Conjecture Unrefereed

Free 2-3-6 button machines

No dated manuscript catalogued; standalone verification archive linked above. On this site since · last revised · Verify & co-author
Kind
Half known, half conjecture: the "if" half is Klarner's 1982 theorem; the "only if" half is checked at every ratio of height up to 100
Status
The "only if" half holds up under agent review; not proved; unrefereed
Prior work
The "if" half is Klarner's (1982); the converse was not found in the searches logged
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

Pick numbers aa, bb, cc and make three buttons:

  • 2: x→2x+ax \to 2x + a;
  • 3: x→3x+bx \to 3x + b;
  • 6: x→6x+cx \to 6x + c.

Pressing buttons in a row (a word, written 2.6 for "press 2, then 6") builds a map x→Nx+Tx \to Nx + T. Two different words that build the same map are a collision. The choice of a,b,ca,\allowbreak b,\allowbreak c is free when it has no collision at all.

Everything depends on one number, the ratio r=(5a−c)/(2a−b)r = (5a - c)/(2a - b): renaming the number line keeps rr and every collision, and two choices with the same rr are renamed copies of each other.

Klarner's theorem (1982), not claimed here. If rr is 3/23/2, 10/310/3 or −5/4-5/4, the buttons are free. These are his six triples (0,2,3)(0,\allowbreak 2,\allowbreak 3), (1,0,2)(1,\allowbreak 0,\allowbreak 2), (0,3,10)(0,\allowbreak 3,\allowbreak 10), (2,1,0)(2,\allowbreak 1,\allowbreak 0), (2,0,15)(2,\allowbreak 0,\allowbreak 15) and (1,6,0)(1,\allowbreak 6,\allowbreak 0) and their renamed copies. Run backwards, their buttons cut a stick into a half, a third and a sixth that fit exactly (1/2+1/3+1/6=11/2 + 1/3 + 1/6 = 1), so every point of the stick has one address.

Conjecture (the "only if" half). Every other rational rr has a collision. So the buttons 2x+a2x + a, 3x+b3x + b, 6x+c6x + c are free if and only if rr is 3/23/2, 10/310/3 or −5/4-5/4. (Every irrational rr is free, since a collision is a linear equation with whole-number coefficients that fixes a rational rr; so the claim is about rational rr.)

Check it yourself

  • A collision at a=b=c=1a = b = c = 1 (r=4r = 4). The word 2.6 gives 2x+12x + 1, then 6(2x+1)+1=12x+76(2x + 1) + 1 = 12x + 7. The word 3.2.2 gives 3x+13x + 1, then 6x+36x + 3, then 12x+712x + 7. From x=1x = 1: 1→3→191 \to 3 \to 19 and 1→4→9→191 \to 4 \to 9 \to 19.
  • A longer one at a,b,c=1,0,8a,\allowbreak b,\allowbreak c = 1,\allowbreak 0,\allowbreak 8 (r=−3/2r = -3/2). The words 3.6.2.6.2.3 and 2.3.3.3.2.3.2.2 (6 and 8 presses) both build 1296x+6631296x + 663; both take 1 to 1959.
  • Each pair of short words fixes one ratio. 6 against 2.3 collide exactly when r=1r = 1; 6 against 3.2 when r=2r = 2; 2.6 against 6.2 when r=0r = 0; 3.6 against 6.3 when r=5/2r = 5/2; 2.3 against 3.2 when 2a=b2a = b ("rr is infinite").
  • A free choice, 2x2x, 3x+23x + 2, 6x+36x + 3 (r=3/2r = 3/2). Undo each button: y→y/2y \to y/2, y→(y−2)/3y \to (y - 2)/3, y→(y−3)/6y \to (y - 3)/6. They carry the stick [−1,0][-1,\allowbreak 0] onto [−1/2,0][-1/2,\allowbreak 0], [−1,−2/3][-1,\allowbreak -2/3] and [−2/3,−1/2][-2/3,\allowbreak -1/2]: a half, a third and a sixth, end to end, with no overlap.

