ON THE DIOPHANTINE EQUATION (x+1)k +(x+2)k +...+(lx)k = yn 7 GO¨KHANSOYDAN 1 0 To my wife and my daughter 2 n Abstract. Let k,l ≥2 be fixed integers. In this paper, firstly, we prove a that all solutions of the equation (x+1)k+(x+2)k+...+(lx)k =yn in J 0 integers x,y,n with x,y ≥1,n≥2 satisfy n<C1 where C1 =C1(l,k) is an effectively computable constant. Secondly, we prove that all solutions 1 of this equation in integers x,y,n with x,y ≥ 1,n ≥ 2,k 6= 3 and l ≡ 0 ] (mod 2) satisfy max{x,y,n} < C2 where C2 is an effectively computable T constant dependingonly on k and l. N . h t a 1. Introduction m [ In 1956, J.J. Scha¨ffer [14] considered the equation 1 v (1.1) 1k +2k +...+xk = yn. 6 6 He proved that for fixed k ≥ 1 and n ≥ 2, (1.1) has at most finitely many 4 2 solutions in positive integers x and y, unless 0 . (k,n) ∈ {(1,2),(3,2),(3,4),(5,2)}, 1 0 where, in each case, there are infinitely many such solutions. 7 Scha¨ffer’s proof used an ineffective method due to Thue and Siegel so his 1 : result is also ineffective. This means that the proof does not provide any algo- v rithm to find all solutions. Applying Baker’s method, K. Gy˝ory, R. Tijdeman i X and M. Voorhoeve [6] proved a more general and effective result in which the r exponent n is also unknown. a Let k ≥ 2 and r be fixed integers with k ∈/ {3,5} if r = 0, and let s be a square-free odd integer. In [6], they proved that the equation s(1k +2k +...+xk)+r = yn Date: January 11, 2017. 2010 Mathematics Subject Classification. Primary 11D61; Secondary 11B68. Key words and phrases. Bernoulli polynomials, high degree equations. 1 2 GO¨KHANSOYDAN in positive integers x,y ≥ 2, n ≥ 2 has only finitely many solutions and all these can be effectively determined. Of particular importance is the special case when s = 1 and r = 0. They also showed that for given k ≥ 2 with k ∈/ {3,5}, equation (1.1) has only finitely many solutions in integers x,y ≥ 1, n ≥ 2, and all these can be effectively determined. The following striking result is due to Voorhoeve, Gy˝ory and Tijdeman [17]. Let R(x) be a fixed polynomial with integer coefficients and let k ≥ 2 be a fixed integer such that k ∈/ {3,5}. In [17], same authors proved that the equation 1k +2k +...+xk +R(x)= byn in integers x,y ≥ 2, n ≥ 2 has only finitely many solutions, and an effective upper bound can be given for n. Later, various generalizations and analogues of the results of Gy˝ory, Tijdeman and Voorhoeve have been established by several authors [1], [2], [3], [4], [5], [8], [11], [16]. For a survey of these results we refer to [7] and the references given there. Here we present the result of B. Brindza [2]. For brevity let us set S (x) = k 1k +2k +...+xk, A= Z[x], κ = (k+1) p (p prime). Let (p−1)|(k+1)! Y F(y) = Q yn+...+Q y+Q ∈ A[y]. n 1 0 Consider the equation (1.2) F(S (x)) = yn k in integers x,y ≥ 2, n ≥ 2. Let Q (x) = κiK (x) where K (x) ∈ Z[x] for i i i i = 2,3,...,m. In [2], Brindza proved that if Q (x) ≡ 0 (mod κi), for i = i 2,3,...,m; Q (x) ≡ ±1 (mod 4) and k ∈/ {1,2,3,5}, then all solutions of (1.2) 1 satisfy max{x,y,n} < C , where C is an effectively computable constant 1 1 depending only on F and k. Recently C. Rakaczki [12] gave a generalization of the results of Gy˝ory, Tijdeman and Voorhoeve and an extension of the result of Brindza to the case when the polynomials Q (x) are arbitrary constant polynomials. i Let F(x) bea polynomial with rational coefficients and d6= 0 bean integer. Suppose that F(x) is not an n-th power. In [12], Rakaczki showed that the equation F(S (x)) = dyn k has only finitely many integer solutions x,y ≥ 2, n ≥ 2, which can be effec- tively determined provided that k ≥ 6. Let k > 1, r,s 6= 0 be fixed integers. Then apart from the cases when (i) k = 3 and either r = 0 or s+64r = 0, and (ii) k = 5 and either r = 0 or ON THE DIOPHANTINE EQUATION (x+1)k+(x+2)k+...+(lx)k =yn 3 s−324r =0, Rakaczki proved that the equation s(1k +2k +...