The C0-inextendibility of some spatially flat FLRW spacetimes Authors: Eric Ling Published: 2026 ABSTRACT AbstractUtilizing some of Sbierski’s recent C0-inextendibility techniques [18], we prove the C0-inextendibility of a class of spatially flat FLRW spacetimes without particle horizons. Introduction In [18], Sbierski proved the C0-inextendibility of a class of FLRW spacetimes without particle horizons. His main results, Theorems 1.5 and 1.6, apply to FLRW spacetimes which are spatially spherical and spatially hyperbolic, respectively. In this paper, we complement his results by establishing the C0-inextendibility of a class of spatially flat FLRW spacetimes, also without particle horizons. As in [18], we work in the class of simply connected models, where I is an open interval and the warping function a:I→(0,∞) is called the scale factor. Our definitions, notations, and conventions follow [18] which we briefly review. A Ck Lorentzian manifold (M, g) is one where the metric g is of regularity class Ck. Regardless of the regularity class of the metric, the manifold M is always assumed be smooth. A Ck spacetime is a time-oriented Ck Lorentzian manifold (M, g) which is connected. A C0 extension of a smooth spacetime (M, g) is an isometric embedding where (M~,g~) is a connected C0 Lorentzian manifold with the same dimension as M and ι(M)⊊M~ is a proper open subset. Given a C0 extension, the past boundary ∂-ι(M) is the subset of the topological boundary ∂ι(M) given by the set of points p∈M~\ι(M) such that there is a C1 timelike curve γ:[a,b]→M~ such that γ(a)=p and γ((a,b])⊂ι(M) and ι-1∘γ|(a,b] is future-directed in M. The future boundary ∂+ι(M) is defined time-dually. If no C0 extension of a smooth spacetime (M, g) exists, then we say (M, g) is C0-inextendible. It’s known that if a C0 extension of a smooth spacetime (M, g) does exist, then either the future or past boundary is nonempty. We say (M, g) is past C0-inextendible if the past boundary of any C0 extension is empty. Our main theorem is the following. Theorem 1.1 Let (M, g) be a spatially flat simply connected FLRW spacetime with dimension n+1≥3. Suppose the scale factor a(t) is smooth and satisfies the following. (i)a(t) is defined for all t∈R and a(t)→0 as t→-∞, a(t) is defined for all t∈R and a(t)→0 as t→-∞, (ii)∫t11a(s)ds→∞ as t→-∞, ∫t11a(s)ds→∞ as t→-∞, (iii)a(t)∫t11a(s)ds→∞ as t→-∞. a(t)∫t11a(s)ds→∞ as t→-∞. Then (M, g) is past C0-inextendible. Remarks If the scale factor satisfies (i), then t=-∞ is sometimes referred to as the “big bang." If the scale factor satisfies (i), then t=-∞ is sometimes referred to as the “big bang." The theorem generalizes to the case that a(t) is only defined on an interval I=(-∞,tf) for some tf∈R; the proof is the same. These spacetimes are called “past eternal" since t→-∞. The theorem generalizes to the case that a(t) is only defined on an interval I=(-∞,tf) for some tf∈R; the proof is the same. These spacetimes are called “past eternal" since t→-∞. If the scale factor satisfies (ii), then (M, g) is said to “not possess particle horizons." If the scale factor satisfies (ii), then (M, g) is said to “not possess particle horizons." Condition (ii) in Theorem 1.1 is redundant since it’s implied by (i) and (iii). Condition (ii) in Theorem 1.1 is redundant since it’s implied by (i) and (iii). If the scale factor satisfies (i) and (ii), but, instead of (iii), it satisfies a(t)∫t11a(s)ds→c∈(0,∞) as t→-∞, then (M, g) is known to be past C0-extendible [5]. For example, take a(t)=et. If the scale factor satisfies (i) and (ii), but, instead of (iii), it satisfies a(t)∫t11a(s)ds→c∈(0,∞) as t→-∞, then (M, g) is known to be past C0-extendible [5]. For example, take a(t)=et. If (M, g) in Theorem 1.1 is also future complete, then it’s known that it’s also future C0-inextendible [4]; therefore (M, g) is C0-inextendible in this case. If (M, g) in Theorem 1.1 is also future complete, then it’s known that it’s also future C0-inextendible [4]; therefore (M, g) is C0-inextendible in this case. Example Consider a(t)=e--t for t≤-1, arbitrarily extended to a smooth positive function for t>-1. Then a(t) satisfies (i) - (iii) of Theorem 1.1. Therefore the corresponding spatially flat simply connected FLRW spacetime (M, g) is past C0-inextendible. Note that, at first glance, it might seem like (M, g) is past timelike complete since t can limit to -∞. If that were the case, then the main result in [4] would already suffice to prove past C0-inextendibility of (M, g), rendering Theorem 1.1 redundant. However, (M, g) is not past timelike complete. While the vertical timelike geodesic t↦(t,u01,⋯,u0n) is past complete for any fixed point (u01,⋯,u0n) in Rn, any other timelike geodesic—specifically those with an initial tangent vector not parallel to -∂t—will be past incomplete. The investigation of low-regularity inextendibility results is motivated in part by a desire to understand and classify the strength of gravitational singularities. See the nice discussion in [16]. This investigation began with Sbierski’s seminal work [15] on the C0-inextendibility of the positive-mass Schwarzschild spacetime1. Since then other results have been found [2–4, 6–10, 12–14, 16–19]. Nevertheless, the understanding of low-regularity inextendibility is far from complete. The proof Characterization of TIFs In this section, we characterize the TIFs of the spatially flat FLRW spacetimes appearing in Theorem 1.1. Along the way, we will prove the C0-extendibility result mentioned in remark (5) after Theorem 1.1. Consider the spatially flat FLRW spacetime (M^,g^) written out in comoving coordinates: Introducing spherical normal coordinates (t^,r,ω) centered at the origin ui=0, we can rewrite the metric as where dΩ2 is the round metric on the unit sphere Sn-1. We introduce new spherical coordinates (T,R,ω) defined via The spherical coordinates (T,R,ω) produce the following diffeomorphism where The metric in (T,R,ω)-coordinates is Therefore (M^,g^) is conformal to the open subset D of the Einstein static universe, save for the t^-line at r=0 which is in one-to-one correspondence with the T-line at R=0 along -π2-1+dRdT2+sin2(R(T))|ω′(T)|n-12. Therefore |ω′(T)|Sn-1<1sin(R(T)). Recall that R(T)→R∗=T∗+π2 as T→T∗; hence R∗∈(0,π) and so sin(R∗)>0. By continuity, there are numbers ε,δ>0 such that sin(R∗)-ε>0 and 1/sin(R(T))<1/(sin(R∗)-ε) for all T∈(T∗,T∗+δ). Therefore 1/sin(R(T)) is integrable in a right-sided neighborhood of T∗. (i)This is clear since T is a time function.