Lonely runner conjecture: newer results and the limits of numerical grids

The lonely runner conjecture asks a precise question about separation on a circular track. Its appeal is that the statement is accessible while the proof problem is difficult. But two distinctions are essential: a numerical experiment is not automatically a proof, and the frontier reported in an older explanation is not a permanent mathematical boundary.

The earlier version of this article repeatedly said the conjecture was established only for seven runners. That was already inconsistent with the April reporting it cited. Subsequent primary papers provide further dated developments, so repeating that boundary would misinform readers about both the news and the mathematics.

State the problem with one counting convention

Let N runners start together on a track of circumference one, each with a different constant speed. The conjecture says that every runner has some time at which their circular distance from every other runner is at least 1/N. Equality is allowed. The time may differ from runner to runner.

The circular distance is the shorter distance around the loop. Positions zero and nine-tenths, for example, are one-tenth apart, not nine-tenths apart. This is why ordinary subtraction without reducing positions around the circle gives the wrong test.

To follow one runner, subtract that runner's speed from all the speeds. The chosen runner becomes stationary at zero. For every other relative speed u, the condition is that the distance from tu to the nearest integer is at least 1/N. Some papers call the number of moving relative speeds k, making N equal to k+1. Comparing a k in one formulation with an N in another can create an apparent discrepancy of one runner.

The research record extends beyond seven

The April WIRED account, reprinted from Quanta, described the advance to eight, nine and ten runners. The primary paper Nine and ten lonely runners, by Tanupat Trakulthongchai, presents computer-assisted proofs building on Rosenfeld's approach.

Sungkawichai and Trakulthongchai's subsequent paper, first submitted in April 2026 and revised in September, presents computer-assisted proofs for eleven, twelve and thirteen runners. Its abstract uses k moving speeds and k+1 total runners, matching the distinction above.

Jaan Allikvere's September 24 revision, titled Fourteen and fifteen lonely runners, presents proofs of those two cases and links public code and certificates. This is a description of the identified preprint and version, not a claim that UnliveNews independently checked the full computation or established the conjecture for every N.

Finite cases remain different from a general theorem. Extending a method to another specified number does not, by induction alone, prove every larger case. The mechanism that justifies an induction step would itself need a proof.

An example can be checked exactly

Take three runners with speeds zero, one and two. At time one-third their positions are zero, one-third and two-thirds. Every pair has circular distance one-third. Thus each runner meets the required threshold at that time for this particular speed tuple.

This calculation is an exact witness: the fractions can be checked without rounding. It proves the property for the example. It does not prove the three-runner case for every possible choice of speeds, much less the unrestricted conjecture.

For integer relative speeds, the configuration repeats after one unit of time. That bounds the time interval needed to analyze that tuple. It does not transform a finite set of sampled times into an exhaustive analysis of the interval between samples.

Why a fine numerical grid can miss equality

Consider just two runners with relative speed one. Their distance reaches the required one-half exactly at time one-half modulo one. A grid whose points do not include that time can fail to reach the threshold, however close some samples come.

That failure is not a counterexample. It is a limitation of the sample. Floating-point rounding can also make a value appear slightly above or below an exact boundary. A plotted maximum therefore needs a different evidentiary description from a verified rational witness.

For a fixed rational tuple, exact interval reasoning can describe where each distance condition holds and intersect those intervals. A rigorous finite computation requires justified coverage, exact arithmetic or controlled error, and a reason omitted cases cannot matter. Calling a grid sufficiently dense does not supply those arguments.

Computer assistance needs a mathematical bridge

The interesting research development is not simply that computers ran more laps. The identified papers combine mathematical restrictions with finite computations. The restrictions explain why checking a particular collection of cases can establish more than a collection of examples.

This also limits the earlier article's engineering analogies. An idealized common-start, constant-speed model is not a guarantee for a real network or scheduler with changing rates, offsets and delays. The model is valuable on its own terms: a clean question about simultaneous separation, a growing set of dated results and a clear demonstration that experimentation and proof answer different questions.