On center of mass and foliations by constant spacetime mean curvature surfaces for isolated systems in General Relativity

We propose a new foliation of asymptotically Euclidean initial data sets by 2-spheres of constant spacetime mean curvature (STCMC). The leaves of the foliation have the STCMC-property regardless of the initial data set in which the foliation is constructed which asserts that there is a plethora of STCMC 2-spheres in a neighborhood of spatial infinity of any asymptotically flat spacetime. The STCMC-foliation can be understood as a covariant relativistic generalization of the CMC-foliation suggested by Huisken and Yau. We show that a unique STCMC-foliation exists near infinity of any asymptotically Euclidean initial data set with non-vanishing energy which allows for the definition of a new notion of total center of mass for isolated systems. This STCMC-center of mass transforms equivariantly under the asymptotic Poincar\'e group of the ambient spacetime and in particular evolves under the Einstein evolution equations like a point particle in Special Relativity. The new definition also remedies subtle deficiencies in the CMC-approach to defining the total center of mass suggested by Huisken and Yau which were described by Cederbaum and Nerz.


Introduction and goals
In General Relativity, isolated (gravitating) systems are individual or clusters of stars, black holes, or galaxies that do not interact with any matter or gravitational radiation outside the system under consideration. Intuitively, they should have a total center of mass which should in a suitable sense behave as a point particle in Special Relativity. In this paper, we suggest a definition of total center of mass for suitably isolated systems and argue that this center of mass notion indeed behaves as a point particle in Special Relativity in a suitable sense (meaning it transforms equivariantly under the asymptotic Poincaré group of the ambient spacetime). In particular, we will show that the center of mass notion we suggest evolves in time under the Einstein evolution equations like a point particle in Special Relativity.
The main idea of our approach is to modify the definition of center of mass given by Huisken and Yau [26] for asymptotically Euclidean Riemannian manifolds -using an asymptotic foliation by 2-spheres of constant mean curvature (CMC), see Section 2by staging it in a Lorentzian (spacetime) setting or in other words by staging it in asymptotically Euclidean initial data sets. More specifically, we will prove existence and uniqueness of an asymptotic foliation by 2-spheres of constant spacetime mean curvature under optimal asymptotic decay assumptions. Here, "spacetime constant mean curvature (STCMC)" means that the co-dimension 2 mean curvature vector H of each 2-sphere has constant Lorentzian length H. It is well known that this STCMC-condition can be reformulated in terms of initial data sets, namely as the product of the inner and outer "expansions" (or "null mean curvatures", see Remark 5.2) with respect to any given null frame along a 2-surface. On the other hand, the STCMC-condition is naturally independent of the initial data set in which the foliation is constructed. Our result thus asserts that there is a plethora of STCMC-surfaces in a neighborhood of spatial infinity of any asymptotically flat spacetime.
Furthermore, the new construction of a center of mass will be shown to remedy the subtle deficiencies of the Huisken and Yau approach [26] described by Cederbaum and Nerz [10]. Last but not least, we will provide an asymptotic flux integral formula for the center of mass extending that of Beig andÓ Murchadha [4]. The analytic techniques in our proofs rely on those developed by Metzger [31] and Nerz [36,37].
Concluding this introduction, we would like to point out that the notion of spacetime mean curvature of 2-surfaces in initial data sets has independently been considered in other contexts, both before and after the results of this paper had been announced. For example, the inverse spacetime mean curvature flow has been studied by Bray, Hayward, Mars, and Simon in [6] and by Hangjun Xu in his thesis [42].
The STCMC-condition is (trivially) satisfied by marginally outer/inner trapped surfaces (MOTS/MITS), extremal surfaces (see e.g. [21]), and generalized apparent horizons (see e.g. [30], [7]), with spacetime mean curvature H = 0 in all those cases. More generally, 2-surfaces with constant spacetime mean curvature are critical points for the area functional inside the future-directed null-cone, with mean curvature vector pointing in the direction in which the expansion of the surface is extremal. The aforementioned generalized apparent horizons have outer area minimizing property which is appealing in the view of spacetime Penrose Inequality. We would like to point the reader to the interesting work by Carrasco and Mars [9] giving insights into the (over-)generality of H = 0 as a condition for a horizon. In a recent paper of Cha and Khuri [11], the area A of the outermost STCMC-surface with H = 2 appears in the conjectured Penrose Inequality m ≥ A /16π expected to hold for an asymptotically anti-de Sitter initial data set of mass m satisfying the dominant energy condition.
Because of the spacetime geometry nature of the STCMC-condition, we expect that STCMC-surfaces and STCMC-foliations will have a number of applications beyond the definition of a center of mass of an isolated system as well as beyond the setting of asymptotically Euclidean initial data sets. For example, a special subfamily of STCMC-surfaces foliating a null hypersurface implicitly appears in recent work by Klainerman and Szeftel [28], where they arise as surfaces with both constant outer and constant inner expansion.
Structure of the paper. In Section 2, we will summarize the necessary definitions and notations as well as more details on the background and existing work on the total center of mass of isolated systems. In Section 3, we will state our main results and very briefly explain the strategy of our proofs. The following sections will be dedicated to the more technical components of the proof with Section 4 focusing on a priori estimates for STCMC-surfaces, Section 5 discussing the linearization of spacetime mean curvature, Section 6 asserting existence of the STCMC-foliation, Section 7 introducing the coordinate expression of the center of mass associated with the STCMC-foliation, and Section 8 proving the claimed law of time evolution under the Einstein evolution equations. Appendix A collects results such as Sobolev Inequalities on 2-surfaces, while Appendix B studies STCMC-surfaces in normal geodesic coordinates. Finally, in Section 9, we will discuss an exemplary initial data set highlighting the differences between the newly suggested notion of center of mass and the existing one suggested by Huisken and Yau. Max Planck Institute for Mathematics for allowing us to collaborate in stimulating environments.
CC is indebted to the Baden-Württemberg Stiftung for the financial support of this research project by the Eliteprogramme for Postdocs. The work of CC is supported by the Institutional Strategy of the University of Tübingen (Deutsche Forschungsgemeinschaft, ZUK 63). CC also thanks the Fondation des Sciences Mathématiques de Paris for generous support. The work of AS was supported by Knut and Alice Wallenberg Foundation and Swedish Research Council (Vetenskapsrådet). We also thank the Crafoord Foundation for generous support.

Preliminaries
Recall that an initial data set for the Einstein equations is a tuple (M 3 , g, K, µ, J) where (M 3 , g) is a smooth Riemannian manifold and K is a smooth symmetric (0, 2)tensor field on M 3 playing the role of the second fundamental form of M 3 in an ambient Lorentzian spacetime. The (scalar) local energy density µ and the (1-form) local momentum density J defined on M 3 can be read off from the constraint equations Scal − |K| 2 + (tr K) 2 = 2µ (1a) div(K − (tr K)g) = J.
Here, tr, div, and | · | denote the trace, the divergence, and the tensor norm with respect to g, respectively, and Scal denotes its scalar curvature. Sometimes we will find it convenient to use the conjugate momentum tensor π := (tr K)g − K.
The constraint equations (1) arise as a consequence of the Gauss-Codazzi-Mainardi equations from the Einstein equations Ric − 1 2 Scal g = T satisfied by a given spacetime (M 1,3 , g) with energy-momentum tensor T, where g is the Riemannian metric induced by the Lorentzian metric g on the spacelike hypersurface M 3 and K is the induced second fundamental form. Letting η denote the timelike future unit normal to the initial data set (M 3 , g, K, µ, J), the energy and momentum density are derived from T via µ = T(η, η), and J = T(η, ·), and the stress tensor S on M 3 is defined by S = T(·, ·). The constraint equations (1) thus necessarily hold on any spacelike hypersurface (or "initial data set") (M 3 , g, K, µ, J) in the spacetime (M 1,3 , g). In order to model an "isolated system", we will assume that the ambient spacetime (M 1,3 , g) with its energy-momentum tensor T and the choice of initial data set (M 3 , g, K, µ, J) are such that the initial data set is "asymptotically Euclidean", a notion made precise in the following standard definition. Definition 2.1 (Asymptotically Euclidean initial data sets). Let ε ∈ (0, 1 2 ] and let (M 3 , g, K, µ, J) be a smooth initial data set. Assume there is a smooth coordinate chart x : M 3 \B → R 3 \B R (0) defined in the region exterior to a compact set B ⊂ M 3 . We say that I := (M 3 , g, K, µ, J) is a C 2 1 /2+ε -asymptotically Euclidean initial data set (with respect to x) if there is a constant C = C(I, x) such that, in the coordinates x = (x 1 , x 2 , x 3 ) ∈ R 3 \ B R (0), we have the pointwise estimates for all x ∈ R 3 \B R (0) and for all i, j, k, l ∈ {1, 2, 3}. Here, slightly abusing notation, we have silently pushed forward all tensor fields on M 3 (including scalars) and written g ij := ( x * g) ij as well as K ij := ( x * K) ij , etc. The Kronecker delta δ ij denotes the components of the Euclidean metric with respect to the coordinates x. By another slight abuse of notation, we will refer to the above constant C as C I , suppressing the dependence on the chart x.
Asymptotically Euclidean initial data sets are well-known to possess well-defined total energy, linear momentum, and mass. More precisely, if I = (M, g, K, µ, J) is a C 2 1 /2+ε -asymptotically Euclidean initial data set for any ε > 0 (naturally extending the definition to ε > 1 2 ), its (ADM-)energy E and its (ADM-)linear momentum P = (P 1 , P 2 , P 3 ) are given by respectively, where dµ δ denotes the area measure induced on the coordinate sphere {| x | = r} by the Euclidean metric δ and ADM stands for Arnowitt-Deser-Misner [1].
The quantities E and P are well-defined under the asymptotic conditions imposed here for arbitrary ε > 0 [3,16] -meaning the expressions converge and E is asymptotically independent of the chart x while P is asymptotically covariant under chart deformations in a suitable way. From them, one defines the (ADM-)mass by whenever this expression makes sense, that is whenever the energy-momentum 4vector (E, P ) is causal with respect to the Minkowski metric of Special Relativity.
Remark 2.2 (Bounds on ε). For ε ≤ 0 in the above definition, one can find an asymptotic chart x (meaning a coordinate transformation outside a compact set) on the canonical Euclidean initial data set I Eucl. = (R 3 , δ, K ≡ 0, µ ≡ 0, J ≡ 0) with respect to which I Eucl. is C 2 1 /2+ε -asymptotically Euclidean but the expression E does not vanish as it should for Euclidean space, see Denisov and Soloviev [19]. This explains the suggestive notation of the decay order as 1 /2 + ε.
On the other hand, if ε > 1 2 for an initial data set I, a simple computation shows that it has E = P = 0 which is non-desirable in the context of discussing the center of mass and asymptotic foliations by constant mean curvature. This explains why we exclude this case in Definition 2.1.
2.1. Center of mass. We now proceed to discussing the total center of mass of an asymptotically Euclidean initial data set I = (M 3 , g, K, µ, J) with energy E = 0. The assumption E = 0 is both technical (as many definitions of center of mass explicitly divide by E) and physically reasonable when considering the center of mass.
First, let us remark that our field knows many definitions of center of mass for isolated systems. The first definitions were given in terms of asymptotic flux integral expressions in coordinates, similar to those of energy and linear momentum above, see (6) below and the text surrounding it. In 1996, Huisken and Yau [26] proved existence and uniqueness of a foliation by constant mean curvature 2-spheres near infinity of an asymptotically Euclidean Riemannian manifold of positive energy E > 0 and related it to a definition of center of mass in a way described below and in more detail in Section 7. More recently, Chen, Wang, and Yau [13] suggested a new definition of center of mass for isolated systems which is constructed from optimal isometric embeddings into the flat Minkowski spacetime of Special Relativity. For a brief, non-complete summary of other definitions of center of mass, please see [10].
