Olympiad Maths Prep

Library / /2 of 3

Number theory Difficulty 7.7 National olympiad, round 2 Prove it Balkan Mathematical Olympiad

Find all polynomials P(x)P(x) with integer coefficients such that there exists a positive integer NN such that for all positive integers n>Nn > N we have P(n)>0P(n) > 0 and
n+P(n)nP(n)+P(n)n. n + P(n) \mid n^{P(n)} + P(n)^n.

Solution

The solutions are P(n)=1P(n) = 1 and P(n)=2snknP(n) = 2^s n^k - n for 0s30 \le s \le 3 and k1k \ge 1, excluding P(n)=0P(n) = 0.

Firstly, we denote Q(n)=P(n)+nQ(n) = P(n) + n. We then have Q(n)>nQ(n) > n for all n>Nn > N. Note that the condition can be rewritten as
nQ(n)n+(Q(n)n)n0(modQ(n)) n^{Q(n)-n} + (Q(n) - n)^n \equiv 0 \pmod{Q(n)}
nQ(n)n+(n)n0(modQ(n)) n^{Q(n)-n} + (-n)^n \equiv 0 \pmod{Q(n)}
Let Q(0)=a0Q(0) = a_0. Assume for the sake of contradiction that a00a_0 \neq 0 and let n>Nn > N be an odd positive integer such that (a0,n)=1(a_0, n) = 1. Notice that (Q(n),n)=1(Q(n), n) = 1, since if some prime pp divides both nn and Q(n)Q(n), it would also divide a0a_0, contradiction. Therefore we have
nn(nQ(n)2n1)0(modQ(n))    nQ(n)2n10(modQ(n)), n^n(n^{Q(n)-2n} - 1) \equiv 0 \pmod{Q(n)} \implies n^{Q(n)-2n} - 1 \equiv 0 \pmod{Q(n)},
that is
nQ(n)2n1(modQ(n))(1) n^{Q(n)-2n} \equiv 1 \pmod{Q(n)} \quad (1)
Let pQ(n)p \mid Q(n) be an odd prime (we know that pp doesn't divide nn). It follows that ordp(n)Q(n)2n\text{ord}_p(n) \mid Q(n) - 2n. Let n0>Nn_0 > N be any odd integer. By the Chinese remainder theorem there exists an integer m>Nm > N such that
mn(modp) m \equiv n \pmod{p}
mn0(modp1). m \equiv n_0 \pmod{p-1}.
Therefore, pQ(m)p \mid Q(m). Since n0n_0 is odd and pnp \nmid n, it follows that mm is odd and pmp \nmid m. Therefore we can conclude that (in a similar fashion as we have done for nn):
mQ(m)2m1(modp), m^{Q(m)-2m} \equiv 1 \pmod{p},
implying ordp(m)Q(m)2m\text{ord}_p(m) \mid Q(m) - 2m. Since mn(modp)m \equiv n \pmod{p} and mn0(modp1)m \equiv n_0 \pmod{p-1}, it follows that ordp(n)Q(n0)2n0\text{ord}_p(n) \mid Q(n_0) - 2n_0.
Therefore
ordp(n)Q(m)2m, \text{ord}_p(n) \mid Q(m) - 2m,
for all odd m>Mm > M, but due to the periodicity of the polynomial QQ modulo ordp(n)\text{ord}_p(n), this in fact means that the above holds for any odd integer mm. Now we have that
ordp(n)gcd(Q(1)2,Q(3)6,,Q(2i+1)2(2i+1),). \text{ord}_p(n) \mid \text{gcd}(Q(1) - 2, Q(3) - 6, \dots, Q(2i + 1) - 2(2i + 1), \dots).
Denote the previous gcd with GG. Let pGp \mid G be an odd prime. It follows that pQ(p)2pp \mid Q(p) - 2p, implying pa0p \mid a_0. Let kk be a large enough positive integer. Then vp(Q(a0k)2a0k)=vp(a0)v_p(Q(a_0^k) - 2a_0^k) = v_p(a_0). Therefore vp(G)vp(a0)v_p(G) \le v_p(a_0) and so it follows that G2ta0G \mid 2^t a_0 for some integer t0t \ge 0.
Coming back to our original nn we get that
nG1(modp). n^G \equiv 1 \pmod{p}.