Worked examples

Figure 1. Top: for 2x2x, 3x+23x + 2, 6x+36x + 3 the backward buttons cut the stick [−1,0][-1,\allowbreak 0] into a third, a sixth and a half, and each piece again the same way; the pieces of every row add up to length 1, and each piece has exactly one address. Bottom: for 2x+12x + 1, 3x+13x + 1, 6x+16x + 1 (r=4r = 4) the pieces overlap and leave a gap, and the pieces with addresses 2.6 and 3.2.2 are the same interval, [−2/3,−3/5][-2/3,\allowbreak -3/5]: the collision 12x+712x + 7.Full size
Figure 2. Two collisions step by step, with the value at x=1x = 1 on each row. Top: 2.6 and 3.2.2 at r=4r = 4. Bottom: the 6-press and 8-press words at r=−3/2r = -3/2 that both build 1296x+6631296x + 663.Full size

Status and who checked it

  • Proved, and not claimed here: the "if" half (Klarner 1982; a ping-pong argument on the stick).
  • Proved (checked line by line by a reviewer agent): rr decides freeness, and every irrational rr is free.
  • Checked by computer (complete searches, each collision re-checked in exact arithmetic):
  • every reduced ratio p/qp/q with ∣p∣|p| and qq at most 100: 12,176 ratios, and all but the three free ones collide (12,173 collisions); the longest needs an 18-letter word, at r=−99/79r = -99/79;
  • 1,460 more ratios crowding in on the three free values (denominators 101 to 300): all collide. Near those values the collisions grow long (the largest smallest-collision multiplier is 2,176,782,336, at r=−321/257r = -321/257) but never stop appearing;
  • 400 random ratios with denominators 101 to 1000: all collide;
  • at each of 3/23/2, 10/310/3 and −5/4-5/4, no collision among the 96,959,662 words with multiplier up to 10810^8.
  • Open: the "only if" half. Nobody has a proof; the search is finite.
  • Who checked it: a falsifier agent and an adversarial reviewer agent, each with its own code. The reviewer confirmed the claim with three corrections, all carried here: a missed neighbouring paper (below), a narrower statement about the density of orbit sets, and a count (4,405 ratios of height at most 60, not 4,406). No human referee has read it.

Prior work

Searches run. A falsifier agent and a reviewer agent each kept a prior-work log (OEIS, arXiv, the web, and Lagarias's annotated 3x+1 bibliography); both logs pass the literature-search gate. On 2026-11-19 a keyed CORE search made 7 requests and got 23 records: 6 relatives, none with the same claim or rule. The converse was not found in the searches logged. On 2026-11-22, a Google Scholar sweep (5 queries) found no prior work stating this claim.

Known parts, credited to their authors.

  • Klarner (1982) proved the "if" half and found the six triples. His paper was read only through Lagarias's bibliography (entry 97), not in full.
  • Crampin and Hilton (unpublished, via Lagarias) used the collision 2.6=3.2.22.6 = 3.2.2 for 2x+12x + 1, 3x+13x + 1, 6x+16x + 1 to answer an Erdős question about density; that a collision gives density zero is their argument.
  • Konyagin, Protasov and Talambutsa (2025) proved the matching "free if and only if the pieces tile" statement for buttons that all share one prime slope. Their footnote says freeness "is still open for integer affine functions". This is the nearest relative.
  • Kolpakov and Talambutsa, Shamazov and Talambutsa, and Kalmynin (density zero for Klarner's six) are relatives; none states or asks this converse.

Still owed: the full text of Klarner 1982; Lagarias's 2016 survey, section 8.9, the likeliest place for a remark on the converse; Guy's Unsolved Problems in Number Theory; Math Stack Exchange and MathOverflow.

Further reading