Flux integral definition. The most prominent flux integral notion of center of mass C BÓM = (C 1 BÓM , C 2 BÓM , C 3 BÓM ) for asymptotically Euclidean initial data sets was introduced by Beig andÓ Murchadha [4] as the asymptotic flux integral a definition going back in parts to Regge and Teitelboim [39]. See Szabados [41] for valuable critical comments on this definition, and see Section 7 for a covariant generalization of this formula following from our work. The center of mass integral C BÓM will in general not converge for initial data sets I = (M 3 , g, K, µ, J) which are merely C 2 1 /2+ε -asymptotically Euclidean with respect to some chart x and have E = 0. It will however converge once one assumes that the initial data set satisfies certain asymptotic symmetry conditions in the given chart x, as for example the Regge-Teitelboim conditions introduced in [39], see [4,17,24] and Definition 2.5 below. We also point out that the expression for C BÓM does not explicitly depend on the second fundamental form K of the initial data set.
Definitions via foliations. Several authors define the center of mass of an initial data set I = (M 3 , g, K, µ, J) via a foliation by 2-spheres near infinity. Following Cederbaum and Nerz [10], we will call such definitions "abstract" in contrast to the more explicit "coordinate definitions" of center of mass, see below.
The first abstract definition of center of mass was given in 1996 by Huisken and Yau [26], who proved existence and uniqueness of a foliation near the asymptotic end of an asymptotically Euclidean Riemannian manifold by closed, stable 2-spheres of constant mean curvature, the CMC-foliation. This goes back to an idea of Christodoulou and Yau [15]. In 2006, Metzger [31] considered a foliation by 2-spheres of constant null mean curvature (also called constant expansion) and concluded that this foliation is not fully suitable for defining a center of mass. For a more detailed review of foliations suggested to study in this context and of recent progress in terms of necessary and sufficient asymptotic decay conditions, please see [10].
Huisken and Yau [26] also assign a coordinate center to their foliation. It is constructed from the abstract CMC-center as a "Euclidean center" of the CMC-foliation as follows: First, any closed, oriented 2-surface Σ ֒→ R 3 has a Euclidean coordinate center c (Σ) defined by Picking a fixed asymptotically flat coordinate chart x : , this definition can naturally be extended to closed, oriented 2-surfaces Σ ֒→ M 3 \ B by pushing Σ forward to R 3 and identifying c (Σ) := c ( x(Σ)), slightly abusing notation. We will also call this center Euclidean center of Σ (with respect to x). This naturally extends to asymptotic foliations: Definition 2.4 (Coordinate center of a foliation). Let I = (M 3 , g, K, µ, J) be a C 2 1 /2+ε -asymptotically Euclidean initial data set for a chart x : M 3 \ B → R 3 \ B R (0). Let {Σ σ } σ>σ 0 be a foliation of the asymptotic end M 3 \ B of M 3 with area radius r(Σ σ ) = |Σ σ | /4π of Σ σ diverging to ∞ as σ → ∞. Denote by c (Σ σ ) the Euclidean coordinate center of the leaf Σ σ with respect to x. Then the (Euclidean) coordinate center C = (C 1 , C 2 , C 3 ) of the foliation {Σ σ } σ>σ 0 (with respect to the asymptotic chart x) is given by in case the limit exists. Otherwise, we say that the coordinate center of the foliation {Σ σ } σ>σ 0 diverges (with respect to the asymptotic chart x).
The vector C can be pictured to describe a point in the target R 3 of the asymptotically flat coordinate chart x : M 3 \ B → R 3 \ B R (0), but it need not lie in the image of x, and indeed will often lie inside B R (0). This means it cannot necessarily be pulled back into M 3 . Moreover, C depends on the choice of asymptotic chart x -at least a priori.
Coming back to the CMC-foliation constructed by Huisken and Yau [26], it is well-known that the coordinate center C HY of the CMC-foliation of a suitably asymptotically flat Riemannian manifold (M 3 , g) or initial data set (M 3 , g, K, µ, J) of nonvanishing energy E with respect to a given asymptotic chart x coincides with the Beig-Ó Murchadha center of mass vector C BÓM defined by (6), provided that some additional symmetry assumptions are satisfied, see Huang [25], Eichmair and Metzger [20], and Nerz [34]. The most optimal result to date [36, Theorem 6.3] states that for C 2 1 /2+ε -asymptotically Euclidean Riemannian manifolds with E = 0 satisfying the C 2 1+ε -Regge-Teitelboim condition (see Definition 2.5 below), we have C HY = C BÓM whenever both definitions converge, and that divergence of one implies divergence of the other. Again, let us point out that the construction of C HY does not explicitly depend on the second fundamental form K of the initial data set under consideration. We furthermore note that the product E C BÓM is sometimes referred to as the "center of mass charge" in the literature, even when E = 0. We will not follow this usage here.
In this paper, we construct a novel geometric foliation {Σ σ } σ>σ 0 of the asymptotically flat end M 3 \ B of a given C 2 1 /2+ε -asymptotically Euclidean initial data set (M 3 , g, K, µ, J) with non-vanishing energy E = 0, namely a foliation with "constant spacetime mean curvature (STCMC)"-leaves, see Section 3. The general approach to define the coordinate center of a foliation {Σ σ } σ>σ 0 described above will then be applied to this new foliation to obtain a new definition of the coordinate center of mass of an initial data set as well as a coordinate expression analogous to and extending (6), see Section 7.

Miscellannea.
Here we collect some other definitions for future reference.
Regge-Teitelboim condition for initial data sets. With the exception of the later part of Section 7, we will not assume that the initial data sets under consideration satisfy any asymptotic symmetry assumptions, in particular the Regge-Teitelboim conditions. However, it will be useful in our discussion to refer to those conditions which is why we define them here. Definition 2.5 (Regge-Teitelboim conditions for initial data sets). We say that a C 2 1 /2+ε -asymptotically Euclidean initial data set I = (M 3 , g, K, µ, J) satisfies the C 2 γ+ε -Regge-Teitelboim conditions for γ > 1 2 (with respect to the given chart x with respect to which it is C 2 1 /2+ε -asymptotically Euclidean) if there is a constant C = C(I, x, γ) such that holds for all x ∈ R 3 \ B R (0) and for all i, j, k, l ∈ {1, 2, 3}. Here, as usual, we have denoted the even and odd parts of any continuous function f : Remark 2.6 (Regge-Teitelboim conditions for Riemannian manifolds). We say that a C 2 1 /2+ε -asymptotically Euclidean Riemannian manifold (M 3 , g) satisfies the C 2 γ+ε -(Riemannian) Regge-Teitelboim conditions on R 3 \ B R (0) for γ > 1 2 if the above inequalities are satisfied for π ≡ K ≡ 0, i.e. if (9a) holds and if |Scal odd | ≤ C| x | − 5 Weighted Sobolev spaces. In this paper, we use the following definition of Sobolev spaces, which is well-suited for keeping track of fall-off rates of different quantities associated with our foliation. Suppose that (Σ, g Σ ) is a closed (compact without boundary), oriented 2-surface in an asymptotically Euclidean 3-manifold (M 3 , g) of suitable regularity. For p ∈ [1, ∞), the Lebesgue space L p (Σ) is defined as the set of all measurable functions f : Σ → R such that their L p -norm , the Sobolev norms are defined as follows: where r := |Σ| /4π is the area radius of Σ. The Sobolev space W k,p (Σ) is the set of all functions with finite W k,p -norm. This definition naturally extends to the case of tensor fields on Σ. Appendix A in particular collects some Sobolev Inequalities for functions on 2-surfaces (Σ, g Σ ) embedded in Euclidean space.
3. Main results, motivation, and the strategy of the proof Given a 2-dimensional surface Σ in an initial data set (M 3 , g, K, µ, J), we denote its mean curvature inside the Riemannian manifold (M 3 , g) with respect to the outward pointing unit normal 1 by H and set P := tr Σ K, as usual. The spacetime mean curvature (STMC) of Σ is defined by the length of the spacetime mean curvature vector H We will suggestively write H σ τ to denote the spacetime mean curvature of a surface called Σ σ τ etc, H to denote the spacetime mean curvature of a surface called Σ etc., whenever the initial data set inducing the intrinsic and extrinsic geometry on the surface is clear from context.
In this paper we prove the following theorems.
Here, A(a, b, η) is an a priori class of "asymptotically centered" spheres introduced in Section 4. It has been shown in particular by Brendle and Eichmair [8] that such an a priori condition is necessary to obtain uniqueness of CMC-surfaces in general, see the discussion in Section 6.3. As STCMC-surfaces generalize CMC-surfaces, their observation applies here, too.
We also obtain a coordinate expression C STCMC for the STCMC-center of mass, see below. It differs from the Beig-Ó Murchadha formula C BÓM given in (6) by a term Z, as stated in the following theorem.
Theorem 7.5 (STCMC-coordinate expression). Let I = (M 3 , g, K, µ, J) be a C 2 1 /2+εasymptotically Euclidean initial data set with respect to an asymptotic coordinate chart x : M 3 \ B → R 3 \ B R (0) and decay constant C I , with non-vanishing energy E = 0. Assume in addition that for all x ∈ R 3 \ B R (0) and that g satisfies the Riemannian C 2 3 /2+ε -Regge-Teitelboim condition. Then the coordinate center C STCMC of the unique foliation by surfaces of constant spacetime mean curvature is well-defined if and only if the correction term limits exist for i = 1, 2, 3. In this case, we have where C BÓM is the Beig-Ó Murchadha center of mass and Z = (Z 1 , Z 2 , Z 3 ), or equivalently An example of an initial data set with C STCMC = C BÓM or in other words with Z = 0 will be discussed in Section 9. The above formula for C STCMC allows to compute the STCMC-center of mass of an initial data set explicitly, once an asymptotic chart x has been picked. However, as the assumptions of Theorem 7.5 suggest, this formula cannot be expected to always converge. See Conjecture 7.9 and the text above of it for a discussion of when the coordinate expression for C STCMC should converge, without reference to C BÓM and Z and without any Regge-Teitelboim conditions nor additional decay assumptions on K.
We get the following theorem on the time evolution of the STCMC-foliation and center of mass. The full covariance of the STCMC-foliation under the Poincaré group is discussed in Section 8.2. Theorem 8.1 (Time evolution of STCMC-foliation). Let (R × M 3 , g) be a smooth, globally hyperbolic Lorentzian spacetime satisfying the Einstein equations with energy momentum tensor T. Suppose that, outside a set of the form R×K, Assume that I 0 = ({0} × M 3 , g, K, µ, J) ֒→ (R × M 3 , g) is a C 2 1 /2+ε -asymptotically Euclidean initial data set with respect to the coordinate chart x and with E = 0, and suppose additionally that K = O 1 (| x | −2 ) with constant C I as | x | → ∞. Now consider the C 1 -parametrized family of C 2 1 /2+ε -asymptotically Euclidean initial data sets with respect to x which starts from I(0) = I 0 , and which exists for all t ∈ (−t * , t * ) for some t * > 0. Assume furthermore that the constants C I(t) are uniformly bounded on (−t * , t * ), without loss of generality such that C I(t) ≤ C I 0 .
