Explicit witnesses Unrefereed

Zaremba's conjecture at bound 5 for powers of 7

No dated manuscript catalogued; standalone verification archive linked above. On this site since · last revised · Verify & co-author
Kind
Computation (explicit witnesses, checked by exact Euclid), with a proved doubling lemma
Status
Witnesses checked by exact arithmetic and held by a falsifier agent; unrefereed
Prior work
Komatsu (2005) has odd parts 1 to 11; bound 6 for every power of 7 is Shulga's (2023); bound 5 beyond odd part 11 is not in the sources read
On this page
  1. The statement
  2. Check it yourself
  3. Worked examples
  4. Status and who checked it
  5. Prior work
  6. Further reading

The statement

Zaremba's conjecture says that every qq has an aa, coprime to qq, for which all the quotients of Euclid's algorithm on (a,q)(a,\allowbreak q) (the partial quotients of a/qa/q) are at most 5. Bound 6 is known for every power of 7; bound 5 for every power of 7 is open.

Result. Bound 5 holds for every 7n7^n whose exponent nn has odd part at most 41, and for n=43n = 43. The first open exponent is 45.

Two proved facts turn finitely many witnesses into infinitely many exponents. Call a witness S1 when its quotient sequence begins and ends with a quotient from 2 to k−1k - 1, at bound kk.

  • Doubling. An S1 numerator of 7c7^c at bound kk, for kk from 3 to 5, gives a numerator of 7c⋅2j7^{c \cdot 2^j} at bound kk for every j≥0j \ge 0. So one S1 numerator per odd part settles every exponent with that odd part.
  • "S1 is kept" under the folds used to build longer witnesses.

Check it yourself

S1 numerators aa at bound 5 for 7n7^n, nn odd (run Euclid on (a,7n)(a,\allowbreak 7^n) and every quotient is at most 5):

nn numerator aa number of quotients
1 2 2
3 72 5
5 3469 9
7 187440 13
9 8453376 22
11 410505146 23
13 25426309987 32
15 995852075430 38
17 73173037382640 41
19 4402150398387787 46
21 207554712365340118 51

For example, 2/7=[0;3,2]2/7 = [0; 3,\allowbreak 2] and 72/73=[0;4,1,3,4,4]72/7^3 = [0; 4,\allowbreak 1,\allowbreak 3,\allowbreak 4,\allowbreak 4]. The witnesses for odd parts 23 to 41, and the 47 witnesses for n=43n = 43, are longer and are listed in the same way in the source.

Worked examples

How the odd parts were covered.

  • Odd parts 1 to 11: Komatsu (2005), whose 11 printed expansions were recomputed and match.
  • Odd parts 13 to 21: numerators found by search (the table above).
  • Odd parts 23 to 29: S1 numerators from a near-fold probe.
  • Odd parts 31 to 41: S1 numerators built as two-step fold lines.
  • n=43n = 43: 47 numerators, folds by 797^9 and 777^7 of middle words of 7177^{17} and 7187^{18}, each checked twice. None is S1, so n=86n = 86 is not settled by doubling.

At smaller bounds, the same way: bound 4 holds for odd part at most 15 (first open n=17n = 17), bound 3 for odd part at most 13 (first open n=15n = 15), and bound 2 fails at 777^7, as Komatsu states.

Status and who checked it

  • Checked by computation, exactly. Every numerator is checked by Euclid's algorithm on a/7na/7^n in exact integers. The n=43n = 43 numerators were checked by the job that found them, by a separate script, and again by a falsifier agent's own program.
  • Proved. The folding lemma, "S1 is kept" and doubling, with their proofs held by a falsifier agent.
  • Open. Bound 5 for every power of 7 is one infinite statement; each single exponent, 45 included, is a finite search that settles that exponent only. Unrefereed.

Prior work

Credited. Komatsu (2005) gives bound 3 for 7c⋅2n7^{c \cdot 2^n} with cc in {1,3,…,11}\{1,\allowbreak 3,\allowbreak \ldots,\allowbreak 11\}. Elias Dubno (2024, Amer. Math. Monthly / arXiv:2310.11279) states and proves the S1 boundary condition (1<a1,an<A1 < a_1,\allowbreak a_n < A), repeated doubling, and preservation of S1 under folds (applied there to powers of 12 and 18). Shulga (2023/2024, Theorem 1.2, arXiv:2310.09801) gives bound rad⁡(q)−1\operatorname{rad}(q) - 1 for every qq other than 2n2^n and 3n3^n, yielding bound 6 for every power of 7. The folding lemma is classical (Mendès France 1973). For almost every qq, Huang proved bound 5, after Bourgain and Kontorovich's bound 50.

Searches run. A literature search and a 2026-11-22 Google Scholar sweep (5 queries) found no source that settles bound 5 for powers of 7 beyond Komatsu's odd parts 1 to 11; Ayadi and Komatsu (2024) add nothing for powers of 7 in the abstract and secondary summaries read. Bound 5 beyond odd part 11 is not found in the searches logged.

Still owed. Full-text verification of paywalled Ayadi and Komatsu (2024, Experimental Mathematics), and checking a 2026 preprint by Zhang (arXiv:2605.02518), which claims Zaremba's conjecture with a large absolute constant (leaving constant 5 open if it stands).

Further reading