WorksPhysics and Astronomy
Gravitational collapse in Lorentz-violating gravity and the spin-0 mode
Abstract
Einstein-Aether theory is a theory of gravity with a preferred frame, in which Lorentz symmetry is broken by a dynamical unit timelike vector field. The theory is closely related to Horava gravity, and in spherically symmetric collapse the two produce the same dynamics. Both propagate a scalar (spin-0) mode besides the tensor mode and, in Einstein-Aether theory, the vector mode. The physical nature of the spin-0 mode has recently been called into question following the post-Newtonian analysis of Taherasghari and Will, whose post-Newtonian calculation of radiation reaction in compact binaries finds no trace of its speed in the energy loss, suggesting that the mode may not be a dynamical, energy-carrying wave. We test this with fully nonlinear evolutions of scalar-field collapse in spherical symmetry, where the spin-0 mode is the only radiative degree of freedom. The collapse emits an outgoing spin-0 wave that falls off as the inverse of the radius, travels at the speed predicted by the linear theory, and carries positive energy to infinity. The static aether configuration left around the black hole has negative energy instead, so that the horizon mass exceeds the ADM mass. For large couplings, collapse does not produce a black hole: we recover the transient white holes found by Akhoury, Garfinkle and Gupta, but at higher resolution the collapsing shell keeps steepening, not hidden behind any trapped surface, which may signal a naked singularity.
Introduction
Lorentz invariance has been tested to high precision in the matter sector [25, 26]. The bounds on its violation in the gravitational sector are weaker, because there are few consistent frameworks in which such violations can be parametrised beyond the weak field [3, 24, 26]. Einstein–æther theory [1, 2, 3] provides such a framework. It couples general relativity (GR) to a dynamical vector field, the æther, constrained to be timelike and of unit norm. The æther therefore selects a preferred frame at every point, breaking local boost invariance, while preserving general covariance. The action contains every term quadratic in the first derivatives of the æther, weighted by four dimensionless couplings, so the theory is the most general two-derivative theory of this kind. The linear theory about flat space propagates a spin-1 mode and a spin-0 mode besides the spin-2 mode of GR, with speeds fixed by the couplings [4].
Lorentz-violating gravity is also motivated by approaches to quantum gravity. Hořava [7] formulated a theory on a preferred foliation, with higher spatial derivatives but only second order in the time derivatives, which scales anisotropically in the ultraviolet and is power-counting renormalisable. Its projectable version, in which the lapse depends only on time, has been shown to be perturbatively renormalisable [11], whereas a general proof of perturbative renormalizability for the non-projectable theory is still lacking [67, 68]. The consistent non-projectable theory [8, 10] contains a scalar field with timelike gradient, the khronon, whose level sets are the leaves of the foliation. Its low-energy limit is Einstein–æther theory with the æther constrained to be the unit normal to the khronon foliation [9, 12]. In spherical symmetry the æther is necessarily hypersurface orthogonal, so every spherically symmetric solution of Einstein–æther theory also solves the low-energy Hořava equations. The converse also holds, when regularity at the universal horizon or at the centre is imposed [9, 10, 42, 44, 45].
The couplings of both theories are constrained by theory and by observation. Stability of the linear modes requires positive squared speeds [4] and positive energy [5, 41]. The observation of ultra-high-energy cosmic rays requires that no gravitational mode propagate slower than light, since the cosmic rays would otherwise lose their energy to gravitational Cherenkov radiation [27]. In the solar system the theory differs from GR only through the preferred-frame parameters and of the parametrised post-Newtonian formalism [23], which are tightly bounded [24]. The æther rescales the gravitational constant that drives the cosmological expansion relative to the one measured locally. The primordial abundances of light elements bound this rescaling [28], and the cosmic microwave background and large-scale structure give further bounds [29]. Compact objects acquire a dependence of their mass on their velocity relative to the æther, measured by their sensitivities [22, 30, 31]. This violates the strong equivalence principle and produces dipole gravitational radiation [6, 22, 30, 31]. The orbital decay of binary pulsars is sensitive to this dipole, and their timing has become the main probe of the couplings [21, 22, 30, 31, 35]. Finally, the coincident detection of GW170817 and GRB 170817A [18] fixed the speed of the spin-2 mode to that of light to within a part in . This requires the combination of the coupling parameters to vanish [14, 19, 20, 34]. The surviving parameter space is significantly restricted, but some directions remain weakly constrained due to parameter degeneracies [14, 19, 21].
These theories also alter the causal structure of black holes. Each mode propagates along the null cone of its own effective metric and is therefore associated with a distinct horizon. The horizon of a faster mode lies inside that of a slower one, so that a stationary black hole is bounded by a nested set of horizons. Static spherical solutions are singled out by regularity at the spin-0 horizon [38, 44], slowly rotating ones have been obtained to first order in the spin [49], and rapidly rotating ones have been constructed numerically by imposing regularity at each mode horizon [47]. In Hořava gravity the higher spatial derivatives make the dispersion relations anisotropic, schematically, so the group velocity grows without bound in the ultraviolet and no finite speed limits the propagation of signals [42, 45]. Static solutions nevertheless contain a universal horizon [44, 45]: a leaf of the preferred foliation at constant radius, located inside the mode horizons, which traps even arbitrarily fast signals [43, 46, 48]. The dynamical properties of these surfaces and whether they form during gravitational collapse cannot be determined from stationary solutions alone.
Gravitational collapse in these theories has been studied mostly in spherical symmetry. Garfinkle, Eling and Jacobson (GEJ) [36] evolved a massless scalar field, initially a spherical shell, in Einstein–æther theory. For moderate couplings, a black hole with a regular spin-0 horizon formed and settled to the static solution of Ref. [38]; at large couplings, the fields developed steep gradients, which GEJ read as a possible naked singularity. Akhoury, Garfinkle and Gupta [37] revisited this case using both the GEJ gauge and a second, zero-shift radial gauge, ingoing initial data and improved boundary conditions, and found instead a transient white hole: an anti-trapped surface that forms and later disperses. Bhattacharyya, Coates, Colombo and Sotiriou [42] showed that in spherical symmetry, with regularity at the centre or at the universal horizon, hypersurface-orthogonal Einstein–æther theory and low-energy Hořava gravity have identical dynamics, so the GEJ simulations also describe Hořava collapse. They interpreted the late-time vanishing of the preferred-frame lapse in those simulations as the dynamical formation of a universal horizon. Such formation had been shown for a thin shell collapsing in the cuscuton model [50], a scalar field with infinite sound speed that coincides with a special case of low-energy Hořava gravity, and was later found in the version of Hořava gravity that survives the GW170817 and solar-system bounds [13], where the metric couples exactly as in GR while the khronon foliation asymptotes to a limiting leaf. Whether universal horizons form generically, and whether gravitational collapse excites additional propagating gravitational degrees of freedom, are therefore central questions in these theories, and motivate the present work.
A related question concerns the dynamical status of the spin-0 mode. Recently Taherasghari and Will [32, 33] applied the post-Minkowskian expansion and the direct integration of the relaxed field equations to Einstein–æther theory, two methods that underlie high-order post-Newtonian results in GR. Since the unit-norm constraint spoils the usual form of the relaxed equations, they redefined the metric and æther variables to obtain flat-space wave equations with sources quadratic in the fields, solved them in the near zone to 2.5 post-Newtonian order [32], and derived the equations of motion of a compact binary, including radiation reaction at 1.5 and 2.5 post-Newtonian order, and its energy loss [33]. They found that the dipole term, the leading effect in the orbital decay, depends only on the speed of the transverse æther waves. Previous works by Foster [6, 30] and Yagi, Blas, Barausse and Yunes [31] instead computed the far-zone energy flux, and found that the dipole flux depends also on the speed of the longitudinal mode, i.e. the spin-0 speed of Eq. (3) below. In the relaxed equations of Ref. [33], by contrast, the metric potentials propagate at the speed of light, the æther potential at , and no wave equation carries .
The authors emphasize that the two calculations are conceptually distinct: one describes radiation reaction in the near zone, while the other yields the radiative flux in the far zone. Their equivalence relies on the energy-balance argument, which has been subject to only limited explicit verification even in GR and whose validity in alternative theories of gravity cannot be taken for granted. Consequently, they leave open whether energy balance fails in this theory or one of the two approaches is flawed. Their result, however, raises a further question: is the spin-0 mode a dynamical, energy-carrying wave propagating at its own speed, as the linear analysis indicates [4, 5], or is its apparent contribution to the radiated energy an artefact of the far-zone calculation? In the latter case, the dipole bounds [21, 22, 30, 31, 35] would have to be revisited, and the theory would differ qualitatively from its linearized description [4, 6].