Assume the foliation I(t) has initial lapse N = 1 as | x | → ∞ with decay measuring constant denoted by C N and initial shift X = 0, and suppose furthermore that the initial stress tensor S of There is a constant t > 0, depending only on ε, C I 0 , C N , and E(0) such that the following holds: If the initial data set I 0 has well-defined STCMC-center of mass C STCMC (0) then the STCMC-center of mass C STCMC (t) of I(t) is also well-defined for |t| < t. Furthermore, the initial velocity at t = 0 is given by Moreover, we have that d dt t=0 E = 0 and d dt t=0 P = 0.
3.1. Strategy of the proofs of Theorems 6.2 and 6.12. The underlying structure of the proofs of Theorems 6.2 and 6.12 presented in Section 6 and several of the lemmas proved in the same section is a method of continuity inspired by Metzger [31,32]. Given an initial data set I = (M 3 , g, K, µ, J), we will consider the one-parameter family of initial data sets I τ = (M 3 , g, τ K, µ τ , τ J), τ ∈ [0, 1], with µ τ given through the constraint equations (1) as For τ = 0, we thus consider the Riemannian manifold (M 3 , g) with 2µ = 2µ 0 = Scal while for τ = 1, we study the original initial data set I = (M 3 , g, K, µ, J) with µ = µ 1 . It is straightforward to see that if the original initial data set I is C 2 1 /2+ε -asymptotically Euclidean with respect to an asymptotical chart x : M 3 \ B → R 3 \ B R (0) then all initial data sets I τ are also C 2 1 /2+ε -asymptotically Euclidean with respect to the same chart and comparable constants. In particular, the Riemannian manifold (M 3 , g) is C 2 1 /2+ε -asymptotically Euclidean in this chart. This is what will allows us to drop the explicit mention of the chart in the proofs. Moreover, we note that the energy E τ computed for the initial data set I τ = (M 3 , g, τ K, µ τ , τ J) does in fact not depend on τ and can and will thus be called E. We globally assume in this paper that E = 0 and we will fix the background Riemannian manifold (M 3 , g) once and for all.
For second fundamental form K = 0, the desired STCMC-foliation coincides with the classical CMC-foliation. From Nerz' work [36], we thus know that the theorems and lemmas we will prove for initial data sets hold in the Riemannian setting under the Riemannian version of our assumptions, see also Remark 2.3. In other words, we know that our claims hold for τ = 0 in the method of continuity approach described above. In Section 6, we will recall Nerz' corresponding theorems in our notation.
As usual, we will appeal to the Implicit Function Theorem in order to show openness of the interval in the method of continuity. Closedness follows from a standard convergence argument.

A priori estimates on STCMC-surfaces
When deforming the foliation by 2-surfaces of constant mean curvature to the foliation by 2-surfaces of constant spacetime mean curvature, we need to keep track of how the geometry of the leaves changes. For this, following [32] and [36], we will now introduce an a priori class of closed, oriented 2-surfaces having the properties that their "area radius", "coordinate radius", and "mean curvature radius" as defined below are comparable in a certain sense.
In this section, we will not make explicit reference to the asymptotic coordinate chart x : M 3 \ B → R 3 \ B R (0) in most estimates, however the asymptotic coordinates x will be used in order to compute the coordinate radius and the center of mass of a given 2-surface Σ ֒→ M 3 (or " Σ ֒→ I "). We will always and mostly tacitly assume that Σ ֒→ M 3 \ B so that it lies in the domain of the asymptotic coordinate chart. , and z i : respectively, where dµ δ denotes the area element on Σ induced by the Euclidean metric δ. Given constants a ∈ [0, 1), b ≥ 0, and η ∈ (0, 1], we say that Σ belongs to the a priori class of (M 3 , g)-asymptotically centered surfaces, if its area radius r, coordinate center z, coordinate radius | x |, and mean curvature H satisfy the following estimates where γ denotes the genus of Σ.
Remark 4.2. We will use the same a priori classes in the context of asymptotically Euclidean initial data sets I = (M 3 , g, K, µ, J), where the definition of A(a, b, η) only depends on the Riemannian manifold part (M 3 , g). This will later be important when we consider families of initial data sets of the form I τ = (M 3 , g, τ K, µ τ , τ J), see Section 6 and (12).
Example 4.4. Let (M 3 , g) be a C 2 1 /2+ε -asymptotically Euclidean manifold with nonvanishing energy E = 0. Then the unique leaves of the constant mean curvature foliation {Σ σ } σ>σ 0 constructed in [36] are asymptotically centered in this sense. More specifically, there are constants b > 0 and σ 0 > 0 depending only on C I such that Σ σ ∈ A(a = 0, b, η = ε) for σ > σ 0 . See [36, Section 5] for details. Proposition 4.5. Suppose that a ∈ [0, 1), b ≥ 0, η ∈ (0, 1], and assume that 0 ≤ a ≤ a, 0 ≤ b ≤ b, and η ≤ η ≤ 1. Let I = (M 3 , g, K, µ, J) be a C 2 1 /2+ε -asymptotically Euclidean initial data set. Then there exist constants σ and C depending only on ε, a, b, η, and C I such that the following a priori conclusions hold for any closed, oriented 2-surface Σ ֒→ I with Σ ∈ A(a, b, η): Suppose that Σ has constant spacetime mean curvature H ≡ 2 /σ in I for some σ > σ. Then Σ is a topological sphere and the tracefree partÅ of its second fundamental form satisfies Furthermore, there exists a function f : as well as a conformal parametrization ψ : where Id denotes the trivial embedding (S 2 r ( z ), g S 2 r ( z ) ) ֒→ (R 3 , δ). The conformal factor u : Finally, the Euclidean distance to the coordinate origin | x | (on Σ), the area radius r, and the spacetime mean curvature radius σ are comparable in the following sense: Remark 4.6. The conclusions of this theorem are mostly the same as those in [36, Proposition 4.4], only for STCMC-rather than CMC-surfaces. However, we cannot directly refer to this result because, roughly speaking, it assumes that the mean

whereas the relation
recalling P = tr Σ K, and the definition of the a priori class A(a, b, η) -which coincides with that in [36] -, only ensure via the second inequality in (15) that which does not a priori give us H − 2 σ = O(r − 3 2 −ε ). We will thus need to extend the result and its proof to our setting.
Proof. Within this proof, C will always be a generic constant depending only on σ, a, b, η, and C I . With Remark 4.6 in mind, we need to improve the estimate in (22). For this purpose, we first note that by the definition of (M 3 , g) being C 2 1 /2+ε -asymptotically Euclidean and by the second inequality in (15), we have Combining this with the last inequality of (15), we conclude by the Gauss equation and the Gauss-Bonnet Theorem that We are now in a position to apply the result of De Lellis and Müller [18, Theorem 1.1] (see also [32, Section 2.3] where this result is reformulated in a scale invariant form) to conclude that Σ is a topological sphere, with a conformal parametrization ψ : S 2 r ( z ) → Σ and the conformal factor u : In order to prove that σ and r are comparable, we estimate , where we have used (2) and the second inequality in (15) in the last line. Here, we have (2.4 (15), and (22).

) in [32]), and
Summing up, we conclude that To prove that | x | and r are comparable, note that by the first inequality in (15) and For such large radii, we thus elementarily find with the help of the first inequality in (15). By (23a) and the Sobolev Inequality in the form of Lemma A.2, it follows that |ψ − Id | ≤ C| x | 1− η 2 . Combining this with the above inequalities, we conclude that, on Σ, we have provided that the area radius r of Σ is sufficiently large. Via (24), we can alternatively state that (25) holds if the spacetime mean curvature radius σ satsifies σ > σ for a suitably large σ only depending on ε, a, b, η, and C I .
Bootstrapping. With these new bounds (24) and (25) at hand, we can apply [36, Proposition 4.1] with κ chosen as 3 2 + ε > 1, η chosen as our η 2 > 0, and c 1 , c 2 chosen as our generic constant C. As all the estimates going into verifying the assumptions from [36, Proposition 4.1] hold pointwise in our case, the assumptions are indeed satisfied for any p > 2. Note that the existence of the uniform Sobolev Inequality assumed in [36, Proposition 4.1] is well-established in our setting, and goes back to [26,Proposition 5.4] which holds for surfaces in asymptotically Euclidean manifolds with general asymptotics as described in Section 2. Again via (24), this gives us (16) for σ > σ, with suitably enlarged σ only depending on ε, a, b, η, and C I .
Finally, now that we have a pointwise bound on the tracefree part of the second fundamental formÅ accompanying the pointwise estimate (22) for the mean curvature H, it follows that Σ is the graph of a function f ∈ W 2,∞ (S 2 r ( z )) such that (17) holds for σ > σ, for again suitably enlarged σ only depending on ε, a, b, η, and C I , see e.g. [36, Corollary E.1], which adapts [18, Theorem 1.1] to our setting. To be more precise, [36, Corollary E.1] is only not stated invariantly under scaling but with |Σ| = 4π, but it is straightforward to adapt it to include the area radius for our purposes. This finishes the proof of Proposition 4.5.

The linearization of spacetime mean curvature
In this section, we will introduce the spacetime mean curvature map H in a given initial data set I = (M 3 , g, K, µ, J). We will analyze its properties in a neighborhood of a given 2-surface Σ having constant spacetime mean curvature. We will show that the linearization of the map H is invertible when the linearization is computed with respect to normal variations within the given initial data set I. This will later be used to ensure that the CMC-foliation of (M 3 , g) constructed in [36] can be pushed via a method of continuity to an STCMC-foliation of I.
Throughout this section, we will assume that I = (M 3 , g, K, µ, J) is a C 2 1 /2+εasymptotically Euclidean initial data set with non-vanishing energy E = 0 and with fixed asymptotic coordinates x. Furthermore, it will be assumed that Σ is a fixed 2surface of constant spacetime mean curvature H(Σ) ≡ 2 /σ which has sufficiently large mean curvature radius σ and which for some fixed a ∈ [0, 1), b ≥ 0, and η ∈ (0, 1] belongs to the a priori class A(a, b, η), see Definition 4.1. In this setting, we know from Proposition 4.5 that Σ is a topological sphere, and that its coordinate radius, area radius, and mean curvature radius are comparable as stated in (20), (21). This in particular implies that (26) for σ > σ, where σ and the constants hidden in the O-notation only depend on ε, a, b, η, C I . 5.1. Stability operators associated with prescribed (spacetime) mean curvature surfaces. In a neighborhood of Σ, we introduce normal geodesic coordinates y : Σ × (−ξ, ξ) → M 3 for some ξ > 0, such that y(·, 0) = Id Σ , and ∂y ∂t = ν Σt , with ν Σt being the outward unit normal to Σ t := y(Σ, t). For a function f ∈ C ∞ (Σ) with |f | < ξ, we define the graph of f over Σ as Then, slightly abusing notation, let H : C ∞ (Σ) → C ∞ (Σ) be the operator which assigns to a function f the spacetime mean curvature H(f ) of graph f (with respect to the fixed initial data set I). The linearization of this map H is computed in the following lemma.
. Then the linearization L H of the spacetime mean curvature map at Σ is given by where △ Σ , ∇ Σ denote the Laplacian and covariant gradient on (Σ, g Σ ), respectively.
