We prove a splitting theorem for cosmological spacetimes which admit a globally p-harmonic Cauchy temporal function for some p<1, without making any use of timelike lines. In Bartnik's context of cosmological and timelike geodesically complete spacetimes, we establish a Liouville-type unboundedness result which shows that globally p-harmonic temporal functions, p<1, are necessarily Cauchy temporal. Moreover, using the theory of elliptic PDEs, we show that locally Lipschitz p-harmonic temporal functions with p<1 are in fact smooth. Altogether, our work reduces the resolution of Bartnik's cosmological splitting conjecture to the global weak solvability of the p-d'Alembert equation for some p<1 in the class of locally Lipschitz temporal functions.
Introduction
Background
1.1. Background. The singularity theorems of Hawking and Penrose [43, 29, 31] show that singularities, i.e., incomplete causal geodesics, are a generic occurrence in spacetimes. We refer to the standard texts [30, 42, 3] for their accessible treatment. Among the classical theorems, the Hawking–Penrose singularity theorem [31] is the most refined, makes the weakest a priori assumption on the global causality and is able to accommodate a variety of initial and boundary conditions to conclude the existence of singularities. In turn, it makes use of a global curvature nontriviality assumption called the genericity condition, which requires nontrivial tidal forces to exist somewhere along every inextendible causal geodesic. In the special setting of cosmological spacetimes, i.e., spacetimes with compact Cauchy surfaces satisfying the strong energy condition, the Hawking–Penrose theorem takes the following form:
Theorem 1.1 (Cosmological Hawking–Penrose singularity theorem). Let (M,g) be a globally hyperbolic spacetime with compact Cauchy surfaces satisfying the strong energy condition, i.e., Ric(v,v)≥0 for all v∈TM timelike, as well as the genericity condition, i.e., for every inextendible causal geodesic γ there exists a parameter t such that the tidal force operator R:Tγ(t)M→Tγ(t)M, v↦R(v,γ˙(t))γ˙(t), is not the zero endomorphism. Then there exist incomplete causal geodesics in M.
It is natural to ask just how generic the genericity condition actually is, i.e., what spacetimes remain complete in its absence? In 1988, Bartnik [2] famously conjectured that product spacetimes of the form (R×Σ,dt2−h), where (Σ,h) is a compact Riemannian manifold of nonnegative Ricci curvature, are the only complete cosmological spacetimes. More precisely:
Conjecture 1.2 (Bartnik’s cosmological splitting conjecture). Let (M,g) be a globally hyperbolic spacetime with compact Cauchy surfaces satisfying the strong energy condition. If (M,g) is timelike geodesically complete, then it is isometric to a product spacetime (R×Σ,dt2−h), where (Σ,h) is a compact Riemannian manifold of nonnegative Ricci curvature.
The conjecture is known to hold under the stronger assumption of nonnegative timelike sectional curvature, and thus in spacetime dimension 2 since timelike sectional and Ricci curvatures coincide, due to the work of Ehrlich–Galloway [16]. Their approach has recently been extended to prove a Bartnik-type cosmological splitting theorem in the synthetic setting of Lorentzian length spaces [18].
The principal difficulty is the following: Under the assumptions of Bartnik’s conjecture, it is elementary to construct a causal line. If the latter is timelike, the claim follows from the Lorentzian splitting theorem [17, 19, 40]. Because of this observation, a variety of rather natural additional conditions have been imposed to guarantee timelikeness of the line, we refer to Galloway’s excellent survey for a detailed discussion of these developments [20]. While the plethora of contributions leads us to believe that Bartnik’s splitting conjecture is correct, a number of interesting causality-theoretic properties have been recently observed for any potential counterexample to it by Minguzzi [38].