Fully nonlinear spherical collapse is a natural place to address these questions. In spherical symmetry the spin-2 and spin-1 modes are absent, so the spin-0 mode is the only gravitational degree of freedom that can radiate, and its fall-off, speed and energy can be measured directly on the full solution rather than in a post-Newtonian expansion. We therefore return to the GEJ problem, starting from their couplings and extending them to a superluminal spin-0 speed, which separates the spin-0 wave from the scalar field, and to the large- case of Akhoury, Garfinkle and Gupta. These couplings lie far outside the observational bounds and serve as a laboratory for the dynamics of the theory rather than its phenomenology. Our results show that the spin-0 mode propagates dynamically at its characteristic speed and carries positive energy to infinity. This provides a different picture from that suggested by the post-Newtonian analysis of Refs. [32, 33], where the longitudinal mode does not appear in the radiation-reaction terms. We return to this apparent tension and discuss a possible origin of the discrepancy in § 1.4 E.
The paper is organised as follows. Section II reviews Einstein–æther theory, the GEJ formulation, our numerical method, and the couplings, initial data and diagnostics used. Section III B establishes the convergence of the free evolution. Section IV studies the Misner–Sharp mass profile and the horizon. § 1.4 identifies the spin-0 mode and measures its fall-off, its speed and the energy it carries and discusses the tension with the post-Newtonian results. § 1.5 treats collapse at large couplings and Section VII summarizes the results. Appendix A collects the convergence tests while Appendix B provides the technical details underlying the argument presented in § 1.4 E. We use units with unless otherwise stated, and adopt a metric signature . The code, the scripts that produce all figures and numbers of this paper, and symbolic checks of the equations are publicly available [65].
Einstein–æther theory
Einstein–æther theory is general relativity (GR) coupled to a dynamical, unit-norm, timelike vector field , the æther [2, 3, 15, 51]. Because is nowhere vanishing, it selects a preferred frame at every point and breaks local Lorentz invariance while keeping the theory generally covariant. It is the most general such theory with at most two derivatives acting on the fields. The action is
with
and a Lagrange multiplier that enforces the unit-norm condition. The four dimensionless couplings are also indicated through the combinations and . When the æther is hypersurface-orthogonal, as it necessarily is in spherical symmetry, the twist term can be removed and the theory coincides with the infrared limit of Hořava gravity [7, 8, 9, 10, 12, 13, 14].
Linearised about flat space with a constant æther, the theory has five propagating degrees of freedom: two spin-2, two spin-1 and one spin-0 mode [4], with squared speeds
The energy densities of the linear modes were computed by Eling [5] with the Weinberg and Einstein energy–momentum pseudotensors; that of the spin-0 mode has the sign of [3, 5], so the spin-0 mode carries positive energy only if . In spherical symmetry only the spin-0 mode survives: it is the only channel through which the spacetime can radiate apart from the matter itself.
Observationally, the parameter space is now tightly constrained [23, 27, 30, 31, 34, 35]. The coincident detection of GW170817 and GRB 170817A [18] bounds the spin-2 speed and hence [19, 20, 21]. The couplings used in this paper are outside this range. We choose them deliberately, as a numerical laboratory: two of them are the coupling sets of Refs. [36, 37], allowing a direct comparison, the third gives a superluminal spin-0 mode that separates from the scalar field, and all of them make the æther sector strongly nonlinear, which is the hardest test of the scheme. Our results should be read as statements about the dynamics of the theory, not as predictions for viable couplings.
Collapse in Einstein–æther theory
Black holes in Einstein–æther theory have a richer causal structure than in GR: each mode has its own horizon, and the stationary solutions are fixed by regularity at the spin-0 horizon [38, 39, 44, 45, 46, 47, 48, 49, 50]. We build on GEJ [36], who showed that such black holes form dynamically in scalar-field collapse. Using the GEJ formulation and a second radial gauge, Akhoury, Garfinkle and Gupta [37] later argued that at large , collapse can produce a transient white hole rather than a black hole. In the following, we review our numerical formulation and test the convergence of our set up.
Formulation and numerical method
Since in spherical symmetry the æther is hypersurface-orthogonal, GEJ [36] write it as , i.e., the spacetime foliation is chosen to match the æther-orthogonal slices. This is known as unitary gauge [10]. The lapse is therefore not a gauge variable, but is fixed by the æther through the kinematic identity
We work in spherical symmetry, with a massless scalar field normalised as in GEJ [36], so that the field equations read , i.e. is times a canonically normalised field. The spherically symmetric metric can be written as
where is the radial shift and is the areal radius, defined such that the symmetry spheres have area . We denote by the spatial metric induced on the æther-orthogonal slices, which determines the proper spatial line element . In this form of the metric measures proper radial distance along the slices, i.e. .
Following GEJ [36], the freely evolved variables are
namely the scalar, its momentum along the æther, the trace of the extrinsic curvature, the areal radius, and the radial æther acceleration. The shear of the extrinsic curvature, defined by and , the lapse and the shift are recovered on each slice. The 3+1 field equations in unitary gauge, for general matter and without symmetry, are given in Ref. [40] (their , , are our , , , with ). In spherical symmetry, with a prime for , the evolution equations are
The variables , and follow from
and Eq. (4), integrated outward from and from the regularity condition , and inward from at large radius. The Hamiltonian constraint is not imposed; multiplied by , it reads
and we monitor it during the evolution.
Free evolution of . Although is fixed on each slice by the lapse through Eq. (4), we evolve it with the radial æther equation combined with the momentum constraint [36]. Because in unitary gauge the æther equation follows from the metric equations [17], this keeps the Hamiltonian constraint satisfied as long as is rebuilt from with Eq. (4). We check this by relaxing Eq. (4) to . Near the centre, the constraint of Eq. (9) behaves as , and for data with
which vanishes only for .
The slicing is singularity avoiding. Integrating Eq. (4) inward from at large radius gives . Once the horizon has formed, is non-negative inside it, so decreases monotonically inward there. The lapse does not collapse at the apparent horizon, where it stays of order unity, but at a smaller radius inside it, where it drops by many orders of magnitude over a narrow range of (by this happens at , with ); inside that radius it is exponentially small all the way to the centre. Every source term in the evolution equations is proportional to . The advection terms are proportional to instead, but Eq. (8) and together with the monotonicity of give
so also vanishes wherever the lapse has collapsed. The fields inside that radius therefore freeze before the singularity is reached, as in the original GEJ simulations [36].
This freezing has a geometric interpretation. In static black holes the leaves of the æther foliation do not cross the universal horizon but accumulate on it from outside, so the preferred time diverges there and the lapse of the æther slicing vanishes [44, 48]. Accordingly, Ref. [42] reads the vanishing of the lapse in the GEJ simulations as the dynamical formation of a universal horizon, which in Hořava gravity is also a Cauchy horizon [43]. On this reading the frozen region is the interior of the forming universal horizon, which the æther slicing cannot penetrate. The freezing must not spread to the region we want to study, and we check this directly in the baseline collapse run. The lapse at the apparent horizon stays of order unity, so the horizon and everything outside it continue to evolve and the growth of the horizon can be followed to late times. When the interior freezes while the exterior keeps moving, slices can stretch, building steep gradients at the edge of the frozen region that eventually cannot be resolved. Here saturates instead of growing, so the gradients stay bounded. Because the singularity is never reached, there is nothing to excise: the grid extends to throughout, with no interior boundary.
Discretization. Spatial derivatives are second-order centered finite differences on a uniform grid of points, , with regularity at imposed through parity; time integration is by iterated Crank–Nicolson [63], with fourth-order Kreiss–Oliger dissipation [62]. At the outer boundary, conditions are imposed only on the incoming characteristic fields of the spin-0 mode and of the scalar, and only while these fields enter the grid. This treatment relies on the characteristic analysis of the spin-0 sector and is described in § 1.4 B. Because, in the gauge, the grid falls inward ( is nearly uniform across the exterior and of order ), the outer boundary shrinks in areal radius at roughly one areal unit per unit time (Figure 1(a)). For the long runs of § 1.4 we therefore use an arbitrary Lagrangian–Eulerian [64] map , with a quintic taper supported for (the outer 60% of the grid) and chosen to hold fixed. This changes coordinates only: with the map switched off, the code reproduces the fixed-grid run, and with it switched on, the apparent-horizon radius agrees with the fixed grid (Figure 1(b)).
Figure 1. Moving-mesh validation (, , ). (a) Outer boundary against . On the fixed grid, the areal radius of the outer boundary, , shrinks with time (red); the moving mesh holds it fixed (blue, solid) by advecting the boundary coordinate outward (blue, dashed). (b) The areal radius of the apparent horizon against on the fixed grid (red open circles) and on the moving mesh (blue dots); the two agree, showing that the map only changes the coordinates. Note that reaches its maximum at and then slowly decreases, as discussed in Sec. IV.
Couplings and initial data. Unless stated otherwise, , , (so ), the values of GEJ [36] for which . These will be referred to as baseline couplings. To separate the spin-0 signal from the scalar, we also study the case (). The initial data are a time-symmetric scalar shell, a variant of that of GEJ [36] (whose profile is ) multiplied by ,
with as in GEJ [36] (hence ), and and (the GEJ values), a shell centred at with peak value . At the baseline amplitude it collapses to a black hole.
