Wiki
Wiki

Research notes on every problem, and a library of the papers behind them. Built from the open erdos repository.

Updated

Chamberland 2015 averaging structure

../

theorem_2_2: For odd q and r, the generating function of the n-th iterates of the qx+r map is a rational function P(x)/(1-x^{2^n})^2 with P of degree 2^{n+1}-1 divisible by x, given explicitly by the first 2^n iterates.

theorem_2_3: The rational generating function of the n-th iterates of the qx+r map at x equals that of the qx-r map at 1/x, so one function encodes the qx+r map on the positive and on the negative integers.

theorem_2_4: For odd q, r and m, n >= 1, the integral of f_{n,q,r}(x) x^{m-1} around the circle |x| = 2 equals 2 pi i times the n-th iterate of the qx-r map at m; for the 3x+1 map and m = 1 the value is 2 pi i for every n.

theorem_3_1: For odd q, r and n >= 1, the generating function of the n-th iterates of the qx+r map is a sum over the 2^n-th roots of unity s of double-pole and simple-pole terms whose coefficients A_{n,q,r}(s), B_{n,q,r}(s) are explicit finite sums.

theorem_3_2: For odd q > r > 0, the generating function of the (n+1)-th iterates of the qx+r map at x^q is expressed through that of the n-th iterates at x^{2q} and at mu^k x^2 for the q-th roots of unity mu^k, generalizing Berg and Meinardus.

theorem_4_1: At a fixed 2^N-th root of unity s, the double-pole coefficient B_{n,q,r}(s), and for s other than 1 the residue coefficient A_{n,q,r}(s), are multiplied by (q+1)/4 at each step n >= N, so they are constant when q = 3; A_{n,q,r}(1) is given in closed form and equals -rn/4 when q = 3.

theorem_4_2: For m >= 1 the n-th iterate of the qx+r map at m equals the sum, over the 2^n-th roots of unity s, of s^{-m} times m B_{n,q,r}(s) - A_{n,q,r}(s).


Marc Chamberland, Averaging Structure in the 3x+1 Problem, J. Number Theory 148 (2015) 384-397 (author's version; 14 pp.).

Studies, for the maps Tq,rT_{q,r} (n/2n/2 for even nn, (qn+r)/2(qn+r)/2 for odd nn, with q,rq,r odd), the generating functions fn,q,r(x)=∑k≥1Tq,r(n)(k)xkf_{n,q,r}(x)=\sum_{k\ge1}T_{q,r}^{(n)}(k)x^k of the nn-th iterates. They are rational of the form Pn,q,r(x)/(1−x2n)2P_{n,q,r}(x)/(1-x^{2^n})^2, so their poles are at most double, at the 2n2^n-th roots of unity (Theorem 2.2, p. 3); they satisfy fn,q,r(x)=fn,q,−r(1/x)f_{n,q,r}(x)=f_{n,q,-r}(1/x) (Theorem 2.3, p. 4) and a contour identity (Theorem 2.4, p. 5), and have a partial-fraction expansion with coefficients An,q,r(s)A_{n,q,r}(s), Bn,q,r(s)B_{n,q,r}(s) at each root ss (Theorem 3.1, pp. 6-7). Theorem 3.2 (p. 8) generalizes Berg and Meinardus's recursion from fnf_n to fn+1f_{n+1} to odd q>r>0q>r>0. Theorem 4.1 (p. 9) proves Bn+1,q,r(s)=q+14Bn,q,r(s)B_{n+1,q,r}(s)=\frac{q+1}{4}B_{n,q,r}(s) for s2N=1s^{2^N}=1 and n≥Nn\ge N, and the same for AA when s≠1s\ne1, so these coefficients are frozen when q=3q=3, while An,3,r(1)=−rn/4A_{n,3,r}(1)=-rn/4 changes with nn; Theorem 4.2 (p. 12) recovers Tq,r(n)(m)T_{q,r}^{(n)}(m) from them. The paper uses this only heuristically: for q=3q=3 it suggests that every orbit is bounded, and for q≥5q\ge5 that orbits diverge (Section 4, pp. 12-13); nothing is proved about the 3x+1 conjecture. Relevance: Its results are identities for the generating functions of the iterates of qx+r maps, whose polar coefficients stay fixed in n when q = 3, except the residue at x = 1, supporting only a heuristic for bounded orbits in problem 1135.

Source: PDF. The copy read for this card is the author's version, which prints no notice and whose hosting URL is not recorded; the journal version's Crossref record (DOI 10.1016/j.jnt.2014.09.024, read 2026-10-02) names only Elsevier's text-and-data-mining and open-archive user licenses, no Creative Commons license, and the publisher's page could not be read on 2026-10-02 (ScienceDirect returned HTTP 403), none of which governs that author's version; the term is unstated.

Bears on. #1135: with (q,r)=(3,1)(q,r)=(3,1) the map T3,1T_{3,1} is the problem's map, and Theorems 4.1 and 4.2 (pp. 9, 12) show that its generating functions' polar coefficients stay fixed in nn, except the residue at x=1x=1, and recover each iterate from them; the paper draws from this only a heuristic (pp. 12-13) that every orbit is bounded, and proves nothing about whether orbits reach 1 (theorem_4_1, theorem_4_2).

Results. Theorem 2.2 (p. 3, with Theorem 2.1): fn,q,rf_{n,q,r} is rational, Pn,q,r(x)/(1−x2n)2P_{n,q,r}(x)/(1-x^{2^n})^2 with an explicit numerator; Theorem 2.3 (p. 4): fn,q,r(x)=fn,q,−r(1/x)f_{n,q,r}(x)=f_{n,q,-r}(1/x); Theorem 2.4 (p. 5): ∮∣x∣=2fn,q,r(x)xm−1dx=2πiT−,q,r(n)(m)\oint_{|x|=2}f_{n,q,r}(x)x^{m-1}dx=2\pi iT_{-,q,r}^{(n)}(m); Theorem 3.1 (pp. 6-7): the partial-fraction expansion with coefficients An,q,r(s)A_{n,q,r}(s), Bn,q,r(s)B_{n,q,r}(s); Theorem 3.2 (p. 8): fn+1,q,r(xq)f_{n+1,q,r}(x^q) through fn,q,rf_{n,q,r}, for odd q>r>0q>r>0; Theorem 4.1 (p. 9): the coefficients scale by (q+1)/4(q+1)/4 in nn, with An,q,r(1)A_{n,q,r}(1) in closed form; Theorem 4.2 (p. 12): Tq,r(n)(m)T_{q,r}^{(n)}(m) as a sum over the 2n2^n-th roots of unity.

No file of this source is held: no license on record permits its redistribution, and the card cites the edition it names above.