Proof. This follows from the definition of spacetime mean curvature H = √ H 2 − P 2 and the well-known formulas for ∂H(V(·,t)) ∂t t=0 and ∂P (V(·,t)) ∂t t=0 , see Metzger [32, The map L H naturally extends to a bounded mapping L H : W 2,2 (Σ) → L 2 (Σ). In Section 5.3, we will prove that this mapping has a bounded inverse, for which it is convenient to rewrite the above expression for L H in the form Since the denominator is clearly bounded and bounded away from zero by our assumptions on Σ, the (bounded) invertibility of L H : W 2,2 (Σ) → L 2 (Σ) will follow once we show that L : Remark 5.2. Recall that the H ±P -stability operator L H±P of the map H ±P (surfaces of constant expansion or null mean curvature) is given by . As it turns out, the analytic properties of L H±P imply that constant expansion foliations do not provide an adequate notion of center of mass, in contrast to the STCMC-foliation studied here. The main difference is that the contribution of the second fundamental form K in the H ± P -stability operator is large, while it is rescaled by a factor P /H in the STCMC-stability operator. The largeness of the contribution of K in the H ± P -stability operator will cause the geometric centers of the surfaces of the foliation to drift away in the direction of the linear momentum P in general, see Metzger [32, Section 7]. This can only be avoided by imposing very fast fall-off conditions on K to ensure that P = 0. Furthermore, a certain smallness assumption on K is also directly required to ensure the invertibility of L H±P , and hence the existence of the constant expansion foliation, see [37, Theorem 3.1]. As a consequence of the factor P /H in the STCMC-stability operator, no smallness assumption on K will be needed to ensure the existence of the foliation by surfaces of constant spacetime mean curvature. Furthermore, we will see that the leaves of this foliation do not translate as their spacetime mean curvature approaches zero, provided that the standard asymptotic symmetry conditions are imposed.
As we will see, the operator L, and consequently the operator L H , is in many respects similar to the standard (CMC-)stability operator of Σ, namely to This operator has been intensively studied, see e.g. [2].

5.2.
Eigenvalues and eigenfunctions of −∆ Σ . In preparation for proving the invertibility of the operator L, we summarize the spectral properties of the operator −∆ Σ , the Laplacian calculated with respect to the metric g Σ induced by (Σ, g Σ ) ֒→ (M 3 , g). For this, let us first consider the operator −∆ S 2 r , the Laplacian calculated with respect to the standard round metric δ S 2 r on S 2 The eigenvalues of −∆ S 2 r are l(l + 1) /r 2 for l ≥ 0, and the eigenspace corresponding to l(l + 1) /r 2 is given by the space of homogeneous harmonic polynomials of degree l restricted to S 2 r , see e.g. [12,Chapter II.4]. In particular, the first non-zero eigenvalue of −∆ S 2 r is 2 /r 2 , the corresponding eigenspace is spanned by the restrictions to S 2 r of the coordinate functions x 1 , x 2 , x 3 on R 3 . In the following, we enumerate the eigenvalues of −∆ S 2 r counting their multiplicity by and we denote the associated complete L 2 (S 2 r )-orthonormal system of eigenfunctions by {f δ i } ∞ i=0 . Without loss of generality, we may assume that the chosen enumeration is such that Note that the tracefree part of the Hessian of each of these functions vanishes, and that we have where ∇ S 2 r denotes the gradient with respect to δ S 2 r . In order to describe the eigenvalues and eigenfunctions of the operator −∆ Σ , note that by Proposition 4.5 there is a vector z ∈ R 3 and a conformal parametrization where r is the area radius of (Σ, g Σ ). As all spheres of radius r in Euclidean space are isometric, we can easily "translate" ψ to a conformal parametrization ψ : S 2 r → Σ such that where r still denotes the area radius of (Σ, g Σ ).
We will now describe a complete orthonormal system in L 2 (Σ) consisting of the . ., again counted with multiplicity. The eigenfunctions f i will be chosen so that ψ * f i is asymptotic to f δ i for each i = 1, 2, . . . . For simplicity of notation, in what follows we will identify f i : Σ → R with its pullback ψ * f i without further ado. This enumeration and identification will also allow us to prove useful estimates for the eigenvalues and eigenfunctions of −∆ Σ . 1], and consider a 2-surface Σ ֒→ I such that Σ ∈ A(a, b, η) with respect to I. Then there exist constants C > 0 and σ > 0 depending only on ε, a, b, η, and C I such that if Σ has constant spacetime mean curvature H ≡ 2 /σ for σ > σ, then there is a complete orthonormal system in L 2 (Σ) consisting of the eigenfunctions . ., counted with multiplicity, and such that for i = 1, 2, 3 the following estimates hold • Furthermore, λ 0 = 0 and Proof. By (31) and Lemma A.2 we have Applying the Rayleigh Theorem (see e.g. [12, Chapter II.5]), we see from the above estimate and (21) that where the infimum is taken over all f ∈ W 1,2 (Σ) with´Σ f dµ = 0 and f L 2 (Σ) = 1.
Of course, the O-term constant and the lower bound on σ coming from this calculation only depend on ε, a, b, µ, and C I . We will now construct the respective eigenfunctions f i , for which we will use the fact that these functions are solutions to the equation where λ δ i = 2 /r 2 for i = 1, 2, 3. Noting that the right hand side of the equation equals −∆ S 2 r f i − λ δ i f i , and using integration by parts it is straightforward to check that it is orthogonal in L 2 (S 2 r ) to any element in the kernel of the self-adjoint differential operator in the left hand side. Thus, by the Fredholm Alternative [5, Appendix I, Theorem 31], for every i = 1, 2, 3 there is a unique solution f i − f δ i ∈ W 2,2 (S 2 r ) orthogonal in L 2 (S 2 r ) to the linear space spanned by f δ i , i = 1, 2, 3. Note that we may without loss of generality assume that −ε ) as a consequence of the Gauss equation (see e.g. (43) below), in view of (21) and Lemma A.
With the above estimates at hand, it is now straightforward to check that whenever σ > σ, for suitably large C > 0 and σ > 0 depending only on ε, a, b, ν, and C I . This defines the eigenfunctions f i , i = 1, 2, 3, up to applying the Gram-Schmidt process in the case of multiple eigenvalues. Note that (38) with (21), (31), and the properties of the functions f δ i implies (34) and (35).
We can now give a more detailed characterization of the lowest eigenvalues λ i , i = 1, 2, 3. More specifically, in the following lemma we show that these eigenvalues are computed in terms of the Hawking mass of Σ in the initial data set I. We will drop the explicit reference to Σ later and will write m H instead of m H (Σ). This lemma and its proof are very similar to [36, Lemma 4.5], but rephrased in the spacetime setting. 1], and that Σ ∈ A(a, b, η) with respect to I is a surface with Hawking mass m H (Σ). Then there exist constants C > 0 and σ > 0, depending only on ε, a, b, η, |E|, and C I such that if Σ has constant spacetime mean curvature H ≡ 2 /σ for σ > σ then the following estimates hold and Proof. A polarized version of the standard Bochner formula (see e.g. Proposition 33 (3) in [38,Chapter 3]) in dimension 2 applied to the eigenfunctions f i and f j for i, j = 1, 2, 3 reads Integrating this identity, using the Divergence Theorem on the closed surface Σ, integrating by parts, and recalling (34), we obtain for some constant C > 0 and all σ > σ > 0, with C and σ only depending on ε, a, b, η, and C I . Next, the Gauss equation combined with (16) and (26) gives us possibly enlarging C > 0 and σ > 0 without introducing new dependencies.
Substituting this into (42) and using (35), (32) together with the fact that our initial data set is C 2 1 /2+ε -asymptotically Euclidean we conclude that, by partial integration, we get When i = j, i, j = 1, 2, 3, this gives us (41) once we recall that Scal = O(σ −3−ε ) as a consequence of Definition 2.1 with possibly enlarged C > 0 and σ > 0. In the case i = j, i, j = 1, 2, 3, one arrives at (40) by combining (44), (21), (32), and the fact that our initial data set is C 2 1 /2+ε -asymptotically Euclidean, as well as using This last inequality follows from (43), the Gauss-Bonnet Theorem, and the definition of r, σ, and m H (Σ) with possibly enlarged C > 0 and σ > 0. This proves the claims of the lemma. (45), this lemma remains valid if we replace the Hawking mass m H (Σ) by the Geroch mass m H (Σ) (also sometimes referred to as "(Riemannian) Hawking mass") given by The same remark will hold true for the subsequent results. However, we choose to use m H (Σ), and not m H (Σ), throughout to emphasize the spacetime nature of our result.

5.3.
Invertibility of the operator L. Section 5.2 above provides the following description of the eigenvalues of the Laplacian −∆ Σ : • for i = 1, 2, 3 the eigenvalues λ i are characterized by formula (40), It turns out to be useful to decompose functions h ∈ L 2 (Σ) with respect to the L 2 (Σ)complete orthonormal system {f 0 , f 1 , f 2 , f 3 , . . . } of eigenfunctions corresponding to −∆ Σ when analyzing Lh and the L 2 (Σ)-adjoint L * h (the latter being of interest as we are aiming for a Fredholm Alternative argument). More specifically, it is useful to split any given function h ∈ L 2 (Σ) into its mean value its translational part and the difference part 3 Proposition 5.6. Let I = (M 3 , g, K, µ, J) be a C 2 1 /2+ε -asymptotically Euclidean initial data set with energy E. Suppose that a ∈ [0, 1), b ≥ 0, η ∈ (0, 1], and that Σ ∈ A(a, b, η) with respect to I is a surface with non-vanishing Hawking mass m H (Σ) = 0. Then there exist constants C > 0 and σ > 0, with C depending only on ε, a, b, η, and C I and σ in addition depending on |E| in (54), such that if Σ has constant spacetime mean curvature H ≡ 2 /σ for σ > σ, the following estimates hold for any h, h 1 , h 2 ∈ W 2,2 (Σ). The same estimates apply to the L 2 (Σ)-adjoint L * . Moreover, the Hawking mass m H (Σ) and the energy E are related by In particular, the operator L : W 2,2 (Σ) → L 2 (Σ) is invertible as long as the energy E of the initial data set does not vanish and σ is sufficiently large, depending only on ε, a, b, η, and C I .
Proof. In this proof, C > 0 and σ > 0 denote generic constants that may vary from line to line, but depend only on ε, a, b, η, and C I , and, in the case of (54), also on |E|.
Proving (51). Arguing as above, by Proposition 4.5, Lemma 5.3, and our decay assumptions on the initial data set, we have that for any i, j = 1, 2, 3. It then follows by Lemma 5.4 that In particular, we see that (51) holds for h t 1 , h t 2 ∈ {f 1 , f 2 , f 3 }. The general case follows by bilinearity and by the Cauchy-Schwarz Inequality on R 3 .
Proving (53). We will use the following manifest relation for the linear operator L Arguing similarly to how we argued above, we now integrate by parts and use Proposition 4.5, (26), and (36) giving λ i > 5 /σ 2 for i = 4, 5, . . . , and the asymptotic decay conditions on I to estimate from below the expression Here, the factor 7 4 < 3 is chosen for later convenience. Hence by a Cauchy-Schwarz Inequality on´Σ Note also that h 0 is a constant, so that and thus where again, the factor 7 4 < 2 is chosen for later convenience. Using (58), integration by parts and finally Young's Inequality, one can also check with the same decay arguments as above that Combing this estimate with (57) and (59), (53) follows from (56) once we recall that Proving (55). To see that E and m H (Σ) are as close as claimed, we recall the wellknown fact that the Geroch mass m H (Σ) of sufficiently round large surfaces in a C 2 1 /2+ε -asymptotically flat initial data set I is close to the energy E of I. More specifically, Lemma A.1 in [36] (relying on [23] and [33]) and (45) imply that Thus m H (Σ) = 0 if E = 0 as long as σ > σ with C > 0 and σ > 0 sufficiently large, depending only on ε, a, b, η, and C I .