(ii)Define u=T+R. Let ∇ denote the gradient with respect to the metric for the Einstein static universe. Then ∇u=-∂T+∂R is past-directed null. Hence u∘γ(t) is a strictly increasing function. Therefore u∘γ(t) converges as t→-∞ and hence so does R∘γ(t). Let u∗ and R∗ denote their limits. Since γ is past-inextendible within M1, it follows that R∗=T∗+π2 and hence u∗=2T∗+π2. If T∗=-π2, then R∗=0 and the conclusion follows from the geometry of the Einstein static universe.(iii)Assume T∗∈(-π2,π2). Reparameterize γ by T and represent γ(T) by the curve T↦(T,R(T),ω(T)). Since γ is timelike, we have 0>-1+dRdT2+sin2(R(T))|ω′(T)|n-12. Therefore |ω′(T)|Sn-1<1sin(R(T)). Recall that R(T)→R∗=T∗+π2 as T→T∗; hence R∗∈(0,π) and so sin(R∗)>0. By continuity, there are numbers ε,δ>0 such that sin(R∗)-ε>0 and 1/sin(R(T))<1/(sin(R∗)-ε) for all T∈(T∗,T∗+δ). Therefore 1/sin(R(T)) is integrable in a right-sided neighborhood of T∗. (i)This is clear since T is a time function. This is clear since T is a time function. (ii)Define u=T+R. Let ∇ denote the gradient with respect to the metric for the Einstein static universe. Then ∇u=-∂T+∂R is past-directed null. Hence u∘γ(t) is a strictly increasing function. Therefore u∘γ(t) converges as t→-∞ and hence so does R∘γ(t). Let u∗ and R∗ denote their limits. Since γ is past-inextendible within M1, it follows that R∗=T∗+π2 and hence u∗=2T∗+π2. If T∗=-π2, then R∗=0 and the conclusion follows from the geometry of the Einstein static universe. Define u=T+R. Let ∇ denote the gradient with respect to the metric for the Einstein static universe. Then ∇u=-∂T+∂R is past-directed null. Hence u∘γ(t) is a strictly increasing function. Therefore u∘γ(t) converges as t→-∞ and hence so does R∘γ(t). Let u∗ and R∗ denote their limits. Since γ is past-inextendible within M1, it follows that R∗=T∗+π2 and hence u∗=2T∗+π2. If T∗=-π2, then R∗=0 and the conclusion follows from the geometry of the Einstein static universe. (iii)Assume T∗∈(-π2,π2). Reparameterize γ by T and represent γ(T) by the curve T↦(T,R(T),ω(T)). Since γ is timelike, we have 0>-1+dRdT2+sin2(R(T))|ω′(T)|n-12. Therefore |ω′(T)|Sn-1<1sin(R(T)). Recall that R(T)→R∗=T∗+π2 as T→T∗; hence R∗∈(0,π) and so sin(R∗)>0. By continuity, there are numbers ε,δ>0 such that sin(R∗)-ε>0 and 1/sin(R(T))<1/(sin(R∗)-ε) for all T∈(T∗,T∗+δ). Therefore 1/sin(R(T)) is integrable in a right-sided neighborhood of T∗. Assume T∗∈(-π2,π2). Reparameterize γ by T and represent γ(T) by the curve T↦(T,R(T),ω(T)). Since γ is timelike, we have 0>-1+dRdT2+sin2(R(T))|ω′(T)|n-12. Therefore |ω′(T)|Sn-1<1sin(R(T)). Recall that R(T)→R∗=T∗+π2 as T→T∗; hence R∗∈(0,π) and so sin(R∗)>0. By continuity, there are numbers ε,δ>0 such that sin(R∗)-ε>0 and 1/sin(R(T))<1/(sin(R∗)-ε) for all T∈(T∗,T∗+δ). Therefore 1/sin(R(T)) is integrable in a right-sided neighborhood of T∗. Recall D is given by (2.2). Let Ri=limt→-∞R∘γi(t). Since Ri=Ti+π2, the endpoints (Ti,Ri,ωi) lie on the boundary of D. Then (i) - (iii) follow from the geometry of the Einstein static universe. Recall D is given by (2.2). Let Ri=limt→-∞R∘γi(t). Since Ri=Ti+π2, the endpoints (Ti,Ri,ωi) lie on the boundary of D. Then (i) - (iii) follow from the geometry of the Einstein static universe. □ The geometric obstruction Only properties (i) and (ii) of Theorem 1.1 were used to establish Proposition 2.1. The next proposition makes use of (iii) and highlights the main geometric obstruction to C0-extendibility. It’s an adaptation of [18, Prop. 3.13] to the