A different approach to Bartnik’s conjecture is to find a CMC (i.e., constant mean curvature) Cauchy surface, which immediately implies the splitting [2 Cor. 1]. However, in general, such CMC Cauchy surfaces are difficult to construct (cf. [15, 21, 22]), and there are examples (even in vacuum) of globally hyperbolic spacetimes with compact Cauchy surfaces which satisfy the strong energy condition but which do not contain any CMC Cauchy surfaces [2, 12, 33, 34]. Let us mention the recent work of Galloway–Ling [23] which shows the existence of a CMC Cauchy surface using mean curvature flow methods under the additional assumption that there exists an expanding Cauchy surface.
Aside from the CMC Cauchy surface approach to Bartnik’s conjecture, far fewer works have tried to establish splitting directly without relying on the existence of a timelike line and thus the Lorentzian splitting theorem (for results of this flavor, see the works of Galloway–Vega [25, 26]).
Main results
The aim of this article is to approach Bartnik’s conjecture from an elliptic point of view, by making use of the degenerate elliptic p-d’Alembertian
□pf:=−div(∣∇f∣gp−2∇f),(1)
where p<1. This operator first appeared in the work of Mondino–Suhr [39] and its (degenerate) ellipticity properties were discovered and studied in a very general metric spacetime setting by Beran, Braun, Calisti, Gigli, McCann, Rott, Sämann and the author in [4] (see also Braun’s followup work [6], the surveys by Braun [7] and McCann [36] and the note by Galloway, McCann and the author [24]). They were subsequently used by Braun, Gigli, McCann, Sämann and the author to give elliptic proofs of the Lorentzian splitting theorems [9] which are much closer to the proof of the Riemannian splitting theorem of Cheeger–Gromoll [11]. The strategy was adapted to low regularity metrics to prove a splitting theorem for (C1,α-weighted)C1-Lorentzian metrics by the same set of authors [8] and to Finsler spacetimes in order to establish a vastly improved splitting theorem by Caponio, Ohta and the author in [10].
Let us outline our program and discuss our main results. First, we fix our terminology on temporal functions. We use the Lorentzian signature (+,−,…,−) throughout, as indicated in Subsection 1.4.
Definition 1.3 (Temporal function). Let (M,g) be a spacetime, X its time orientation vector field and Ω⊆M open. A locally Lipschitz function f:Ω→R is called temporal if for every compact set K⊆Ω there exists cK>0 such that
min(g(∇f,∇f),g(X,∇f))≥cK(2)
a.e. on K. If Ω=M, a temporal function f:M→R is called Cauchy temporal if each level set {f=s} is a Cauchy hypersurface.
If f∈C1(Ω), the definition reduces to the usual notion of temporal function which requires ∇f to be future timelike.
Definition 1.4 (Weakly p-harmonic temporal function). Let (M,g) be a spacetime, p<1, Ω⊆M open and f:Ω→R a locally Lipschitz temporal function. Then f is called weakly p-harmonic if
∫Ωg(∣∇f∣gp−2∇f,∇φ)dvolg=0(3)
for every φ∈Cc∞(Ω).
Our first main result, proved in § 3, establishes the smoothness of p-harmonic temporal functions:
Theorem 1.5 (Smoothness of p-harmonic temporal functions). Let (M,g) be a spacetime, Ω⊆M open, p<1 and f:Ω→R a locally Lipschitz temporal function. If f is weakly p-harmonic, then f∈C∞(Ω) and it is classically p-harmonic in the sense that
□pf=−div(∣∇f∣gp−2∇f)=0.(4)
Our arguments rely on results from the theory of elliptic PDEs [28]: First, we use de Giorgi–Nash–Moser estimates to show that f∈C1,α for some α∈(0,1), then Schauder estimates boost that regularity to C2,α. From here, classical elliptic regularity gives smoothness. When comparing to regularity results for solutions of the p-Laplace equation in the Riemannian setting [14, 32], where C1,α can in general not be improved, we benefit from the fact that local boundedness and local uniform future timelikeness of ∇f restrict it to the domain of uniform ellipticity of □p, making a bootstrap all the way to smoothness possible. In the Riemannian setting, of course, p-harmonic functions are smooth on the open set where their gradient does not vanish, but the latter can in general not be excluded.
