A machine that performs one task each day may need to revisit several tasks indefinitely. Pinwheel scheduling asks whether all required revisit frequencies can be met by a single daily assignment. The requirement concerns every consecutive block of days: a good long-run average alone does not prevent a task from waiting too long. This distinction makes the problem relevant to the mathematical study of perpetual service and maintenance, rather than just finite job sequencing. The source submission, Section 1, places pinwheel scheduling among perpetual scheduling problems and explains its relation to machine maintenance.
For integer periods, the density threshold of five sixths was conjectured by Chan and Chin and proved by Kawamura using computer assistance; the source records both the original STOC 2024 result and its 2026 full version. Allowing real periods changes the service condition: each task must satisfy requirements at all positive multiplicities. Fujiwara, Miyagi and Ouchi studied this model and conjectured the same threshold in DMTCS 2026. The fixed target here is the real-period theorem with a quantitative bound on schedule length, as stated in the October 2026 submission by Fujiwara, Gutowski, Ouchi and Wolff. The submission label does not assert acceptance.
A real-period instance is a tuple of arbitrary real numbers with and . Task identity is its position in the tuple: equal periods still give distinct tasks. A schedule assigns exactly one task to every day. Here days start at zero, with day representing the source's day .
Let count executions of task on the integer days of . Validity means
The density is . The potential is , omitting one largest-period task. Potential is zero when . These quantities are defined from periods, independently of schedule existence.
A schedule is periodic with period when is a positive integer and for every day. A word is a nonempty finite sequence of task labels containing every task. Infinite repetition gives . Its dominance interval consists of the positive densities for which task satisfies every window requirement at period . The dominance box is ; its coordinates need not be ordered. These objects follow Sections 1 and 3 of the source.
The paper's two main results are the general bound, refTheorem 3, and the bound for minimum period at least four, refTheorem 4. The platform allows one goal, so Theorem 3 remains the goal and Theorem 4 is a milestone of equal standing.
The goal is Theorem 3, with both existential witnesses and the strict bound retained:
The targets supporting Theorem 3 are the source's six statements. Period monotonicity preserves the same valid schedule when one period increases. Unfolding preserves schedulability when a task of period is replaced by any positive number of distinct tasks of period , followed by reordering.
For the interval target, let , let count task in one copy, and let be its largest cyclic distance from an occurrence to its -th following occurrence. Equation (2) and its endpoint rule assert
The interval is if any gap term attains the minimum, and otherwise. Equivalently, positive belongs to it exactly when and every retained inequality holds.
The next three milestone targets are
Their provenance is Lemmas 5–9 and Section 3 of the source. The nine-task assumption bounds potential, permitting total density as large as .
The Section 5 milestones, with Lemma 12 proved in Appendix A, were added on 2026-11-15 so that the mission covers every theorem proved in the paper.
The theorem guarantees an infinite schedule with a finite representation whose length is explicitly bounded by the number of tasks. It covers irrational periods and the density boundary itself. A statement giving only an unspecified infinite schedule would leave this finite representation bound unestablished. The source's unschedulable integer instances have densities approaching from above, so its introduction also explains why a larger universal density guarantee is impossible. See Section 1 and Theorem 3.
The mathematical result is proved in the submission; the remaining work is a Lean proof, including the finite domination assertions and the quantitative periodicity obligations. The proposed development contributes an arbitrary-real window model, reusable task transformations, and an exact interface for cyclic gaps and dominance endpoints. The related integer/rational pinwheel declarations found in the existing library do not provide this real-period bounded-length result.
Checking average frequencies does not control empty blocks between executions. In the source's word AABACB, task A occupies half the days, but the consecutive letters CB prevent validity at period two. Its admissible density is strictly less than , while the other two coordinates include their upper endpoints. Even checking only the once-in-a-window requirement misses the higher-multiplicity requirements for real periods. These examples and the exact open faces of dominance boxes are documented in Section 3. A complete formal proof must therefore account for all multiplicities and for equality at each boundary.
The Lean namespace is RealPinwheel.FiveSixths. Instances use Fin n positions and ℝ periods, with nonemptiness, lower bound one and ordering in the instance structure. Schedules are total functions from natural days to task positions. Raw validity has no ordering requirement, permitting label-preserving period replacement. Tagged lists retain task origins through sorting and replication. Cyclic gaps include every starting occurrence and every wrap across a word boundary. Natural ceilings are used only on positive arguments in the mathematical statements.
Validity is the explicit all-window counting predicate: it is never defined through density, a certificate checker, coverage of a domain, or the theorem's conclusion. No rationality restriction or external certificate-success hypothesis is allowed.
The required infrastructure comprises finite counting, real arithmetic and ceilings, finite words, task relabelling, finite minima and maxima, and exact real-domain coverage. The paper's nine certificate families contain 14,287 words. Its Table 1 gives maximum lengths for dimensions one through nine. Section 2 supplies a factor at most two for each periodic unfolding; Section 4 uses the exact ratio . These are obligations of the known proof, not extra assumptions of the target. No planned target contains big-O or a failure probability, so there are no asymptotic constants to instantiate. Contributions establishing these facts in Lean are welcome; the independent theorem and its supporting branch are also part of this mission.
e7ef67a2bb85a8d0e82b92c0f835f71d06f5f99fef1b2f4fe36df8b6e83d60f3.namespace RealPinwheel.FiveSixths
/-- Theorem 3, printed p. 2 (PDF p. 3); proof printed p. 11 (PDF p. 12).
The length is a positive period, not necessarily the least one. -/
theorem five_sixths_periodic {n : ℕ} (A : Instance n)
(hD : density A ≤ (5 : ℝ) / 6) :
∃ (L : ℕ) (S : ℕ → Fin n),
0 < L ∧ L < 4 * 2 ^ n ∧
(∀ t : ℕ, S (t + L) = S t) ∧ Valid A.period S := by sorry
end RealPinwheel.FiveSixths
Let be any nonempty real-period pinwheel instance with . The periods may be irrational and repeated period values still label distinct tasks. If , then there exist a positive integer and a schedule assigning exactly one task to each day such that
and
The integer is a period of the schedule and need not be its least period. The result supplies both a valid infinite schedule and a strict dimension-dependent bound on a finite word representing it. Neither conclusion is conditional on certificate data.
Differences from the source:
No open leaves. Every sub-goal is proved or awaiting decomposition.