spatially flat setting. Proposition 2.2 Let (M, g) be as in Theorem 1.1 and define (M1,g1) by (2.4). Let γ1 and γ2 be two past-directed past-inextendible timelike curves parameterized by t. Let q∈M1 be a point such that the images of γ1 and γ2 are contained in M1\J-(q,M1). Let ω denote the angular coordinates on M1 in spherical normal coordinates. Suppose ω1=limt→-∞ω∘γ1(t) exists and similarly for ω2. If ω1≠ω2, then where Σt is the spacelike hypersurface in M1 of constant t (in comoving coordinates) and the distance is with respect to the induced Riemannian metric on Σt\J-(q,M1). Proof In spherical normal coordinates, the metric is g=-dt2+a(t)2(dr2+r2dΩ2). Using the property that there are no particle horizons, there is a point p∈J-(q,M1) such that r(p)=0. (This can be proven analytically but it’s easy to see geometrically since M1 is conformal to D within the Einstein static universe.) Consequently, Σt\J-(q,M1)⊂Σt\J-(p,M1), and so it suffices to show Since Σt\J-(p,M1)={(t,r,ω)∣r>∫tt(p)1a(s)ds}, any spacelike curve joining γ1(t) to γ2(t) lying in Σt\J-(p,M1) will have length atleast Since ω1≠ω2, the term dSn-1(ω∘γ1(t),ω∘γ2(t)) is bounded below by a positive number for all sufficiently negative t. Hence condition (iii) in Theorem 1.1 implies that the above quantity diverges to ∞ as t→-∞. □ Proof of Theorem 1.1 Let (M, g) be as in Theorem 1.1 and define (M1,g1) by (2.4). To prove (M, g) is past C0-inextendible, it suffices to show that (M1,g1) is past C0-inextendible by the following lemma. Lemma 2.3 Let (M, g) be a spacetime. Suppose there is a proper open subset M′⊂M such that for every future-directed and past-inextendible timelike curve γ:(a,b]→M, there is a t∈(a,b) such that γ((a,t])⊂M′. Then (M, g) is past C0-inextendible if (M′,g|M′) is past C0-inextendible. Proof We prove the contrapositive. Suppose ι:(M,g)→(M~,g~) is a C0 extension with ∂-ι(M)≠∅. By definition there is a p~∈M~\ι(M) and a timelike curve γ~:[0,1]→M~ with γ~(0)=p~ such that γ:=ι-1∘γ~|(0,1] is a future-directed past-inextendible timelike curve in M. Define ι′:(M′,g|M′)→(M~,g~) via ι′=ι|M′. By assumption, there is a t∈(0,1) such that γ((0,t])⊂M′. Therefore p~∈∂-ι′(M′). □ Proofthat(M1,g1)ispastC0-inextendible_: The proof follows the same strategy as the proof of [18, Thm. 1.6]. Seeking a contradiction, suppose there is a continuous extension ι:(M1,g1)→(M~,g~) such that ∂-ι(M1)≠∅. For notational simplicity, we identify M1 with its image ι(M1), viewing M1 as a proper open submanifold of M~. There is a point p∈∂-M1. Using global hyperbolicity of (M1,g1), along with the assumption that the spacetime dimension satisfies n+1≥3, we can apply the time-reversed version of [18, Prop. 2.5] to produce a past boundary chart ϕ=(x0,x1,⋯,xn):V→(-ε0,ε0)×(-ε1,ε1)n centered around p such that the graphing function f:(-ε1,ε1)n→(-ε0,ε0) is differentiable at the origin and ∂1f(0)=0; hence the graph of f has a spacelike direction at the origin. Recall that each point on the graph of f is a point belonging to ∂-M1 and everything above the graph of f lies in M1, i.e., V>:=ϕ-1∘{x0>f(x1,⋯,xn)}⊂M1. Moreover, the graph of f is achronal with respect to smooth future-directed timelike curves in V, where the time-orientation on V is determined by ∂0. Moreover, we can assume that g~ on V satisfies gn≺g~≺gw, where Here g~≺gw means g~(X,X)≤0 