While solving the p-d’Alembert equation in not a part of this article, we refer to [13], where p-harmonic temporal functions constructed as minimizers of the Lorentzian p-energy on suitable domains in Minkowski spacetime are indeed shown to be temporal and locally Lipschitz, hence smooth by our Theorem 1.5.
Our next main result and certainly the one requiring the most substantial body of arguments is a Liouville-type unboundedness result for p-harmonic temporal functions in cosmological spacetimes which are timelike geodesically complete, i.e., precisely in Bartnik’s setting.
Theorem 1.6 (Liouville-type theorem for p-harmonic temporal functions). Let (M,g) be a globally hyperbolic, timelike geodesically complete spacetime with compact Cauchy surfaces satisfying the strong energy condition. If f:M→R is a p-harmonic smooth temporal function, then f is Cauchy temporal.
The proof, which is the content of § 4, is lengthy and involved. It makes full use of every assumption made (see Subsection 4.2). A useful tool in the proof, also relevant for the next main result, is the divergence theorem, which, due to compactness of Cauchy surfaces, allows us to fix a background Cauchy temporal function and exploit divergence-freeness of the vector field ∣∇f∣gp−2∇f in order to reduce spacetime integrals to integrals on level sets of that function. We refer to [1 Thm. 1.3.16] for a version of the Lorentzian divergence theorem that we will use in various places.
Our final main result, established in § 5, is a cosmological splitting theorem under the assumption that there exists a globally p-harmonic Cauchy temporal function.
Theorem 1.7 (Cosmological splitting via p-harmonicity). Let (M,g) be a globally hyperbolic spacetime with compact Cauchy surfaces satisfying the strong energy condition. If there exists a p-harmonic Cauchy temporal function f:M→R, then (M,g) splits isometrically as (R×Σ,dt2−h), where (Σ,h) is a compact Riemannian manifold of nonnegative Ricci curvature. In fact, Σ={f=0} and the splitting map is the flow of ∇f (upon normalization).
In the elliptic proof of the Lorentzian splitting theorems [9], a p-harmonic temporal function is provided by the Busemann function associated with the given complete timelike line, whose gradient is of unit norm. Consequently, the p-Bochner–Ohta identity (see Theorem 2.7 below) immediately implied that the Hessian of the Busemann function vanishes. In our cosmological splitting result above, no assumption on ∣∇f∣g is necessary. The aforementioned divergence-freeness of the vector field J together with compactness of Cauchy surfaces allows us to show global concavity of a certain positive function constructed from integrals along the level sets of f, which are thus constant and this leads to the vanishing of the additional term in the p-Bochner–Ohta identity, allowing us to conclude that Hessf=0. From here, it is easy to conclude that the flow of ∇f (upon normalizing by the a posteriori constant ∣∇f∣g) must be the desired isometry to a product spacetime.
Altogether, our main results allow us to give the following equivalent reformulation of Bartnik’s conjecture in terms of the existence of global weak solutions of the p-d’Alembert equation:
Theorem 1.8 (Bartnik’s conjecture ⇔ weak solvability of p-d’Alembert equation). Let (M,g) be a cosmological, timelike geodesically complete spacetime with compact Cauchy surfaces satisfying the strong energy condition. Then the following are equivalent:
(i) (M,g) is isometric to a product spacetime (R×Σ,dt2−h), where (Σ,h) is a compact Riemannian manifold with nonnegative Ricci curvature.
(ii) There exists p<1 and a locally Lipschitz temporal function f:M→R that is weakly p-harmonic.
The direction (ii)⇒(i) follows from our main results, i.e., Theorems 1.5, 1.6 and 1.7. The direction (i)⇒(ii) is of course trivial, since the projection onto R is p-harmonic for every p in a product spacetime.
