Wiki
Wiki

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

Updated


This proves L126. Only the increment bounds 1≤ΔhQ≤21\le\Delta h_Q\le2 are needed of the large path in the finite construction.

One representation at each smaller dyadic scale

Put Q=2kQ=2^k, V=2QV=2Q and n=Q2+QV+hQ(V)n=Q^2+QV+h_Q(V). For a dyadic qq in the stated range,

16Q≤q≤Q/256,m=Q/q∈N.16\sqrt Q\le q\le Q/256,\qquad m=Q/q\in\mathbb N.

Choose an integer of the form y=mt−q+1y=mt-q+1 in the interval [7q/10,18q/25][7q/10,18q/25]. This is possible because the interval has length q/50q/50, whereas the spacing m=Q/qm=Q/q is at most q/256q/256. This also defines an integer tt, and gives

q2+qy=Qt+q,0<t≤43q225Q≤43q6400<q128.(1)q^2+qy=Qt+q,\qquad 0<t\le\frac{43q^2}{25Q}\le\frac{43q}{6400}<\frac q{128}. \tag{1}

In particular V−tV-t is an integer in [0,4Q][0,4Q]. Write

Δ=hQ(V)−hQ(V−t),H=q−Δ.\Delta=h_Q(V)-h_Q(V-t),\qquad H=q-\Delta.

The increment bounds give t≤Δ≤2tt\le\Delta\le2t, so H≤qH\le q and H≥(3157/3200)qH\ge(3157/3200)q. Our choice of yy therefore ensures

43y≤2425q<H≤q<32y.(2)\frac43y\le\frac{24}{25}q<H\le q<\frac32y. \tag{2}

There is an admissible path with hq(y)=Hh_q(y)=H. To construct it, put b=H−yb=H-y and c=3y−2Hc=3y-2H. Equation (2) implies 0≤c≤b0\le c\le b. Concatenate cc copies of the word 112112 and b−cb-c copies of the word 1212, in any order. The resulting word has length 3c+2(b−c)=y3c+2(b-c)=y and sum 4c+3(b−c)=H4c+3(b-c)=H. It contains neither 111111 nor 2222, including at the concatenation boundaries. Extend it by repetitions of 1212 and truncate at length 4q4q. This defines the required admissible hqh_q.

Use the bare points, with cluster shifts both zero:

aq=Q2+Q(V−t)+hQ(V−t)∈AQ,bq=q2+qy−hq(y)∈Bq.a_q=Q^2+Q(V-t)+h_Q(V-t)\in A_Q,\qquad b_q=q^2+qy-h_q(y)\in B_q.

Equations (1) and hq(y)=q−Δh_q(y)=q-\Delta give the exact integer equality

aq+bq=Q2+QV+hQ(V−t)+Δ=n.(3)a_q+b_q=Q^2+QV+h_Q(V-t)+\Delta=n. \tag{3}

Thus no discarded carry or approximate equality is involved.

Distinct pairs and an infinite family

For these small summands,

1710q2−q≤bq≤4325q2.\frac{17}{10}q^2-q\le b_q\le\frac{43}{25}q^2.

The intervals are disjoint at successive dyadic values of qq: the next lower endpoint is (34/5)q2−2q>(43/25)q2(34/5)q^2-2q>(43/25)q^2. Hence all bqb_q are distinct, so (3) gives distinct ordered pairs in the actual sets, even after taking unions. The number of available dyadic scales is

(k−8)−(⌈k/2⌉+4)+1=⌊k/2⌋−11.(k-8)-(\lceil k/2\rceil+4)+1=\lfloor k/2\rfloor-11.

For the infinite assertion choose dyadic scales Ql→∞Q_l\to\infty with 16Ql+1>Ql16\sqrt{Q_{l+1}}>Q_l. At step ll, prescribe any admissible large path at QlQ_l and apply the finite construction. Every smaller scale used in this step is larger than Ql−1Q_{l-1}, so none of its paths has been assigned earlier. After these assignments choose arbitrary admissible paths on the unused scales, including all 3⋅2j−13\cdot2^{j-1}. No earlier path is changed, and taking further unions preserves each displayed set of distinct pairs.

For Ql=2klQ_l=2^{k_l}, the target satisfies 3Ql2≤nl≤3Ql2+4Ql≤4Ql23Q_l^2\le n_l\le3Q_l^2+4Q_l\le4Q_l^2. Consequently

rA,B(nl)≥⌊kl/2⌋−11≥14log⁡2nl−13.r_{A,B}(n_l)\ge\lfloor k_l/2\rfloor-11 \ge\tfrac14\log_2 n_l-13.

The nln_l tend to infinity, proving the claim. □\square