---
title: Null Distance in Lorentzian Geometry
url: https://www.emergentmind.com/topics/null-distance
type: topic
---

# Null Distance in Lorentzian Geometry

Null distance is a distance construction on a time-oriented Lorentzian manifold \((M,g)\) that depends on a chosen time function \(\tau\) and measures the minimal total variation of \(\tau\) along piecewise causal “zig-zag” curves. Introduced by Sormani and Vega, it yields a symmetric pseudometric that is conformally invariant, agrees with the usual time separation on causal pairs, and under local anti-Lipschitz hypotheses becomes a genuine metric inducing the manifold topology. A central theme of the subject is that, with suitable choices of \(\tau\), the pair \((d_\tau,\tau)\) can recover the causal order, characterize global hyperbolicity through metric completeness, and support Gromov–Hausdorff and intrinsic-flat convergence theories for spacetimes and Lorentzian length spaces [1508.00531] [2209.15610] [2106.05393].

## 1. Definition and formal construction

Let \((M,g)\) be a time-oriented Lorentzian manifold and let \(\tau:M\to\mathbb R\) be a continuous time function, meaning that \(\tau\) is strictly increasing along every future-directed causal curve. A piecewise causal curve \(\gamma:[a,b]\to M\) is broken into finitely many future- or past-directed causal segments, with break-points \(\gamma(a)=x_0,x_1,\dots,x_m=\gamma(b)\). Its null length is
\[
L_\tau(\gamma)=\sum_{i=1}^m \bigl|\tau(x_i)-\tau(x_{i-1})\bigr|.
\]
The null distance is then defined by
\[
d_\tau(p,q)=\inf\Bigl\{L_\tau(\gamma):\gamma\text{ piecewise causal from }p\text{ to }q\Bigr\}.
\]
The literature also writes \(\hat d_\tau\) or \(d(\cdot,\cdot;\tau)\) for the same construction [2306.17128] [1508.00531] [2511.05847].

This definition extends beyond smooth spacetimes. In a Lorentzian (pre-)length space \(X=(X,d,\ll,\le,\rho)\), one again defines piecewise causal paths, the same null-length functional, and an induced null distance \(d_\tau\). In this setting the construction becomes part of the synthetic Lorentzian program, where it provides a metric-space avatar of causal geometry beyond the manifold category [2106.05393].

The basic intuition, made explicit in later expositions, is that \(d_\tau\) measures the shortest “time-jump” needed by a causal zig-zag from \(p\) to \(q\). That interpretation is precise because the admissible curves are causal on each segment but may alternate between future and past directions, while the cost is the total absolute change in the chosen time function [2306.17128].

## 2. Pseudometric structure, definiteness, and temporal regularity

For any time function \(\tau\), the null distance is always finite, symmetric, and satisfies the triangle inequality by concatenation of piecewise causal curves. One has
\[
d_\tau(p,p)=0,\qquad d_\tau(p,q)=d_\tau(q,p),\qquad d_\tau(p,r)\le d_\tau(p,q)+d_\tau(q,r).
\]
It is therefore a pseudometric in general, not necessarily a metric [2306.17128] [1508.00531].

A universal lower bound is
\[
d_\tau(p,q)\ge |\tau(q)-\tau(p)|,
\]
so \(d_\tau(p,q)=0\) implies \(\tau(p)=\tau(q)\). If \(p\le q\), then
\[
d_\tau(p,q)=\tau(q)-\tau(p),
\]
and in the smooth setting this quantity is also bounded below by the Lorentzian distance \(d_L(p,q)\). Thus the null distance agrees with the time increment on causal pairs, while possible failures of causal encoding arise only in the converse direction [2106.05393] [2511.05847].

Definiteness is controlled by the local anti-Lipschitz condition. A time function \(\tau\) is locally anti-Lipschitz if for every \(p\in M\) there is a neighborhood \(U\), a Riemannian metric \(h\) on \(U\), and in some formulations a constant \(C>0\), such that
\[
x\le y \;\Longrightarrow\; \tau(y)-\tau(x)\ge C\,d_h(x,y)
\]
for all \(x,y\in U\). Sormani–Vega’s criterion states that \(d_\tau\) is a true distance on \(M\) if and only if \(\tau\) is locally anti-Lipschitz; in that case the topology induced by \(d_\tau\) agrees with the manifold topology [1508.00531] [2306.17128] [2507.07158].

For \(C^1\) functions, local anti-Lipschitzness is equivalent to temporality: a \(C^1\) function \(f\) is locally anti-Lipschitz if and only if its gradient \(\nabla f\) is everywhere past-directed timelike. More generally, if \(\nabla\tau\) exists almost everywhere, is past-pointing timelike there, and is locally bounded away from the light cones, then \(\tau\) is locally anti-Lipschitz. In local coordinates this is reflected by inequalities of the form
\[
g(\nabla f,X)\ge C\|X\|_h
\]
for all future causal \(X\), which integrate to the anti-Lipschitz estimate along causal curves [1508.00531] [2507.07158].