Discussion and outlook
Based on Theorem 1.8, let us formulate the following conjecture, which is equivalent to Bartnik’s:
Conjecture 1.9. Let (M,g) be a globally hyperbolic, timelike geodesically complete spacetime with compact Cauchy surfaces satisfying the strong energy condition. Then there exists p<1 and a locally Lipschitz temporal function f:M→R which is weakly p-harmonic.
Conjecture 1.9 may prove to be a simpler approach to Bartnik’s conjecture in comparison with previous ones. We remark that naive Busemann-like approaches to construct p-harmonic functions do not work in Bartnik’s setting, see the work of Galloway, McCann and the author [24] for a detailed discussion. On suitable local domains, such as between level sets of a fixed background Cauchy temporal function, we expect to be able to find p-harmonic temporal functions as minimizers of the Lorentzian p-energy f~↦∫ΩHp(df~)dvolg, where Hp is the p-Hamiltonian (cf. Definition 2.5) subject to certain well-chosen boundary conditions, see [13] for results in this direction on Minkowski spacetime. The major remaining challenge is to construct all of these local p-harmonic temporal functions in a way that their temporality and local Lipschitzness is uniform, as one can then pass to the limit as follows:
Proposition 1.10 (Global p-harmonicity in the limit). Let (M,g) be a spacetime, p<1, (Ωn) a nested exhaustion of M by open sets, and fn:Ωn→Rp-harmonic. If (fn) is locally uniformly Lipschitz and (∇fn) is locally uniformly future timelike, i.e., for every K⊆M compact there exists cK>0 such that
n≥nKinfmin(g(∇fn,∇fn),g(X,∇fn))≥cK,(5)
where nK is such that K⊆ΩnK and X is the time orientation, then for every o∈M there exists a p-harmonic temporal function f:M→R such that, up to subsequences, fn−fn(o)→f in Cloc1.
Proof. We can fix some o∈M (w.l.o.g. o∈Ω1) and assume that fn(o)=0, otherwise replace fn by fn−fn(o), which are still locally uniformly Lipschitz and locally uniformly temporal functions that are p-harmonic on Ωn.
The arguments given in the proof of Theorem 3.1, specifically (32), allow us to show that (dfn) is locally uniformly α-Hölder for some α∈(0,1) as a consequence of local uniform Lipschitzness, local uniform temporality and p-harmonicity. Thus, we may apply the Arzelá–Ascoli theorem to conclude the existence of a C1-function f:M→R such that, up to subsequences fn→f in Cloc1. Local uniform future timelikeness of (∇fn) implies that ∇f is future timelike and bounded, i.e., f is temporal. The weak formulation of p-harmonicity is easily seen to pass to the limit under Cloc1-convergence, so we conclude that f is p-harmonic. □
Let us formulate the existence of the sequence (fn) as above explicitly as a conjecture, which is what we think will be the result to be targeted if one approaches Bartnik’s conjecture from the point of view we have presented here.
References
[1]Bär, C., Ginoux, N., and Pfäffle, F. Wave equations on Lorentzian manifolds and quantization. EMS Lectures in Mathematics and Physics. European Mathematical Society (EMS), Zürich, 2007.DOI
[2]Bartnik, R. Remarks on cosmological spacetimes and constant mean curvature surfaces. Comm. Math. Phys. 117, 4 (1988), 615–624.DOI
[3]Beem, J. K., Ehrlich, P. E., and Easley, K. L. Global Lorentzian geometry, second ed., vol. 202 of Monographs and Textbooks in Pure and Applied Mathematics. Marcel Dekker, Inc., New York, 1996.