Diagnostics and physical quantities. We use the expansions of the outgoing and ingoing radial null congruences orthogonal to the symmetry spheres, with null normals ( the unit radial vector), rescaled by and with the sign of the ingoing one reversed, so that both are positive on an ordinary sphere:
The Misner–Sharp mass [52, 53] is
the apparent horizon is the outermost surface where , and the ADM mass is read where is flat. Because the æther energy density has a long negative tail, this requires , beyond the support of the scalar. On the apparent horizon , so Eq. (14) gives : the horizon mass is fixed by the areal radius of the horizon, and the horizon mass and the horizon area grow or shrink together.
Convergence
We test the convergence of the code in two regimes. In the first, scattering, a shell falls through the regular centre, bounces and disperses without forming a horizon. This tests the formulation and the treatment of the centre in a spacetime that is nonlinear but regular everywhere. In the second, collapse, the shell forms a black hole and the interior freezes. This tests the region outside the horizon, from which the results of the following sections are extracted.
Both tests follow the same procedure. We evolve the same initial data on three grids, with spacings , and , and compare the results. The grids are nested: every point of the coarsest grid is also a point of the two finer ones, so that the differences between resolutions are taken pointwise, with no interpolation. For this reason both tests use the fixed grid, without the moving mesh. The convergence order for a quantity is
where is computed with grid spacing and time step , so that both are halved together, and is the norm over on the coarsest grid, in a window given below for each test. The scheme is second order in space and time, so the error of should scale as , and the expected result is . Equivalently, the difference between the two finer grids, multiplied by 4, should coincide with the difference between the two coarser ones. The figures of Appendix A show the norms in this form.
The Hamiltonian constraint provides an independent check. It vanishes for the exact solution, so its value on the grid is itself a measure of the numerical error, and its norm should decrease by a factor 4 each time is halved. Its order is therefore of the ratio of its norms on two successive grids, averaged over the two pairs of grids. The constraint itself contains the term , which is at a regular centre and cannot be evaluated reliably there on the grid. We therefore monitor , Eq. (9), which is finite at : on a regular slice and are odd in and the other fields even, so near the centre every term of vanishes except , and . The constraint at is then the regularity condition .
Scattering. We evolve the initial data of Eq. (12) with three amplitudes, , and , on the grids (, ). In all three cases the shell falls through the centre, bounces and disperses. The field grows stronger with . We measure its strength by the minimum of the central lapse during the bounce, which is , and for the three amplitudes. We do not go beyond . For stronger shells the coarsest grid no longer resolves the bounce, and a trapped region a few cells wide appears at the centre, a resolution artefact discussed in § 1.5. We compute the norms over , which contains the centre and the outgoing shell. We average the orders over –14, just after the bounce. Table I lists the orders, and Appendix A shows the norms against time.
| (weak) | 2.35 | 2.52 | 2.17 | 2.16 | 2.92 | 2.16 | 1.95 |
| (strong, ) | 2.62 | 2.46 | 2.20 | 2.14 | 3.36 | 2.12 | 1.90 |
| (strongest, ) | 2.81 | 1.99 | 1.45 | 1.47 | 2.42 | 1.42 | 1.49 |
| Collapse, | 2.62 | 2.55 | 2.34 | 2.38 | 2.62 | 2.42 | 2.24 |
Table 1. Measured convergence order (target 2). First three rows: scattering, nested grids (), norm over , orders averaged over –14; is the minimum of the central lapse. Last row: collapse, nested grids (), norm outside the apparent horizon, over , orders averaged over –28.
For and , all fields and the constraint converge at second order. For , the scalar field and the areal radius still converge at second order. The æther variables , and and the constraint, instead, converge only with –1.5. This loss of order appears after the bounce, when the shell is most compressed at the centre, and our coarsest grid no longer fully resolves it. The convergence indeed improves with resolution. At , the constraint norm decreases by a factor 1.8 between the two coarser grids and by 2.5 between the two finer ones, approaching the factor 4 of second order.
All three scattering amplitudes lie well below the black-hole threshold . We locate it by bisection in on four grids, from () to (). We identify black-hole formation with the appearance of an apparent horizon of finite areal radius when the shell reaches the centre (Figure 2). Unlike the trapped region a few cells wide that the coarsest grid shows for stronger scattering shells, this horizon has the same areal radius on every grid that forms it. The smallest amplitude that forms it decreases with resolution, from 0.135 on the coarsest grid to 0.125 on the two finest. Therefore lies between 0.120 and 0.125. Above , the areal radius of the horizon grows with .
Figure 2. Black-hole formation threshold for the time-symmetric shell (, , ) on four grids: areal radius of the apparent horizon that forms when the shell reaches the centre () against the amplitude of the initial pulse, set to zero where no horizon forms.
Collapse. For the collapse test we take the baseline shell, , on the grids with , so that as in the scattering test, and compare the three runs every two time units up to . An apparent horizon is already present at . Inside it, the lapse collapses and the fields freeze, as described in Sec. III A. The frozen interior develops steep gradients that a fixed grid cannot resolve. It does not affect the exterior, however, and the exterior is the region we study. We therefore take the norms strictly outside the apparent horizon, over , where is the largest of the coordinate radii of the horizon on the three grids.
We average the orders over the outputs from to and list them in the last row of Table I. They lie between 2.24 and 2.62, so every evolved field, and the constraint converge at second order. For the constraint, Table II in Appendix A lists the ratios of its norms on successive grids at selected times. For exact second order these ratios equal 4. On the coarser pair of grids they stay close to 4. On the finer pair they are close to 4 up to and then rise to , so on this pair the constraint violation decreases faster than second order.
This happens throughout the window, so it is not caused by a localised feature.
The areal radius of the apparent horizon also converges at second order. At , the differences between its values on successive grids shrink by a factor 4.3. Assuming second order, its Richardson limit is .
Finally, we check that the ADM mass is conserved. We read it at , where is flat (Sec. III A). The outer edge of the fixed grid falls inward, and this sphere leaves the grid at . We therefore read the mass on the moving mesh at the finest resolution (), where the outer edge stays at . Between and , on this sphere grows by 0.07%. This is neither a truncation error nor an effect of the outer boundary: at the outer edge of the fixed grid is at , and on that sphere the three fixed grids agree to and the fixed grid and the moving mesh agree to . Nor is it a change of , the limit of at large : on the spheres – the change falls off as (it is 1.2% at ), and its extrapolation to is about 0.01%.
The mass function and the horizon
Figure 3 shows the Misner–Sharp mass at , when the apparent horizon is largest. Outside the horizon, decreases outward from (Richardson limit of , Sec. III B) to at large radius, monotonically apart from a small feature just outside the horizon. The decrease is slow, which is why must be read at (Sec. III A). The exterior therefore carries a net energy .
Figure 3. Misner–Sharp mass (blue) against the areal radius at in the baseline collapse. Outside the apparent horizon (red, dashed) decreases outward from (red dot) to (green, dotted). The narrow spike at , inside the apparent horizon in the frozen interior, is a discretisation artefact.
In spherical symmetry the mass function obeys [53]
where and are the energy density and radial energy current measured by the æther, is the unit radial vector and is the total (scalar plus æther) stress tensor. The expansions of Eq. (13) are , so a sphere is untrapped () when . The dominant energy condition requires the energy-flux vector to be causal and future-directed, i.e. . The bracket in Eq. (16) is then non-negative on untrapped spheres: outside the apparent horizon, can decrease outward only where the dominant energy condition is violated [53].
The outward decrease of in Figure 3 therefore requires a violation of the dominant energy condition outside the horizon. Equation (16) is linear in and the scalar obeys the condition, so the violation comes from the æther, whose contribution to the exterior energy is at most . Indeed, the æther energy density is negative everywhere outside the horizon except across the outgoing scalar shell ( at ), and decays outward in the long negative tail noted in Sec. III A.
In GR with matter obeying the dominant energy condition, integrating Eq. (16) from the horizon to infinity gives the spherical Penrose inequality [53, 54, 55], . Here instead
a horizon more massive than the spacetime that contains it. Similarly, the area theorem [56, 57], in its form for apparent horizons [58, 59], would forbid from decreasing, whereas in our run peaks at and then decreases slowly, from 12.6 to 11.8 by on the grid of Figure 1(b). The area theorem relies only on the null energy condition, which the æther stress tensor does not satisfy in general [37].
The spin-0 mode
Principal part and characteristic variables
With as an extra variable, Eqs. (7) are first order in time and space, and the eigenvalues and eigenvectors of their principal part give the characteristic speeds and variables. The equations for , and contain no derivatives of or , so the principal symbol is block-triangular and the spin-0 speeds are those of the block, whose principal part is
with and
The diagonal entries describe advection with the æther observers, ; the off-diagonal ones couple the two fields as in a first-order wave equation, whose squared speed is that of the linear spin-0 mode, Eq. (3). The eigenvalues give characteristics , i.e. speeds relative to the æther in its proper time and proper radial distance (), whatever the lapse and shift. Franchini, Herrero-Valea and Barausse [13] found the same speeds for the infrared limit of Hořava gravity in spherical symmetry, with the same slicing but zero shift and , using derivatives of the metric functions as first-order variables: their spin-0 characteristics, , again correspond to a speed relative to the æther [9].
