Unofficial archive — problems, solutions & results © IMC, reproduced with permission.

IMC / 2011 / Problems / Day 2, P9

IMC 2011 · Day 2 · P9

Let f(x)f(x) be a polynomial with real coefficients of degree nn. Suppose that f(k)f(m)km\dfrac{f(k) - f(m)}{k - m} is an integer for all integers 0k<mn0 \le k < m \le n. Prove that aba - b divides f(a)f(b)f(a) - f(b) for all pairs of distinct integers aa and bb.

(Fedor Petrov, St. Petersburg)

Solution 1 of 2 (official)

We need the following

Lemma. Denote the least common multiple of 1,2,,k1, 2, \dots, k by L(k)L(k), and define hk(x)=L(k)(xk)(k=1,2,).h_k(x) = L(k) \cdot \binom{x}{k} \quad (k = 1, 2, \dots). Then the polynomial hk(x)h_k(x) satisfies the condition, i.e.\ aba - b divides hk(a)hk(b)h_k(a) - h_k(b) for all pairs of distinct integers aa, bb.

Proof. It is known that (ak)=j=0k(abj)(bkj).\binom{a}{k} = \sum_{j=0}^{k} \binom{a-b}{j} \binom{b}{k-j}. (This formula can be proved by comparing the coefficient of xkx^k in (1+x)a(1+x)^a and (1+x)ab(1+x)b(1+x)^{a-b} (1+x)^b.) From here we get hk(a)hk(b)=L(K)((ak)(bk))=L(K)j=1k(abj)(bkj)=(ab)j=1kL(k)j(ab1j1)(bkj).h_k(a) - h_k(b) = L(K) \left( \binom{a}{k} - \binom{b}{k} \right)

= L(K) \sum_{j=1}^{k} \binom{a-b}{j} \binom{b}{k-j} = (a-b) \sum_{j=1}^{k} \frac{L(k)}{j} \binom{a-b-1}{j-1} \binom{b}{k-j}. On the right-hand side all fractions L(k)j\frac{L(k)}{j} are integers, so the right-hand side is a multiple of (a,b)(a, b).

The lemma is proved.

Expand the polynomial ff in the basis 1,(x1),(x2),1, \binom{x}{1}, \binom{x}{2}, \dots as f(x)=A0+A1(x1)+A2(x2)++An(xn).(1)\tag{1} f(x) = A_0 + A_1 \binom{x}{1} + A_2 \binom{x}{2} + \dots + A_n \binom{x}{n}. We prove by induction on jj that AjA_j is a multiple of L(j)L(j) for 1jn1 \le j \le n. (In particular, AjA_j is an integer for j1j \ge 1.) Assume that L(j)L(j) divides AjA_j for 1jm11 \le j \le m-1. Substituting mm and some k{0,1,,m1}k \in \{0, 1, \dots, m-1\} in (1), f(m)f(k)mk=j=1m1AjL(j)hj(m)hj(k)mk+Ammk.\frac{f(m) - f(k)}{m - k} = \sum_{j=1}^{m-1} \frac{A_j}{L(j)} \cdot \frac{h_j(m) - h_j(k)}{m - k} + \frac{A_m}{m - k}. Since all other terms are integers, the last term Ammk\frac{A_m}{m-k} is also an integer. This holds for all 0k<m0 \le k < m, so AmA_m is an integer that is divisible by L(m)L(m).

Hence, AjA_j is a multiple of L(j)L(j) for every 1jn1 \le j \le n. By the lemma this implies the problem statement.

Solution 2 of 2 (official)

The statement of the problem follows immediately from the following claim, applied to the polynomial g(x,y)=f(x)f(y)xyg(x, y) = \dfrac{f(x) - f(y)}{x - y}.

Claim. Let g(x,y)g(x, y) be a real polynomial of two variables with total degree less than nn. Suppose that g(k,m)g(k, m) is an integer whenever 0k<mn0 \le k < m \le n are integers. Then g(k,m)g(k, m) is a integer for every pair k,mk, m of integers.

Proof. Apply induction on nn. If n=1n = 1 then gg is a constant. This constant can be read from g(0,1)g(0, 1) which is an integer, so the claim is true.

Now suppose that n2n \ge 2 and the claim holds for n1n - 1. Consider the polynomials g1(x,y)=g(x+1,y+1)g(x,y+1)andg2(x,y)=g(x,y+1)g(x,y).(1)\tag{1} g_1(x, y) = g(x+1, y+1) - g(x, y+1) \quad \text{and} \quad g_2(x, y) = g(x, y+1) - g(x, y). For every pair 0k<mn10 \le k < m \le n-1 of integers, the numbers g(k,m)g(k, m), g(k,m+1)g(k, m+1) and g(k+1,m+1)g(k+1, m+1) are all integers, so g1(k,m)g_1(k, m) and g2(k,m)g_2(k, m) are integers, too. Moreover, in (1) the maximal degree terms of gg cancel out, so degg1,degg2<degg\deg g_1, \deg g_2 < \deg g. Hence, we can apply the induction hypothesis to the polynomials g1g_1 and g2g_2 and we thus have g1(k,m),g2(k,m)Zg_1(k, m), g_2(k, m) \in \mathbb{Z} for all k,mZk, m \in \mathbb{Z}.

In view of (1), for all k,mZk, m \in \mathbb{Z}, we have that

(a) g(0,1)Zg(0, 1) \in \mathbb{Z};

(b) g(k,m)Zg(k, m) \in \mathbb{Z} if and only if g(k+1,m+1)Zg(k+1, m+1) \in \mathbb{Z};

(c) g(k,m)Zg(k, m) \in \mathbb{Z} if and only if g(k,m+1)Zg(k, m+1) \in \mathbb{Z}.

For arbitrary integers k,mk, m, apply (b) k|k| times then apply (c) mk1|m - k - 1| times as g(k,m)Zg(0,mk)Zg(0,1)Z.g(k, m) \in \mathbb{Z} \Leftrightarrow \dots \Leftrightarrow g(0, m-k) \in \mathbb{Z} \Leftrightarrow \dots \Leftrightarrow g(0, 1) \in \mathbb{Z}. Hence, g(k,m)Zg(k, m) \in \mathbb{Z}. The claim has been proved.

Similar problems

IMC 2001 · Day 1 · P4killeravg 0.3/10 · solved 2% · near-0 95% · disc 0.32
IMC 2006 · Day 2 · P11mediumavg 3.8/10 · solved 35% · near-0 55% · disc 0.53
IMC 2007 · Day 1 · P1easyavg 9.7/10 · solved 97% · near-0 0% · disc 0.16