[4]Beran, T., Braun, M., Calisti, M., Gigli, N., McCann, R. J., Ohanyan, A., Rott, F., and Sämann, C. A nonlinear d’Alembert comparison theorem and causal differential calculus on metric measure spacetimes. Preprint, arXiv:2408.15968 (2024).arxiv.org/abs/2408.15968
[5]Bernal, A. N., and Sánchez, M. Smoothness of time functions and the metric splitting of globally hyperbolic spacetimes. Comm. Math. Phys. 257, 1 (2005), 43–50.DOI
[6]Braun, M. Exact d’Alembertian for Lorentz distance functions. Calc. Var. Partial Differential Equations 65, 9 (2026), Paper No. 262.arxiv.org/abs/2408.16525
[7]Braun, M. New perspectives on the d’Alembertian from general relativity. An invitation. Indag. Math. (N.S.) 37, 3 (2026), 843–891.arxiv.org/abs/2501.19071
[8]Braun, M., Gigli, N., McCann, R. J., Ohanyan, A., and Sämann, C. A Lorentzian splitting theorem for continuously differentiable metrics and weights. Preprint, arXiv:2507.06836 (2025).arxiv.org/abs/2507.06836
[9]Braun, M., Gigli, N., McCann, R. J., Ohanyan, A., and Sämann, C. An elliptic proof of the splitting theorems from Lorentzian geometry. Comm. Amer. Math. Soc. 6 (2026), 1087–1112.arxiv.org/abs/2410.12632
[10]Caponio, E., Ohanyan, A., and Ohta, S.-I. Splitting theorems for weighted Finsler spacetimes via the p-d’Alembertian: Beyond the Berwald case. J. Reine Angew. Math. 838 (2026), 221–253.DOI
[11]Cheeger, J., and Gromoll, D. On the structure of complete manifolds of nonnegative curvature. Ann. of Math. (2) 96 (1972), 413–443.
[12]Chruściel, P. T., Isenberg, J., and Pollack, D. Initial data engineering. Comm. Math. Phys. 257, 1 (2005), 29–42.DOI
[13]de Philippis, G., Gigli, N., McCann, R. J., Ohanyan, A., Picerni, M., and Zanardini, M. Regularity of Lorentz p-harmonic functions. in preparation (2026).
[14]DiBenedetto, E. C^{1+α} local regularity of weak solutions of degenerate elliptic equations. Nonlinear Anal. 7, 8 (1983), 827–850.
[15]Dilts, J., and Holst, M. When do spacetimes have constant mean curvature slices? Preprint, arXiv:1710.03209 (2017).arxiv.org/abs/1710.03209
[16]Ehrlich, P. E., and Galloway, G. J. Timelike lines. Classical Quantum Gravity 7, 3 (1990), 297–307.DOI
[17]Eschenburg, J.-H. The splitting theorem for space-times with strong energy condition. J. Differential Geom. 27, 3 (1988), 477–491.DOI
[18]Flores, J. L., Herrera, J., and Solis, D. A low regularity approach to Bartnik’s conjecture. Classical Quantum Gravity 42, 21 (2025), Paper No. 215020, 15.arxiv.org/abs/2412.08967
[19]Galloway, G. J. The Lorentzian splitting theorem without the completeness assumption. J. Differential Geom. 29, 2 (1989), 373–387.DOI
[20]Galloway, G. J. Existence of CMC Cauchy surfaces and spacetime splitting. Pure Appl. Math. Q. 15, 2 (2019), 667–682.arxiv.org/abs/1902.08803
[21]Galloway, G. J., and Ling, E. Existence of CMC Cauchy surfaces from a spacetime curvature condition. Gen. Relativity Gravitation 50, 9 (2018), Paper No. 108, 7.DOI
[23]Galloway, G. J., and Ling, E. A CMC existence result for expanding cosmological spacetimes. Trans. Amer. Math. Soc. 378, 11 (2025), 8055–8069.arxiv.org/abs/2410.16619
[24]Galloway, G. J., McCann, R. J., and Ohanyan, A. Failure of local equi-Lipschitzness for families of Lorentz distances to Cauchy surface foliations. Gen. Relativity Gravitation 58, 9 (2026), Paper No. 105.arxiv.org/abs/2606.21997
[25]Galloway, G. J., and Vega, C. A chronal limits, Lorentzian spheres, and splitting. Ann. Henri Poincaré 15, 11 (2014), 2241–2279.arxiv.org/abs/1211.2460
[26]Galloway, G. J., and Vega, C. Hausdorff closed limits and rigidity in Lorentzian geometry. Ann. Henri Poincaré 18, 10 (2017), 3399–3426.DOI
[27]Gerhardt, C. Curvature flows in semi-Riemannian manifolds. In Surveys in differential geometry. Vol. XII. Geometric flows, vol. 12 of Surv. Differ. Geom. Int. Press, Somerville, MA, 2008, pp. 113–165.arxiv.org/abs/0704.0236
[28]Gilbarg, D., and Trudinger, N. S. Elliptic partial differential equations of second order. Classics in Mathematics. Springer-Verlag, Berlin, 2001. Reprint of the 1998 edition.
