Wiki
Wiki

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

Updated


Source. Theorem 5.1 and Remark 5.1, preprint p. 11, with the opening of Section 5 (p. 11); Theorem 5.2 (p. 11, proof p. 12), Examples 5.1--5.2 (p. 12) and Example 5.3 (p. 13) for the related constructions. Read on the rendered pages. The paper is cited by its record on the source card.

Statement

For a sequence (bn)n≥1(b_n)_{n\ge1} and a positive integer kk put TN=∑n≥NbnkN−nT_N=\sum_{n\ge N}b_nk^{N-n} (N≥1N\ge1). Let k>1k>1 be an integer and (bn)n≥1(b_n)_{n\ge1} a sequence of positive integers such that T=∑n≥1bnk−nT=\sum_{n\ge1}b_nk^{-n} converges and bn≤(1−1/k)Tn+1b_n\le(1-1/k)T_{n+1} for all nn. Let S∈(T/(k+1),T]S\in(T/(k+1),T]. Then there are an∈{k,k+1,…,k2}a_n\in\{k,k+1,\ldots,k^2\} with

S=∑n=1∞bna1⋯an.S=\sum_{n=1}^{\infty}\frac{b_n}{a_1\cdots a_n}.

Remark 5.1 (p. 11): every nondecreasing sequence of positive integers bnb_n with TT convergent satisfies the condition, so the theorem applies to all such sequences. The opening of Section 5 records that Hančl and Tijdeman had the case k=2k=2, an∈{2,3,4}a_n\in\{2,3,4\}, for nondecreasing bnb_n and S∈(T/2,T)S\in(T/2,T). The proof chooses ana_n greedily from the position of the current remainder SnS_n relative to TnT_n and sets Sn+1=anSn−bnS_{n+1}=a_nS_n-b_n.

  • Theorem 5.2 (p. 11): for an integer k>1k>1 and positive integers bnb_n with T=∑bnk−nT=\sum b_nk^{-n} convergent and TN+1≥(k+1)bNT_{N+1}\ge(k+1)b_N for N>1N>1, every S∈(k2T/(k+1)2,T]S\in(k^2T/(k+1)^2,T] is ∑bn/(a1⋯an)\sum b_n/(a_1\cdots a_n) with an∈{k,k+1}a_n\in\{k,k+1\}; Example 5.1 (p. 12) gives bn=[(k−13)n]b_n=[(k-\frac13)^n] as a sequence meeting these conditions.
  • Example 5.2 (p. 12): with bn=(n−2)!b_n=(n-2)! for n≥2n\ge2, every S∈(14675,83]S\in(\frac{146}{75},\frac83] is ∑n≥2bn/(a2⋯an)\sum_{n\ge2}b_n/(a_2\cdots a_n) with an∈{n,n+1}a_n\in\{n,n+1\}; the paper calls this in some sense a counterpart to Theorems 3.1, 4.1 and 4.2.
  • Example 5.3 (p. 13): for integers d>c>1d>c>1, 0<ϵ<(cd−c)/(d2−c)0<\epsilon<(cd-c)/(d^2-c) and bn=(d−1)nb_n=(d-1)^n, every S∈(c(d−1)(d−1+ϵ)/(d2ϵ),(d−1)/ϵ]S\in\bigl(c(d-1)(d-1+\epsilon)/(d^2\epsilon),(d-1)/\epsilon\bigr] is ∑bn/(a1⋯an)\sum b_n/(a_1\cdots a_n) with an∈{c,d}a_n\in\{c,d\}.

The paper closes Section 5 (p. 13) with two questions it calls open: for an integer k≥2k\ge2, positive integers bnb_n with ∑bnk−n\sum b_nk^{-n} convergent and distinct integers a,b≥ka,b\ge k, whether there is a fixed interval each of whose points is ∑bn/(a1⋯an)\sum b_n/(a_1\cdots a_n) with every an∈{a,b}a_n\in\{a,b\}, and whether there are infinitely many such representations.

Role

The paper presents these constructions (p. 2) as showing that the results of Sections 3 and 4 do not hold without growth restrictions. For instance, with k=2k=2 and bn=pnb_n=p_n, the nnth prime, Remark 5.1 applies, so every number in (T/3,T](T/3,T], where T=∑pn/2nT=\sum p_n/2^n, is ∑pn/(a1⋯an)\sum p_n/(a_1\cdots a_n) for some an∈{2,3,4}a_n\in\{2,3,4\}; this specialization is the corpus's, and it says nothing about the rationality of TT itself.

Bears on. No catalog problem directly.