Time-Constrained Distance in Temporal Metrics

• Time-constrained distance is a framework where temporal rules are embedded into distance definitions, ensuring only time-admissible paths or events contribute to the computed metric.
• The methodology spans diverse domains such as Lorentzian geometry, transportation, temporal graphs, timed automata, and phase-change memory by applying unique max-min and sliding-window strategies.
• Key implications include establishing metric definiteness via local anti-Lipschitz conditions, optimizing worst-case route times, and enabling rigorous analyses of collision prediction and system conformance.

In the supplied literature, time-constrained distance is not a single standardized invariant but a family of constructions in which temporal structure restricts how separation is measured. The temporal ingredient may be a time function on a spacetime, a departure-time window on a transportation system, waiting-time bounds in a temporal graph, timestamps in a timed trace, or a sliding rewrite window in a memory array. In each case, the central operation is similar: admissible objects are first filtered by a temporal rule, and only then is an infimum, supremum, or metric-like quantity formed (Nigri, 9 Jul 2025, Halpern, 2015, Carnevale et al., 12 Feb 2026, Rosenmann, 2019, Analooee et al., 2019, Qin et al., 2012).

1. Comparative framework

Setting	Basic object	Distance principle
Lorentzian geometry	Piecewise-causal curve	Infimum of null length
Transportation	Route over departure time $t\in P$	Max over $t$ of quickest-route times
Temporal graphs	Temporal path	Optimize EA, LD, FT, ST, MH, or MW
Timed automata	Timed trace / timed language	Sup-inf of timestamp deviation
Collision prediction	Moving objects in $(x,y)$	Time until contact
Phase-change memory	Rewrite sequence	Sliding-window cumulative Hamming cost

A common pattern is that the temporal rule is not an accessory constraint added after the fact. It is part of the definition of the quantity itself. In Lorentzian geometry, only piecewise-causal curves contribute to $\widehat d_\tau$. In transportation, route times are first minimized at fixed departure time and then maximized over the observation window. In timed automata, traces of different untimed shape are placed at distance $\infty$. In phase-change memory, the relevant “distance” is cumulative rewrite cost over every time window of length $\alpha$ and every spatial window of $\beta$ cells (Nigri, 9 Jul 2025, Halpern, 2015, Rosenmann, 2019, Qin et al., 2012).

The literature also separates genuine metrics from more general temporal costs. The null distance $\widehat d_\tau$ is in general only a semi-metric, but becomes a metric exactly under the local anti-Lipschitz condition. Halpern’s travel-time function is constructed to satisfy the metric axioms. By contrast, the six temporal distances in D-TMB are optimization objectives on temporal paths, and the TD collision framework defines a time-to-contact quantity that may take value $+\infty$ (Nigri, 9 Jul 2025, Halpern, 2015, Carnevale et al., 12 Feb 2026, Analooee et al., 2019).

2. Null distance, temporal functions, and Lorentzian causality

For a spacetime $(M,g)$ and a continuous time function $t$0, the null distance is defined from piecewise-causal curves. If $t$1 is piecewise causal, with partition $t$2, its null length is

$t$3

and the induced null distance is

$t$4

Equivalently,

$t$5

For any piecewise-causal $t$6 from $t$7 to $t$8, one has $t$9, hence $(x,y)$0. In particular, $(x,y)$1 implies $(x,y)$2 (Nigri, 9 Jul 2025).

Definiteness is controlled exactly by the local anti-Lipschitz condition. A time function $(x,y)$3 is locally anti-Lipschitz if for every $(x,y)$4 there is a neighborhood $(x,y)$5, a Riemannian metric $(x,y)$6 on $(x,y)$7, and a constant $(x,y)$8 such that

$(x,y)$9

The associated proposition states that $\widehat d_\tau$0 is definite, hence a true metric, if and only if $\widehat d_\tau$1 is locally anti-Lipschitz; when this holds, $\widehat d_\tau$2 induces the manifold topology (Nigri, 9 Jul 2025).

Temporal functions provide a particularly important class. A $\widehat d_\tau$3 function $\widehat d_\tau$4 is temporal if $\widehat d_\tau$5 is everywhere past-directed timelike. Sormani and Vega showed that the class of $\widehat d_\tau$6 temporal functions coincides with that of $\widehat d_\tau$7 locally anti-Lipschitz time functions. On a regular level set

$\widehat d_\tau$8

the induced metric

$\widehat d_\tau$9

is Riemannian. If $\infty$0 is smooth and $\infty$1 on $\infty$2, then the main comparison theorem gives

$\infty$3

The proof outline uses normal coordinates, a local Minkowski model, two null segments through an “$\infty$4-slim” cone, and a Lebesgue-number-lemma/chaining argument (Nigri, 9 Jul 2025).

