ( $K=0$ ) trumpet this is exactly $R_0=3M/2$ , where $\alpha\to0$ ; the proper distance down the throat is infinite, so $R=0$ is never on the grid. The 1+log trumpet is the close hyperbolic cousin, with $R_0\approx1.3M$ . Why: a lower-bounded areal radius with infinite proper depth is precisely a geometry a finite grid can hold without ever resolving the singularity.

Step 5. Conclude why punctures can move. Because the gauge possesses a regular stationary attractor, a boosted copy of it translates rigidly; the damping $\eta$ in the Gamma-driver sets how fast transients settle as the hole orbits, which is why $\eta\sim1\text{–}2/M$ is chosen for stability. Why: this stationary-attractor property, not any one equation in isolation, is the deep reason moving punctures work.

Limiting cases

Case	Result	Meaning
Geodesic slicing	$\alpha=1,\ \beta^i=0$	Normal observers are geodesics; $\partial_t K\ge\tfrac13 K^2$ focuses them into a caustic and the code crashes within a light-crossing time. Clean pedagogy, useless for black holes.
Maximal slicing	$K=0\Rightarrow D^iD_i\alpha=\alpha K_{ij}K^{ij}$	Elliptic equation for $\alpha$ ; strongly singularity-avoiding, keeps Schwarzschild slices spacelike forever, but needs a global solve each step — costly and hard to parallelize.
1+log slicing	$\partial_t\alpha-\beta^i\partial_i\alpha=-2\alpha K$	Hyperbolic; mimics maximal slicing at a fraction of the cost with gauge speed $\sqrt{2\alpha}$ (superluminal coordinate mode, harmless). The standard production slicing.
Gamma-driver shift	$\partial_t\beta^i=\tfrac34 B^i,\ \partial_t B^i=\partial_t\tilde\Gamma^i-\eta B^i$	Second-order hyperbolic driver chasing $\tilde\Gamma^i$ to kill slice-stretching; damping $\eta\sim1\text{–}2/M$ tunes puncture motion and stability.
Damped harmonic (GHG)	$H_\mu=\mu_0\big(\ln\tfrac{\sqrt g}{\alpha}\big)n_\mu-\mu_2\,\tfrac1\alpha\gamma_{\mu i}\beta^i$	The generalized-harmonic analogue: gauge source functions chosen to damp coordinate distortions, used in excision-based spectral binary evolutions.

Common trapGauge is free, but bad gauge crashes good physics

The single most common conceptual error is to treat $\alpha$ and $\beta^i$ as if Einstein's equations determine them. They do not — the contracted Bianchi identities guarantee that four of the ten field equations are constraints with no second time derivative, so the four functions $\alpha,\beta^i$ are pure gauge, chosen by the modeller. The practical trap is the flip side: because they are free, a careless choice (geodesic slicing) will drive coordinate observers straight into the singularity and destroy an otherwise perfect simulation. A subtler trap is forgetting that gauge changes hyperbolicity. Pairing a slicing with the wrong shift, or with the wrong formulation, can render the coupled metric-plus-gauge system only weakly hyperbolic and therefore ill-posed — even though each piece looks reasonable in isolation. The constants $3/4$ in the Gamma-driver and $f=2/\alpha$ in 1+log are not decoration; they are tuned so the gauge characteristic speeds stay real and distinct from the physical ones. Gauge is not physics, but the wrong gauge will happily crash the right physics.

Worked examples

1 · How long before geodesic slicing dies (fully worked)

Estimate the time at which geodesic slicing of a Schwarzschild black hole crashes, and convert it to physical units for a $1\,M_\odot$ and a $60\,M_\odot$ hole.

Step 1. Use the focusing result. A momentarily static observer released at the throat, areal radius $R=2M$ , follows a radial geodesic to $r=0$ . The cycloid solution $r=(R_0/2)(1+\cos\eta)$ , $\tau=\sqrt{R_0^3/8M}\,(\eta+\sin\eta)$ reaches $r=0$ at $\eta=\pi$ , giving

\tau=\pi\sqrt{\frac{R_0^3}{8M}}\bigg|_{R_0=2M}=\pi\sqrt{\frac{8M^3}{8M}}=\pi M.

Step 2. Because geodesic slicing marches the grid along exactly these normal geodesics with $\alpha=1$ , the foliation itself hits the singularity at coordinate time $t=\tau=\pi M$ . This sits comfortably inside the focusing bound $t_\ast\le 3/K_0$ from the derivation — the exact spacetime collapses even faster than the crude inequality guarantees. The crash time is $t=\pi M$ .

