% Note: I admit having used GPT Codex in the writing of the LaTeX Code. \documentclass[12pt]{article} \usepackage{amsmath,amssymb,amsthm} \usepackage{geometry} \usepackage{hyperref} \geometry{margin=1in} \title{A Sequence with Doubling Ratio and Full-Density Subset Sums} \author{Enrique Barschkis} \date{January 21, 2026} \newtheorem{lemma}{Lemma} \begin{document} \maketitle \begin{abstract} We construct a nondecreasing sequence of integers $A=\{a_1\le a_2\le\cdots\}$ with \[ \lim_{n\to\infty}\frac{a_{n+1}}{a_n}=2 \] such that for every cofinite subsequence $A'\subseteq A$, the finite subset-sum set \[ P(A')=\left\{\sum_{x\in B}x: B\subseteq A' \text{ finite}\right\} \] has asymptotic density $1$ in $\mathbb{N}$. \end{abstract} \section{Problem} Does there exist a nondecreasing sequence $A$ of integers with \[ \lim_{n\to\infty}\frac{a_{n+1}}{a_n}=2 \] such that $P(A')$ has density $1$ for every cofinite subsequence $A'\subseteq A$? \section{Construction by Blocks} \subsection{Block lengths} Define \[ k_n := 4+\left\lceil \log_2\log_2(n+16)\right\rceil \qquad (n\ge0). \] Then $k_n\to\infty$ slowly and $2^{k_n}\asymp \log n$ as $n\to\infty$. \subsection{Block scales} Let $M_0:=10$ and define recursively \[ M_{n+1}:=\left\lfloor \left(2^{k_n}-\tfrac32\right)M_n \right\rfloor. \] Since $M_n\ge10$, we have \begin{equation} (2^{k_n}-2)M_n0$, and hence \[ M_{N+1} \ge c \cdot M_{n_0}\prod_{n=n_0}^N\left(a_n+\tfrac12\right). \] Hence \begin{equation} \label{eq:final_bound} % <--- Added label here \frac{E_N}{M_{N+1}} \le \frac{C_{n_0}}{c} \left(\prod_{n=n_0}^N\frac{a_n}{a_n+\tfrac12}\right) \left(\sum_{j=0}^{M_{n_0}-1}\frac{S_N^j}{j!}\right). \tag{5} \end{equation} where $C_{n_0} := (M_{n_0}+1)/M_{n_0}$. Since \[ \log\left(\frac{a_n}{a_n+\tfrac12}\right) =-\log\left(1+\frac{1}{2a_n}\right) \le -\frac{1}{4a_n} \] for all $n$ (using $a_n\ge 2^{k_n}-2\ge 62$), we have \[ \prod_{n=n_0}^N\frac{a_n}{a_n+\tfrac12} \le \exp\left(-\frac14 S_N\right). \] Because $2^{k_n}\asymp\log n$, we have $a_n\asymp\log n$ and therefore \[ S_N=\sum_{n=n_0}^N\frac{1}{a_n}\asymp\sum_{n=n_0}^N\frac{1}{\log n}\to\infty. \] The exponential decay in \eqref{eq:final_bound} dominates the fixed polynomial % <--- Changed \label to \eqref factor $\sum_{j=0}^{M_{n_0}-1}S_N^j/j!$, so \[ \frac{E_N}{M_{N+1}}\to 0, \] and $P(T)$ has asymptotic density $1$. \section{Conclusion} The constructed sequence $A$ satisfies \[ \lim_{n\to\infty}\frac{a_{n+1}}{a_n}=2, \] and for every cofinite subsequence $A'\subseteq A$, the set $P(A')$ has density $1$. Hence the answer to the original problem is \emph{yes}. \section*{Acknowledgments} The author would like to thank Terence Tao and Wouter van Doorn, as the proof presented here builds upon their ideas. \end{document}