. As a consequence, using (50), (53), and Young's Inequality, we obtain for any 0 < α < 1, e.g. α = 1 2 , that provided that σ > σ, where now σ may actually depend on E as we used (55) in the last step. Thus (54) holds in case h d 2 . In this case, where by (51), (55), and because h t and h d are L 2 (Σ)-orthogonal, we see that for σ > σ, σ suitably large depending now in addition on E, we have where we used that ε ≤ 1 2 by definition. Moreover, using the Cauchy-Schwarz Inequality and the assumption h d 2 Further, arguing once more as above with the explicit form of L in (28), using the asymptotic decay conditions of I, (16), (26), and integration by parts, one confirms that using again the assumption h d 2 L 2 (Σ) ≤ σ − 1 2 −κ h 2 L 2 (Σ) as well as h t L 2 (Σ) ≤ h L 2 (Σ) in the last step. It then follows from (60), the Cauchy-Schwarz Inequality, and the bounds on κ that (54) also holds when h d 2 . Combining Case 1 and Case 2, we have thus shown (54). To conclude the proof, it only remains to show that L * also satisfies the estimates (50)-(54) and that L is invertible provided that the initial data set I has non-vanishing energy E = 0 and the spacetime mean curvature radius σ of Σ is sufficiently large, σ > σ, σ depending on E.
Invertibility of L and estimates on L * . The operator L is not self-adjoint, but its L 2 (Σ)-adjoint L * has very similar structure, differing only in the last term. In Lh, Going back to the proofs of (50)-(54), we see that all of them work if we replace L by L * modulo exchanging the performance of partial integration with the decay estimate (61) and vice versa. This, in particular (54), implies that the L 2 (Σ)-kernel of L * is trivial, and hence L : W 2,2 (Σ) → L 2 (Σ) is invertible by the Fredholm Alternative [5, Appendix I, Theorem 31], as long as m H (Σ) = 0 which is guaranteed from E = 0 and (55). The Fredholm Alternative applies as L is clearly a linear elliptic operator as its symbol is that of the Laplacian −∆ Σ and because Σ is compact.
Proof. Note by (43) we have Scal Σ = 2 −ε ), hence, in the view of (21), Lemma A.3 applies to Σ. Combined with the Cauchy-Schwartz Inequality and (21), this result gives us Recalling the definition of the operator L (see (28)) and the fall-off properties of the initial data set, we further find that Combining (62) with this estimate and (54) we thereby obtain This proves the estimate for h. The estimate for h d is proven similarly, using (53) instead of (54).
This result is a starting point for proving the following theorem, which is essentially the main result of this paper. For the sake of clarity of exposition, we provide the proof of the following theorem right away, saving the verification of some preliminary lemmas for later. We state Theorem 6.2 here in a notation convenient for its proof. Theorem 6.2 (Existence of STCMC-foliation). Let I 1 = (M 3 , g, K, µ, J) be a C 2 1 /2+εasymptotically Euclidean initial data set with non-vanishing energy E = 0. Then there is a constant σ I 1 > 0 depending only on ε, C I 1 , and E, a compact set K 1 ⊂ M 3 , and a bijective C 1 -map Ψ 1 : (σ I 1 , ∞) × S 2 → M 3 \ K 1 such that each of the surfaces Σ σ 1 := Ψ 1 (σ, S 2 ) has constant spacetime mean curvature H(Σ σ 1 ) ≡ 2 /σ 1 provided that σ > σ I 1 . Remark 6.3. As the proof of Theorem 6.2 will show, the surfaces Σ σ 1 are in fact asymptotically centered in the sense of Definition 4.1, more specifically, Σ σ 1 ∈ A(0, b I 1 , η I 1 ) for all σ > σ I 1 , with constants b I 1 > 0, η I 1 ∈ (0, 1], and σ I 1 > 0 defined in the proof of Theorem 6.2, and depending only on ε, C I 1 , and E. Proof. The family of closed, oriented 2-surfaces {Σ σ 1 } σ>σ I 1 will be constructed via a method of continuity, see also Section 3. Roughly speaking, we will deform the constant (automatically spacetime) mean curvature foliation {Σ σ 0 } σ>σ I 0 of the initial data set I 0 from Theorem 6.1 along the curve of initial data sets {I τ } τ ∈[0,1] , where I τ := (M 3 , g, τ K, µ τ , τ J) is as described in Section 3.1, arriving at the foliation of the initial data set I 1 by constant spacetime mean curvature surfaces {Σ σ 1 } σ>σ I 1 . In order to make this idea more precise, we introduce the following construction.
By Theorem 6.1, we know that for every σ > σ I 0 there is a closed, oriented 2-surface Σ σ 0 ֒→ M 3 with constant spacetime mean curvature H(Σ σ 0 ) ≡ 2 /σ with respect to the initial data set I 0 . Furthermore, the proof of this result in [36] shows that there are constants b I 0 ≥ 0 and 1 ≥ η I 0 > 0 such that Σ σ 0 ∈ A(0, b I 0 , η I 0 ) for all σ > σ I 0 . We recall from [36] that b I 0 and η I 0 only depend on ε, C I 0 , and E which can be restated as saying that they only depend on ε, C I 1 , and E by our construction. Set b I 1 := 4b I 0 > b I 0 and η I 0 > η I 1 := η I 0 4 > 0. From Section 5 and by the definition of b I 1 and η I 1 , we know that there are constants C and σ depending only on ε, C I 1 , and E such that the operator L : W 2,2 (Σ) → L 2 (Σ) is invertible whenever Σ ∈ A(0, b I 1 , η I 1 ) is a surface of constant spacetime mean curvature H(Σ) ≡ 2 /σ with respect to the initial data set I 1 for σ ≥ σ, and whenever in addition the estimates of Proposition 5.6 and Corollary 5.7 are available on Σ. Without loss of generality, we may also assume that C and σ are such that the regularity result in Proposition 4.5 as well as a supplementary result stated in Lemma 6.8 (see Section 6.2 below) apply with a = a = 0, b = b I 1 2 , b = b I 1 , η = 2η I 1 , η = η I 1 . We set σ I 1 := max{σ, 4σ I 0 }, and note that by their definition σ I 1 , b I 1 , and η I 1 only depend on ε, C I 1 , and E. Now fix σ * > σ I 1 for the rest of the argument until we start discussing the foliation property when applying Lemma 6.10. Let Y σ * ⊆ [0, 1] be the maximal subset such that there is a C 1 -map with the following properties for every τ ∈ Y σ * : (i) The surface Σ σ * τ := F σ * (τ, S 2 ) has constant spacetime mean curvature H(Σ σ * τ ) ≡ 2 /σ * with respect to the initial data set I τ .
Maximality of Y σ * is understood here as follows: if the above conditions are satisfied for some Y σ * ⊆ [0, 1] and a map F σ * : Note that for τ = 0, Condition (i) is ensured by the assumptions in Theorem 6.1. The same is true for Condition (iii) once one takes into account that A(0, b I 0 , η I 0 ) ⊆ A(0, b I 1 , η I 1 ). However, Condition (ii) is not automatically satisfied for τ = 0 as we do not even know whether the map F σ * exists. The following lemma ensures that Y σ * contains an interval [0, τ 0 ) for some τ 0 > 0. In particular, Condition (ii) is satisfied a posteriori for τ = 0. More generally, this result shows that Y σ * is open around any τ * ∈ Y σ * such that Σ σ * τ * ∈ A(0, b, η) for 0 ≤ b < b I 1 and η I 1 < η ≤ 1.
Recall that the surface Σ σ * τ * is a graph over some round sphere by our assumptions and by Proposition 4.5, recalling again the a priori bounds on σ * , b, and η. As Σ σ * τ was defined as a graph over Σ σ * τ * for every τ ∈ U τ * , composition of these two graphical representations gives us that Σ σ * τ is parametrized over a round sphere.
Proof. Closedness of Y σ * can be addressed by following the arguments given in [36, Lemma 5.6] and [37, Lemma 3.14], as the necessary preliminaries are available in the form of Lemma 6.6 and Lemma 6.7 below. Alternatively, one may rely on a more standard method used in [32, Proof of Proposition 6.1], which we describe below. The Sobolev spaces we use throughout the paper are weighted, however, for a given closed, oriented 2-surface, the weighted Sobolev norms are equivalent to the traditional unweighted ones; we will thus switch to the usual unweighted ones for this proof in order to allow us to use standard results on Sobolev spaces on 2-surfaces.
Let {τ n } ∞ n=1 ⊂ Y σ * be a sequence such that lim n→∞ τ n = τ ∈ [0, 1] and let Σ σ * τn ∈ A(0, b I 1 , η I 1 ) be a surface with constant spacetime mean curvature H(Σ σ * τn ) ≡ 2 /σ * with respect to the initial data set I τn . By Proposition 4.5 we know that there are functions f n : S rn ( z n ) → R such that Σ σ * τn = graph f n where r n and z n are the area radius and the coordinate center of Σ σ * τn . By the first inequality of (15) and by (21), we know that the sequences {r n } ∞ n=1 and { z n } ∞ n=1 are uniformly bounded, so we may assume (up to passing to a subsequence) that lim n→∞ r n = r and lim n→∞ z n = z. Consequently, in view of (17), we may assume that there is a sequence {f n } ∞ n=1 , such thatf n : S r ( z ) → R and Σ σ * τn = graphf n . Again in the view of (17), we may assume that this sequence is uniformly bounded in W 2,∞ (S 2 r ( z )) and hence in C 1,β (S 2 r ( z )) for any 0 < β < 1. Recalling that Σ σ * τn are surfaces of constant spacetime mean curvature, we see that the functionsf n satisfy a linear elliptic PDE of the form with uniformly bounded coefficients a βγ n , b β n , F n ∈ C 0,β (S 2 r ( z )), see Appendix B for details. A standard argument using Schauder estimates (see e.g. [22,Theorem 9.19] and [27, Theorem 10.2.1]) allows us to conclude that the functionsf n ∈ C 2,β (S 2 r ( z )) are uniformly bounded in C 2,β (S 2 r ( z )), and consequently, up to passing to a subsequence, we may assume that {f n } ∞ n=1 converges in C 2,α (S 2 r ( z )) to a limit f ∈ C 2,α (S 2 r ( z )) for some fixed 0 < α < 1. As a consequence of (64) and C 2,α -convergence, we see that Σ σ * τ := graph f has constant spacetime mean curvature H(Σ σ * τ ) ≡ 2 /σ * .
Finally, we confirm that Σ σ * τ = graph f ∈ A(0, b I 1 , η I 1 ) by passing to the limit in the respective inequalities of (15) for Σ σ * τn = graphf n ∈ A(0, b I 1 , η I 1 ). Again, this is possible in the view of the C 2,α -convergence of the graph functions.
6.2. Supplementary lemmas. We will now prove the supplementary lemmas that were used in the proof of Theorem 6.2 above. Lemma 6.6. Let I = (M 3 , g, K, µ, J) be a C 2 1 /2+ε -asymptotically Euclidean initial data set with non-vanishing energy E = 0, with x : M 3 \ B → R 3 \ B R (0) denoting the asymptotic coordinate chart. Assume in addition that K satisfies the potentially stronger decay assumptions |K| ≤ C I | x | −δ−ε for some δ ≥ 3 2 and all x ∈ R 3 \ B R (0). 1] be an open subset of [0, 1] and define I τ as in the proof of Theorem 6.2 for each τ ∈ U. Let a ∈ [0, 1), b ≥ 0, η ∈ (0, 1] be fixed. Then there exist constants σ > 0 and C > 0, depending only on ε, δ, a, b, η, C I , and E such that the following holds for any σ > σ: Assume there exists a C 1 -map F σ : U × S 2 → M 3 such that for every τ ∈ U the surface Σ σ τ := F σ (τ, S 2 ) is in the a priori class A(a, b, η) and has constant spacetime mean curvature H(Σ σ τ ) ≡ 2 /σ with respect to the initial data set I τ . Assume further that F σ is a normal variation map in the sense that there exists a continuous lapse function u = u σ τ : and Lu = O(σ 1−2δ−2ε ).