The left eigenvectors of give the characteristic variables
which obey , with for , outgoing relative to the æther, and for . They become exact characteristic variables of the full system far from the source, where the terms coupling them to and , proportional to and , fall off. The outgoing and ingoing scalar combinations are , which move along .
The spin-0 wave far from the source
Far from the source the spacetime is nearly flat and the fields are small, so the spin-0 equations can be solved exactly. We solve them in the static frame of flat space, and zero shift, with the areal radius as radial coordinate. Our gauge differs from this frame only by a rigid translation of the radial labels on each slice, , which keeps and gives the shift . Since far out, : a point at fixed falls inward in areal radius at about unit speed (Sec. III A), and a time derivative at fixed areal radius becomes in our coordinates. Keeping only the terms linear in the fields, the and equations of Eq. (7) become and ; the scalar drops out, since it enters them only quadratically. Together they give the wave equation , and an outgoing pulse is
where the profile is set by the source and is its integral. Besides the falloff of any spherical wave, has a part, which builds up as the pulse goes by and remains behind it. In terms of the characteristic variables of Eq. (20),
The radiation is carried by , and is the spin-0 waveform we extract. The ingoing variable has no part, but it is not zero: it changes while the pulse goes by. This matters for the outer boundary condition below.
At we impose regularity: , , and are even in , and , and are odd. At the outer boundary a condition is needed only for the fields that enter the grid there, for the spin-0 pair and for the scalar; the outgoing fields simply leave. Physically, no radiation should come in from infinity. At finite radius, however, a purely outgoing wave has a nonzero entering field, one power of below the radiation: by Eq. (22), and for a scalar wave . Setting these fields to zero at , their value at infinity, therefore reflects part of the wave, by the ratio of the two falloffs, i.e. about the pulse width over . Instead we impose non-reflecting conditions, which require the fields at to behave as for an outgoing wave: and must be constant along the outgoing characteristics, of speed and . For the spin-0 pair this is equivalent to , the form we implement [16]; on the collapse background, with the time derivative at fixed areal radius divided by the lapse and for , it reads
For the scalar we impose that be constant along the outgoing light rays; for a linear wave on flat space this is equivalent to , the analogue of Eq. (23), and the entering field again changes at a rate set by its part.
These conditions reflect very little. A spin-0 pulse crossing a flat-space boundary at and , for and , is reflected into below of the incident , against () and () when is set to zero. A scalar pulse is reflected with of its amplitude.
The spin-0 waveform and its speed
We analyse two collapses of the same initial shell: one with the baseline couplings, for which and the spin-0 and scalar signals travel together, and one with , for which and the two signals separate (Sec. III A). We first extract the spin-0 waveform from the outgoing characteristic variable and check that it falls off as . We then measure the speed of the signal in two independent ways: with a linear probe pulse on the collapse background, and with the burst in the Misner–Sharp luminosity, used as a clock between spheres.
At fixed , shows a broad outgoing hump followed by a narrow pulse (Figure 4(a)). To isolate the pulse, we split on each sphere into a slowly varying background and a fast remainder , smoothing in time with a Gaussian kernel:
Figure 4. Outgoing spin-0 characteristic in the baseline collapse (), on spheres of fixed areal radius. (a) against (solid) and its background (dashed), a Gaussian running mean of of width at fixed . (b) The pulse against retarded time . Its amplitude is the same on every sphere, as expected from the falloff of Eq. (22).
and . With a width , the remainder isolates the pulse , whose amplitude depends on but is the same on every sphere for any (Figure 4(b)), as expected from Eq. (22). To compare spheres we use the retarded time
the flat-space of Eq. (21) with replaced by the Schwarzschild tortoise coordinate , which accounts approximately for the slowing of the wave near the black hole. The narrow pulse drifts slowly in (Figure 4(b)), which we attribute to this approximation.
To test the spin-0 propagation, we add to the initial data a linear probe pulse of (amplitude , centred at ), isolate it as the difference with the run without it, and measure the speed of the peak of its relative to the æther, . As shown in § 1.4 A, the outgoing spin-0 characteristic moves at exactly relative to the æther, measured in its proper time and proper radial distance, so this measurement requires no approximation of the background. Figure 5 shows during the evolution. The scatter at early times, , comes from the splitting of the initial pulse, which has both and components, into its outgoing and ingoing parts; afterwards, the deviation stays below . The median speed agrees with to for and to for .
Figure 5. Relative deviation of the speed of a linear spin-0 probe, measured relative to the æther on the background of the collapse, against , for (blue circles) and (red squares); is the coordinate radius of the peak of the probe. The dashed line is and the dotted lines are the medians of the measured speeds, which agree with to 0.01% and 0.02%, respectively.
To carry out an independent measurement of the speed of the nonlinear signal emitted by the collapse, without using the characteristic variables, we use the Misner–Sharp luminosity through a sphere of fixed areal radius, (its decomposition into æther and scalar fluxes is given in § 1.4 D). For , shows on every sphere a sharp burst at , followed by a broad oscillation (Figure 6(a)); the burst coincides with the narrow pulse of . We take the arrival time of the burst on each of the spheres – and plot its lag with respect to the innermost sphere against the tortoise distance between the spheres (Figure 6(b)). A signal moving at speed in tortoise distance gives a straight line of slope . The arrival times lie on a line, and a linear fit gives , below , which we attribute to the Schwarzschild tortoise coordinate approximation in Eq. (25). The burst is therefore carried by the spin-0 mode, not by the scalar field, which propagates at the speed of light.
Figure 6. (a) Misner–Sharp luminosity through spheres of constant , against retarded time , for . It is dominated by the æther Coulomb term (§ 1.4 D). (b) The burst as a clock: its arrival lag at each sphere (dots, coloured as in (a)) against the tortoise distance from the innermost sphere. A linear fit (blue) gives the speed 1.241, 2% below (black, dashed), which we attribute to the Schwarzschild tortoise approximation. The dotted line is the speed of the scalar, .
Mass conservation and radiated energy
We now turn to the energy carried by the wave. We first check that, at every time, the change of the Misner–Sharp mass on each sphere is accounted for by the flux of the stress-energy tensor through it. We then show that most of this flux is the æther Coulomb field, which is linear in the wave amplitude, and remove it to obtain the radiated energy.
The time derivative of the Misner–Sharp mass at fixed areal radius equals the flux of the stress-energy tensor through the sphere [53]. Using the field equations , one obtains the luminosity , split into æther and scalar-field contributions as
with outward, , , and as in Sec. IV for each component X=\AE,\psi, and the radial pressure measured by the æther. For , the two sides of Eq. (26) agree to within of the peak flux at all times on every sphere from to $200 ({xref}Fig. 7`). The scalar flux is small, |F_{\psi}/F_{\AE}|\lesssim0.02, against æther-flux peaks of –, mostly the Coulomb contribution discussed below. However, the luminosity gives a confusing answer for the radiated energy. As the pulse crosses a sphere, swings by an amount that grows with radius, and behind the pulse it settles above (Figure 8(b), thin lines), so that is negative on every sphere. We explain this behaviour below.

Figure 8. The æther Coulomb field left behind by the spin-0 pulse (), on spheres with constant against retarded time . (a) ; during the pulse its amplitude grows with (roughly as ). (b) (thin lines) and (thick lines), with defined in Eq. (28). Removing the Coulomb term reduces the swing and settles below at all radii.
Let us first note that the spin-0 front sets up the acceleration of the æther around a black hole, absent from the initial data: behind the pulse relaxes to the same value on every sphere (Figure 8(a)), the term of Eq. (21). We call such a contribution the æther Coulomb field.
At large radius, where , the Hamiltonian constraint reduces to
with . To leading order the linear term is a total derivative, so during the pulse, when the amplitude of grows with , the swing of grows with radius and dominates the luminosity (Figure 6(a)). Being linear in the amplitude, it is not radiated energy, and we remove it, defining the corrected mass and the cumulative radiated energy as:
where is the energy that has crossed the sphere up to retarded time . Behind the pulse settles below on every sphere, Figure 8(b). This discussion agrees with the definition of the quasi-local version of the total energy of an asymptotically flat Einstein–æther spacetime [5, 41],
which gives in spherical symmetry, positive for a class of solutions when [41]. For the superluminal case with , differs from the energy radiated to infinity by corrections of order , because the spheres lie at finite radius. We therefore extrapolate to infinity with the two-parameter fit over –200, and the scalar flux gives . We obtain
so the spin-0 wave carries off 9% of the ADM mass, a positive energy as required for the linear spin-0 mode when [3, 5].
