Maths Olympiad Prep

Library / /201 of 397

Algebra Difficulty 5.9 AIME, harder Prove it Taiwan

Consider the sequence a0,a1,a2,a_0, a_1, a_2, \dots, where an=2n+2n/2a_n = 2^n + 2^{\lfloor n/2 \rfloor}. Prove that infinitely many terms in the sequence can be written as the sum of two or more distinct terms. Also, infinitely many terms cannot be written as such a sum.

Solution

Call a nonnegative integer representable if it equals the sum of several (possibly 0 or 1) distinct terms of the sequence. We say that two nonnegative integers bb and cc are equivalent (written as bcb \sim c) if they are either both representable or both non-representable.

One can easily compute (or check by induction) that
Sn1:=a0+a1++an1=2n+2n/2+2n/23. S_{n-1} := a_0 + a_1 + \dots + a_{n-1} = 2^n + 2^{\lceil n/2 \rceil} + 2^{\lfloor n/2 \rfloor} - 3.
In particular, S2k1=22k+2k+13S_{2k-1} = 2^{2k} + 2^{k+1} - 3. Note that, if n3n \ge 3, then 2n/2>32^{\lceil n/2 \rceil} > 3, so
Sn1>2n+2n/2=an. S_{n-1} > 2^n + 2^{\lceil n/2 \rceil} = a_n.
Also notice that Sn1an=2n/23<anS_{n-1} - a_n = 2^{\lceil n/2 \rceil} - 3 < a_n. The main tool of the solution is the following claim.

Claim 1. If Sn1an<b<anS_{n-1} - a_n < b < a_n for some n3n \ge 3, then bSn1bb \sim S_{n-1} - b.

Proof. If bb is representable, then since b<anb < a_n, bb must be the sum of some subset of {a0,a1,,an1}\{a_0, a_1, \dots, a_{n-1}\}. So Sn1bS_{n-1} - b is the sum of the complement.
Conversely, notice that Sn1an<Sn1b<anS_{n-1} - a_n < S_{n-1} - b < a_n, so similarly, if Sn1bS_{n-1} - b is representable, then so is bb. \square

Proof. Similar to the proof of Claim 1, if ana_n can be represented as the sum of two or more distinct terms, then ana_n must be the sum of some subset of {a0,a1,,an1}\{a_0, a_1, \dots, a_{n-1}\}. So Sn1anS_{n-1} - a_n is the sum of the complement. Conversely, since Sn1an<anS_{n-1} - a_n < a_n, so if representable, Sn1anS_{n-1} - a_n is represented by a subset of {a0,a1,,an1}\{a_0, a_1, \dots, a_{n-1}\} and hence ana_n is representable by the complement. \square

By Claim 2, it suffices to find infinitely many representable and non-representable numbers of the form 2t32^{t-3}

Claim 3. For every t3t \ge 3, we have 2t24t632^t \sim 2^{4t-6} - 3 and 24t6>2t32^{4t-6} > 2^t - 3.

Proof. Since S2t3a2t2=2t13<2t3<a2t2S_{2t-3} - a_{2t-2} = 2^{t-1} - 3 < 2^t - 3 < a_{2t-2}, by Claim 1, we have
2t3S2t3(2t3)=22t2. 2^t - 3 \sim S_{2t-3} - (2^t - 3) = 2^{2t-2}.
And since S4t7a4t6=22t33<22t2<a4t6S_{4t-7} - a_{4t-6} = 2^{2t-3} - 3 < 2^{2t-2} < a_{4t-6}, by Claim 1, we have
22t2S4t722t2=24t63. 2^{2t-2} \sim S_{4t-7} - 2^{2t-2} = 2^{4t-6} - 3.
Therefore, 2t324t532^t - 3 \sim 2^{4t-5} - 3. The inequality follows from t3t \ge 3 \square

Since 233=5=a0+a12^3 - 3 = 5 = a_0 + a_1 is representable, and easy to check that
273=125S6125=24S424=17 2^7 - 3 = 125 \sim S_6 - 125 = 24 \sim S_4 - 24 = 17
is non-representable. So by Claim 3, there are infinitely many representable and non-representable numbers of the form 2t32^t - 3.

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 translated into English from zh; metadata (topic, difficulty) added by this project.