The cosmological-time application turns this comparison into a singularity theorem. Cosmological time is defined by

$\infty$5

When $\infty$6 is regular, Andersson, Galloway, and Howard showed that it is a locally anti-Lipschitz time function, that $\infty$7 exists almost everywhere and is past-directed timelike with unit norm, and that through each $\infty$8 there is a unique future-directed unit-speed geodesic generator $\infty$9 with $\alpha$0. Since $\alpha$1, one obtains

$\alpha$2

If $\alpha$3 as $\alpha$4, then the initial singularity $\alpha$5 is a single point $\alpha$6, confirming a conjecture of Sakovich and Sormani (Nigri, 9 Jul 2025, Sakovich et al., 2024).

A common misconception is that null distance is automatically metric because it is defined by an infimum over curve lengths. The cited result rules this out: without local anti-Lipschitz control, $\alpha$7 may vanish for distinct points (Nigri, 9 Jul 2025).

3. Travel-time metrics and collision-time fields

In transportation, Halpern defined a travel-time metric on a point-set $\alpha$8 over a closed time-interval $\alpha$9. For each ordered pair $\beta$0 and departure time $\beta$1, $\beta$2 is the set of admissible routes from $\beta$3 to $\beta$4, each route $\beta$5 carrying an empirical travel time $\beta$6. The consistency axioms are identity, positivity, existence, and composition. The quickest-route travel time at fixed $\beta$7 is

$\beta$8

The metric is then defined by the max-min formula

$\beta$9

Its construction is explicitly designed to satisfy non-negativity, identity of indiscernibles, symmetry, and the triangle inequality; the latter is derived from the time-shifted composition law

$\widehat d_\tau$0

whenever the departure and shifted arrival times remain in $\widehat d_\tau$1 (Halpern, 2015).

The max-min step is essential. A pure min-min aggregate over departure times can violate the triangle inequality, as shown in the paper’s three-node example with $\widehat d_\tau$2 h and $\widehat d_\tau$3 min. The max-min construction restores metric behavior by taking the worst quickest-route time in each direction (Halpern, 2015).

A different use of temporal distance appears in collision prediction and path planning. In the TD framework, the relative TD between two moving objects $\widehat d_\tau$4 and $\widehat d_\tau$5 is defined as “the time interval that must elapse for them to come into contact. If they never meet, TD = $\widehat d_\tau$6.” For a moving polygon $\widehat d_\tau$7, the TD to a point $\widehat d_\tau$8 is the minimum over its edge TD functions; for a moving circle,

$\widehat d_\tau$9

The configuration-time field is assembled from obstacle TD functions and $+\infty$0 penalties: $+\infty$1 A Route Function

$+\infty$2

is then used to carve out a safe corridor $+\infty$3, while collision warning is based on

$+\infty$4

Here the quantity is operational rather than metric: it is a time-to-contact field for dynamic free-space construction and collision alarms (Analooee et al., 2019).

4. Temporal graphs, waiting-time constraints, and broadcast distances

A temporal graph may be presented as $+\infty$5, where $+\infty$6 is a finite vertex set, $+\infty$7 is the lifetime, and $+\infty$8 is a set of time-stamped edges. A temporal walk

$+\infty$9

requires $(M,g)$0 and $(M,g)$1, and it is a temporal path if the vertices are pairwise distinct. The $(M,g)$2-restless variant adds the waiting-time bound

$(M,g)$3

RESTLESS TEMPORAL PATH is NP-complete already for any fixed $(M,g)$4, even when $(M,g)$5 and each edge appears exactly once. It is W[1]-hard parameterized by the vertex-deletion distance of the underlying graph to a disjoint union of paths, and therefore W[1]-hard parameterized by the feedback vertex number or the pathwidth. Positive results are obtained for path length, feedback-edge number, and timed feedback vertex number (Casteigts et al., 2019).