Proof. In this proof, C > 0 and σ > 0 denote generic constants that may vary from line to line, but depend only on ε, δ, a, b, η, C I , and E. The surfaces Σ σ τ have constant spacetime mean curvature H(Σ σ τ ) ≡ 2 /σ in the initial data set I τ . For clarity, we will write this constant spacetime mean curvature with an explicit reference to the initial data set I τ as H(Σ σ τ , I τ ) ≡ 2 /σ for all τ ∈ U. Hence ∂ τ H(Σ σ τ , I τ ) = 0, which gives us the following linear elliptic PDE on the closed surface Σ σ τ for the a priori only continuous lapse function u = u σ τ : Σ σ τ → R: where the elliptic operator L is (up to a certain factor) the linearization of the spacetime mean curvature operator for the surface Σ σ τ in the initial data set I τ , defined in (28). Then Proposition 5.6 implies that u ∈ W 2,2 (Σ σ τ ), and that such a u is unique. Together with (26) and P = O(σ −δ−ε ), (67) implies that As a consequence, by Corollary 5.7 and (50), we get In order to estimate u t Using the Cauchy-Schwarz Inequality, integration by parts, and (35), together with (21), and (68) we obtain ˆΣ σ τ . Combining the last three estimates with a Triangle Inequality, it follows that Recalling (69), we now get a W 2,2 -estimate for u d , namely From this, as a consequence of (68) and Corollary 5.7, we also have Lemma 6.6 enables us to prove the following result.
Lemma 6.7. Let I = (M 3 , g, K, µ, J) be a C 2 1 /2+ε -asymptotically Euclidean initial data set with non-vanishing energy E = 0, with x denoting the asymptotic coordinate chart. Let ∅ = U ⊆ [0, 1] be an open subset of [0, 1] and define I τ as in the proof of Theorem 6.2 for each τ ∈ U. Let a ∈ [0, 1), b ≥ 0, η ∈ (0, 1] be fixed. Then there exist constants σ > 0 and C > 0, depending only on ε, a, b, η, C I , and E such that the following holds for any σ > σ: Assume there exists a C 1 -map F σ : U × S 2 → M 3 such that for every τ ∈ U the surface Σ σ τ := F σ (τ, S 2 ) is in the a priori class A(a, b, η) and has constant spacetime mean curvature H(Σ σ τ ) ≡ 2 /σ with respect to the initial data set I τ . Assume further that F σ is a normal variation map in the sense explained in Lemma 6.6. Then Proof. In this proof, C > 0 and σ > 0 denote generic constants that may vary from line to line, but depend only on ε, a, b, η, and C I , and E. Let u : Σ σ τ → R denote the lapse function as in Lemma 6.6. Then Lemma 6.6 applied with δ = 3 2 and the Sobolev Embedding Theorem in the form of Lemma A.2 imply that |∂ τ F σ | = |u| ≤ Cσ 1−ε . Then the elementary estimate In order to prove (71), we first recall that the mean curvature of Σ σ τ satisfies (26). The first variation of area formula, the fact that the eigenfunctions used to span L 2 (Σ σ τ ) are L 2 (Σ σ τ )-orthogonal so that in particulaŕ Σ σ τ u t dµ = 0, combined with Lemma 6.6 for δ = 3 2 lead to where we also used the Cauchy-Schwarz Inequality.
Proof. We drop the explicit reference to σ for notational convenience, as σ will not be modified in this proof. Let r τ and z τ denote the area radius and the coordinate center of Σ τ , respectively, and let (slightly abusing notation) x τ denote the restriction of the coordinate vector x to Σ τ , where x denotes the asymptotic coordinate chart. The mean curvature of Σ τ is denoted by H τ . In this proof σ > 0, C > 0 and constants involved in the O-notation may vary from line to line but depend only on ε, a, a, b, b, η, η, C I , and E.
We first show that there exists η τ = η + O(σ −ε ) such that the second inequality describing the fact that Σ τ ∈ A(a, b τ , η τ ) in (15) holds, namely Since Σ τ ∈ A(a, b, η) for all τ ∈ U, by the Mean Value Theorem combined with (71) and (21) we have Similarly, combining (70) with (20) and (21) we conclude that Since Σ τ 0 ∈ A(a, b, η) we have (r τ 0 ) 2+η ≤ | x τ 0 | 5 2 +ε , which in the view of (74) and (75) can be written as (73) follows. Note that by (20) we have We will now apply a similar method and adjust the value of the constant b τ so that the first and the third inequality in (15) hold with η τ as defined in (76). First, we deal with (77). Since Σ τ 0 ∈ A(a, b, η) we have Combining this with (74) and (72) we obtain which in the view of (21) may be further rewritten as Recall that by our definition (76) of η τ we have η = η τ − log rτ (1 − Cσ −ε ). Hence Consequently, we have Next, we address (78). We recall that . Consequently, we may compute using the variation of area formula and (66) with δ = 3 2 that Again, since Σ τ 0 ∈ A(a, b, η) we havê As before, we use the Mean Value Theorem, (80), and (74) to rewrite this aŝ which in the view of (21) may be further rewritten aŝ

Consequently, we havê
Στ Together, the inequalities (79) and (81) imply the existence of the constant b τ = b + O(σ − min{2ε−η,ε} ) such that (77) and (78) hold. This concludes the proof as we can choose a τ = a as the above computations show. Lemma 6.10. Under the assumptions of Theorem 6.2, there exists a constant σ > 0, depending only on ε, a, b, η, C I 1 , and E, and a compact set K ⊂ M 3 such that the map Ψ 1 : (σ, ∞) × S 2 → M 3 \ K defined by (65) is a bijective C 1 -map.
Proof. To prove the claim, we need to show that Ψ 1 is C 1 , injective, and surjective onto a suitably chosen exterior region of M 3 . We already proved in Theorem 6.2 that F σ and thus Ψ 1 is C 1 with respect to the S 2 -component. The differentiability with respect to σ can be proven following the Implicit Function Theorem argument of Lemma 6.4, where the graphical spacetime mean curvature map is to be interpreted as a function of σ ∈ (σ, ∞) instead of as a function of τ ∈ [0, 1]. This is to be viewed in light of the uniqueness results in Section 6.3.
Instead, we argue as in the proof of Lemma 6.6. For v := u − 1, the above computation shows that Lv = O(σ − 5 2 −ε ), which in combination with Corollary 5.7 and (50) gives In addition, for i = 1, 2, 3, by adding a rich zero and using the orthogonality of v d and f i , we have by (51), and Ric(ν, ν)x i dµ = O(σ −ε ) which is fully Riemannian and thus directly carries over to our spacetime context), and integration by parts, we obtain as in Lemma 6.6 that Combining these estimates, we get, again grouping terms as in Lemma 6.6, that v t is strictly positive for all σ > σ. This shows that Ψ 1 is indeed injective.
6.3. Uniqueness of the STCMC surfaces. We close this section by proving that the constant spacetime mean curvature surfaces are unique in the a priori class of asymptotically centered surfaces A(a, b, η). As Brendle and Eichmair [8] constructed examples of asymptotically Euclidean Riemannian manifolds with "off-center" (i.e. not included in the a priori class) CMC-surfaces provided Scal ≥ 0 is violated, we will restrict our uniqueness statements to the a priori class -at least when not assuming the dominant energy condition µ ≥ |J| g . To the best knowledge of the authors, it is not known whether such examples can also be constructed if the dominant energy condition or its Riemannian analog Scal ≥ 0 are satisfied.
Proof. We rely on the same type of argument as in the proof of Theorem 6.2. Fix a surface Σ σ 1 as in the assumptions, with σ > σ I 1 . We now drop the explicit reference to σ for notational convenience, as σ will not be modified in this proof. Let Z ⊆ [0, 1] be a maximal subset such that there is a C 1 -map Φ : Z × S 2 → M 3 with the following properties for all τ ∈ Z: (i) Φ(1, S 2 ) = Σ 1 , (ii) Σ τ := Φ(τ, S 2 ) has constant spacetime mean curvature H(Σ τ ) ≡ 2 /σ with respect to the initial data set I τ , where I τ is defined as in the proof of Theorem 6.2, (iii) ∂ τ Φ is orthogonal to Σ τ . Maximality is understood as in the proof of Theorem 6.2. Arguing as in the proof of Theorem 6.2, we conclude that Z = [0, 1] and that there are constants a I 1 ∈ [0, 1), b I 1 ≥ 0 and η I 1 ∈ (0, 1] such that Σ τ ∈ A(a I 1 , b I 1 , η I 1 ) for every τ ∈ [0, 1], if σ I 1 suitably large, depending only on ε, a, b, η, C I 1 , and E. In particular, we see that Σ 0 ∈ A(a I 1 , b I 1 , η I 1 ) is a surface with constant mean curvature H(Σ 0 ) ≡ 2 /σ with respect to I 0 4 . By Theorem 6.11, such a surface is unique in this class. By the method of continuity approach and the local uniqueness in the Implicit Function Theorem, the map Φ is uniquely determined also by its start value Φ(0, S 2 ) = Σ 0 . It follows directly that Σ 1 = Φ(1, S 2 ) is uniquely determined by its spacetime mean curvature in I 1 .

The coordinate center of the STCMC-foliation
Let {Σ σ } σ>σ be a foliation of a C 2 1 /2+ε -asymptotically Euclidean initial data set I = (M 3 , g, K, µ, J) for which Σ σ grows to the round sphere at infinity as σ → ∞. Then we may define the coordinate center of this foliation as the limit lim σ→∞ z σ , where z σ = z (Σ σ ) is the coordinate center of Σ σ as defined in Definition 4.1, provided that this limit exists (in R 3 ). We would like to draw the attention of the reader to the fact that, while the foliations considered here do not depend on the choice of asymptotic coordinates x, the coordinate centers z σ and as a consequence also their limit, do depend on x. We will discuss the subtle consequences of this within this section, too.
Let us first consider the case of the CMC-foliation: In this case, {Σ σ } σ>σ I 0 is the unique foliation of a given C 2 1 /2+ε -asymptotically Euclidean manifold (M 3 , g) or initial data set I 0 = (M 3 , g, K ≡ 0, Scal, J ≡ 0) by surfaces of constant mean curvature constructed in [26,32,36] and discussed in Section 6 above. Under the additional assumption that (M 3 , g) satisfies the Riemannian C 2 1+ε -Regge-Teitelboim condition (see Definition 2.5), the coordinate center of this foliation is well-defined if and only if the Beig-Ó Murchadha center of mass C BÓM given by (6) is well-defined as was shown by [25,36]. They also show that in this case, one has Now suppose that {Σ σ } σ>σ I 1 is the unique foliation of a C 2 1 /2+ε -asymptotically Euclidean initial data set I = (M 3 , g, K, µ, J) by surfaces of constant spacetime mean curvature as constructed in Theorem 6.2. One cannot in general expect that (85) also holds for the STCMC-foliation because the foliation is defined in terms of K, whereas the Beig-Ó Murchadha center of mass is a purely Riemannian quantity, i.e. independent of K. One can only expect that (85) will hold if K falls off very fast, in particular faster than the optimal decay assumed in this paper. In this section we will confirm that this is indeed the case. 7.1. A variational formula for STCMC-surfaces. The following proposition generalizes [36, Proposition 6.5] to the spacetime case.