From (1) and Lifting The Exponent lemma it follows that
vp(nQ(n)2n1)=vp((nG)Q(n)2nG1)=vp(nG1)+vp(Q(n)2nG)vp(Q(n)). v_p(n^{Q(n)-2n}-1) = v_p((n^G)^{\frac{Q(n)-2n}{G}} - 1) = v_p(n^G - 1) + v_p(\frac{Q(n)-2n}{G}) \ge v_p(Q(n)).
As pQ(n)p \mid Q(n) and pnp \nmid n, we get vp(Q(n)2nG)=0v_p(\frac{Q(n)-2n}{G}) = 0.
If 2Q(n)2 \mid Q(n) Lifting The Exponent lemma for p=2p = 2 yields v2(nQ(n)2n1)=v2(n21)+v2(Q(n)2n)1v2(Q(n))v_2(n^{Q(n)-2n}-1) = v_2(n^2-1)+v_2(Q(n)-2n)-1 \ge v_2(Q(n)). If 4Q(n)4 \mid Q(n), it follows that 2Q(n)2n2 \nmid Q(n)-2n, yielding v2(n21)v2(Q(n))v_2(n^2-1) \ge v_2(Q(n)). If 2Q(n)2 \nmid Q(n), v2(n21)v2(Q(n))v_2(n^2-1) \ge v_2(Q(n)) still holds.
If 2Q(n)2 \mid Q(n), it follows that 2G2 \mid G, and we get vp(nG1)vp(Q(n))v_p(n^G - 1) \ge v_p(Q(n)) for all primes pQ(n)p \mid Q(n), implying Q(n)nG1Q(n) \mid n^G - 1 (2).
As GG is independent of nn, we get
Q(n)nG1(2) Q(n) \mid n^G - 1 \qquad (2)
for every odd integer n>Nn > N such that (a0,n)=1(a_0, n) = 1.
We say that a polynomial with integer coefficients is *primitive* if the greatest common divisor of its coefficients is 1.
Lemma. Let P(x),Q(x)Z[x]P(x), Q(x) \in \mathbb{Z}[x] be primitive such that Q(n)P(n)Q(n) \mid P(n) for infinitely many positive integers nn. Then there exists a polynomial F(x)Z[x]F(x) \in \mathbb{Z}[x] such that P(x)=Q(x)F(x)P(x) = Q(x)F(x).
Proof. By the polynomial division algorithm there exist polynomials F(x),R(x)Q[x]F(x), R(x) \in \mathbb{Q}[x] such that P(x)=Q(x)F(x)+R(x)P(x) = Q(x)F(x) + R(x) and deg(R(x))<deg(Q(x))\deg(R(x)) < \deg(Q(x)). Dividing the expression by Q(x)Q(x) yields
P(x)Q(x)=F(x)+R(x)Q(x). \frac{P(x)}{Q(x)} = F(x) + \frac{R(x)}{Q(x)}.
Let n1,n2,n_1, n_2, \dots be a sequence of positive integers such that Q(ni)P(ni)Q(n_i) \mid P(n_i) and let P(ni)=diQ(ni)P(n_i) = d_iQ(n_i) for diZd_i \in \mathbb{Z}, for all iNi \in \mathbb{N}. Let DD be the least common multiple of the denominators of the coefficients of F(x)F(x). Then we can write F(ni)=fiDF(n_i) = \frac{f_i}{D} and R(ni)Q(ni)=ri\frac{R(n_i)}{Q(n_i)} = r_i for fiZ,riQf_i \in \mathbb{Z}, r_i \in \mathbb{Q}. We get the following equation for every ii:
di=fiD+ridiDfiD=ri. d_i = \frac{f_i}{D} + r_i \\ \frac{d_i D - f_i}{D} = r_i.
As limiri=0\lim_{i \to \infty} r_i = 0, there has to exist a large enough positive integer jj such that 0<rj<1D0 < |r_j| < \frac{1}{D}. Therefore R(x)0R(x) \equiv 0 and P(x)=Q(x)F(x)P(x) = Q(x)F(x). We are left to prove that FF has integer coefficients.
Let aa be the greatest common divisor of the numerators of the coefficients of F(x)F(x) and let bb be the least common multiple of the denominators of the coefficients of F(x)F(x). We can assume (a,b)=1(a, b) = 1. Therefore F(x)=abF1(x)F(x) = \frac{a}{b}F_1(x) for some primitive polynomial F1(x)Z[x]F_1(x) \in \mathbb{Z}[x]. Now we get baP(x)=Q(x)F1(x)\frac{b}{a}P(x) = Q(x)F_1(x). Since baP(x)\frac{b}{a}P(x) is a polynomial with integer coefficients and (a,b)=1(a, b) = 1, it follows that aa divides all coefficients of P(x)P(x). Therefore a=1a = 1 (since PP is primitive).
As Q(x)Q(x) and F1(x)F_1(x) are primitive, bP(x)bP(x) is primitive by Gauss' lemma. Therefore b=1b = 1 and F(x)Z[x]F(x) \in \mathbb{Z}[x], as desired. \square
Let Q(x)=dQ1(x)Q(x) = dQ_1(x), where dd is a positive integer (possibly d=1d = 1) and Q1(x)Z[x]Q_1(x) \in \mathbb{Z}[x] is primitive. Assume there exists an odd prime pdp \mid d and let k1(modp)k \equiv 1 \pmod{p} be an even integer such that