A broader optimization framework is D-Temporal Multi-Broadcast (D-TMB), which studies six temporal distances on temporal paths $(M,g)$6:

Distance	Definition
EA	$(M,g)$7
LD	$(M,g)$8
FT	$(M,g)$9
ST	$t$00
MH	$t$01
MW	$t$02

For $t$03, the objective is to minimize $t$04; for $t$05, it is to maximize $t$06. For a single source, EA-TMB and LD-TMB are in polynomial time, while FT, ST, MH, and MW are NP-complete and hard to approximate. For multiple sources, if feasibility is not assumed a priori, then no finite-factor approximation exists unless $t$07, even with just two sources. Feasibility itself is NP-complete for $t$08, $t$09, and $t$10 (Carnevale et al., 12 Feb 2026).

The tractable cases are structurally specific. If $t$11 for every edge, then EA-TMB and LD-TMB are in polynomial time for arbitrary $t$12. On a tree with $t$13 on all $t$14-to-$t$15 paths, EA-TMB and LD-TMB can also be solved in polynomial time by merging single-source solutions (Carnevale et al., 12 Feb 2026).

A plausible implication is that waiting-time constraints and temporal-distance objectives alter complexity for different reasons. In restless paths, the combinatorial difficulty comes from bounding local waiting. In D-TMB, the difficulty comes from choosing labels so that a global worst-case temporal distance is optimized (Casteigts et al., 2019, Carnevale et al., 12 Feb 2026).

5. Timed automata and conformance distance

For timed automata, the relevant distance is defined on timed traces and then lifted to languages. A finite timed trace over $t$16 is

$t$17

with $t$18 and $t$19. The max-metric on timed traces is

$t$20

For timed languages $t$21, the conformance distance is

$t$22

and the symmetrized distance is

$t$23

One has $t$24, and if $t$25, then $t$26 in the Euclidean topology (Rosenmann, 2019).

Rosenmann’s construction digitizes nondeterministic timed automata into deterministic discretized automata. Starting from an augmented region automaton $t$27 with an absolute-time clock $t$28, one builds a discretized timed automaton $t$29 over the time domain $t$30, with one clock $t$31 reset at every transition. If a region edge has weight $t$32 and the source and target fractional parts induce $t$33, the corresponding transition in $t$34 has guard $t$35 or $t$36. The resulting approximation satisfies

$t$37

After the $t$38 speed-up used in the paper’s inclusion-gap theorem, if $t$39 and $t$40, then

$t$41

Hence

$t$42

Under a bounded-cycle restriction, the paper also gives a decision procedure for whether the distance is infinite by analyzing untimed witnesses and positive-delay simple cycles in a time-product construction (Rosenmann, 2019).

The critical distinction here is between exact language inclusion and quantitative conformance. The distance formalism does not merely ask whether an implementation is accepted by a specification; it asks how far each timed trace can be from the closest matching trace when exact inclusion fails (Rosenmann, 2019).

6. Windowed rewrite distance and adjacent delayed-distance notions

In phase-change memory, the temporal restriction is imposed on cumulative rewrite cost. A binary memory state at time $t$43 is $t$44, and the rewrite cost from $t$45 to $t$46 is the Hamming distance

$t$47

A sequence satisfies the $t$48-constraint if, for every time window of length $t$49 and every spatial window of $t$50 contiguous cells,

$t$51

The paper explicitly interprets this as a time-constrained distance on sequences: $t$52 with the requirement $t$53 on every length-$t$54 segment. This is not a single-step Hamming constraint; it is an aggregated sliding-window constraint in both time and space. The general upper bound is

$t$55

and one-dimensional special cases reduce to window-weight-limited constraints such as

$t$56

The paper also gives explicit lower-bound constructions, including the trivial block-partitioning rate $t$57 (Qin et al., 2012).

A related but distinct delayed-distance notion appears in time-delay cosmography. In a strong lens system, the observed time delay $t$58 between two images satisfies

$t$59

with

$t$60

This is not a metric on points or traces. It is a cosmological distance inferred from observed delays. Under the mass-sheet transformation $t$61, the Fermat potential difference scales as $t$62, so the inferred time-delay distance rescales as

$t$63

The cited work shows that this degeneracy can be broken in a cosmology-model-independent way by combining lensing with SN Ia and BAO to constrain $t$64, yielding model-independent distance measurements in time-delay cosmography under mass-sheet transformations (Chen et al., 2020).

Taken together, these constructions show that time-constrained distance can denote a metric, a semi-metric, a worst-case route functional, a conformance gap, a time-to-collision field, or a sliding-window cost. The unifying feature is not a single formula but a recurrent research strategy: temporal admissibility is built into the distance definition itself, and both geometry and complexity change as a consequence.