Then there are constants C > 0 and σ > 0 depending only on ε, b, η, C I such that provided that σ > σ.
Proof. Since the coordinate center of a surface is invariant under tangential diffeomorphisms (along Σ), we may without loss of generality assume that F is a normal variation of Σ, such that in particular (∂ s F )| s=0 = uν holds on Σ. By definition, Using the variation of area formula and adding rich zeroes in the third and the forth lines, we compute, dropping the explicit reference to δ in the denominator, Now subtract and add the component (ν δ ) i of the δ-outward unit normal ν δ to Σ in the bracket of the second term and recall the fact that Σ can be written as a graph over S 2 r ( z ) with graph function f satisfying f W 2,∞ = O(r 1 2 −ε ) by Proposition 4.5. Then, by comparability of | x | and r as established in Proposition 4.5, we find, recalling a = 0 in our case, Applying the Cauchy-Schwarz Inequality to the above identity for (∂ s z i s ) | s=0 and using (87), Lemma A. 1, and (26) to estimate the individual terms, respectively, we obtain (86). 7.2. STCMC-center of mass. In Section 6, we constructed the unique STCMCfoliation of a C 2 1 /2+ε -asymptotically Euclidean initial data set I = (M 3 , g, K, µ, J), i.e. the unique foliation by surfaces {Σ σ 1 } σ>σ 1 of constant spacetime mean curvature H(Σ σ 1 ) ≡ 2 /σ, provided it has non-vanishing energy E = 0. This was achieved by deforming the constant mean curvature foliation {Σ σ 0 } σ>σ 0 of the C 2 1 /2+ε -asymptotically Euclidean manifold (M 3 , g) from Theorem 6.1 along the curve of initial data sets {I τ } τ ∈[0,1] , where I τ = (M 3 , g, τ K, µ τ , τ J) is as described in Section 3.1. We will now apply Proposition 7.1 to find how the coordinate center of a leaf changes under this particular deformation. As a result, we prove Lemma 7.2 relating the respective coordinate centers z σ 0 and z σ 1 of the surfaces Σ σ 0 and Σ σ 1 . Note that for the proof of this result, it is necessary to assume that the fall-off rate of K is K = O(| x | −2 ), which is faster than we originally assumed in Definition 2.1 and in particular faster than one needs for existence and uniqueness of the foliation. See also Conjecture 7.9 below.
Lemma 7.2. Let I = (M 3 , g, K, µ, J) be an STCMC-foliated C 2 1 /2+ε -asymptotically Euclidean initial data set with non-vanishing energy E = 0. Assume in addition that , with x the asymptotic chart. Then there exist constants C > 0, σ > 0 depending only on ε, C I , and E such that for i = 1, 2, 3 we have for all σ > σ where (z σ 0 ) i and (z σ 1 ) i denote the components of the coordinate centers z σ 0 and z σ 1 of the STCMC-surfaces Σ σ 0 and Σ σ 1 with respect to I 0 and I 1 as defined above, respectively. Remark 7.3. Instead of assuming K = O(| x | −2 ), one could also assume Regge-Teitelboim conditions on K by carefully tracking all even and odd parts or identify other sufficient decay conditions, as necessary for what one wants to do. For our purposes, it is enough to assume K = O(| x | −2 ). dµ δ converges. This in particular shows that K can in a sense "compensate" for the diverging coordinate center of the CMCfoliation. See Section 9 for more details on this.
Proof. First, note that the constants C Iτ are uniformly bounded by the constant C I . Second, pick constants b ≥ 0 and η ∈ (0, 1] that will remain fixed in this argument and always use the class A(0, b, η) in what follows. Also, C > 0 and σ > 0 denote generic constants that may vary from line to line, but depend only on ε, C I , and |E| (as well as on our global choice of b and η). From now on we assume that σ > σ is fixed. For our choice of σ and for τ ∈ [0, 1], we let z i τ = (z σ τ ) i , i = 1, 2, 3, denote the components of the coordinate center z (Σ σ τ ) of the unique surface Σ σ τ of constant spacetime mean curvature H(Σ σ τ ) ≡ 2 /σ in the initial data set I τ (see Section 6 for details). Since the index σ is assumed to be fixed, it will be suppressed in the remainder of this proof.
According to Proposition 7.1, the variation of the coordinate center with respect to τ is given by the formula where u τ is the respective lapse function for an arbitrary τ ∈ [0, 1] and ν τ is the outward pointing unit normal to Σ τ ֒→ (M 3 , g). In order to pass from (90) to (89), we will apply Lemma 6.6 with δ = 2 − ε ≥ 3 2 . By this, we have that Next, by (33) and Lemma A.1 we have where f i τ denotes the i-th eigenfunction of the operator −∆ Στ , see Section 5.2. Then we may rewrite (90) by a Cauchy-Schwarz Inequality and Lemma 6.6 with δ = 2 − ε as 2 +ε . At the same time, Proposition 5.6 implies that where m H is the Hawking mass of Σ τ with respect to I τ . Recall that |m H −E| ≤ Cσ −ε by Proposition 5.6 so that m H = 0 follows from E = 0. Thus, Note that a computation in the proof of Lemma 6.6 shows that Consequently, in view of (67) and (91), and the Cauchy-Schwarz Inequality, (92) is equivalent to Definition 7.7. We suggest to call the expression C mBÓM := C BÓM + Z the modified Beig-Ó Murchadha center of mass.
In Section 9, we will give an example that shows that the contribution of the correction term Z is indeed relevant and fixes a problem of the CMC-center of mass uncovered in [10].
Remark 7.8. It is not obvious which decay conditions on I (e.g. versions of Regge-Teitelboim, faster decay assumptions on K, etc.) are sufficient to ensure convergence of the correction term Z without forcing it to vanish entirely. This will be studied in detail in our forthcoming work. More importantly, sufficient conditions for convergence of C STCMC that do not force vanishing of Z in accordance with the example studied in Section 9 will also be studied in our forthcoming work.
We conjecture the following sufficient conditions, in line with Bartnik's [3] and Chruściel's [16] corresponding results for convergence of ADM-energy and ADMlinear momentum. Conjecture 7.9. We conjecture that the coordinate expression we derived will converge for asymptotic coordinates x if µx i ∈ L 1 (M 3 ).

Time evolution and Poincaré covariance
of the STCMC-center of mass 8.1. Evolution. In this section, we will study the evolution of the coordinate center of the unique foliation by surfaces of constant spacetime mean curvature under the Einstein evolution equations. We will show that the STCMC-center of mass has the same evolution properties as a point particle in special relativity, evolving according to the formula d dt Note that the analogous formula is valid for the CMC-center of mass and also for Chen-Wang-Yau's center of mass, although under stronger decay assumptions, see [34] and [13], respectively.
Assume that I 0 = ({0} × M 3 , g, K, µ, J) ֒→ (R × M 3 , g) is a C 2 1 /2+ε -asymptotically Euclidean initial data set with respect to the coordinate chart x and with E = 0, and suppose additionally that K = O 1 (| x | −2 ) with constant C I as | x | → ∞. Now consider the C 1 -parametrized family of C 2 1 /2+ε -asymptotically Euclidean initial data sets with respect to x which starts from I(0) = I 0 , and which exists for all t ∈ (−t * , t * ) for some t * > 0. Assume furthermore that the constants C I(t) are uniformly bounded on (−t * , t * ), without loss of generality such that C I(t) ≤ C I 0 .
Assume the foliation I(t) has initial lapse N = 1 + O 2 (| x | − 1 2 −ε ) as | x | → ∞ with decay measuring constant denoted by C N and initial shift X = 0, and suppose furthermore that the initial stress tensor S of There is a constant t > 0, depending only on ε, C I 0 , C N , and E(0) such that the following holds: If the initial data set I 0 has well-defined STCMC-center of mass C STCMC (0) then the STCMC-center of mass C STCMC (t) of I(t) is also well-defined for |t| < t. Furthermore, the initial velocity at t = 0 is given by Moreover, we have that d dt t=0 E = 0 and d dt t=0 P = 0. Remark 8.2. In fact, one gets more information about the evolution of the STCMCcenter of mass from the proof of Theorem 8.1: Not only the coordinate expression C STCMC evolves according to (97), but also the individual leaves of the STCMCfoliation evolve in a way more and more close to a translation in direction P /E according to formula (106).
Remark 8.3. We expect that the evolution of the leaves of the STCMC-foliation as well as the evolution of C STCMC can actually be understood when replacing the condition K = O 1 (| x | −2 ) by more natural conditions related to integrability criteria on the constraints when integrated against x. We will investigate this in our forthcoming work, see also Remark 7.8 and Conjecture 7.9.
Remark 8.4. It is straightforward to prove a version of this theorem allowing for nonvanishing shift. As this is not of primary interest here and can also be fixed by a suitable gauge, we will not go in this direction.
Proof. Throughout this proof, dotted quantities like for exampleĖ will denote time derivatives at t = 0, e.g.Ė = d dt E t=0 . Moreover, σ > 0, t > 0, and C > 0 denote generic constants that may vary from line to line, but depend only on ε, C I 0 , and E(0) as well as on C N (0), the constant in the O-term of N.
The Einstein evolution equations with zero shift are given at t = 0 bẏ where Ric is the Ricci tensor of the spacetime (R × M 3 , g), which is completely determined by the stress-energy tensor T through the Einsteins equations Ric − 1 2 Scal g = T. Here, we used that Scal = − tr g T = O(| x | − 5 2 −ε ) and thus Ric ij = O(| x | − 5 2 −ε ). In particular, we see from the ADM-formulas (3) and (4) respectively that the energy and linear momentum satisfyĖ = 0 and˙ P = 0, with respect to this variation. Also, as E(0) = 0 and E(t) is continuous, the initial data sets I(t) have a unique foliation by surfaces of constant spacetime mean curvature near infinity for |t| < t and mean curvature radii σ > σ. This foliation depends in a C 1 -fashion on t for |t| < t which can be seen as follows: Perform a method of continuity procedure around t = 0 as in the proof of Theorems 6.2, 6.12 or in the proof of Lemma 6.10. Note that, as the initial shift X was chosen to vanish, X = 0, we in fact know that the STCMC-surface variation is normal at t = 0. This gives you an STCMC-foliation of I(t) for each |t| < t which depends on t in a C 1 -fashion. By Theorem 6.12, this family of foliations must coincide with the one studied here and must thus depend on t in a C 1 -fashion. Now fix σ > σ and let Σ t denote the unique leaf of the STCMC-foliation with constant spacetime mean curvature H(Σ t , I(t)) ≡ 2 /σ in the initial data set I(t), where |t| < t. For this, we use the product rule to see that The first term on the right shows how the spacetime mean curvature changes if Σ t is considered to be a varying surface in the initial data set I(0). As the initial shift vanishes and we thus have an initially normal variation, this term exactly gives our well-known linearization L H u, where the operator L H is as defined in Lemma 5.1 and u is the initial lapse function of the normal variation. The second term on the right shows how the spacetime mean curvature changes if Σ is considered to be a fixed surface in the varying initial data set I(t). More precisely, this term is where H = H(0) and P = P (0). In order to computeḢ = d dt t=0 H(Σ 0 , I(t)), we introduce geodesic normal coordinates in a neighborhood U ⊂ M 3 of Σ 0 , with y n such that ∂ n is the outer unit normal to the level set {y n = const.}, in particular ∂ n = ν on Σ 0 , and y α , α = 1, 2, are some coordinates on Σ 0 transported to U along the flow generated by ∂ n . Note that in this case g nn = 1, g nα = 0, and in U, for all α, β, γ = 1, 2. We will now drop the index on Σ 0 and just write Σ instead for notational convenience. We use the standard formula for the variation of the second fundamental form when the ambient metric is changing (see e.g. Section 3 in [29] 5 ) and compute, using first (98) and (99), second the decay properties of N and the decay estimate for the second fundamental form A = H 2 g Σ +Å, with g Σ the metric induced on Σ by I(t), namely |A| ≤ C σ from Proposition 4.5, third adding some rich zeros, fourth because J = O(σ −3−ε ) by assumption, fifth by (101), (102), (103), and finally (26), and Proposition 4.5 to =ġ αβ A αβ + g αβȦ , where J is the momentum density defined on page 3. Further, let η denote the timelike future unit normal vector field to M 3 ֒→ (R × M 3 , g). Then it is straightforward to check that, by (98), (99), the decay assumptions on the initial data set and on N, as well as the definition of µ and S from page 3 Summing up and multiplying by 1 − ( P H ) 2 , it follows from (100) that Lu = −Ḣ + P HṖ = div Σ K(·, ν) where the operator L is given by (28). This uniquely defines u ∈ W 2,2 (Σ) by the invertibility of L, see Proposition 5.6 as the right hand side is bounded and thus in L 2 (Σ). In order to compute the initial velocity of I(0), we first need to compute the initial velocity˙ z of the Euclidean coordinate center We remind the reader that we chose coordinates x which do not depend on t. Relying on Proposition 7.1, we will now compute the variation of the coordinate center,˙ z, starting from the variation formula The idea is to argue as in the proof of Lemma 7.2 and pass from u to Lu, and from ν i to f i in (105). For this we note that since I is C 2 1 /2+ε -asymptotically Euclidean, Passing to the limit when σ → ∞ we obtain the result.

