Maths Olympiad Prep

Library / Putnam /47 of 60

Putnam

Algebra Difficulty 8.1 Shortlist Find the answer

Assume that (an)n≥1(a_n)_{n\geq 1} is an increasing sequence of positive real numbers such that lim⁡an/n=0\lim a_n/n=0. Must there exist infinitely many positive integers nn such that an−i+an+i<2ana_{n-i}+a_{n+i}<2a_n for i=1,2,…,n−1i=1,2,\ldots,n-1?

A number or a short expression. Fractions can be typed as 3/2, and spacing doesn't matter.

Solution

Yes, there must exist infinitely many such nn. Let SS be the convex hull of the set of points (n,an)(n, a_n) for n≥0n \geq 0. Geometrically, SS is the intersection of all convex sets (or even all halfplanes) containing the points (n,an)(n, a_n); algebraically, SS is the set of points (x,y)(x,y) which can be written as c1(n1,an1)+⋯+ck(nk,ank)c_1(n_1, a_{n_1}) + \cdots + c_k(n_k, a_{n_k}) for some c1,…,ckc_1, \dots, c_k which are nonnegative of sum 1.

We prove that for infinitely many nn, (n,an)(n, a_n) is a vertex on the upper boundary of SS, and that these nn satisfy the given condition. The condition that (n,an)(n, a_n) is a vertex on the upper boundary of SS is equivalent to the existence of a line passing through (n,an)(n, a_n) with all other points of SS below it. That is, there should exist m>0m>0 such that
a k < a n + m(k-n) k 1.\text{a k < a n + m(k-n) k 1.}

We first show that n=1n=1 satisfies ( ).\text{( ).} The condition ak/k→0a_k/k \to 0 as k→∞k \to \infty implies that (ak−a1)/(k−1)→0(a_k - a_1)/(k-1) \to 0 as well. Thus the set {(ak−a1)/(k−1)}\{(a_k-a_1)/(k-1)\} has an upper bound mm, and now ak≤a1+m(k−1)a_k \leq a_1 + m(k-1), as desired.

Next, we show that given one nn satisfying ( ),\text{( ),} there exists a larger one also satisfying ( ).\text{( ).} Again, the condition ak/k→0a_k/k \to 0 as k→∞k \to \infty implies that (ak−an)/(k−n)→0(a_k-a_n)/(k-n) \to 0 as k→∞k \to \infty. Thus the sequence {(ak−an)/(k−n)}k>n\{(a_k-a_n)/(k-n)\}_{k>n} has a maximum element; suppose k=rk = r is the largest value that achieves this maximum, and put m=(ar−an)/(r−n)m = (a_r -a_n)/(r-n). Then the line through (r,ar)(r, a_r) of slope mm lies strictly above (k,ak)(k, a_k) for k>rk > r and passes through or lies above (k,ak)(k, a_k) for k<rk< r. Thus ( )\text{( )} holds for n=rn=r with mm replaced by m−ϵm-\epsilon for suitably small ϵ>0\epsilon > 0.

By induction, we have that ( )\text{( )} holds for infinitely many nn. For any such nn there exists m>0m>0 such that for i=1,…,n−1i=1, \dots, n-1, the points (n−i,an−i)(n-i, a_{n-i}) and (n+i,an+i)(n+i, a_{n+i}) lie below the line through (n,an)(n, a_n) of slope mm. That means an+i<an+mia_{n+i} < a_n + mi and an−i<an−mia_{n-i} < a_n - mi; adding these together gives an−i+an+i<2ana_{n-i} + a_{n+i} < 2a_n, as desired.

Want a route through all this instead of an archive? The track puts 2,604 problems in a working order, from Junior Challenge level to the IMO shortlist.

Source: Omni-MATH, licensed Apache-2.0. Statement reproduced verbatim; metadata (topic, difficulty) added by this project.