[29]Hawking, S. W. The occurrence of singularities in cosmology. J. Proc. Roy. Soc. London Ser. A 294 (1966), 511–521.DOI
[30]Hawking, S. W., and Ellis, G. F. R. The large scale structure of space-time, anniversary ed. Cambridge Monographs on Mathematical Physics. Cambridge University Press, Cambridge, 2023. With a foreword by Abhay Ashtekar.
[31]Hawking, S. W., and Penrose, R. The singularities of gravitational collapse and cosmology. Proc. Roy. Soc. London Ser. A 314 (1970), 529–548.
[32]Lewis, J. L. Regularity of the derivatives of solutions to certain degenerate elliptic equations. Indiana University Mathematics Journal 32, 6 (1983), 849–858.
[33]Ling, E., and Ohanyan, A. Examples of cosmological spacetimes without CMC Cauchy surfaces. Lett. Math. Phys. 114, 4 (2024), Paper No. 96, 27.DOI
[35]McCann, R. J. Displacement convexity of Boltzmann’s entropy characterizes the strong energy condition from general relativity. Camb. J. Math. 8, 3 (2020), 609–681.arxiv.org/abs/1808.01536
[36]McCann, R. J. Trading linearity for ellipticity: a nonsmooth approach to Einstein’s theory of gravity and the Lorentzian splitting theorems. Preprint, arXiv:2501.00702 (2025).arxiv.org/abs/2501.00702
[37]Minguzzi, E. Causality theory for closed cone structures with applications. Rev. Math. Phys. 31, 5 (2019), 1930001, 139.DOI
[38]Minguzzi, E. Bartnik’s splitting conjecture and the cosmological wall: a Cauchy Busemann function approach. Preprint, 2610.17092 (2026).arxiv.org/abs/2610.17092
[39]Mondino, A., and Suhr, S. An optimal transport formulation of the Einstein equations of general relativity. J. Eur. Math. Soc. (JEMS) 25, 3 (2023), 933–994.DOI
[40]Newman, R. P. A. C. A proof of the splitting conjecture of S.-T. Yau. J. Differential Geom. 31, 1 (1990), 163–184.DOI
[41]Ohanyan, A., and Sálamo Candal, M. Timelike Ricci curvature lower bounds via optimal transport for Orlicz-type Lorentzian costs. Preprint arXiv:2604.22538 (2026).arxiv.org/abs/2604.22538
[42]O’Neill, B. Semi-Riemannian geometry, vol. 103 of Pure and Applied Mathematics. Academic Press, Inc. [Harcourt Brace Jovanovich, Publishers], New York, 1983. With applications to relativity.
[43]Penrose, R. Gravitational collapse and space-time singularities. Phys. Rev. Lett. 14 (1965), 57–59.DOI