Figure 9 follows the corrected mass in time. Once the Coulomb term is removed, the rate (panel (a)) is more than an order of magnitude smaller than the uncorrected luminosity of Figure 6(a), and the cumulative energy (panel (b)) is nearly the same on all spheres once the pulse has passed; its extrapolation to infinity, shown once the pulse has crossed all spheres, settles on the value of Eq. (30). Before the front arrives, the freely falling æther makes slightly negative; then overshoots its asymptotic value before settling to the total radiated energy. Integrating Eq. (27) inward from a sphere not yet reached by the wave shows that, up to corrections, is the integral of the quadratic and higher-order field energy outside . The transient overshoot is a consequence of the fact that this quadratic energy density is not positive definite, due to shear contributions from the æther stress tensor, yielding , which is negative for .
The spin-0 mode in the post-Newtonian expansion
Ref. [32] derived the Einstein–æther field equations to 2.5PN order, and Ref. [33] found that the resulting energy flux does not depend on the spin-0 propagation speed ; this was interpreted as evidence that the spin-0 mode is non-radiative. Here we point out a subtlety in the interpretation of the truncated field equations: a non-local field redefinition involving inverse Laplacians can shift the PN order at which the propagation speed appears explicitly in a given field equation. A single-field example of this mechanism is given in Sec. B1.
In this subsection and in Appendix B we adopt the notation of Ref. [32]. In particular, , , , and denote the PN potentials defined here, and should not be confused with the lapse, and used above. As in Refs. [32, 33], where is assumed because of the GW170817 bound, the PN analysis is carried out for , for which the spin-0 speed of Eq. (3) reduces to the speed of Ref. [32], .
In the spherically symmetric scalar sector of the PN-expanded Einstein–æther equations of Ref. [32], eliminating the auxiliary potentials yields an equation for the lapse perturbation that contains an explicit -dependent term at 1PN order. Truncating the field redefinition of Ref. [32] at first order in the PN parameter removes this term, but the dependence on reappears at the next order: the first surviving -dependent term in the transformed lapse equation is 2PN,
Figure 7. Test of the flux identity, Eq. (26), for on the spheres , 100 and 200 (columns). Top row: energy flux through the sphere against : F_{\ae} (blue), computed from the æther stress tensor, (red), their sum F_{\ae}+F_{\psi} (thick grey) and (black, dashed). Bottom row: the relative residual \epsilon=(F_{\ae}+F_{\psi}+dm/dt|_{\Theta})/\max|dm/dt|_{\Theta}|, in units of , which decreases outward.
where , counts PN orders, and the coefficient is given in Eq. (57). The derivation is given in Appendix B.
The reason why reappears becomes transparent on a linear mode with . Defining and , the reduced lapse operator before the redefinition, [Eq. (50)], and the diagonal lapse operator of Eq. (4.13a) of Ref. [32], represented on this mode by , are related exactly by
Figure 9. Energy carried by the spin-0 wave (), on spheres of constant against retarded time . (a) Rate and its extrapolation to infinity (black). (b) Energy that has crossed each sphere (left axis; on the right axis) and its extrapolation from a fit (black). The extrapolations are shown only after the pulse has crossed the spheres, outside the shaded interval. The horizontal grey band is the value of Eq. (30).
where . The factor containing the scalar propagation speed is an inverse propagator,
Removing the dependence from the lapse equation to all orders would therefore require the whole PN series of this inverse propagator, each term of which carries one more PN order and one more inverse Laplacian. The first-order redefinition retains only the first term, and the first neglected one reappears at 2PN as the term in Eq. (31). Moreover, is singular at the physical scalar pole : its cancellation of the pole in the reduced lapse operator must not be interpreted as the removal of a physical degree of freedom by an admissible, globally invertible field redefinition.
For the first-order truncation of Eq. (4.7) of Ref. [32], the explicit dependence of the lapse equation on is thus removed at 1PN and reappears at 2PN. We expect the same mechanism to operate when the full redefinition (4.7) is used, as in Ref. [32]: each additional order retained in the transformation reproduces one more term of the PN series of the inverse propagator in Eq. (33), and therefore pushes the first explicit appearance of to higher PN order, plausibly beyond the 2.5PN order at which the energy flux was computed in Ref. [33]. The absence of from the flux at that order would then be a consequence of the non-local choice of variables, rather than evidence that the spin-0 mode is non-radiative. In other words, the disappearance of the propagation speed from the field equations at a given PN order, after a non-local field redefinition, does not imply that the corresponding mode does not radiate.
This was one of the main motivations for the analysis presented in this section, in which we tracked the spin-0 field directly in nonlinear evolutions of spherical collapse, without truncating a PN expansion or performing non-local field redefinitions. The two analyses cover different regions of parameter space: the PN calculation above assumes , whereas our simulations have . In particular, the run used to measure the radiated energy has , so that the spin-0 mode propagates only because . The comparison is therefore not a quantitative one between energy fluxes; it concerns the mechanism by which a propagating spin-0 mode can disappear from a truncated PN expansion while, as our simulations show, still carries energy to infinity.
Large : transient white holes
Akhoury, Garfinkle and Gupta [37] found that for large the collapse of a scalar shell can produce an anti-trapped surface instead of a trapped one, the defining feature of a white hole. They reported that the anti-trapped region is transient and that the fields eventually disperse, but also that sharp features appear which their grids could not fully resolve. We revisit this case at higher resolution, with the numerical scheme of Sec. III A.
We take , and , which keeps . The shell is the one of Eq. (12), made initially ingoing as in Ref. [37]:
so that initially moves inward at the speed of light with and . Besides , we use the ingoing expansion of Eq. (13). Where the sphere is anti-trapped: ingoing light rays diverge, as in a white hole. In GR the energy conditions forbid anti-trapped surfaces from forming out of regular data; the negative energy density of the æther (Sec. IV) removes this obstruction.
Whether a white hole or a black hole forms depends on the amplitude. At the data are initially untrapped everywhere but crosses zero by and a black hole forms. With , after a brief dip of below zero at –17, an anti-trapped region appears at and grows, its outer edge moving out with the escaping pulse, until it splits at . The inner part shrinks after and disappears by , while a thin anti-trapped layer stays around the peak of the escaping pulse and travels outward with it (Figure 11(a)). The layer is present on all our grids and narrows with the pulse, from about 0.9 at to 0.06 at on the refined grids, which agree on its width to within 20%; its depth, , still grows with resolution. This reproduces the transient white hole of Ref. [37], whose anti-trapped surface first appears at a similar time (); their Fig. 11 also shows trapped surfaces at higher amplitude. The two amplitude normalisations differ, so the thresholds cannot be compared directly.

Figure 11. The transient white hole at , . (a) Areal radius against , on the uniform grid with , of the edges of the anti-trapped region (, purple dots), of the inner shell (peak of , black solid), of the escaping pulse (largest local maximum of outside the shell, green), of the inner front (steepest slope of at $1.5 < \Theta < 5$, yellow) and of the outgoing light ray leaving the inner shell at (blue dashed). The dotted vertical line marks , when the inner shell steepens, and the grey line , the profile of Figure 10(a). The ray joins the escaping pulse by . The outermost branch of purple dots is a thin anti-trapped layer, less than one unit wide (its two edges overlap at this scale), that travels with the peak of the escaping pulse. (b) Width (full width at half maximum) and (c) height, , of the peak of on the inner shell at $t = 30 (Figure 10(c)) against the grid spacing at the shell, for uniform grids (blue circles) and grids refined around the shell (red squares), with a power-law fit for each group.
On coarse grids the white-hole profiles exhibit three narrow features, each with a different physical origin. Figure 10(a) shows them on one profile, at , and Figure 11(a) follows their areal radius in time: the escaping pulse (green), the inner shell (black) and the inner front (yellow). The part of the scalar pulse that escapes is compressed as it moves outward through the white-hole geometry, its width shrinking roughly as . We resolve it by adding a moving refinement to the coordinate map of Sec. III A. The number of grid points does not change: the computational coordinate is first reparametrised so that the points cluster in a region, a few units wide, where the spacing is eight times finer, and are drawn from a band around it, while the grid elsewhere is left exactly as it was. This is not adaptive mesh refinement: there are no nested grids, and the refinement factor is fixed. The centre of the region follows the peak of through a feedback controller, and the pulse is resolved as a thin but regular wave (Figure 10(b)). Following it further, however, would need a resolution growing exponentially in time. The compression is visible in the outgoing light rays, which converge on the pulse: the ray that leaves the inner shell at behind the peak, is 0.3 behind it at and 0.1 at (Figure 11(a)).
Figure 10. The three sharp features of the white-hole run (, ), shown in . (a) against at (grey line in Figure 11(a)) on the uniform grid with . (b) The escaping pulse at –, while its peak moves from to , on the grid with a region eight times finer that follows it, against the distance from the peak. (c) The inner shell at , against the distance from its peak, on the three uniform grids (blue; , , ) and on two grids refined around the shell (red; and ). (d) The inner front at on the three uniform grids.
The part of the pulse that does not escape piles up, as an ingoing wave, on a sphere of areal radius , and at it steepens beyond what any of our grids resolves (Figure 10(c)). Figure 11(b) measures the width and height of the peak of Figure 10(c) at on seven grids: three uniform, and four with a refined region of the same kind, eight or sixteen times finer. Since the shell forms abruptly, with no earlier peak to follow, this region is placed at the radius where the uniform runs show the shell forming and is held there from until . In each group of runs, the width of the peak shrinks roughly as , its height grows as and the energy density as . This is consistent with a singularity forming at but does not prove it: the growth slows on the finest grids, and on the finest run the width levels off once the refinement is centred on the shell, which may indicate a finite width of that order. A regular structure narrower than our finest spacing cannot be excluded.