A further regularity class used in the global theory is that of weak temporal functions: continuous functions that strictly increase along causal curves and are both locally Lipschitz and locally anti-Lipschitz. On compact sets, the corresponding null distances are locally bi-Lipschitz equivalent to any auxiliary Riemannian metric, and null distances arising from two weak temporal functions are bi-Lipschitz equivalent on compact subsets [2209.15610].

## 3. Encoding causality and the role of global hypotheses

The central causality statement is
\[
p\le q \quad\Longleftrightarrow\quad d_\tau(p,q)=\tau(q)-\tau(p).
\]
The implication \(p\le q \Rightarrow d_\tau(p,q)=\tau(q)-\tau(p)\) is immediate from the definition; the converse is the substantive issue. It can fail for general time functions, even on causally well-behaved spacetimes, so global causal hypotheses on the level sets of \(\tau\) are essential [2306.17128] [2209.15610].

A first global result, due to Sakovich–Sormani, states that if \(\tau\) is a proper, locally anti-Lipschitz time function—equivalently, all level sets \(\{\tau=c\}\) are compact Cauchy hypersurfaces—then \(d_\tau\) encodes causality on all of \(M\). Burtscher–García-Heveling proved a related theorem: if \((M,g)\) is globally hyperbolic and \(\tau\) is locally anti-Lipschitz with all level sets Cauchy, then the same equivalence holds globally [2306.17128].

Galloway strengthened these assumptions by showing that Cauchy level sets can be replaced by the strictly weaker condition of future causal completeness. If each level set \(S_c=\{\tau=c\}\) is future causally complete, then \(S_c\) is in fact a future Cauchy hypersurface, \(M\) is globally hyperbolic, and \(d_\tau\) encodes causality. An immediate corollary is that compact level sets already suffice to obtain both global hyperbolicity and global causality encoding [2306.17128].

Null distance also yields an existential metric characterization of global hyperbolicity:
\[
(M,g)\text{ is globally hyperbolic}\quad\Longleftrightarrow\quad \exists\,\tau \text{ such that }(M,d_\tau)\text{ is complete}.
\]
This is a metric-space analogue of Hopf–Rinow that is specific to null distance. The statement is existential rather than universal: the same work exhibits examples where a genuine Cauchy temporal function does not yield a complete null metric, so completeness depends on the choice of \(\tau\) [2209.15610].

These results correct two common misunderstandings. First, not every time function makes \(d_\tau\) a metric; local anti-Lipschitzness is exactly the missing condition. Second, not every causally well-behaved choice of \(\tau\) encodes causality globally; the level-set hypotheses are sharp in the sense that non-Cauchy choices can fail even on globally hyperbolic spacetimes [2209.15610].

## 4. Distinguished time functions and geometry of level sets

Regular cosmological time is a particularly important choice of \(\tau\). If
\[
\tau_{\Cos}(x)=\sup\{d_L(y,x):y\ll x\}
\]
is finite everywhere and tends to \(0\) along every past-inextendible causal curve, then Andersson–Galloway–Howard’s theory implies that \(\tau_{\Cos}\) is continuous, locally Lipschitz, locally anti-Lipschitz, and has
\[
\nabla\tau_{\Cos}\cdot\nabla\tau_{\Cos}\le -1
\]
almost everywhere; in the smooth regular setting one has \(g(\nabla\tau_g,\nabla\tau_g)=-1\). Consequently its null distance is a bona fide metric inducing the manifold topology, and in the globally hyperbolic theory it encodes causality globally [1508.00531] [2209.15610] [2511.05847].

Another canonical source of null distance is the surface function associated to a \(C^1\) Cauchy hypersurface \(S\subset M\) in a globally hyperbolic spacetime:
\[
\tau_S(x)=
\begin{cases}
d_L(S,x),&x\in I^+(S),\\
0,&x\in S,\\
-d_L(x,S),&x\in I^-(S).
\end{cases}
\]
This function is continuous, locally Lipschitz, locally anti-Lipschitz, and satisfies
\[
\nabla\tau_S\cdot\nabla\tau_S=-1
\]
wherever the gradient exists. Hence \(d(\cdot,\cdot;\tau_S)\) is again a metric inducing the manifold topology [2511.05847].

For smooth temporal functions there is also a comparison between null distance and the intrinsic Riemannian geometry of level sets. If \(f\) is smooth and temporal, \(M_t=f^{-1}(t)\) is a regular spacelike level set with induced metric \(h_t\), and
\[
g(\nabla f,\nabla f)=-C^2
\]
with \(C>0\) constant along \(M_t\), then
\[
\widehat d_f(p,q)\le C\, d_{h_t}(p,q),\qquad p,q\in M_t.
\]
Applied to a smooth regular cosmological time \(\tau_g\), for which \(\|\nabla\tau_g\|_g\equiv 1\), this becomes
\[
\widehat d_{\tau_g}(p,q)\le d_{h_t}(p,q).
\]
This estimate confirms a conjecture of Sakovich–Sormani: if the Riemannian diameters of the level sets \(\tau_g^{-1}(t)\) shrink to zero as \(t\to0^+\), then the extended initial slice \(\overline{\tau_g}^{-1}(0)\) in the metric completion of \((M,\widehat d_{\tau_g})\) consists of exactly one point \(p_{BB}\) [2507.07158].