k>Nk > N. Then kQ(k)k+(k)k2(modp)k^{Q(k)-k} + (-k)^k \equiv 2 \pmod{p}, so p=2p = 2, which is a contradiction. Therefore d=2ld = 2^l for some integer l0l \ge 0.
Let P1(x)=xG1P_1(x) = x^G - 1. Since Q1(n)P1(n)Q_1(n) | P_1(n) for infinitely many positive integers nn and P1(x),Q1(x)P_1(x), Q_1(x) are primitive, by the above lemma there exists a polynomial F(x)Z[x]F(x) \in \mathbb{Z}[x] such that P1(x)=Q1(x)F(x)P_1(x) = Q_1(x)F(x). In particular, 1=P1(0)=Q1(0)F(0)-1 = P_1(0) = Q_1(0)F(0), so Q1(0)Q_1(0) is either 1 or 1-1. Therefore a0a_0 is either 2l2^l or 2l-2^l.
Let n>Nn > N be an even positive integer. If pQ(n)p \mid Q(n) and pnp \mid n, it follows that p=2p = 2. Additionally, v2(Q(n))=v2(d)=lv_2(Q(n)) = v_2(d) = l. The original condition yields
2l(nQ(n)2n+1)0(modQ(n)) 2^l(n^{Q(n)-2n} + 1) \equiv 0 \pmod{Q(n)}
nQ(n)2n1(modQ1(n)) n^{Q(n)-2n} \equiv -1 \pmod{Q_1(n)}
n2(Q(n)2n)1(modQ1(n)) n^{2(Q(n)-2n)} \equiv 1 \pmod{Q_1(n)}
Let pQ1(n)p \mid Q_1(n) be an odd prime (pnp \nmid n). Picking n0n_0 divisible by p1p-1 such that n0n(modp)n_0 \equiv n \pmod{p} yields
na01(modp). n^{a_0} \equiv -1 \pmod{p}.
If a0<0a_0 < 0, i.e. a0=a0a_0 = -|a_0|, then
na0+1=na0+1=na0+1na00(modp) n^{|a_0|} + 1 = n^{-a_0} + 1 = \frac{n^{a_0} + 1}{n^{a_0}} \equiv 0 \pmod{p}
In either case we have
na01(modp) n^{|a_0|} \equiv -1 \pmod{p}
n2l1(modp). n^{2^l} \equiv -1 \pmod{p}.
Therefore ordp(n)=2l+1\text{ord}_p(n) = 2^{l+1}. Using Lifting The Exponent lemma analogously as before (without the case p=2p = 2 as 2Q1(n)2 \nmid Q_1(n)), we get:
vp(n2(Q(n)2n)1)=vp((n2l+1)2(Q(n)2n)2l+11)=vp(n2l+11)+vp(2(Q(n)2n)2l+1). v_p(n^{2(Q(n)-2n)} - 1) = v_p((n^{2^{l+1}})^{\frac{2(Q(n)-2n)}{2^{l+1}}} - 1) = v_p(n^{2^{l+1}} - 1) + v_p(\frac{2(Q(n)-2n)}{2^{l+1}}).
Since vp(2(Q(n)2n)2l+1)=0v_p(\frac{2(Q(n)-2n)}{2^{l+1}}) = 0 (because pQ1(n)p \mid Q_1(n) so pQ(n)p \mid Q(n) and pnp \nmid n), we get that vp(n2(Q(n)2n)1)=vp(n2l+11)vp(Q1(n))v_p(n^{2(Q(n)-2n)} - 1) = v_p(n^{2^{l+1}} - 1) \ge v_p(Q_1(n)) and so:
n2l+11(modQ1(n)) n^{2^{l+1}} \equiv 1 \pmod{Q_1(n)}
(n21)(n2+1)0(modQ1(n)) (n^2 - 1)(n^2 + 1) \equiv 0 \pmod{Q_1(n)}
n2l1(modQ1(n)), n^{2^l} \equiv -1 \pmod{Q_1(n)},
where the last equation holds because (n21,Q1(n))=1(n^2 - 1, Q_1(n)) = 1 (since any prime pp that divides Q1(n)Q_1(n) also divides n2+1n^2 + 1, and pp is odd).
Thus Q1(n)n2+1Q_1(n) \mid n^2 + 1 for all even positive integers n>Nn > N. As Q1(x)Q_1(x) and x2+1x^2 + 1 are primitive, by the lemma there exists a polynomial F1(x)Z[x]F_1(x) \in \mathbb{Z}[x] such that x2+1=Q1(x)F1(x)x^2 + 1 = Q_1(x)F_1(x). Since x2+1x^2 + 1 is known to be irreducible and Q1(x)Q_1(x) is eventually positive, it must hold that Q1(x)=x2+1Q_1(x) = x^2 + 1.
We return to odd values of nn once again. If l1l \ge 1, we obtain G=2G = 2 (it is immediate for l>1l > 1 and for l=1l = 1 it follows from GQ1(1)G \mid Q_1(1)).
Let n>Nn > N be an odd positive integer. Divisibility (2) yields
n21(mod2ln2l+2l), n^2 \equiv 1 \pmod{2^l n^{2^l} + 2^l},

