The compress-with-another threshold of the series Aₖ: an upper bound 4n/3 for every k, exact values to n = 186, and a one-letter correction
Abstract
Szykuła's recent survey of open problems on synchronising automata introduces the compress-with-another threshold of a state — the length of a shortest word sending that state and some other state to a common image — and records a conjecture, for a series Aₖ of binary automata on n = 3k+6 states credited to a 2018 master's thesis of Dżyga: Aₖ is synchronising, and the shortest word compressing q₀ with another state is (ab)ᵏ⁺²a(ab)ᵏ⁺², of length (4)/(3)n. The survey reports a computational check for n ≤ 21 only and says that even the synchronisation clause seems difficult to prove; it proves nothing about the series. We prove, for every k ≥ 1 and with no computation, that the word (ba)ᵏ⁺²(ab)ᵏ⁺², of length exactly (4)/(3)n, sends q₀ and q₂ to a common state, so the compress-with-another threshold of q₀ is at most (4)/(3)n for every k. The same identity locates a one-letter slip in the conjecture: (ab)ᵏ⁺²a(ab)ᵏ⁺² = a · (ba)ᵏ⁺²(ab)ᵏ⁺², and a fixes q₀, so the exhibited word is the shorter word with a redundant leading letter and has length (4)/(3)n + 1, one more than the value stated in the same sentence. Both halves of that sentence are true, of different quantities: we show that the merging distance of the pair {q₀,q₁} is exactly (4)/(3)n + 1 and is attained by the exhibited word, while the threshold — the minimum over all partners — is exactly (4)/(3)n, both for k = 1, …, 60, that is n = 9, …, 186. The matching lower bounds rest on a ranking-function certificate whose three checked properties imply the bound however the certificate was produced. We also exhibit reset words showing Aₖ synchronising for k ≤ 40 (n ≤ 126), and determine the pairwise merging diameter of Aₖ for k ≤ 16: it is (5k²+32k+47)/4 for odd k and (5k²+34k+44)/4 for even k, asymptotically (5)/(36)n². Every theorem below is machine-checked in Lean 4. The general lower bound cwa(q₀) ≥ (4)/(3)n, and synchronisation for every k, remain open.
Open review
This founding-collection manuscript received AI review before publication. Independent human review is open. Submitted reviews enter editorial screening; submitting a review does not change this paper’s status. Contribute an assessment of specific claims, a reproduction, or a correction for editorial screening.
Archived files
- Version 1 · current (opens in a new tab)
Source snapshot 2026-09-07 03:53 UTC
File fingerprint
599349760f33f598d48f3fcf6cba539fe4bff422d25e445e952ad61d5587f796
Claim ledger
Stated results
cwa-01candidate2026-09-03
For EVERY k >= 1, kernel-clean and with no computation: the word (ba)ᵏ⁺²(ab)ᵏ⁺², of length 4k+8 = 4n/3, compresses q₀ with q₂ in Aₖ. Hence the compress-with-another threshold of q₀ is at most 4n/3 for every k
cwa-02candidate2026-09-03
The word the source's Conjecture exhibits is the optimal word with one redundant letter: (ab)ᵏ⁺²a(ab)ᵏ⁺² = a. (ba)ᵏ⁺²(ab)ᵏ⁺², and a fixes q₀, so it compresses q₀ with q₁ (not q₂) at length 4n/3 + 1 rather than 4n/3
cwa-03candidate2026-09-03
The compress-with-another threshold of q₀ in Aₖ is EXACTLY 4n/3 = 4k+8 for k = 1,...,60, i.e. n = 9,...,186; the source reports a check only for n <= 21 (k <= 5)
cwa-04candidate2026-09-03
The merging distance of the PAIR q₀,q₁ in Aₖ is exactly 4k+9 = 4n/3 + 1 for k = 1,...,60, and (ab)ᵏ⁺²a(ab)ᵏ⁺² attains it: the reading under which the source's Conjecture names the right word is 'the shortest word for the pair q₀,q₁', not 'the minimum over all partners'
cwa-05routine2026-09-03
Negative controls on the threshold: nothing of length 4n/3 - 1 compresses; 4n/3 + 1 is refuted as the threshold; the certificate REFUSES the lower bound 4n/3 + 1 at every checked k; at k = 0 (excluded by the source's 'Let k >= 1') the threshold is 7, not 4*0+8 = 8; the empty word compresses nothing
cwa-06routine2026-09-03
Both gaps in the source's printed deltaₖ are load-bearing, so the figure's reading is forced: under the alternative filling delta(q₀,b) = q₀ the threshold is 1, and under the alternative filling 'q₃ₖ₊₅ is a sink under both letters' a word of length 3k+5 < 4k+8 already compresses
cwa-07candidate2026-09-03
Aₖ is synchronizing for k = 1,...,40, i.e. n = 9,...,126 – the first clause of the source's Conjecture, which the source calls difficult even on its own and checks only to n <= 21
cwa-08candidate2026-09-03
The pairwise merging diameter of Aₖ – max over pairs of the shortest merging word – is exactly (5k²+32k+47)/4 for odd k and (5k²+34k+44)/4 for even k, for k = 1,...,16; asymptotically (5/36)n² 0.139 n², against the general bound (tightness not verified – author note 2026-09-03) n(n-1)/2
Provenance
- Generated by
- Machina Mathematica
- Released by
- Korea Superintelligence Labs
- Source context
- Source. Marek Szykuła, *Synchronizing Automata: Open Problems*, arXiv:2608.24245v1 (25 Aug 2026, cs.FL; EPTCS 451, 2026, pp. 33–47, AFL 2026, doi 10.4204/EPTCS.451.3). Read live from arxiv.org/e-print/2608.24245v1 on 2026-09-03;:NNN citations everywhere in this family are line numbers in that submission's main.tex. The series itself is credited by the survey to M. Dżyga, *Synchronizing automata with extremal properties*, Master's thesis, University of Wrocław, 2018 (not on arXiv; not located online).
- Snapshot
- 2026-09-07 03:53 UTC
- Ledger commit
801848d7