🧮 Brain Teaser

The Polynomial That Divides Its Own Composition

Let ff be a nonconstant polynomial with positive integer coefficients. Prove that for every positive integer nn,

nf(n)    nf(f(n)+1).n \mid f(n) \implies n \mid f(f(n) + 1).

Wait — prove the stronger claim from Putnam 2007 B-1:

Let ff be a nonconstant polynomial with positive integer coefficients. Prove that for any positive integer nn, f(n)f(n) divides f(f(n)+1)f(f(n)+1) if and only if n=1n = 1.

polynomialdivisibilitynumber theorypositive coefficientsmodular arithmetic

Answer: The Polynomial That Divides Its Own Composition

Key Idea / Intuition

The magic here is a single observation: for any polynomial ff with integer coefficients and any integers a,ba, b, we have (ab)(f(a)f(b))(a - b) \mid (f(a) - f(b)). So if we let a=f(n)+1a = f(n) + 1 and b=nb = n, then (f(n)+1n)(f(f(n)+1)f(n))(f(n) + 1 - n) \mid (f(f(n)+1) - f(n)). This lets us control f(f(n)+1)f(f(n)+1) modulo f(n)f(n). The "positive integer coefficients" condition then prevents accidental cancellation and pins down why only n=1n=1 works.


Formal Proof / Solution

Setup and the key divisibility lemma.

For any polynomial ff with integer coefficients and integers a,ba, b: (ab)(f(a)f(b)).(a - b) \mid (f(a) - f(b)).

This follows because akbk=(ab)(ak1++bk1)a^k - b^k = (a-b)(a^{k-1} + \cdots + b^{k-1}), so each term of f(a)f(b)f(a) - f(b) is divisible by (ab)(a-b).


Step 1: Reduce f(f(n)+1)f(f(n)+1) modulo f(n)f(n).

Apply the lemma with a=f(n)+1a = f(n) + 1 and b=nb = n:

(f(n)+1n)(f(f(n)+1)f(n)).\bigl(f(n) + 1 - n\bigr) \mid \bigl(f(f(n)+1) - f(n)\bigr).

So: f(f(n)+1)f(n)(modf(n)+1n).f(f(n)+1) \equiv f(n) \pmod{f(n)+1-n}.

But we want to work modulo f(n)f(n) itself. Apply the lemma differently: set a=f(n)+1a = f(n)+1, b=1b = 1:

(f(n))(f(f(n)+1)f(1)).(f(n)) \mid (f(f(n)+1) - f(1)).

That is: f(f(n)+1)f(1)(modf(n)).f(f(n)+1) \equiv f(1) \pmod{f(n)}.


Step 2: Use this to analyze when f(n)f(f(n)+1)f(n) \mid f(f(n)+1).

From Step 1: f(n)f(f(n)+1)    f(n)f(1).f(n) \mid f(f(n)+1) \iff f(n) \mid f(1).

(Since f(f(n)+1)f(1)(modf(n))f(f(n)+1) \equiv f(1) \pmod{f(n)}, divisibility of f(f(n)+1)f(f(n)+1) by f(n)f(n) is equivalent to f(n)f(1)f(n) \mid f(1).)


Step 3: Since coefficients are positive integers, ff is strictly increasing on positive integers.

Because ff has positive integer coefficients and is nonconstant, write f(x)=adxd++a1x+a0f(x) = a_d x^d + \cdots + a_1 x + a_0 with ai0a_i \geq 0, ad1a_d \geq 1, d1d \geq 1. For positive integers n1n \geq 1:

f(n)na1+a01,f(n) \geq n \cdot a_1 + a_0 \geq 1,

and more importantly f(n)f(1)1f(n) \geq f(1) \geq 1 with equality only if n=1n = 1 (since each term ainiai1ia_i n^i \geq a_i \cdot 1^i for n1n \geq 1, and at least one term is strictly increasing).

Precisely: for n2n \geq 2, since ad1a_d \geq 1 and d1d \geq 1: f(n)ndn2>1=f(1) only if f(1)=1,f(n) \geq n^d \geq n \geq 2 > 1 = f(1) \text{ only if } f(1)=1,

but in general f(n)>f(1)f(n) > f(1) for n>1n > 1 because all coefficients are positive:

f(n)f(1)=k=1dak(nk1)ad(nd1)nd1n11>0.f(n) - f(1) = \sum_{k=1}^{d} a_k(n^k - 1) \geq a_d(n^d - 1) \geq n^d - 1 \geq n - 1 \geq 1 > 0.

So for n2n \geq 2: f(n)>f(1)1f(n) > f(1) \geq 1, which means f(n)f(1)f(n) \nmid f(1) (a larger positive integer cannot divide a smaller positive integer).


Step 4: Conclusion.

  • If n=1n = 1: f(n)=f(1)f(n) = f(1) and we need f(1)f(1)f(1) \mid f(1), which is obviously true. ✓

  • If n2n \geq 2: f(n)>f(1)1f(n) > f(1) \geq 1, so f(n)f(1)f(n) \nmid f(1), hence f(n)f(f(n)+1)f(n) \nmid f(f(n)+1). ✗

Therefore, f(n)f(f(n)+1)f(n) \mid f(f(n)+1) if and only if n=1n = 1. \blacksquare


The elegant summary:

f(f(n)+1)f(1)(modf(n)),f(f(n)+1) \equiv f(1) \pmod{f(n)},

and for n2n \geq 2, f(n)>f(1)f(n) > f(1) because all coefficients are positive, so no divisibility can occur.

Source: Putnam 2007, Problem B-1

Type: PutnamSource: Putnam 2007, Problem B-1Edit on GitHub ↗