8.2.
Poincaré-covariance and accordance with Special Relativity. As we have seen before, whether or not a given 2-surface is STCMC is in fact independent of a choice of slice (as well as of a choice of coordinates). In this sense, STCMC-surfaces are covariant in the sense of General Relativity. The role of the initial data set then is to select a unique family of STCMC-surfaces near the asymptotic end of the spacetime, forming its abstract STCMC-center of mass. In this sense, STCMC-foliations and the associated (abstract) center of mass are Poincaré-covariant.
We will now discuss the transformation behavior of the STCMC-coordinate center under the asymptotic Poincaré group of the ambient spacetime -assuming vanishing angular momentum in the boost case. Dealing with angular momentum and treating the boost case more adequately will be left for our future work. Let I = (M 3 , g, K, µ, J) be an initial data set which is C 2 1 /2+ε -asymptotically Euclidean with respect to asymptotic coordinates x and has E = 0.
Euclidean motions. Consider the coordinates y := O x + T , with O an orthogonal rotation matrix and T ∈ R 3 a translation vector. In other words, y arises from x through a Euclidean motion. Then, for each leaf Σ σ of the STCMC-foliation constructed in Theorem 6.2, we find that the Euclidean center of Σ σ with respect to the y-coordinates is given by Thus, the STCMC-coordinate center C y STCMC = lim σ→∞ 1 |Σ σ | δˆΣ σ y dµ δ with respect to the coordinates y converges if and only the STCMC-coordinate center x dµ δ converges with respect to the coordinates x converges and if they converge, we find C y STCMC = O C x STCMC + T as one would expect from Euclidean Geometry, Newtonian Gravity, and from the description of the spacetime position of a point particle in Special Relativity.
Time translations. The transformation behavior of C STCMC under asymptotic time translation corresponds to its evolution behavior under the Einstein equations. In other words, Theorem 8.1 tells us under the additional assumption which corresponds precisely to the instantaneous law of motion of a point particle in Special Relativity.
Boosts. The last constituent of the asymptotic Poincaré group of the spacetime are of course the asymptotic boosts. In a given (asymptotic region of a) spacetime (R × M 3 , g = −N 2 (dx 0 + X i dx i )(dx 0 + X j dx j ) + h ij dx i dx j ) with asymptotic coordinates x α = (x 0 , x i ) and suitably decaying lapse N, shift X, and tensor h, a boosted initial data set I = (M 3 , g, K, µ, J) ֒→ (R ×M 3 , g) is any spacelike hypersurface arising as the set {y 0 = 0} with respect to a boosted coordinate system y α := Λ α β x β , y α = (y 0 , y ), meaning that the matrix Λ is a boost. If the lapse N, the shift X, and the tensor h decay suitably fast in space and time coordinate directions, the boosted initial data set {y 0 = 0} = I is in fact C 1 1 /2+ε -asymptotically Euclidean with respect to y. It is thus reasonable to ask how the STCMC-coordinate centers of the initial data sets {x 0 = 0} and {y 0 = 0} are related (if they converge). The corresponding question was addressed by Szabados [41] for the BÓM-center of mass although from a slightly different perspective. Of course, we expect that the STCMCcoordinate center boosts like the position of a point particle in Special Relativity, namely at least in the absence of angular momentum. This can easily be verified for example for a boosted slice (over the canonical slice) in the Schwarzschild spacetime where in fact the centers both coincide with the center of symmetry 0. Similarly, if one first spatially translates the coordinates on the Schwarzschild spacetime and then considers a boosted slice, the transformation law will be as in (107). In both of these examples, the deviation Z introduced in Theorem 7.5 in fact vanishes, so that the transformation law (107) already holds for the CMC-BÓM-center of mass. In view of Section 9 below, it is possible to construct examples of boosted slices in the Schwarzschild spacetime by boosting the example discussed below. One can then see that (107) in fact also applies in this case, but only for the STCMC-center of mass, and not for the CMC-BÓM-center of mass. However, the computation is so tedious that we prefer not to show it here as it is not particularly enlightening. A proof of a generalized version of (107) incorporating the angular momentum will be given elsewhere, see also Remarks 7.8,8.3. 9. A concrete graphical example in the Schwarzschild spacetime As briefly sketched in Sections 2, 3 and analyzed in more detail in Section 7, determining the coordinate center of an asymptotic foliation is tricky and depends on choosing suitable coordinates (see also Conjecture 7.9). In [10, Section 6], this was illustrated by explicitly computing the coordinate center of the CMC-foliation of an asymptotically Euclidean "graphical" time-slice in the Schwarzschild spacetime of mass m = 0. This example, to be described in more detail below, satisfies all assumptions in [26], in particular those of Theorem 4.2, but yet its CMC-coordinate center does not converge. Equivalently, its BÓM-center also does not converge. After a brief introduction to the graphical example discussed in [10], we will compute that the STCMC-coordinate center (96) does in fact converge in this example and moreover converges to the origin 0, i.e. to the center of symmetry of the spherically symmetric spacetime as one would expect.
We consider the Schwarzschild spacetime (R × M 3 , g) of mass m = 0 in Schwarzschild coordinates, meaning that M 3 = (max{0, 2m}, ∞) × S 2 ∋ (r, η), g = −N 2 dt 2 + g, g = N −2 dr 2 + r 2 dΩ 2 , where dΩ 2 denotes the canonical metric on S 2 . We will freely switch between polar coordinates (r, η) and the naturally corresponding Cartesian coordinates x defined on M 3 . A graphical time-slice in the (automatically vacuum) Schwarzschild spacetime is an initial data set (M 3 T , g T , K T , µ T ≡ 0, J T ≡ 0) arising as the graph of a smooth function T : M 3 → R "over" the canonical time-slice {t = 0} (in time-direction), meaning that while g T is the Riemannian metric induced on M T ֒→ (R × M 3 , g) and K T is the second fundamental form induced by this embedding with respect to the future pointing unit normal.
Computing the CMC-coordinate center of mass (via the BÓM-center of mass). Clearly, the center of mass of the canonical time-slice {t = 0} of the Schwarzschild spacetime is the coordinate origin, C CMC = C BÓM = 0. We will now compute this vector for graphical time-slices with the asymptotic decay conditions on T chosen such that (M 3 T , g T , K T , µ T ≡ 0, J T ≡ 0) is C 2 1 -asymptotically Euclidean with respect to the coordinates x. To most easily comply with the asymptotic decay conditions specified in Section 2, we will assume that T = O k (r 0 ) as r → ∞, with k ≥ 3. Now let y := x| M T denote the induced coordinates on M T . As computed in [10, Section 6], the metric g T and second fundamental form K T 6 are given by in the coordinates y. A straightforward computation shows that the graphical initial data set (M 3 T , g T , K T , µ T ≡ 0, J T ≡ 0) is indeed C 2 1 -asymptotically Euclidean and in fact has E = m = 0.
When evaluating the BÓM-center of mass surface integral on a finite coordinate sphere with respect to the y-coordinates in M T , using s := | y | and η := y s , we find T ,i η i T ,l − |dT | 2 δ η l dµ δ + O(s −1 ).
We point out that this choice of T ensures that (M 3 , g T , µ T ≡ 0) satisfies the Riemannian C 2 1 2 +ε -Regge-Teitelboim conditions so that [34,Cor. 4.2] or [36, Theorem 6.3] apply and ensure that C CMC = C BÓM or that both diverge. One directly computes from the above expression for C BÓM S 2 s ( 0 ) that which diverges as s → ∞. Hence, the BÓM-and thus also the CMC-coordinate center diverge in this example.
Computing the STCMC-coordinate center of mass (via Formula (96)). In order to check whether the STCMC-coordinate center of the C 2 1 -asymptotically Euclidean initial data set (M 3 T , g T , K T , µ T ≡ 0, J T ≡ 0) converges, one needs to compute the STCMC-leaves Σ σ and the coordinate averages z (Σ σ ) and check whether they converge as σ → 0. However, the proof of Theorem 7.5 asserts that C STCMC converges if and only if the coordinate expression given in (96) converges, or in other words if and only if C STCMC S 2 s ( 0 ) = C BÓM S 2 s ( 0 ) + Z S 2 s ( 0 ) converges as s → ∞, where we recall that, using E = m and (π T ) kl = −(K T ) ij + tr g T K T (g T ) ij , we know that Our (diverging) spacetime correction term Z thus precisely compensates for the divergence occurring in C CMC = C BÓM . Hence the STCMC-coordinate center of the considered graphical slice converges to 0 as desired.
Further, if H L 2 (Σ) is a priori bounded, then the respective trace-free parts of the second fundamental forms satisfy where C also depends on the bound on H L 2 (Σ) .
The following result is a Sobolev Embedding Theorem which holds for a very general class of 2-surfaces. In Section 4 this result is applied to Σ being a large coordinate sphere S 2 r ( z) ֒→ M 3 \ B in the asymptotic end of an asymptotically Euclidean initial data set. Note that in this case one can without loss of generality replace the area radius |Σ| /4π in the formulation of Lemma A.2 by the coordinate sphere's radius r as these two radii are uniformly equivalent. In the subsequent sections, this result is applied to Σ ֒→ M 3 \ B being an asymptotically centered closed 2-surface with constant spacetime mean curvature. Here we are using the fact that (108) is available for large asymptotically centered surfaces in M 3 \B in the form of [26,Proposition 5.4], which applies to asymptotically Euclidean initial data sets with general asymptotics as described in Section 2.
Lemma A.2. Let (Σ, g Σ ) be a closed, oriented 2-surface with area radius r = |Σ| /4π. If there is a constant C S such that for any Lipschitz continuous function f on Σ, the so-called first Sobolev Inequality holds, then we also have the Sobolev Inequality 8 for any f ∈ W 2,2 (Σ).