Maths Olympiad Prep

Library / /2 of 2

Number theory Difficulty 5.6 AIME, harder Prove it United States

Problem:
Let S={a1,,an}S=\{a_{1}, \ldots, a_{n}\} be a finite set of positive integers of size n1n \geq 1, and let TT be the set of all positive integers that can be expressed as sums of perfect powers (including 11) of distinct numbers in SS, meaning
T={i=1naieie1,e2,,en0} T=\left\{\sum_{i=1}^{n} a_{i}^{e_{i}} \mid e_{1}, e_{2}, \ldots, e_{n} \geq 0\right\}
Show that there is a positive integer NN (only depending on nn) such that TT contains no arithmetic progression of length NN.

Solution

Solution:
In general we can assume that each ai>1a_{i}>1, since replacing ai=1a_{i}=1 by some large integer aa creates a set TT containing the original TT as a subset (by setting ei=0e_{i}=0).

We proceed by induction on nn. For the base case n=1n=1, an arithmetic progression of length at least 33 would give a1e1+a1e3=2a1e2a_{1}^{e_{1}}+a_{1}^{e_{3}}=2 a_{1}^{e_{2}} where e3>e2>e1e_{3}>e_{2}>e_{1}. But a1e32a1e2a_{1}^{e_{3}} \geq 2 a_{1}^{e_{2}}, so this is impossible. Thus our result holds for n=1n=1.

Assume the result is true for some n1n-1, and let An1A_{n-1} be a number such that the longest progression when S=n1|S|=n-1 has length less than An1A_{n-1}. Let MM be a large integer that we will choose later. Take an arithmetic progression of length 2M+12M+1, calling the terms b1,b2,,b2M+1b_{1}, b_{2}, \ldots, b_{2M+1}. Note that b2M+12bM+1b_{2M+1} \leq 2 b_{M+1}, since bb is a sequence of positive integers. For each term from bM+1b_{M+1} to b2M+1b_{2M+1} assign to it the maximum power that is part of the sum. Call this value c(bi)c(b_{i}). More explicitly, if bi=j=1najejb_{i}=\sum_{j=1}^{n} a_{j}^{e_{j}}, then c(bi)=max(a1e1,a2e2,,anen)c(b_{i})=\max \left(a_{1}^{e_{1}}, a_{2}^{e_{2}}, \ldots, a_{n}^{e_{n}}\right).

Since bM+1nc(bi)b2M+12bM+1\frac{b_{M+1}}{n} \leq c(b_{i}) \leq b_{2M+1} \leq 2 b_{M+1} for M+1i2M+1M+1 \leq i \leq 2M+1, there are at most j=1n(logai2n+1)n(log22n+1)\sum_{j=1}^{n}\left(\log_{a_{i}} 2n+1\right) \leq n\left(\log_{2} 2n+1\right) different values of c(bi)c(b_{i}). By Van der Waerden's Theorem, there exists a value of MM such that coloring an arithmetic progression of length MM with n(log22n+1)n\left(\log_{2} 2n+1\right) colors yields a monochromatic arithmetic progression of length An1A_{n-1}. In particular, we can take M=W(An1,n(log22n+1))M=W\left(A_{n-1}, n\left(\log_{2} 2n+1\right)\right), where W(n,k)W(n, k) denotes the Van der Waerden number. So, we color bM+1,,b2M+1b_{M+1}, \ldots, b_{2M+1} by their c(bi)c(b_{i}). Subtracting the common perfect power from each term of the monochromatic arithmetic progression obtained gives an arithmetic progression of An1A_{n-1} integers expressible as the sum of perfect powers of distinct numbers in SajS \setminus a_{j}. By the inductive hypothesis, this is a contradiction. So no arithmetic progression of length 2M+12M+1 can be contained in TT, and we can take An=2M+1A_{n}=2M+1.

By induction, we thus have such an AnA_{n} for all nn.

Want a route through all this instead of an archive? The track puts 2,000 problems in a working order, from AMC 10 level to the IMO shortlist.

Source: MathNet, licensed CC-BY-4.0. Statement reproduced verbatim; metadata (topic, difficulty) added by this project.