Step 3. Convert with $GM_\odot/c^3=4.926\ \mu\text{s}$ . For $1\,M_\odot$ , $t=\pi M=\pi\times4.926\ \mu\text{s}=\mathbf{15.5\ \mu s}$ . For a $60\,M_\odot$ hole the crash comes at $60\times15.5\ \mu\text{s}\approx\mathbf{0.93\ ms}$ — less than a single orbit of a binary at merger. A code on geodesic slicing would die before the black holes even touch, which is precisely why singularity-avoiding slicing is non-negotiable.

2 · Gauge speeds of the live slicings (fully worked)

Compare the coordinate propagation speed of the gauge mode for harmonic slicing ( $f=1$ ) and 1+log slicing ( $f=2/\alpha$ ) at a far-field lapse $\alpha=1$ and at a near-puncture lapse $\alpha=0.3$ . Is anything wrong when the speed exceeds $c=1$ ?

Step 1. Use $v_g=\alpha\sqrt{f(\alpha)}$ from equation (4). For harmonic slicing $f=1$ , so $v_g=\alpha$ : at $\alpha=1$ , $v_g=1$ (exactly luminal); at $\alpha=0.3$ , $v_g=0.3$ .

Step 2. For 1+log, $f=2/\alpha$ , so $v_g=\alpha\sqrt{2/\alpha}=\sqrt{2\alpha}$ . At $\alpha=1$ , $v_g=\sqrt2=\mathbf{1.414}$ ; at $\alpha=0.3$ , $v_g=\sqrt{0.6}=\mathbf{0.775}$ . The 1+log gauge mode is superluminal wherever $\alpha>\tfrac12$ (since $\sqrt{2\alpha}>1$ requires $\alpha>\tfrac12$ ) and subluminal in the collapsed region near the puncture.

Step 3. Nothing is wrong. The gauge speed is the coordinate speed of a pure coordinate wave — a rearrangement of the labels, carrying no energy, momentum, or signal. Only the physical characteristic speeds (the light cones and the matter modes) are bounded by $c$ , and they are untouched. What the number does matter for is well-posedness: the $3/4$ in the Gamma-driver exists precisely so the shift's gauge speed matches this $\sqrt{2\alpha}$ and the coupled system stays strongly hyperbolic. Superluminal coordinate modes are routine and harmless; superluminal physical modes would signal an ill-posed system.

3 · Reading the collapse off the 1+log integral

Using $\alpha=1+\ln\gamma$ , estimate the lapse in a cell whose proper volume element $\sqrt\gamma$ has been compressed to $10\%$ of its flat-space value.

Step 1. With $\sqrt\gamma=0.1$ , the determinant ratio is $\gamma=(\sqrt\gamma)^2=0.01$ , so $\ln\gamma=\ln(0.01)=-4.6$ .

Step 2. Then $\alpha=1+\ln\gamma=1-4.6=-3.6$ formally. A negative lapse is unphysical, so in practice a code floors $\alpha$ at a small positive value; the tendency is what matters — strong compression drives $\alpha$ sharply down toward zero, freezing coordinate time exactly where the geometry is trying to run away. This is the collapse of the lapse in a single line of arithmetic, and it is why 1+log is singularity-avoiding.

Physics of the portrait

Gauge conditions were the central practical obstacle of numerical relativity for three decades, and no one framed them more sharply — or hit them first — than the man who ran the earliest black-hole collision on a computer and coined the language the field still uses.

Origins · 1 minLarry Smarr and how this was discovered

Larry Smarr

b. 1948 · gauge conditions and the collapse of the lapse in numerical relativity

Larry Smarr is the American astrophysicist who, in the mid-1970s, performed with Kenneth Eppley the first numerical simulation of the head-on collision of two black holes — a calculation that would have been impossible without a way to keep the coordinates away from the singularities forming inside each hole. Smarr's tool was maximal slicing ( $K=0$ ), and his simulations made vivid the phenomenon that became the field's watchword: the "collapse of the lapse," in which $\alpha$ automatically dives toward zero in the strong-field region, freezing proper time before the grid can reach the curvature blow-up. In 1978 he and James York wrote Kinematical Conditions in the Construction of Spacetime, the paper that codified gauge conditions as a subject in their own right: it introduced maximal slicing and the minimal-distortion shift as the canonical elliptic gauge, arguing that the lapse should be chosen to avoid singularities and the shift to minimize the rate of change of the spatial metric. Those two ideas are the direct ancestors of today's hyperbolic replacements — 1+log slicing inherits the singularity-avoidance of maximal slicing, and the Gamma-driver inherits the distortion-minimizing intent of the minimal-distortion shift. Smarr went on to found the National Center for Supercomputing Applications in 1985, where the graphical web browser Mosaic was later built, but his fingerprint on this article is the recognition that in general relativity the coordinates are not given: they must be constructed, and constructing them well is the whole game.

