Conjecture Unrefereed

The floor-√3 map

No dated manuscript catalogued; standalone verification archive linked above. On this site since · last revised · Verify & co-author
Kind
Conjecture, checked by computer for every start up to 1.3×10101.3 \times 10^{10}
Status
Holds up under agent review; not proved; three smaller facts proved; unrefereed
Prior work
Not found in the searches logged; the rounded-up √3 map (OEIS A264789) and other relatives are in print
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

Take a positive whole number nn.

  • If nn is even, the next number is n/2n/2.
  • If nn is odd, the next number is ⌊n3⌋\lfloor n\sqrt{3} \rfloor: multiply by 3=1.7320508…\sqrt{3} = 1.7320508\ldots and drop the decimals.

1 is odd and ⌊3⌋=1\lfloor \sqrt{3} \rfloor = 1, so the game stops at 1.

Conjecture. Every positive start, followed long enough, reaches 1.

Two odd steps in a row multiply by 3\sqrt{3} twice, so they give exactly 3n−13n - 1 or 3n−23n - 2 (proved). The rule hides a 3n3n-type step, and the factor 3 never cancels against halving: the same clash of 2 and 3 that keeps Collatz open.

Check it yourself

  • Exact arithmetic. ⌊n3⌋\lfloor n\sqrt{3} \rfloor is the integer square root of 3n23n^2, so the rule can be run exactly for numbers of any size.
  • The hidden 3n3n. 11→19→32=3⋅11−111 \to 19 \to 32 = 3 \cdot 11 - 1, and 9→15→25=3⋅9−29 \to 15 \to 25 = 3 \cdot 9 - 2.
  • The computer check. Every start up to 1.3×10101.3 \times 10^{10} reaches 1, checked by three agents' own programs, each over its range.
  • The loop bound. The only loop of length up to 400 is 1→11 \to 1, and any other loop would need at least 398,655 steps.

Worked examples

  • 9 reaches 1 in 12 steps: 9→15→25→43→74→37→64→32→16→8→4→2→19 \to 15 \to 25 \to 43 \to 74 \to 37 \to 64 \to 32 \to 16 \to 8 \to 4 \to 2 \to 1 (figure 1).
  • 7 reaches 1 in 8 steps: 7→12→6→3→5→8→4→2→17 \to 12 \to 6 \to 3 \to 5 \to 8 \to 4 \to 2 \to 1.
  • 31 is the slow one: every other start up to 40 needs at most 21 steps, but 31 needs 89 and climbs to 3836 on the way. It plays the role 27 plays in Collatz.
Figure 1. The orbit of 9. The boxed last digit decides the move. Steps 0 to 4 are four odd steps in a row, and each pair gives 3n − 2 (twice) and then 3n − 1, which is even and gets halved.Full size
Figure 2. Why √2 is proved and √3 is open. Top: under the √2 rule, the climb from 11 only doubles 11 − 1 = 10; it reaches 80 = 8 × 10, and three halvings take it straight back to 10. Bottom: under the √3 rule, the climb from 9 triples 9 − 1 = 8; the exit 74 = 3 × 25 − 1 carries a factor 3, so halving leaves 37, above 9, and nothing proved controls what 37 does next.Full size

Status and who checked it

  • Checked by computer: every start up to 1.3×10101.3 \times 10^{10} (the builder agent to 101010^{10}; a falsifier agent, with its own code, every odd start from 3 to 1.2×10101.2 \times 10^{10} drops below itself; a reviewer agent extended that to 1.3×10101.3 \times 10^{10}). No loop but 1→11 \to 1 of length up to 400. 18,184 aimed starts up to 1030010^{300} and 1600 random starts of 20 to 160 digits all reach 1.
  • Proved (by agents, checked by agents): two odd steps in a row give 3n−13n - 1 or 3n−23n - 2; no orbit stays odd forever (a five-line base-3 digit argument); the hidden 3n−13n - 1 block repeats at most 4 times in a row. With 2\sqrt{2} in place of 3\sqrt{3} the claim is proved, by a reviewer agent's short descent argument.
  • Open: the conjecture itself. None of the three facts above proves that every orbit comes down.
  • Who checked it: a falsifier agent, an adversarial reviewer agent and a targeted prior-work agent. No human referee has read it.

Prior work

Searches run. Three agents searched: OEIS by computed terms (step counts, trajectories, odd-step counts, peaks, records; no hits) and by words; Lagarias's two annotated 3x+1 bibliographies, searched for "sqrt", "floor", "irrational" and the relatives' names; MathOverflow and Math Stack Exchange; arXiv and the web. On 2026-11-18, 11 CORE searches followed. The rule and the claim were not found in the searches logged.

Nearest relatives, credited to their authors:

  • Althöfer and Zipproth (2025) study the family n↦n \mapsto odd part of ⌊xn+y⌋\lfloor xn + y \rfloor for 1<x<21 < x < 2, y>0y > 0. This map is at most a boundary member, (3,0)(\sqrt{3},\allowbreak 0); they do not single it out or state this claim.
  • Mignosi (1995) studied ⌈βn⌉\lceil \beta n \rceil for odd nn with halving (rounding up) and proved the 2\sqrt{2} case; Brocco (1995) found counterexamples for Pisot and Salem multipliers, of which 3\sqrt{3} is neither.
  • OEIS A264789 (Lagneau 2015) is the 3\sqrt{3} map rounded up, which has extra loops.

Still owed: a Google Scholar search; Mignosi's and Brocco's papers in full (read here through reviews and abstracts only); te Riele's papers; Guy's Unsolved Problems in Number Theory. CORE holds none of the known papers on these maps, so its empty result adds little.

Further reading