Conjecture Unrefereed

The climb to a square

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 below 1000012100001^2 = 10,000,200,001
Status
Holds up under agent review; not proved; unrefereed
Prior work
Not found in the searches logged. Without the +1 it is Brocard's walk of 1904, which provably never lands; Weintraub's 2004 note is still unread
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

Any whole number nn sits just above a square: n=s2+dn = s^2 + d, where s2s^2 is the largest square not above nn and dd is the leftover, from 0 to 2s2s. The step replaces nn by n+d+1n + d + 1. As a formula, T(n)=2n+1−⌊n⌋2T(n) = 2n + 1 - \lfloor \sqrt{n} \rfloor^2.

Conjecture. From every whole number n≥0n \ge 0, repeating the step eventually reaches a perfect square. (A start that is already a square counts as landed.)

The number grows at every step, so the climb can never loop; the only question is whether it ever hits a square exactly.

Brocard's walk, credited. Without the "+1", the rule n→n+dn \to n + d was studied by H. Brocard in 1904 and is OEIS A094761, whose entry says the walk "is eventually constant iff n is a perfect square": from a non-square it never lands. The +1 rule is that walk shifted by one: if x+1x + 1 is not a square, T(x)+1=W(x+1)T(x) + 1 = W(x + 1), where WW is Brocard's step. So the conjecture can be restated in Brocard's terms: every walk of his from a non-square m≥2m \ge 2 passes a number k2+1k^2 + 1. The claim here is the +1 form only.

Check it yourself

  • From 7: 7=4+3→11=9+2→14=9+5→20=16+4→257 = 4 + 3 \to 11 = 9 + 2 \to 14 = 9 + 5 \to 20 = 16 + 4 \to 25, a square.
  • From 3: 3=1+2→6=4+2→93 = 1 + 2 \to 6 = 4 + 2 \to 9.
  • The one door onto a square. For n=s2+dn = s^2 + d: if d<sd < s, the next number is s2+(2d+1)s^2 + (2d + 1), whose leftover is odd, so it is not a square. If d≥sd \ge s, the next number is (s+1)2+(2d−2s)(s+1)^2 + (2d - 2s), a square exactly when d=sd = s. So a step lands only from s2+s=s(s+1)s^2 + s = s(s+1): the doors 2, 6, 12, 20, 30, 42, 56, …, each stepping onto the next square.
  • A slow start: 2602 =512+1= 51^2 + 1 takes 78,220 steps and lands on 39252239252^2 = 1,540,719,504; it is the slowest start up to 3000.
  • The slowest start below 1000012100001^2: the climb from 80867280867^2 lands on 43471242310243471242310^2 after 86,942,332,668 steps.

Worked examples

Figure 1. Each row is one block, from s2s^2 to (s+1)2−1(s+1)^2 - 1; a number's column is its leftover dd, so the squares form the left edge. 7 sits in row 2, column 3, so it moves to 7+3+1=117 + 3 + 1 = 11. Every arrow onto the left edge comes from a door s2+ss^2 + s (light blue): 2, 6, 12, 20, 30.Full size
Figure 2. From the same start 7. Left: the climb 7→11→14→20→257 \to 11 \to 14 \to 20 \to 25. Right: Brocard's rule n→n+dn \to n + d, which reads 7, 10, 11, 13, 17, 18, 20, 24, 32, … and never reaches the left edge.Full size

Status and who checked it

  • Checked by computer: every start below 1000012100001^2 = 10,000,200,001 reaches a square, by two independent exact programs (all 200,001 chains born at blocks up to 10510^5 followed to landing, no cap), and every start below 400,040,001 by a third.
  • Proved (short arguments by agents, checked by a falsifier agent and a reviewer agent): the climb always grows; the one door above; the step is one-to-one, and the numbers never stepped onto are exactly the r2−1r^2 - 1 (in the shifted form this is Wenders' 2012 proof of a conjecture of Mathar's in OEIS A094761, credited to them); exactly s+1s + 1 chains pass through block ss and exactly one of them leaves through its door; and no climb can follow a fixed formula in the block size for ever (such eternal families exist for the neighbouring rule n+2dn + 2d, OEIS A094766).
  • Why it is believed, and why it is tense. If one chain in s+1s + 1 lands in block ss at random, every climb lands, but the expected landing is infinite: an "infinite-mean borderline". The data follow that model: of the 100,000 climbs from squares t2t^2, 50,204 land past (2t)2(2t)^2, 9,940 past (10t)2(10t)^2, 1,004 past (100t)2(100t)^2, 110 past (1000t)2(1000t)^2 and 14 past (10000t)2(10000t)^2, against the model's share 1/x1/x.
  • Open: the claim itself, and even a weaker statement such as "a positive share of starts land".
  • Who checked it: a falsifier agent and a reviewer agent, each with its own code; a third agent tried to read Weintraub's note. No human referee has read it.

Prior work

Searches run. The falsifier agent made 41 searches that returned (OEIS 33, arXiv 7, zbMATH 1) and 15 web searches, and searched Lagarias's two annotated 3x+1 bibliographies by hand. The reviewer agent made 27 (OEIS 22, arXiv 4, zbMATH 1) and 5 web searches, and read about 45 OEIS entries citing the square-excess sequence A053186 by hand. On 2026-11-19 a keyed CORE search made 7 requests and got 20 records, none relevant; CORE holds no record of Weintraub's note. The claim was not found in the searches logged. On 2026-11-22, a Google Scholar sweep (6 queries) found no prior work stating this claim.

The main caveat: Weintraub's note is unread. S. H. Weintraub's three-page note "An interesting recursion" (2004) is cited in the OEIS entries for Brocard's walk. Its zbMATH review describes only the rules n→n+r⋅dn \to n + r \cdot d with a fixed whole number r≥2r \ge 2, and names no "+1" form; but a three-page note can hold a remark that a review leaves out, and no legal free copy was reached.

Nearest relatives, credited to their authors.

  • Brocard (1904) and OEIS A094761, A094763, A094764: the walk n→n+dn \to n + d, this rule shifted by one, with the opposite fate.
  • Cloitre (2009), OEIS A156344: "We conjecture sequence is never zero", that is, every start reaches a square, for a different map, x→x⌈x⌉/⌊x⌋x \to x \lceil \sqrt{x} \rceil / \lfloor \sqrt{x} \rfloor. Same shape of claim, different rule; the reviewer agent says any write-up should cite it.
  • OEIS A094765 and A094766 (n+2dn + 2d, which has eternal trajectories); OEIS A159016 and A159258 (a→a+⌊a⌋a \to a + \lfloor \sqrt{a} \rfloor "contains infinitely many squares"); Allouche, Stipulanti and Yao (doubling modulo a fixed odd number).

Still owed: the full text of Weintraub's paywalled note (2004) and of Brocard's 1904 note; the MathSciNet review and citation list of Weintraub (2004); Guy's Unsolved Problems in Number Theory. (Flat-syntax Google Scholar follow-up queries ran on 2026-11-22 with no mathematical hit.)

Further reading