A third, weaker feature forms inside the shell. From part of the ingoing scalar pulse travels as a front, a jump of about 0.3 in , along ingoing light rays, . Where it travels the lapse drops to , and the front nearly stalls at , close to the inner edge of the anti-trapped region (Figure 11(a)). Its steepest slope changes little with resolution at but doubles with each halving of by (Figure 10(d)), so it too steepens beyond our truncation error. The lapse then recovers, the front falls to the centre at and has dispersed by .
No trapped surface exists anywhere on the slice, and the outgoing light ray leaving the inner shell joins the escaping pulse and travels outward with it (Figure 11(a)). We cannot exclude an event horizon at later times, but nothing in our data suggests one. Beyond the evolution does not converge: the shell becomes a thin but resolved structure whose position depends on the grid. Every quantity behind the escaping pulse inherits this uncertainty. In particular, the frozen centre seen in long runs on coarse grids, where the central lapse collapses, is an artefact: the freezing time moves later on every finer grid roughly as , with no sign of converging.
We repeated the amplitude exploration for and , with the same outcome. Weak data disperse (, 0.015 and 0.03 for ), intermediate data form a transient white hole whose inner shell steepens as above, and strong data form a black hole, with no steepening outside the horizon. Near the black-hole threshold the outcome depends on resolution, and coarse grids favour the white hole: for , gives a trapped region that later disappears at , but a persistent black hole at . A trapped region followed by an anti-trapped one does occur, for , : the trapped region is gone by and the anti-trapped region appears at , identically on three grids. Finally, the collapse of the centre into a trapped region seen for is the frozen-centre artefact above: at it occurs at , not at 0.0016, and again at 0.0008, and at it happens later on each finer grid while its size shrinks with the spacing.
Our results broadly agree with earlier work. GEJ [36] already reported, for a single run, that no horizon formed and “the evolution seems to become singular, thus indicating the formation of a naked singularity”. Ref. [37] revisited the case with higher resolution. We agree with the latter that at large a moderately strong shell makes a temporary white hole and stronger shells make black holes, but not on the fate. Ref. [37] concludes that the fields disperse, noting that the sharp features it finds near are not fully resolved with 40,000 and 80,000 points. Those spacings match our two coarsest uniform grids, on which the inner shell also looks like a sharp but finite feature. Refining further shows that it keeps steepening. Whether their feature is ours is not certain: their shell, with , is about 2.5 times wider than ours, those runs use a different radial coordinate, and their feature sits near the centre rather than at . In the white-hole window, therefore, the evolution develops an unresolved and possibly singular feature that no trapped surface hides, and beyond it our runs do not determine the solution.
Conclusion
In spherical symmetry, Einstein–æther theory and low-energy Hořava gravity have the same collapse dynamics, in which both theories propagate a spin-0 mode. Spherical collapse is therefore the cleanest nonlinear setting in which to ask whether the spin-0 mode is a dynamical wave that carries energy away from the system. The code and the scripts behind every figure and number of this paper are public [65].
Our simulations show that the spin-0 mode is indeed a wave. It is the outgoing characteristic field of the evolution system, it carries the whole part of the field that falls off as the inverse areal radius, and on the dynamical black-hole background it propagates at the speed predicted by the linear theory, also when that speed is superluminal. It carries a finite, positive energy to infinity once the Coulomb-like æther acceleration field set up by the wave is subtracted; that field is linear in the wave amplitude, dominates the Misner–Sharp luminosity and can be identified with the surface term of the Einstein–æther energy. For the superluminal coupling we studied, the collapse radiates 14% of the ADM mass: 9% in the spin-0 wave and 5% in the escaping scalar field.
These results do not fully settle the problem in more general settings, such as binary systems: Taherasghari and Will [32, 33] found a near-zone dipole energy loss of compact binaries that depends only on the speed of the transverse æther waves, whereas the spin-0 speed also enters the far-zone flux of Refs. [6, 30, 31]. We have not computed the dipole flux of a binary, and a spherical collapse, which has no spin-1 mode and hence no , cannot decide whether the near-zone radiation-reaction force or the energy flux correctly describes the energy loss of a two-body system. Within the spin-0 sector, however, our results rule out that the mode is non-dynamical or carries no energy. The energy leaves in a wave travelling at , and the Misner–Sharp mass lost through every sphere matches the æther and scalar fluxes computed independently from the stress tensor. If the spin-0 speed does drop out of the energy flux of a binary, the reason must lie in the binary problem itself.
We also find that the static æther left around the black hole has negative energy, so the Misner–Sharp mass decreases outward from the horizon and the dominant energy condition is violated. Inside the horizon the lapse collapses and the fields freeze, as in the GEJ simulations (Sec. III A). Ref. [42] read this freezing as the dynamical formation of a universal horizon, which does form in the collapse of the version of Hořava gravity that survives the GW170817 and solar-system bounds [13]; we have not pursued this identification. At large we recover the transient white holes of Akhoury, Garfinkle and Gupta [37], which the negative energy of the æther allows and GR does not. At higher resolution, however, the inner shell that does not escape steepens beyond what any of our grids resolves, with no trapped surface to hide it. This may be the naked singularity suspected by GEJ [36] or a regular structure finer than our grids. Our runs settle neither this nor the solution beyond that point.
Three research directions follow. First, the energy measurement should be repeated for couplings allowed by GW170817 and binary pulsars, for which the spin-0 speed can differ substantially from the scalar speed, and connected to the sensitivities of compact objects and to binary observables [6, 14, 34, 66]. Second, non-spherical simulations would reach the spin-1 sector, whose speed is the one that appears in the radiation-reaction result of Refs. [32, 33]; the characteristic decomposition of Sec. V provides boundary conditions and extraction variables for them. Third, the large- case calls for adaptive mesh refinement, whose resolution, unlike the fixed factor used here, grows as the shell steepens, to decide whether a naked singularity forms.
Acknowledgments
We acknowledge support from the European Union’s Horizon ERC Synergy Grant “Making Sense of the Unexpected in the Gravitational-Wave Sky” (Grant No. GWSky–101167314). We used Claude (Anthropic) to help write, debug and run the numerical code, to prepare the figures and to edit the text; all results were checked by the authors, who take full responsibility for the content. We thank Ramiro Cayuso for helpful discussions and advice on the numerical methods used in this work.
Convergence tests
This appendix collects the self-convergence data behind Sec. III B, first for the scattering test and then for the collapse test.
Scattering. Figures 12 and 13 show the differences of the evolved fields and of between successive grids for the weakest and the strongest shell. For the two curves overlap, as second-order convergence requires, except at isolated times during the bounce, when a difference passes through an accidental minimum; for
the curves of the æther variables separate after the bounce. Figure 14 shows the Hamiltonian constraint for the three amplitudes of Table I.
Figure 14. Hamiltonian constraint in scattering: over against for , rescaled by and normalised to the coarsest grid (factors 1, 4 and 16). (a) Weak field, ; (b) strong field, ; (c) , the strongest shell. In the weak and strong-field cases the three resolutions collapse onto one curve, i.e. at second order; for they separate after the bounce ().
Collapse
The constraint-norm ratios quoted in Sec. III B are listed in Table II, and the rescaled constraint norm is shown in Figure 15. The ratios must be read on the nested fixed grid: with the moving mesh the three resolutions are no longer coincident point by point, and the stretched outer zone degrades the comparison. The apparent horizon is present from , with , and by the central lapse is below . Because the grid falls inward, the window of the norms drifts: grows from at to at , while the areal radius at decreases from to . At , the areal radius of the apparent horizon is , and on the three grids. On the moving mesh at , the ADM mass read at is at and at .
Figure 15. Hamiltonian constraint in the collapse test (, , fixed grid): outside the apparent horizon, over , against on the nested grids , rescaled by . The rescaled curves coincide if at second order. From the curve of the finest grid lies below the other two: on the finer pair of grids the constraint converges faster than second order.
Figure 13. As Figure 12, for , the strongest shell of the scattering test. The curves of , and separate after the bounce ().
Figure 12. Self-convergence of the evolved fields and in scattering, , on the nested grids (). Each panel shows, against and on a logarithmic scale, the norms over of and . The curves overlap for second-order convergence.
| 4 | 8 | 12 | 16 | 20 | 24 | 28 | |
| 3.69 | 3.80 | 4.69 | 4.53 | 4.26 | 4.05 | 3.92 | |
| 3.95 | 3.97 | 5.59 | 6.26 | 6.12 | 5.95 | 5.82 |
Table 2. Constraint-norm ratios outside the apparent horizon () on the nested fixed grid (, ); a ratio of 4 is exact second order.