implies gw(X,X)<0 for nonzero vectors X. For convenience, we work within the following globally hyperbolic subset U⊂V defined as follows: Let ε>0. Set a,b∈V by ϕ(a)=(-ε,0,⋯,0) and ϕ(b)=(ε,0,⋯,0). Choose ε small enough so that compact closure within V. Then U admits a Cauchy surface, U∩ϕ-1({x0=0}); hence the causal relation J± is closed within U [11, Prop. 3.21]. Moreover, for any two points x,y∈U with y∈I+(x,U), we have Jw+(x,U)∩Jw-(y,U) is compact within U. Set U>:=V>∩U, i.e., U> denotes those points in U that lie above the graph of f. Fix a point s∈U lying on the positive x0-axis. Let q∈U be a point on the graph of f such that 0 such that x1(r)<0 and xi(r)=0 for i=2,⋯,n. Choose |x1(r)| small enough so that r∈I-(s,U). It follows from the achronality of f that r∈I-(s,M1). Since ∂1f(0)=0, we can also ensure that J-(r,U>) does not intersect the sets by choosing r to lie just slightly above the graph of f. We claim that the following two properties hold. I+(p,U)∩I-(s,U) and I+(q,U)∩I-(s,U) are nonempty and neither set is contained in the other. I+(p,U)∩I-(s,U) and I+(q,U)∩I-(s,U) are nonempty and neither set is contained in the other. γ1|(0,1] and γ2|(0,1] are both contained in M1\J-(r,M1). γ1|(0,1] and γ2|(0,1] are both contained in M1\J-(r,M1). (1) follows easily since I+(p,U) is not contained in I+(q,U) nor vice versa. However, (2) is not immediate. To prove (2), note that Proposition 3.11 in [18] generalizes to the spatially flat setting. (Here the simple connectivity of M1 is used.) From this we conclude that no past-directed null geodesic in M1 emanating from r intersects one emanating from s. Now apply the time-reversed version of Proposition 2.6 in [18] (with p playing the role of t~). Therefore However, since J-(r,U>)∩Jw+(p,U)=∅, we have It follows that γ1|(0,1] is contained in M1\J-(r,M1). Replacing p with q in the above argument shows that the same holds for γ2|(0,1]. Thus (2) holds. (M1,g1) is future one-connected; see the proof of [18, Prop. 3.1] or [3, Prop. 2.7]). (Here the simple connectivity of M1 plays an essential role.) Thus we can infer from [16, Lem. 2.12] that and By (1), neither ⋃0<τ<1I+(γ1(τ),M1) nor ⋃0<τ<1I+(γ2(τ),M1) is contained in the other. Set Ti:=limτ→0T∘γi(τ) and ωi:=limτ→0ω∘γi(τ), for i=1,2. Part (ii) of Proposition 2.1(a) implies that T1,T2∈(-π2,π2), while parts (i) and (ii) of Proposition 2.1(b) imply that ω1≠ω2. Then property (2) above puts us in the setting of Proposition 2.2 (with the point r playing the role of q in the statement of Proposition 2.2). Now we establish the contradiction to the conclusion of Proposition 2.2. Recall Σt denotes the smooth spacelike Cauchy surface in M1 of constant t. Since the Cauchy surfaces foliate M1, there is a t0>-∞ such that s′∈I+(Σt,M1) for all -∞ defined by Since U is locally Minkowskian (in particular gn≺g~), we have |∂iht|<2 uniformly in [0,x1(q)]n and for all -∞0 there is a tϵ such that for all t∈(-∞,tϵ), the curve σt lies in the region of the (x0,x1)-plane defined by 0≤x1≤x1(q) and 0)∩Jw+(-s,U), where -s is the point on the negative x0-axis given by x0(-s)=-x0(s). Moreover, since r was chosen such that J-(r,U>) does not intersect the set {x1≥0}, it follows that But then another application of the time-reversed version of Proposition 2.6 in [18] (with -s playing the role of t~) gives Thus {σt}-∞