## 5. Local geometry, rectifiability, and rigidity

Recent work studies null distance not only as an abstract metric but as a locally controlled geometric structure. Uniform Temple charts provide local future-null coordinate systems with optical functions \(\omega_q\) and radial functions \(\lambda_q\) around each point. In a uniform Temple neighborhood \(U_p\), the corresponding Temple maps are homeomorphisms onto sets containing \(U_p\), and the optical functions satisfy uniform gradient bounds with respect to the Riemannianized metric
\[
g_R=g+2\,g(e_0,\cdot)^2,
\]
including
\[
\bigl|\nabla^{g_R}\omega_q\bigr|_{g_R}=\sqrt 2+O(\lambda_q).
\]
This yields uniform Lipschitz control of the local causal geometry [2509.11281].

When \(\tau\) is Lipschitz and locally anti-Lipschitz, these charts imply that \((N,\hat d_\tau)\) is a countably rectifiable metric space. In particular it carries an \((n+1)\)-dimensional Hausdorff measure \(\mathcal H_{\hat d_\tau}^{n+1}\), and the singular set of \(\tau\) has zero measure. The same framework gives a local causality-encoding theorem: around every point \(p\) there is a neighborhood \(U_p\) such that for all \(q,q'\in U_p\),
\[
\hat d_\tau(q,q')=\tau(q')-\tau(q)\quad\Longleftrightarrow\quad q'\in J^+(q).
\]
This is a local reconstruction statement with no global causal-completeness assumption [2509.11281].

Rigidity results show that the null-distance data can determine the Lorentzian metric. For spacetimes of dimension \(n+1\ge 2\) endowed with proper regular cosmological time functions, any bijection preserving both \(\tau\) and \(\hat d_\tau\) is a Lorentzian isometry. A generalized version holds in dimension \(n+1\ge 3\) for Lipschitz time functions satisfying \(|\nabla^{g_i}\tau_i|_{g_i}=1\) almost everywhere, even without a global causality-encoding hypothesis [2208.01975] [2509.11281].

## 6. Convergence theory, model examples, and present directions

One of the original motivations for null distance was to import metric geometry into Lorentzian settings. In Lorentzian length spaces, null distance is compatible with the underlying topology under the same anti-Lipschitz criterion, and for generalized cones \(Y=(I\times X,f)\) with \(\tau(t,x)=t\), uniform convergence \(f_j\to f\) with \(f_j\ge c>0\) implies uniform convergence of the corresponding null distances on bounded balls. If \((X,d)\) is proper, this yields pointed Gromov–Hausdorff convergence; if \(I\) and \(X\) are compact, it yields ordinary Gromov–Hausdorff convergence. In the same framework, timelike curvature lower bounds persist under null-distance GH limits [2106.05393].

For warped product spacetimes, Allen proved optimal null-distance convergence theorems under \(L^1\)-type control and monotone lower bounds on the warping functions. Uniform convergence of warping functions gives uniform convergence of the null distances and therefore both Gromov–Hausdorff and Sormani–Wenger intrinsic-flat convergence. The counterexamples are equally important: if the monotone lower bound or the \(L^1\)-control is dropped, one can obtain smaller-than-expected limits, taxi-type limit metrics, bubbling, or spline-like nonmanifold limits [2306.03165] [1909.04483].

Static spacetimes admit a further extension through a weighted null distance \(\hat d_{w,\tau,g}\). In that setting, a VADB-type theorem gives uniform and Gromov–Hausdorff convergence of null-distance spaces under an \(L^p\)-bound on the rescaled spatial metrics, monotone-from-below comparison, volume convergence, and a uniform boundary-area bound. A conjectural SWIF version has also been formulated for compact globally hyperbolic spacetimes [2510.02237].

Model examples remain instructive. In Minkowski space with \(\tau=t\), the null distance admits explicit formulas; in product form one obtains
\[
d_1\bigl((t,x),(t',x')\bigr)=\max\{|t-t'|,d(x,x')\},
\]
equivalently “spatial distance plus any excess time-separation.” In Minkowski space with one point removed, time slices \(t=t_0<1\) are future causally complete but not future Cauchy, so Galloway’s weakened theorem applies while the older Cauchy-level-set theorem does not. Other examples show that changing the time function can radically change the large-scale metric behavior: on \(\mathbb R^{1,n}\), \(d_t\) is complete whereas \(d_{e^t}\) is not [2106.05393] [2306.17128] [2209.15610].

Current directions, as explicitly identified in the literature, include extending the theory to lower-regularity spacetimes or merely continuous metrics, characterizing geodesics of \(d_\tau\) and their relation to Lorentzian geodesics, adapting null distance to spacetimes with boundary or weaker causality conditions, studying metric completeness of \((M,d_N)\), and using null distance in convergence and stability problems in mathematical relativity [2306.17128] [2511.05847]. A plausible implication is that null distance has become a unifying device: it packages causality, time functions, and conformal information into a metric structure that can be compared, completed, and passed to limits using the tools of modern metric geometry.

Source: https://www.emergentmind.com/topics/null-distance