Therefore l=0l = 0 and Q(x)=x+1Q(x) = x + 1, which is easily seen to be a solution to the problem (it gives P(x)=1P(x) = 1).
Now assume a0=0a_0 = 0. Let Q(x)=xR(x)Q(x) = xR(x) for some R(x)Z[x]R(x) \in \mathbb{Z}[x]. The original condition after cancelling nn rewrites as
R(n)nn1(nnR(n)2n+(1)n). R(n) \mid n^{n-1}(n^{nR(n)}-2n + (-1)^n).
Assume there exists an even integer n>Nn > N and a prime pR(n)p \mid R(n) such that pnp \nmid n. Then pnnR(n)2n+1p \mid n^{nR(n)-2n} + 1. Let n0>Nn_0 > N be an even integer such that p1n0p-1 \mid n_0 and n0n(modp)n_0 \equiv n \pmod{p}. It follows that pn0n0R(n0)2n0+1p \mid n_0^{n_0R(n_0)-2n_0} + 1 implying p2p \mid 2, which is false since nn is even. (3)
Let R(x)=xk1G(x)R(x) = x^{k-1}G(x) for some G(x)Z[x]G(x) \in \mathbb{Z}[x] such that G(0)0G(0) \neq 0 and k1k \ge 1. If G(x)G(x) is non-constant, by Schur's theorem there are infinitely many primes pp dividing G(2mp)G(2m_p) for some mpNm_p \in \mathbb{N}. For large enough such primes pp (and thus also 2mp2m_p), we have pG(0)p \nmid G(0) and thus p2mpp \nmid 2m_p, which is impossible by (3).
Therefore G(x)cG(x) \equiv c for some constant cc. It follows from (3) and eventual positivity of Q(x)Q(x) that c=2sc = 2^s for some integer s0s \ge 0, that is Q(x)=2sxkQ(x) = 2^s \cdot x^k.
We have
nQ(n)n+(n)n0(mod2snk). n^{Q(n)-n} + (-n)^n \equiv 0 \pmod{2^s n^k}.
However, if s4s \ge 4, let n3(mod16)n \equiv 3 \pmod{16} then, Q(n)n5(mod8)Q(n) - n \equiv 5 \pmod{8},
35+(3)38(mod16), 3^5 + (-3)^3 \equiv 8 \pmod{16},
which is a contradiction. Therefore, s3s \le 3.
We will show that all pairs (s,k)(s, k) with 3s03 \ge s \ge 0 and k1k \ge 1 except for s=0,k=1s = 0, k = 1 satisfy the problem's conditions (s=0s = 0 and k=1k = 1 fails since we need to have Q(n)>nQ(n) > n for n>Nn > N).
If nn is even, 2snknn12^s n^k \mid n^{n-1} for large enough positive integers nn.
If nn is odd, nknn1n^k \mid n^{n-1} for large enough positive integers nn and it remains to prove that 2sn2snk2n12^s \mid n^{2^s n^{k-2n}} - 1. For s=0s = 0 the divisibility is obvious and for s1s \ge 1 we have that n2snk2n=n2(2s1nkn)1(mod8)n^{2^s n^{k-2n}} = n^{2(2^{s-1}n^k-n)} \equiv 1 \pmod{8}, implying the claim.
Finally, Q(n)>nQ(n) > n is obviously satisfied for all such s,ks, k and the result follows.

Looking for a route rather than 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 and solution reproduced as published; topic and difficulty added by this site.