Last-digit Collatz (the Ouroboros Conjecture)

Edition 4 · On this site since · last revised · Verify & co-author
Kind
Conjecture, checked by computer for every start through 101510^{15}
Status
Conjecture (not proved): Lean-verified for density-one descent and cycles with at most 3 odd members; checked by computer through 101510^{15}; unrefereed
Prior work
Rule and two-loop claim not found in the searches logged; the loop of 5 appears earlier under a different rule
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 d⋅n+1d \cdot n + 1, where dd is the last decimal digit of nn (so dd is 1, 3, 5, 7 or 9).

Conjecture. Every start, followed long enough, ends in one of two loops: 1→2→11 \to 2 \to 1 or 5→26→13→40→20→10→55 \to 26 \to 13 \to 40 \to 20 \to 10 \to 5. Equivalently, these are the only loops, and no start climbs forever.

Check it yourself

  • The two loops close. 1⋅1+1=21 \cdot 1 + 1 = 2, and 2 halves to 1. 5⋅5+1=265 \cdot 5 + 1 = 26, then 13; 13 ends in 3, so 3⋅13+1=403 \cdot 13 + 1 = 40, then 20, 10 and back to 5.
  • The computer check. Every positive start through one quadrillion (101510^{15}), inclusive, ends in one of the two loops. An independent checker agreed on the final billion starts below the endpoint.
  • How many starts reach 5? For every integer X≥3X \ge 3, at least 0.02134X13/200.02134 X^{13/20} odd starts up to XX reach the loop through 5. The checker verifies integer weights on branches of the inverse tree rooted at 3.
  • How long would another loop have to be? An exact enumeration finds only the two known loops through eight odd members. Combined with the recorded sweep through 101510^{15}, separate bounds require any other loop to have at least 34,905 halvings, 15,496 odd steps and 53,185 total steps.

Worked examples

One step from each nn (not a chain):

nn 1 2 3 4 5 6 7 8 9 10
next 2 1 10 2 26 3 50 4 82 5

Three starts to follow by hand:

  • 31 reaches 1 in 6 steps: 31→32→16→8→4→2→131 \to 32 \to 16 \to 8 \to 4 \to 2 \to 1. The first step is 1⋅31+11 \cdot 31 + 1.
  • 3 reaches 5 in 2 steps: 3→10→53 \to 10 \to 5.
  • 7 reaches 5 in 40 steps, climbing as high as 2,576 on the way (figure 1).

Some starts climb far before they fall: 9,872,887, a 7-digit number, climbs to a 33-digit number and still ends at 5.

Figure 1. The 40 steps from 7 to 5, one row per step. On each odd row the highlighted last digit is the multiplier. After every 3 or 7, the next odd number ends in 5.Full size
Figure 2. Left: starts 1 to 3,000 in rows of 30; dark cells end at 1, light cells in the loop of 5. Only 59 of the 3,000 end at 1. Right: the 349 odd starts up to 1,000,000 that end at 1 all leave remainder 1, 15 or 17 mod 30 (263, 44 and 42 of them).Full size

Status and who checked it

  • Checked by computer: every start through 101510^{15}. Earlier, separate checks reached 1.6×10101.6 \times 10^{10} and 2.5×10142.5 \times 10^{14}.
  • Checked in Lean 4. An algebraic proof covers loops with at most three odd members. The exact search through eight odd members and the large-cycle bounds use additional integer computations.
  • Descent and eventual destination. The set of starts that eventually fall below their initial value has natural density one. Among starts reaching either known loop, the fraction reaching 1 tends to zero.

  • Open: that every orbit is bounded, and that no other loop exists. The conjecture is not proved.

  • Who checked it: a falsifier agent, a second targeted falsifier agent and an adversarial reviewer agent. No human referee has read it.

What is known, and not credited here. The rule reads a decimal digit. In base 4 the same rule ("multiply an odd number by its last base-4 digit and add 1") is the Collatz problem itself, a form Farkas posed in 2005. So the family holds Collatz's problem as its base-4 member; the claim here is about the base-10 member only.

Prior work

Searches run. Four agents searched with their own queries: OEIS by computed terms and by words, Math Stack Exchange and MathOverflow, the web, about 75 queries and pages on Keith Matthews' generalized-3x+13x+1 site, and both of Lagarias's annotated bibliographies. On 2026-11-18, 2 Google Scholar searches and 10 CORE searches followed. On 2026-11-19 a hunt covered Reddit's Collatz forums, AoPS, Stack Exchange sister sites, recreational sources and sources in six other languages. The rule and the two-loop claim were not found in the searches logged.

An earlier occurrence of the loop. The loop 5→26→13→40→20→10→55 \to 26 \to 13 \to 40 \to 20 \to 10 \to 5 was written down earlier, under a different rule, in Augusto Santi's Math Stack Exchange question of 2025-12-10, about a map whose odd steps multiply by 3 and by 5 in turn. That map never reads a digit, so it is a relative: not this rule and not this claim. The two rules share the loop because its odd members, 5 and 13, end in 5 and 3.

Nearest relatives named by the searches: Gurbaxani's "Mapping P4P_4" on Matthews' site (the same ten-branch shape, different multipliers); the general residue-class map theory of Conway (1972), Matthews and Watts, and Leigh; and Farkas's base-4 form above.

Still owed: Guy's Unsolved Problems in Number Theory in print and Wirsching's monograph were not searched.

Further reading