Lorentzian Pre-Length Spaces Overview

• Lorentzian pre-length spaces are synthetic models of spacetime that replace smooth Lorentzian metrics with order-theoretic, causal, and variational structures.
• They unify various formulations by using metric topology, extended time separation, or Lorentzian distance to derive causality and geodesic properties in nonsmooth settings.
• These models enable curvature bounds, optimal transport, and convergence theorems, extending classical Lorentzian geometry to low regularity and discrete contexts.

Lorentzian pre-length spaces are synthetic models of spacetime geometry in which the smooth Lorentzian metric is replaced by order-theoretic, topological, and variational data encoding causality and time separation. In the foundational formulation introduced by Kunzinger and Sämann, a Lorentzian pre-length space is a quintuple $(X,d,\ll,\leq,\tau)$ with a metric topology, chronological and causal relations, and a lower semicontinuous time separation satisfying the reverse triangle inequality; related formalisms use instead an extended time separation function $\ell$ or a Lorentzian distance $d$ as the primitive datum. Across these formulations, the framework is designed to include smooth spacetimes, continuous and causally plain spacetimes, closed cone structures, causal completions, causal sets, and nonsmooth limits, while supporting curvature comparison, geodesic theory, time functions, optimal transport, and convergence theory (Hau et al., 2022, Minguzzi et al., 2022, Mondino et al., 14 Apr 2025).

1. Foundational frameworks

In the standard synthetic formulation, a Lorentzian pre-length space consists of a metric space $(X,d)$, a transitive chronological relation $\ll$, a causal preorder $\leq$ containing $\ll$, and a time separation function $\tau:X\times X\to[0,\infty]$ such that $\tau$ is lower semi-continuous, $\tau(x,z)\geq \tau(x,y)+\tau(y,z)$ for $\ell$0, and $\ell$1 iff $\ell$2. Timelike and causal curves are locally Lipschitz curves monotone with respect to $\ell$3 and $\ell$4, respectively. This abstract package isolates the causal and variational structure of a spacetime without assuming differentiability (Hau et al., 2022).

Several later papers recast the same synthetic agenda in different primitive languages. One approach starts from an extended time separation $\ell$5 satisfying

$\ell$6

defines $\ell$7, $\ell$8, and sets $\ell$9. In that formulation, a Lorentzian pre-length space is a pair $d$0 with a topology finer than the chronological topology (Mondino et al., 14 Apr 2025). A different line, developed by Minguzzi and Suhr and then extended to the unbounded case, takes a Lorentzian distance $d$1 as basic and requires reverse triangle inequality, continuity together with compactness of chronological diamonds, and a distinguishing property that separates points; topology, causal relations, and causal curves are then reconstructed from this two-point function alone (Minguzzi et al., 2022, Bykov et al., 2024).

Formulation	Primitive data	Emphasis
Kunzinger–Sämann type	$d$2	Metric topology plus causal and variational structure
Extended separation type	$d$3	Signed time separation as primary datum
Lorentzian metric-space type	$d$4	Topology, causality, and curves derived from Lorentzian distance

A recurrent misconception is that a manifold structure or an auxiliary positive-definite metric is indispensable. The later abstract approaches explicitly avoid this: one paper states that all topology, causality, and curve classes are inferred from the Lorentzian distance alone, while another emphasizes that no manifold structure is needed to obtain time functions and the classical causal hierarchy (Minguzzi et al., 2022, Burtscher et al., 2021).

2. Curves, intrinsicness, and causal structure

The variational core of the theory is the $d$5-length of a causal curve $d$6,

$d$7

taken over partitions of $d$8. A curve is a maximizer if its $d$9-length equals the time separation of its endpoints, and geodesics are locally maximizing curves. In the Kunzinger–Sämann setting, a Lorentzian length space is a Lorentzian pre-length space in which $(X,d)$0 is recovered as the supremum of $(X,d)$1-lengths of future-directed causal curves; in the Minguzzi–Suhr language, a prelength space requires isocausal curves between chronologically related points, while a length space requires maximal isocausal curves (Barrera et al., 2022, Minguzzi et al., 2022).

The distinction between pre-length and length is substantive. Beran and Rott showed that intrinsicness can be characterized through $(X,d)$2-midpoints, in close analogy with metric midpoint criteria. A space is strictly intrinsic provided it has $(X,d)$3-midpoints, and merely intrinsic provided it has approximate $(X,d)$4-midpoints, under hypotheses involving a locally anti-Lipschitz time function and the null distance $(X,d)$5. Their construction proceeds by dyadic insertion of midpoints and control of a time function along the resulting chain (Beran et al., 2023).

Time functions occupy the causal side of the theory. Burtscher and García-Heveling proved that, for second countable locally compact Lorentzian (pre-)length spaces, $(X,d)$6-causality is equivalent to the existence of a time function. They further established that suitable averaged Geroch volume functions are time functions exactly in the causally continuous case, and that global hyperbolicity is equivalent to the existence of Cauchy time functions and Cauchy sets (Burtscher et al., 2021).