History

1962

— Arnowitt, Deser and Misner recast general relativity as a 3+1 evolution, exposing lapse and shift as freely specifiable functions

1978

— Smarr and York codify maximal slicing and the minimal-distortion shift as gauge conditions for numerical relativity

1995

— Bona, Massó, Seidel and Stela introduce the hyperbolic Bona–Massó slicing family, including 1+log

2006

— the RIT and Goddard groups discover the moving-puncture gauge (1+log slicing plus Gamma-driver shift), cracking the binary-black-hole problem

2015

— GW150914 is decoded with waveform templates built on moving-puncture numerical-relativity simulations

Going further

Texts. Misner, Thorne & Wheeler, Gravitation (1973), Ch. 21 (the ADM action and the 3+1 split) · Wald, General Relativity (1984), Ch. 10 and Appendix E · Carroll, Spacetime and Geometry (2004), Ch. 4 (the ADM decomposition) · Baumgarte & Shapiro, Numerical Relativity (2010), Ch. 4 (gauge conditions) and §13.1 (trumpets) · Alcubierre, Introduction to 3+1 Numerical Relativity (2008), Ch. 4 · Bona, Massó, Seidel & Stela, Phys. Rev. Lett. 75, 600 (1995) · Campanelli, Lousto, Marronetti & Zlochower, Phys. Rev. Lett. 96, 111101 (2006); Baker, Centrella, Choi, Koppitz & van Meter, Phys. Rev. Lett. 96, 111102 (2006); Pretorius, Phys. Rev. Lett. 95, 121101 (2005).

Provenance. A verifier should confirm: (1) the ADM line element and the statement that $\alpha,\beta^i$ never carry a time derivative in Einstein's equations (MTW Ch. 21; Wald App. E; Carroll Ch. 4); (2) the focusing inequality $\partial_t K\ge\tfrac13 K^2$ under geodesic slicing and the Schwarzschild crash at $\tau=\pi M$ , verified here from the radial-infall cycloid ( $\tau=\pi\sqrt{R_0^3/8M}=\pi M$ at $R_0=2M$ ; $15.5\ \mu\text{s}$ per solar mass using $GM_\odot/c^3=4.926\ \mu\text{s}$ ) — Baumgarte & Shapiro Ch. 4; (3) the Bona–Massó family with gauge speed $v_g=\alpha\sqrt{f}$ and the $f=2/\alpha$ integral $\alpha=1+\ln\gamma$ , with $v_g=\sqrt2$ at $\alpha=1$ (Bona et al. 1995; Alcubierre Ch. 4); (4) the Gamma-driver with its $3/4$ coefficient and damping $\eta\sim1\text{–}2/M$ ; (5) the maximal-slicing trumpet with limiting throat $R_0=3M/2$ where $\alpha\to0$ , and $\alpha\approx0.33$ at $R=2M$ , recomputed here from $\alpha=\sqrt{1-2M/R+C^2/R^4}$ with $C=\tfrac{3\sqrt3}{4}M^2$ (Estabrook et al. 1973; Baumgarte & Shapiro §13.1); (6) the moving-puncture and generalized-harmonic breakthrough references. Geometric units $G=c=1$ are used except in the timed examples.

Prerequisites. ADM 3+1 Decomposition of Spacetime · Hamiltonian and Momentum Constraints · BSSN Formulation · Initial-Value (Cauchy) Problem in General Relativity. Leads to. Numerical Relativity and Binary Black Hole Mergers · Black Hole Excision and Puncture Methods · Harmonic and Generalized Harmonic Gauge · Apparent Horizons and Horizon Finders.

Ask the tutor: Ask me to walk the geodesic-slicing focusing bound line by line, integrate the 1+log condition to $\alpha=1+\ln\gamma$ , or explain why the Gamma-driver's $3/4$ and $\eta$ are exactly what keep a moving-puncture evolution strongly hyperbolic.

If the lapse and shift are pure gauge, why does the choice matter at all?

Because "gauge" means physically irrelevant, not numerically irrelevant. The final geometry is independent of how you slice and thread it, but the intermediate numbers a code carries are entirely coordinate-dependent, and a bad choice makes those numbers blow up for reasons that have nothing to do with physics. Geodesic slicing sends coordinate observers into the singularity and produces a caustic; a mismatched slicing–shift pair can make the evolution equations only weakly hyperbolic and therefore ill-posed, so tiny errors amplify without bound. A good gauge keeps the coordinates smooth, the fields bounded, and the system strongly hyperbolic. So the gauge changes nothing about the spacetime and everything about whether you can actually compute it.

What exactly is the "collapse of the lapse"?

