Wiki
Wiki

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

Updated


Source. Theorem 4.1, p. 7, of Daniel J. Bernstein and Jeffrey C. Lagarias, The 3x+1 conjugacy map, Canad. J. Math. 48 (1996) 1154--1169, with label and page as printed in the authors' retypeset manuscript dated 15 February 1996, the edition read for the source card.

Statement

For integers a,ba,b with abab odd, the ax+bax+b function is Ta,b(x)=(ax+b)/2T_{a,b}(x)=(ax+b)/2 for x≡1(mod2)x\equiv1\pmod2 and Ta,b(x)=x/2T_{a,b}(x)=x/2 for x≡0(mod2)x\equiv0\pmod2 (4.1). The ax+bax+b conjugacy map Φa,b:Z2→Z2\Phi_{a,b}:\mathbf Z_2\to\mathbf Z_2 satisfies Φa,b∘S∘Φa,b−1=Ta,b\Phi_{a,b}\circ S\circ\Phi_{a,b}^{-1}=T_{a,b}, with SS the 2-adic shift, and is given for x=∑l2dlx=\sum_l2^{d_l}, 0≤d1<d2<⋯0\le d_1<d_2<\cdots, by Φa,b(x)=−b∑la−l2dl\Phi_{a,b}(x)=-b\sum_la^{-l}2^{d_l} (4.2), citing [2] (p. 7). Φa,b,n\Phi_{a,b,n} is the permutation of Z/2nZ\mathbf Z/2^n\mathbf Z obtained by reducing Φa,b\Phi_{a,b} modulo 2n2^n (p. 7). Cycles σn(x)\sigma_n(x), and the words inert and stable, are as in Section 3 (p. 6), now for Φa,b,n\Phi_{a,b,n}: σn+1(x)\sigma_{n+1}(x) is inert when its length is twice that of σn(x)\sigma_n(x).

Theorem 4.1 (p. 7). For the ax+bax+b conjugacy map Φa,b\Phi_{a,b}, suppose that a cycle σn(x)\sigma_n(x) of Φa,b,n\Phi_{a,b,n} has ∣σn(x)∣≥4|\sigma_n(x)|\ge4.

(i) If a≡1(mod4)a\equiv1\pmod4 and σn(x)\sigma_n(x) is an inert cycle, then σn+1(x)\sigma_{n+1}(x) is an inert cycle.

(ii) If a≡3(mod4)a\equiv3\pmod4 and σn(x)\sigma_n(x) and σn+1(x)\sigma_{n+1}(x) are both inert cycles, then σn+2(x)\sigma_{n+2}(x) is an inert cycle.

The paper adds (p. 7) that its proof shows that in case (i) the weaker hypothesis ∣σn(x)∣≥2|\sigma_n(x)|\ge2 suffices when b≡3(mod4)b\equiv3\pmod4. The case a=3a=3, b=1b=1 of (ii) is Theorem 3.1. For some ax+bax+b maps every cycle eventually becomes stable, and such Φa,b\Phi_{a,b} have no odd periodic points; by Theorem 4.1 the 25x−325x-3 conjugacy map modulo 3232 has odd part consisting of two stable cycles of period 88 (p. 7).

Proof pointer

Section 5, pp. 8--11. Writing ek[i]e_k[i] for bit kk of Φi(x)\Phi^i(x), with ∣σn(x)∣=2j|\sigma_n(x)|=2^j and σn+1(x)\sigma_{n+1}(x) inert, σn+2(x)\sigma_{n+2}(x) is inert exactly when en+1[0]≠en+1[2j+1]e_{n+1}[0]\ne e_{n+1}[2^{j+1}] (5.9). Theorem 5.1 (p. 9) computes en+1[2j+1]−en+1[0]e_{n+1}[2^{j+1}]-e_{n+1}[0] modulo 2 as 1+ab+12 2j+b(a−1)2N1+\tfrac{ab+1}2\,2^j+\tfrac{b(a-1)}2N, with NN a sum of bit counts along the cycle, using the congruence of Lemma 5.1 (p. 8) and two telescoping sums. Corollary 5.1 (pp. 10--11) evaluates this in the two residue classes of aa modulo 4, which gives (i) and (ii) through (5.9).

Dependencies

Lemma 3.1 (p. 6); Lemma 5.1, Theorem 5.1 and Corollary 5.1 (pp. 8--11).

Bears on

  • Problem 1135: background only. The theorem concerns cycles of conjugacy maps modulo powers of 2 and says nothing about orbits of the 3x+13x+1 map on the positive integers.