My Personal Favourite NT Problem from CNO 2019, P5

投稿日:2023/5/16最終更新日:2025/5/18

Find the square in the sequence

Problem Statement

Given any positive integer c>1,c > 1, denote p(c)p(c) as the largest prime factor of c.c. A sequence {an}\{a_n\} of positive integers satisfies a1>1a_1 > 1 and an+1=an+p(an)a_{n+1} = a_n + p(a_n) for all n1.n \geq 1. Prove that there must exist at least one perfect square in the sequence {an}.\{a_n\}.

Solution

We have to show two examples of sequence

a1=7142128354249a1=12152025\begin{aligned} a_1 &= 7 \mapsto 14 \mapsto 21 \mapsto 28 \mapsto 35 \mapsto 42 \mapsto 49 \\ a_1 &= 12 \mapsto 15 \mapsto 20 \mapsto 25 \end{aligned}

We observe, that if p(an)p(a_n) is very large, then it will stay constant for many iterations until we reach the value p(an)2.p(a_n)^2. This observation can be formalized in claim.

Claim 1.

If p(an)2an,p(a_n)^2 \geq a_n, then

nkm:p(ak)=p(an),\forall n \leq k \leq m : p(a_k) = p(a_n),

where m is maximal such that amp(an)2.a_m \leq p(a_n)^2.

Moreover, am=p(an)2a_m = p(a_n)^2

Proof

We will prove the first part of the claim inductively. The base case is clear by f(k)=f(k).f(k) = f(k). By our induction hypothesis, we have for any nk<mn \leq k < m

p(an)ak+p(an)=ak+p(ak)=ak+1.p(a_n) \mid a_k + p(a_n) = a_k + p(a_k) = a_{k + 1}.

Because ak+1p(an)2,a_{k+1} \leq p(a_n)^2, no longer prime divisor may divide

ak+1p(an)an\frac{a_{k+1}}{p(a_n)} \leq a_n

Thus, p(ak+1)=p(an).p(a_{k + 1}) = p(a_n). In the second part, we see that

amp(an)2<am+1=am+p(an),a_m \leq p(a_n)^2 < a_{m + 1} = a_m + p(a_n),

or

p(an)2p(an)<amp(an)2,p(a_n)^2 - p(a_n) < a_m \leq p(a_n)^2,

which implies with p(an)am,p(a_n) \mid a_m, that

am=p(an)2.a_m = p(a_n)^2. \quad\quad\quad\quad\quad \blacksquare

Now, we compare the sequences {anp(an)}\{\frac{a_n}{p(a_n)}\} and {p(an)}.\{p(a_n) \}. When

anp(an)p(an),\frac{a_n}{p(a_n)} \leq p(a_n),

we are immediately done by 1.

We have,

an+1p(an+1)an+1p(an)=an+p(an)p(an)=anp(an)+1.\frac{a{n+1}}{p(a_{n+1})} \leq \frac{a_{n + 1}}{p(a_n)} = \frac{a_n + p(a_n)}{p(a_n)} = \frac{a_n}{p(a_n)} + 1.

So if {anp(an)}\{\frac{a_n}{p(a_n)} \} grows without bound, it hits very integer a1p(a1).\leq \frac{a_1}{p(a_1)}. At least one of those is a prime qq because there are indefinitely many primes. The definition of pp implies

p(an)q=anp(an)p(a_n) \geq q = \frac{a_n}{p(a_n)}