Conjecture Unrefereed

Triple it, or subtract one and flip

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 fraction of height up to 100,000
Status
Holds up under agent review; not proved; unrefereed
Prior work
Rule and claim not found in the searches logged; the kind of problem is in print (Javaheri 2012)
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

Write a positive fraction xx in lowest terms, p/qp/q. Its height is the larger of pp and qq.

  • If x≤1x \le 1, the next fraction is 3x3x (cancel if you can).
  • If x>1x > 1, the next fraction is 1/(x−1)1/(x - 1); for p/qp/q this is q/(p−q)q/(p - q).

Conjecture. Every positive fraction, followed long enough, reaches 1, and so ends in the loop 1→3→1/2→3/2→2→11 \to 3 \to 1/2 \to 3/2 \to 2 \to 1. Equivalently, that loop is the only loop, and no fraction's height grows forever.

Check it yourself

  • The loop closes. 1 triples to 3; 3−1=23 - 1 = 2, flipped is 1/21/2; 1/21/2 triples to 3/23/2; 3/2−1=1/23/2 - 1 = 1/2, flipped is 2; 2−1=12 - 1 = 1, flipped is 1.
  • The computer check. Every fraction of height up to 100,000 (6,079,301,507 fractions) reaches 1, by an exact induction on height.
  • The loop bound. No other loop has 40 steps or fewer. So any other loop has at least 41 steps, and every fraction in it has height above 100,000.

Worked examples

  • 4/7 reaches 1 in 6 steps: 4/7→12/7→7/5→5/2→2/3→2→14/7 \to 12/7 \to 7/5 \to 5/2 \to 2/3 \to 2 \to 1.
  • 2/7 reaches 1 in 9 steps: 2/7→6/7→18/7→7/11→21/11→11/10→10→1/9→1/3→12/7 \to 6/7 \to 18/7 \to 7/11 \to 21/11 \to 11/10 \to 10 \to 1/9 \to 1/3 \to 1.
  • 5 reaches 1 in 17 steps (figure 1).

The only steps that lower a height for good are triplings that cancel a 3, such as 2/9→6/9=2/32/9 \to 6/9 = 2/3. Some fractions climb far before they fall: 1189/15491189/1549 climbs to height 2,433,927 at step 51 and reaches 1 at step 103.

Figure 1. Left: 4/7 reaches 1 in 6 steps, then one turn of the loop follows in grey. Right: 5 reaches 1 in 17 steps. The shaded box is the comparison with 1 that decides the step.Full size
Figure 2. The same rule with 2, 3 or 4 in place of 3, on all 152,231 fractions of height up to 500; each dot is the largest climb (peak height over start height) at that height. With 2 nothing climbs to twice its start, and the claim is easy to prove. With 3 climbs reach 433 times the start and every fraction reaches 1. With 4, 69 starts fall into a second loop, 40 steps long.Full size

Status and who checked it

  • Checked by computer (exact arithmetic): every fraction of height up to 100,000; no other loop of 40 steps or fewer, by two separate exact searches (to length 36, then to length 40, by a different test); 10,000,000 random fractions of heights 10710^7 to 101810^{18} and 95,753 aimed starts, all reaching 1.
  • Proved (short arguments, checked by agent reviewers): a run of flips never raises the height, and every orbit returns to (0,1](0,\allowbreak 1] after finitely many flips; no loop other than the known one goes "flip, flip, …, triple, triple, …" once around.
  • Open: that no height grows forever, and that no other loop exists. The conjecture is not proved; three routes to a proof failed.
  • Who checked it: a falsifier agent that did not build it, then a reviewer agent with no stake, which repeated the counts with its own code. No human referee has read it.

Prior work

Searches run. Three agents searched with their own queries: OEIS by the rule's computed terms (steps from nn, from 1/n1/n, from n/(n+1)n/(n+1), the orbit of 5) and by words, arXiv, Math Stack Exchange and MathOverflow, the web, and both of Lagarias's annotated 3x+1 bibliographies by keyword. On 2026-11-18, 1 Google Scholar search and 9 CORE searches followed, with the list of works citing Javaheri 2012. The rule and its claim were not found in the searches logged.

The kind of problem is known. M. Javaheri (2012) conjectured that a different rule on fractions, also switching at 1 and built from 2 and 3, sends every fraction to 0. No change of variable turns one map into the other: every number has two arrows into it under his map, while above 3 a fraction has only one under this rule. So the rule is one more member of a kind of problem already in print.

Still owed: Javaheri's 2025 paper in J. Number Theory, which cites the 2012 one, could not be read; Guy's Unsolved Problems in Number Theory in print was not searched.

Further reading