Non-local field redefinitions and the spin-0 mode
Below we first give a simple example of the mechanism described in § 1.4 E, using a single propagating scalar, and then derive the results quoted there for the spherically symmetric scalar sector of the PN-expanded equations of Ref. [32]. Throughout, counts PN orders, a tilde denotes the original variables, and untilded symbols denote the variables obtained after the field redefinition.
A single-field example
Consider a scalar field propagating with speed ,
where in the adopted counting. On a mode with , the operator acts as multiplication by
The dependence on encodes the time-dependent propagation correction: each additional power of carries one more PN order and one more inverse Laplacian.
Consider now the non-local field redefinition , which expresses the original field in terms of the redefined field , with
The transformed operator is
The propagation correction is therefore absent through PN and first appears at PN, through a term proportional to . For , for example,
The physical root is still a root of . The polynomial transformation, however, also introduces the roots with ; these are artefacts of the transformation and must not be mistaken for additional physical modes. As for any field redefinition, statements about the mode content hold only where the transformation is well defined and invertible.
Thus, a finite PN truncation of a non-local field redefinition can postpone the explicit appearance of propagation effects without implying that the underlying degree of freedom has disappeared. The rest of this appendix shows that the same mechanism operates in the coupled scalar sector of Einstein–æther theory.
Spherical reduction and the linear scalar mode
We reduce Eqs. (4.5)–(4.6) of Ref. [32] to spherical symmetry, using
so that . Since
the terms in the vector and tensor Laplacians are absorbed, and every operator in Eqs. (4.5)–(4.6) becomes a function of and acting on the scalar potentials. For example, Eqs. (4.5a) and (4.6) become
and the harmonic-gauge conditions read
The remaining reduced equations are obtained in the same way.
On a mode satisfying (for example, ), one has . In analogy with Eq. (36), we introduce
so that ; each power of represents one PN order, together with one inverse Laplacian. With the rescaled amplitudes
the gauge relation gives ; in what follows we drop the hats on \hat{\tilde K}_{\ae} and . The three independent equations, Eqs. (4.5a), (4.5c)_{\mathrm{tr}}, and (4.6) of Ref. [32], for (\tilde N,\tilde B,\tilde K_{\ae}) then take the form
with
Its determinant is
where, for ,
The factors correspond to gauge/constraint modes, while the physical scalar mode is associated with the root . Eliminating and \tilde K_{\ae} yields the reduced equation , with
where
The term linear in splits into the contribution of the diagonal equation (4.13a) of Ref. [32] and an additional -dependent 1PN term,
First-order field redefinition and its interpretation
We now apply Eq. (4.7) of Ref. [32], truncated at first order in . For the lapse, the transformation expresses the original variable in terms of the redefined ones as
together with the corresponding first-order expressions for , , and \widetilde{K}_{\ae}. Writing
the original equations \mathcal{D}_0(\widetilde{N},\widetilde{B},\widetilde{K}_{\ae})^{\mathsf T}=0 become \mathcal{D}_1(N,B,K_{\ae})^{\mathsf T}=0 for the redefined variables, with . At order , the first column of is
which cancels the zeroth-order response of (B,K_{\ae}) to . Hence
where is the spin-1 speed for . These are the symbols of the diagonal operators , , and -c_1\Box^*K_{\ae} in Eqs. (4.13a,d,e) of Ref. [32]. The coefficient of in the lapse equation is now , independent of and hence of : the term in Eq. (52) has been removed.
At the next order, however, the reduced operator acting on the redefined lapse is
with
The first -dependent term in the transformed lapse equation is therefore proportional to
which is 2PN in the adopted counting. The origin of this term follows from the exact relation (32) between , which acts on the original lapse , and the diagonal lapse operator, with . Expanding the inverse propagator (33) in PN orders gives
The first-order redefinition retains only the first correction in this series, so the first neglected power contributes at 2PN. In the notation above,
This is the multi-field counterpart of Eq. (38), whose operator form is
understood on modes for which the inverse Laplacian is defined. A finite PN truncation can therefore shift the order at which the propagation-speed dependence appears explicitly in a chosen equation, whereas a local redefinition, without inverse Laplacians, cannot generate the same series of powers of .
As in the single-field case, the mode content is preserved only where the transformation is invertible. For a finite transformation matrix , the identity
shows that the physical root of is retained whenever is nonzero there, while additional roots may appear where . The exact factor , on the other hand, is singular at the physical pole and does not correspond to an invertible redefinition there (see § 1.4 E).
In summary, in the first-order truncation the explicit dependence of the reduced lapse equation is absent through 1PN and reappears at 2PN. Retaining higher orders in the redefinition is expected to push it to higher PN order, as discussed in § 1.4 E.
References
References
- [1]M. Gasperini, Singularity prevention and broken Lorentz symmetry, Class. Quant. Grav. 4, 485 (1987).DOI
- [2]T. Jacobson and D. Mattingly, Gravity with a dynamical preferred frame, Phys. Rev. D 64, 024028 (2001), arXiv:gr-qc/0007031.DOI
- [3]T. Jacobson, Einstein-aether gravity: a status report, PoS QG-PH, 020 (2007), arXiv:0801.1547.arxiv.org/abs/0801.1547
- [4]T. Jacobson and D. Mattingly, Einstein-aether waves, Phys. Rev. D 70, 024003 (2004), arXiv:gr-qc/0402005.DOI
- [5]C. Eling, Energy in the Einstein-aether theory, Phys. Rev. D 73, 084026 (2006), arXiv:gr-qc/0507059.arxiv.org/abs/gr-qc/0507059
- [6]B. Z. Foster, Radiation damping in Einstein-aether theory, Phys. Rev. D 73, 104012 (2006), arXiv:gr-qc/0602004.arxiv.org/abs/gr-qc/0602004
- [7]P. Hořava, Quantum gravity at a Lifshitz point, Phys. Rev. D 79, 084008 (2009), arXiv:0901.3775.DOI
- [8]D. Blas, O. Pujolàs and S. Sibiryakov, Consistent extension of Hořava gravity, Phys. Rev. Lett. 104, 181302 (2010), arXiv:0909.3525.
- [9]T. Jacobson, Extended Hořava gravity and Einstein-aether theory, Phys. Rev. D 81, 101502 (2010), arXiv:1001.4823.arxiv.org/abs/1001.4823
- [10]D. Blas, O. Pujolàs and S. Sibiryakov, Models of nonrelativistic quantum gravity: the good, the bad and the healthy, JHEP 04, 018 (2011), arXiv:1007.3503.DOI
- [11]A. O. Barvinsky, D. Blas, M. Herrero-Valea, S. M. Sibiryakov and C. F. Steinwachs, Renormalization of Hořava gravity, Phys. Rev. D 93, 064022 (2016), arXiv:1512.02250.DOI
- [12]T. Jacobson, Undoing the twist: the Hořava limit of Einstein-aether theory, Phys. Rev. D 89, 081501 (2014), arXiv:1310.5115.
- [13]N. Francini, M. Herrero-Valea and E. Barausse, Relation between general relativity and a class of Hořava gravity theories, Phys. Rev. D 103, 084012 (2021), arXiv:2103.00929.
- [14]E. Barausse, Neutron star sensitivities in Hořava gravity after GW170817, Phys. Rev. D 100, 084053 (2019); 104, 069903(E) (2021), arXiv:1907.05958.
- [15]O. Sarbach, E. Barausse and J. A. Preciado-López, Well-posed Cauchy formulation for Einstein-aether theory, Class. Quant. Grav. 36, 165007 (2019), arXiv:1902.05130.DOI
- [16]O. Sarbach and M. Tiglio, Continuum and discrete initial-boundary-value problems and Einstein’s field equations, Living Rev. Rel. 15, 9 (2012), arXiv:1203.6443.DOI
- [17]T. Jacobson, Initial value constraints with tensor matter, Class. Quant. Grav. 28, 245011 (2011), arXiv:1108.1496.DOI
- [18]B. P. Abbott et al. (LIGO Scientific, Virgo, Fermi-GBM, INTEGRAL), Gravitational waves and gamma-rays from a binary neutron star merger: GW170817 and GRB 170817A, Astrophys. J. Lett. 848, L13 (2017), arXiv:1710.05834.
- [19]J. Oost, S. Mukohyama and A. Wang, Constraints on Einstein-aether theory after GW170817, Phys. Rev. D 97, 124023 (2018), arXiv:1802.04303.DOI
- [20]A. Emir Gümrükçüoğlu, M. Saravani and T. P. Sotiriou, Hořava gravity after GW170817, Phys. Rev. D 97, 024032 (2018), arXiv:1711.08845.