The framework also extends naturally to causal boundaries. For a globally hyperbolic spacetime $(X,d)$7, the future causal completion $(X,d)$8 can be endowed with a Lorentzian pre-length space structure $(X,d)$9, where points are indecomposable past sets and

$\ll$0

This produces large classes of nonsmooth examples, including generalized Robertson–Walker spacetimes and explicit completions such as de Sitter space (Hau et al., 2022).

3. Curvature bounds, angles, and first variation

Synthetic timelike curvature bounds are formulated by comparison with Lorentzian model spaces of constant curvature. In the triangle-comparison form, for timelike geodesic triangles and interior points $\ll$1 on their edges, one requires $\ll$2 to be bounded above or below by the corresponding model separation. Local or global versions then yield timelike curvature bounds from above or below, in direct analogy with Alexandrov and CAT$\ll$3 geometry (Barrera et al., 2022, Barton et al., 20 Jan 2026).

Angle theory is a major technical ingredient. Beran and Sämann introduced a hyperbolic angle between timelike curves using comparison triangles, together with timelike tangent cones and exponential maps. For timelike curves $\ll$4 with common basepoint $\ll$5,

$\ll$6

They proved symmetry, triangle inequalities for angles, completeness properties of the space of directions under curvature hypotheses, and a characterization of timelike curvature bounds by angle monotonicity. The same work also improved non-branching results: in strongly causal spaces with timelike curvature bounded below, timelike distance realizers cannot branch (Beran et al., 2022).

Barrera, Montes de Oca, and Solis introduced the normalized angle

$\ll$7

proved angle monotonicity and a local Lorentzian Toponogov theorem, and established an Alexandrov convexity property for spaces with timelike curvature bounded below by $\ll$8. In the globally hyperbolic, nonnegatively curved case they obtained a first variation formula relating the initial rate of change of time separation to the normalized angle (Barrera et al., 2022).

Subsequent work unified the proliferating curvature notions. Under mild assumptions, triangle comparison, one-sided triangle comparison, monotonicity, angle comparison, hinge comparison, four-point conditions, and convexity or concavity of a modified time separation function were shown to be equivalent. In particular, causal and timelike curvature bounds were proved equivalent in this generalized setting (Beran et al., 2023).

A major refinement was obtained in the splitting-theorem paper, which extended the first variation formula to Lorentzian pre-length spaces with any timelike curvature bound, either upper or lower and different from $\ll$9: $\leq$0 The proof addresses synthetic issues such as angle continuity, null segments, and limsup/liminf comparison arguments (Barton et al., 20 Jan 2026).

4. Constructions, globalization, and rigidity

Lorentzian pre-length spaces admit several product and gluing constructions. Taxicab and uniform products were introduced to build new Lorentzian pre-length spaces from old ones. For example, if $\leq$1 is Lorentzian and $\leq$2 is metric, the taxicab product uses

$\leq$3

with

$\leq$4

These constructions were used to show that the hyperspace of compact causal diamonds can itself be made into a Lorentzian length space, and that $\leq$5 is geodesic and globally hyperbolic for complete $\leq$6 (Barrera et al., 2023).

Gluing theory provides an analogue of metric amalgamation. For suitable closed non-timelike locally isolating subsets and a structure-preserving identification map, one can define a quotient time separation

$\leq$7

taken over causal chains crossing the glued interface. This yields a synthetic Lorentzian version of Reshetnyak’s gluing theorem: gluing spaces with upper timelike curvature bound preserves that bound under appropriate hypotheses (Beran et al., 2022). A related analysis of the causal ladder showed that chronology, causality, strong causality, distinction, and—under stronger assumptions—global hyperbolicity can survive amalgamation, whereas $\leq$8-causality, reflectivity, and causal simplicity are not generally preserved (Rott, 2022).

Globalization results parallel classical Alexandrov geometry. A Lorentzian analogue of Alexandrov’s Patchwork shows that suitably nice spaces with local upper timelike curvature bound satisfy a corresponding global upper bound. The same paper proves a Bonnet–Myers style result for spaces with global lower timelike curvature bound, giving

$\leq$9

for $\ll$0 under stated regularity and nondegeneracy assumptions (Beran et al., 2023).

The strongest rigidity result currently available is the splitting theorem for non-positively curved Lorentzian spaces. If $\ll$1 has timelike curvature globally bounded above by $\ll$2 and $\ll$3 is a complete timelike line, then the space $\ll$4 of complete timelike lines weakly parallel to $\ll$5, modulo shift reparametrization, has unique synchronised parallel representatives; endowed with the natural distance

$\ll$6

it is a $\ll$7 space, and the union of the parallel lines is isometric to the Lorentzian product $\ll$8. When every point lies on such a line, the splitting is global (Barton et al., 20 Jan 2026).