It is the automatic tendency of a singularity-avoiding slicing to drive the lapse $\alpha$ toward zero in regions of strong curvature. Since $\alpha$ measures how much proper time advances per step of coordinate time, $\alpha\to0$ means coordinate time nearly freezes there. Physically, the normal observers keep falling in the true geometry, but the slices stop advancing into the strong-field region, so the foliation never reaches the singularity and the grid is spared. Maximal slicing produces this through its elliptic equation, whose non-negative source forces $\alpha$ down by the maximum principle; 1+log produces it through $\partial_t\alpha=-2\alpha K$ , since $K$ grows large near a forming singularity. The name comes from Larry Smarr's 1970s black-hole simulations, where the effect was first seen and exploited.

How is the moving-puncture gauge different from excision?

Both keep the singularity off the usable grid, but by opposite philosophies. Excision cuts a hole in the computational domain just inside the apparent horizon and simply does not evolve the interior; because nothing causal can escape the horizon, no boundary condition is needed there, but the excision surface must be tracked and moved as the black hole orbits — delicate bookkeeping. The moving-puncture approach keeps the entire domain, represents the black hole as a coordinate singularity (a puncture), and lets the 1+log slicing and Gamma-driver shift dynamically stretch the interior into a trumpet with a nonzero limiting lapse and an infinitely long throat. The singularity ends up at infinite proper distance, never on the grid, and the puncture can drift freely across a fixed mesh. Excision suits spectral generalized-harmonic codes; punctures suit finite-difference BSSN codes.

Why does the shift need its own second-order equation with an auxiliary field?

A first-order shift condition tied directly to $\tilde\Gamma^i$ tends to overreact: it damps grid distortion but rings, and can destabilize the coupled system. Introducing the auxiliary field $B^i$ and writing $\partial_t\beta^i=\tfrac34 B^i$ , $\partial_t B^i=\partial_t\tilde\Gamma^i-\eta B^i$ turns the shift into a genuinely second-order-in-time, damped hyperbolic system — a driven, damped oscillator chasing the connection functions rather than snapping to them. The damping $\eta$ controls how fast transients settle and how quickly a puncture is allowed to drift; too little and the shift oscillates, too much and it stiffens. The $\tfrac34$ coefficient is chosen so the shift's gauge speed matches the $\sqrt{2\alpha}$ of 1+log, which is what keeps the whole metric-plus-gauge system strongly hyperbolic.

Is the trumpet a real geometry or a coordinate artifact?

It is a coordinate representation — a particular slicing — of a perfectly ordinary black hole, but a very cleverly chosen one. The Schwarzschild spacetime is fixed; the trumpet is what its stationary maximal or 1+log slices look like. Those slices do not cross the horizon into a second asymptotic region (as the maximally extended Kruskal picture would); instead they wrap down an infinitely long throat whose areal radius is bounded below by $R_0$ ( $3M/2$ for maximal slicing), so the singularity at $R=0$ is pushed to infinite proper distance and never appears on the grid. The physics — horizon area, mass, tidal fields outside — is unchanged. What the trumpet buys is a stationary, grid-friendly coordinate picture that a finite computer can hold indefinitely and translate rigidly as the hole moves.

Why is a superluminal gauge speed acceptable but a superluminal physical speed not?

Because the gauge speed describes how coordinate labels rearrange themselves, and labels carry no information. When the 1+log mode propagates at $\sqrt{2\alpha}$ , which exceeds $c$ for $\alpha>\tfrac12$ , all that is moving is the pattern of how you have chosen to name points — you cannot send a message on it, because the receiver would have to already know the naming convention everywhere. The physical characteristics of the system (the light cones of the metric, the sound speed of any matter) are set by the geometry and stay at or below $c$ ; the gauge speeds are extra characteristics introduced by your evolution equations. Well-posedness only requires that all characteristic speeds be real and that the gauge ones not coincide with or invert the physical ones — not that they be subluminal.

Does the same gauge work for the BSSN and generalized-harmonic formulations?

No, and this is one of the deepest lessons of the field: gauge and formulation are a package. The moving-puncture gauge — 1+log plus Gamma-driver — is calibrated for the conformal, first-order-in-time, second-order-in-space BSSN and CCZ4 systems, where its gauge speeds and damping are tuned to keep the coupled equations strongly hyperbolic. The generalized-harmonic formulation instead makes the coordinates themselves obey wave equations, so the gauge is specified through source functions $H^\mu=\Box x^\mu$ ; the singularity-avoiding role of 1+log is played there by damped-harmonic terms. Dropping the moving-puncture gauge into a harmonic code, or vice versa, generally produces an ill-posed system. The right question is never "which gauge is best" in the abstract, but "which gauge keeps this formulation strongly hyperbolic."