- [21]T. Gupta, M. Herrero-Valea, D. Blas, E. Barausse, N. Cornish, K. Yagi and N. Yunes, New binary pulsar constraints on Einstein-æther theory after GW170817, Class. Quant. Grav. 38, 195003 (2021), arXiv:2104.04596.DOI
- [22]K. Yagi, D. Blas, N. Yunes and E. Barausse, Strong binary pulsar constraints on Lorentz violation in gravity, Phys. Rev. Lett. 112, 161101 (2014), arXiv:1307.6219.DOI
- [23]B. Z. Foster and T. Jacobson, Post-Newtonian parameters and constraints on Einstein-aether theory, Phys. Rev. D 73, 064015 (2006), arXiv:gr-qc/0509083.DOI
- [24]C. M. Will, The confrontation between general relativity and experiment, Living Rev. Rel. 17, 4 (2014), arXiv:1403.7377.DOI
- [25]D. Mattingly, Modern tests of Lorentz invariance, Living Rev. Rel. 8, 5 (2005), arXiv:gr-qc/0502097.DOI
- [26]S. Liberati, Tests of Lorentz invariance: a 2013 update, Class. Quant. Grav. 30, 133001 (2013), arXiv:1304.5795.DOI
- [27]J. W. Elliott, G. D. Moore and H. Stoica, Constraining the new aether: gravitational Cherenkov radiation, JHEP 08, 066 (2005), arXiv:hep-ph/0505211.
- [28]S. M. Carroll and E. A. Lim, Lorentz-violating vector fields slow the universe down, Phys. Rev. D 70, 123525 (2004), arXiv:hep-th/0407149.DOI
- [29]B. Audren, D. Blas, M. M. Ivanov, J. Lesgourgues and S. Sibiryakov, Cosmological constraints on deviations from Lorentz invariance in gravity and dark matter, JCAP 03, 016 (2015), arXiv:1410.6514.DOI
- [30]B. Z. Foster, Strong field effects on binary systems in Einstein-aether theory, Phys. Rev. D 76, 084033 (2007), arXiv:0706.0704.DOI
- [31]K. Yagi, D. Blas, E. Barausse and N. Yunes, Constraints on Einstein-Aether theory and Hořava gravity from binary pulsar observations, Phys. Rev. D 89, 084067 (2014), arXiv:1311.7144.DOI
- [32]F. Taherashgari and C. M. Will, Compact binary systems in Einstein-Aether gravity: Direct integration of the relaxed field equations to 2.5 post-Newtonian order, Phys. Rev. D 108, 124026 (2023), arXiv:2308.13243.
- [33]F. Taherashgari and C. M. Will, Compact binary systems in Einstein-Aether gravity. II. Radiation reaction to 2.5 post-Newtonian order, Phys. Rev. D 112, 024013 (2025), arXiv:2506.03843.
- [34]O. Ramos and E. Barausse, Constraints on Hořava gravity from binary black hole observations, Phys. Rev. D 99, 024034 (2019), arXiv:1811.07786.DOI
- [35]M. Vaglio, A. Carleo, A. Susobhanan, E. Barausse, B. Ben Salem, A. Corongiu, P. C. Freire and D. Perrodin, Constraints on Einstein-aether gravity from the precision timing of PSR J1738+0333, arXiv preprint (2026), arXiv:2605.01436.arxiv.org/abs/2605.01436
- [36]D. Garfinkle, C. Eling and T. Jacobson, Numerical simulations of gravitational collapse in Einstein-aether theory, Phys. Rev. D 76, 024003 (2007), arXiv:gr-qc/0703093.DOI
- [37]R. Akhoury, D. Garfinkle and N. Gupta, White holes in Einstein-aether theory, Class. Quant. Grav. 35, 035006 (2018), arXiv:1608.06970.arxiv.org/abs/1608.06970
- [38]C. Eling and T. Jacobson, Black holes in Einstein-aether theory, Class. Quant. Grav. 23, 5643 (2006), arXiv:gr-qc/0604088.arxiv.org/abs/gr-qc/0604088
- [39]C. Eling and T. Jacobson, Spherical solutions in Einstein-aether theory: static aether and stars, Class. Quant. Grav. 23, 5625 (2006), arXiv:gr-qc/0603058.arxiv.org/abs/gr-qc/0603058
- [40]M. Bonetti and E. Barausse, Post-Newtonian constraints on Lorentz-violating gravity theories with a MOND phenomenology, Phys. Rev. D 91, 084053 (2015); 93, 029901(E) (2016), arXiv:1502.05554.DOI
- [41]D. Garfinkle and T. Jacobson, A positive energy theorem for Einstein-aether and Horava gravity, Phys. Rev. Lett. 107, 191102 (2011), arXiv:1108.1835.DOI
- [42]J. Bhattacharyya, M. Colombo and T. P. Sotiriou, Evolution and spherical collapse in Einstein-aether theory and Horava gravity, Phys. Rev. D 93, 064056 (2016), arXiv:1512.04899.DOI
- [43]J. Bhattacharyya, M. Colombo and T. P. Sotiriou, Causality and black holes in spacetimes with a preferred foliation, Class. Quant. Grav. 33, 235003 (2016), arXiv:1509.01558.DOI
- [44]E. Barausse, T. Jacobson and T. P. Sotiriou, Black holes in Einstein-aether and Horava-Lifshitz gravity, Phys. Rev. D 83, 124043 (2011), arXiv:1104.2889.DOI
- [45]D. Blas and S. Sibiryakov, Horava gravity versus thermodynamics: the black hole case, Phys. Rev. D 84, 124043 (2011), arXiv:1110.2195.DOI
- [46]P. Berglund, J. Bhattacharyya and D. Mattingly, Mechanics of universal horizons, Phys. Rev. D 85, 124019 (2012), arXiv:1202.4497.DOI
- [47]A. Adam, P. Figueras, T. Jacobson and T. Wiseman, Rotating black holes in Einstein-aether theory, Class. Quant. Grav. 39, 125001 (2022), arXiv:2108.00005.DOI
- [48]E. Barausse and T. P. Sotiriou, Black holes in Lorentz-violating gravity theories, Class. Quant. Grav. 30, 244010 (2013), arXiv:1307.3359.DOI
- [49]E. Barausse, T. P. Sotiriou and I. Vega, Slowly rotating black holes in Einstein-aether theory, Phys. Rev. D 93, 044044 (2016), arXiv:1512.05894.DOI
- [50]M. Saravani, N. Afshordi and R. B. Mann, Dynamical emergence of universal horizons during the formation of black holes, Phys. Rev. D 89, 084029 (2014), arXiv:1310.4143.DOI
- [51]E. Berti et al., Testing general relativity with present and future astrophysical observations, Class. Quant. Grav. 32, 243001 (2015), arXiv:1501.07274.DOI
- [52]C. W. Misner and D. H. Sharp, Relativistic equations for adiabatic, spherically symmetric gravitational collapse, Phys. Rev. 136, B571 (1964).DOI
- [53]S. A. Hayward, Gravitational energy in spherical symmetry, Phys. Rev. D 53, 1938 (1996), arXiv:gr-qc/9408002.DOI
- [54]R. Penrose, Naked singularities, Ann. N.Y. Acad. Sci. 224, 125 (1973).DOI
- [55]E. Malec and N. O Murchadha, Trapped surfaces and the Penrose inequality in spherically symmetric geometries, Phys. Rev. D 49, 6931 (1994), arXiv:gr-qc/9401014.DOI
- [56]S. W. Hawking, Gravitational radiation from colliding black holes, Phys. Rev. Lett. 26, 1344 (1971).DOI
- [57]S. W. Hawking and G. F. R. Ellis, The Large Scale Structure of Space-Time (Cambridge University Press, Cambridge, 1973).
- [58]S. A. Hayward, General laws of black hole dynamics, Phys. Rev. D 49, 6467 (1994), arXiv:gr-qc/9303006.DOI
- [59]A. Ashtekar and B. Krishnan, Dynamical horizons and their properties, Phys. Rev. D 68, 104030 (2003), arXiv:gr-qc/0308033.DOI
- [62]H.-O. Kreiss and J. Oliger, Methods for the approximate solution of time dependent problems, GARP Publication Series No. 10 (1973).
- [63]S. A. Teukolsky, On the stability of the iterated Crank-Nicholson method in numerical relativity, Phys. Rev. D 61, 087501 (2000), arXiv:gr-qc/9909026.DOI
- [64]C. W. Hirt, A. A. Amsden and J. L. Cook, An arbitrary Lagrangian-Eulerian computing method for all flow speeds, J. Comput. Phys. 14, 227 (1974).
- [65]E. Albertini, R. Cayuso, M. Vaglio and E. Barausse, ae-collapse: numerical code and symbolic checks for gravitational collapse in Einstein-aether theory, https://github.com/enricobarausse/ae-collapse (2026).
- [66]M. Kidney, A. Kovachki and S. Sibiryakov, Black hole sensitivity in Einstein-aether theory, arXiv:2610.00503.
- [67]J. Bellorin, C. Borquez and B. Droguett, Cancellation of divergences in the nonprojectable Horava theory Phys. Rev. D 106, 044055 (2022), arXiv:2207.08938.DOI
- [68]D. Blas, F. Del Porro, M. Herrero-Valea, J. Radkovski and S. Sibiryakov, Quantizing nonprojectable Horava gravity with Lagrangian path integral, Phys. Rev. D 113, 106022 (2026), arXiv:2512.14864