5. Measure, dimension, and optimal transport

McCann and Sämann introduced a Lorentzian analogue of Hausdorff measure built from causal diamonds. For dimension parameter $\ll$9, the diamond pseudo-volume is

$\tau:X\times X\to[0,\infty]$0

and the outer measure $\tau:X\times X\to[0,\infty]$1 is defined by Carathéodory coverings with causal diamonds. The associated geometric dimension

$\tau:X\times X\to[0,\infty]$2

distinguishes spacelike and null subspaces of Minkowski space: spacelike $\tau:X\times X\to[0,\infty]$3-planes have geometric dimension $\tau:X\times X\to[0,\infty]$4, null $\tau:X\times X\to[0,\infty]$5-planes have geometric dimension $\tau:X\times X\to[0,\infty]$6, and null curves have geometric dimension zero. The same work introduced causal doubling and a notion of non-collapsed synthetic spacetimes (McCann et al., 2021).

Cavalletti and Mondino built Lorentzian optimal transport on measured Lorentzian pre-length spaces $\tau:X\times X\to[0,\infty]$7. For $\tau:X\times X\to[0,\infty]$8, the $\tau:X\times X\to[0,\infty]$9-Lorentz-Wasserstein distance is

$\tau$0

with cyclical monotonicity, stability of optimal couplings, and Kantorovich duality established for this causal cost. They then defined synthetic timelike Ricci curvature-dimension conditions $\tau$1 by entropy convexity along future-directed timelike transport geodesics, proved stability under suitable weak convergence, and derived timelike Brunn–Minkowski, Bishop–Gromov, Bonnet–Myers, and sharp Hawking singularity results in the synthetic setting (Cavalletti et al., 2020).

Geometric measure theory has recently been extended further through a Lorentzian coarea inequality. Using the notions of timelike Lipschitz maps, uniformly $\tau$2-controlling maps, the local causal enlargement property, and a causal covering lemma, one obtains coarea-type bounds for Lorentzian Hausdorff measure between Lorentzian pre-length spaces (Kubota, 9 May 2026).

6. Convergence, limits, and conformal transformations

A major recent theme is convergence theory. One direction develops Lorentzian Gromov–Hausdorff convergence using causal diamonds rather than metric balls. Covered Lorentzian pre-length spaces admit a pre-compactness theorem based on controlled covers by small diamonds, and the resulting convergence applies to both smooth globally hyperbolic spacetimes and synthetic Lorentzian spaces. Timelike sectional curvature bounds are stable under this convergence, timelike blow-up tangents can be defined, and faithfully embedded causal sets converging to two smooth globally hyperbolic spacetimes force those spacetimes to be isometric, expressing a version of the causal-set Hauptvermutung (Mondino et al., 14 Apr 2025).

The unbounded theory of Lorentzian metric spaces generalizes earlier bounded constructions by replacing finite timelike diameter with continuity and relative compactness of chronological diamonds. Under countable generation, such spaces are Polish, and Lorentzian (pre)length spaces are stable under the corresponding GH convergence (Bykov et al., 2024).

A second convergence notion, $\tau$3-convergence, treats covered Lorentzian pre-length spaces $\tau$4 through covered GH convergence of the underlying metric spaces combined with uniform convergence of signed time separation functions. Timelike sectional curvature and timelike curvature-dimension bounds are stable under measured $\tau$5-convergence. This framework is particularly effective for generalized Lorentzian cones $\tau$6, where convergence follows from GH convergence of $\tau$7 and $\tau$8 together with uniform convergence of $\tau$9, and it yields sharp curvature and pre-compactness theorems for such cones (Ketterer, 11 May 2026).

Recent local geometry has also moved toward synthetic Jacobi theory. Timelike conjugate points in Lorentzian (pre-)length spaces have been defined in one-sided, symmetric, unreachable, and ultimate forms; in strongly causal smooth spacetimes these notions agree with the classical one, and applications include a timelike Rauch comparison theorem and a Lorentzian Cartan–Hadamard type result (Grant et al., 16 Sep 2025).

Conformal geometry has now entered the synthetic theory as well. For intrinsic, strongly causal Lorentzian pre-length spaces and a continuous conformal factor $\tau(x,z)\geq \tau(x,y)+\tau(y,z)$0, one defines a conformal time separation $\tau(x,z)\geq \tau(x,y)+\tau(y,z)$1 from conformal curve length. The resulting $\tau(x,z)\geq \tau(x,y)+\tau(y,z)$2 is again a Lorentzian pre-length space, giving a synthetic notion of conformal transformation as an equivalence relation. Angles and causality conditions are conformally invariant, global hyperbolicity is characterized by finiteness of $\tau(x,z)\geq \tau(x,y)+\tau(y,z)$3 for all conformal factors, and the Lorentzian Hausdorff measure transforms by the expected $\tau(x,z)\geq \tau(x,y)+\tau(y,z)$4 factor (Manzano et al., 5 Dec 2025).

These developments collectively suggest that Lorentzian pre-length spaces now support a substantial fraction of the structural, variational, and rigidity theory familiar from smooth Lorentzian geometry, but in a form compatible with low regularity, singular limits, and discrete models.