+xk)+r = yn in integers x > 0, y with |y| ≥ 2, and n ≥ 2 has only finitely many solutions which can be effectively determined. Recently, Z. Zhang [18] studied the Diophantine equation (x−1)k +xk +(x+1)k = yn,n > 1, and completely solved it for k = 2,3,4. Now we consider a more general equation. Let G(x) = (x+1)k +(x+2)k +...+(lx)k. In this paper, we are interested in the solutions of the equation (1.3) G(x) = yn in integers x,y ≥ 1 and n ≥ 2. Theorem 1. Let k,l ≥ 2 fixed integers. Then all solutions of the equation (1.3) in integers x,y ≥ 1 and n ≥ 2 satisfy n < C where C is an effectively 1 1 computable constant depending only on l and k. Theorem 2. Let k,l ≥ 2 fixed integers such that k 6= 3. Then all solutions of the equation (1.3) in integers x,y,n with x,y ≥ 1, n ≥ 2, and l ≡ 0 (mod 2) satisfy max{x,y,n} < C where C is an effectively computable constant de- 2 2 pending only on l and k. We organize this paper as follows. In Section 2, firstly, we recall the general results that we will need. Secondly, we give two new lemmas and prove that these lemmas imply our theorems. In Section 3, we discuss the number of solutions in integers x,y ≥1 of (1.3) where n > 1 is fixed k ∈ {1,3} and l ≡ 0 (mod 2) and reformulate this case. In the last section, we give the proofs of Theorems 1 and 2. 2. Auxiliary Results B (lx+1)−B (x+1) Lemma 1. (x + 1)k + (x + 2)k +... + (lx)k = k+1 k+1 k+1 where q 1 1 q q B (x) = xq − qxq−1+ xq−2+... = B xq−i q i 2 6 2 i (cid:18) (cid:19) i=0(cid:18) (cid:19) X is the q-th Bernoulli polynomial with q = k+1. 4 GO¨KHANSOYDAN Proof. It is an application of the equality N−1 1 nk = (B (N)−B (M)) k+1 k+1 k+1 n=M X which is given by Rademacher in [13], pp.3-4. (cid:3) Now we give an important result of Brindza which is an effective version of Leveque’s theorem [9] Lemma 2 (Brindza). Let H(x)∈ Q[x], n H(x) =a xN +...+a = a (x−α )ri, 0 N 0 i i=1 Y with a 6= 0 and α 6= α for i 6= j. Let 0 6= b ∈ Z, 2 ≤ m ∈ Z and define 0 i j t = m . Suppose that {t ,...,t } is not a permutation of the n-tuples i (m,ri) 1 n (a) {t,1,...,1}, t ≥ 1; (b) {2,2,1,...,1}. Then all solutions (x,y) ∈ Z2 of the equation H(x) = bym satisfy max{|x|,|y|} < C, where C is an effectively computable constant de- pending only on H, b and m. Proof. See B. Brindza [2]. (cid:3) Lemma 3 (Schinzel & Tijdeman). Let 0 6= b ∈ Z and let P(x) ∈ Q[x] be a polynomial with at least two distinct zeroes. Then the equation P(x)= byn in integers x,y > 1, n implies that n < C where C = (P,b) is an effectively computable constant. Proof. See A. Schinzel and R. Tijdeman [15]. (cid:3) Lemma 4. For k ∈ Z+ let B (x) be the k-th Bernoulli polynomial. Then the k polynomial B (lx+1)−B (x+1) k+1 k+1 G(x) = k+1 has at least two distinct zeroes. Proof. By Lemma1,we haveG(x) = (lk+1−1)xk+1+(lk−1)xk+...+cxwherec k+1 2 is arational number. Now one can observe thatthecoefficient of xk is nonzero ON THE DIOPHANTINE EQUATION (x+1)k+(x+2)k+...+(lx)k =yn 5 and that x = 0 is a zero of G(x). Let’s also assume that there is no other zero of G(x). Thus we have lk+1−1 G(x) = xk+1 k+1 (cid:18) (cid:19) which is a contradiction. (cid:3) Lemma 5 (Voorhoeve, Gy˝ory and Tijdeman). Let q ≥ 2, R∗(x) ∈ Z[x] and set Q(x) = B (x)−B +qR∗(x). q q Then (i) Q(x) has at least three zeros of odd multiplicity, unless q ∈ {2,4,6}. (ii) For any odd prime p, at least two zeros of Q(x) have multiplicities relatively prime to p. Proof. See M. Voorhoeve, K. Gy˝ory and R. Tijdeman [17]. (cid:3) Lemma 6. For q ≥ 2 let B (x) be the q-th Bernoulli polynomial. Let q (2.1) P(x) = B (lx+1)−B (x+1) q q where l is even. Then (i) P(x) has at least three zeros of odd multiplicity unless q ∈ {2,4}. (ii) For any odd prime p, at least two zeros of P(x) have multiplicities relatively prime to p. Proof. We shall follow the proof of Lemma 5 of [17]. By the Staudt-Clausen theorem(seeRademacher[13],pp.10),thedenominatorsoftheBernoullinum- bers B , B (k = 1,2,...) are even butnot divisible by 4. Choose the minimal i 2k d ∈ N such that both the polynomials d(B (lx+1)−B (x+1)) and dB (x) q q q are in Z[x]. Using the equality B (x+1) = B (x)+qxq−1 (see [13], pp.4-5), q q we have q q (2.2) dP(x) = d (lx+1)q−i −xq−i B −qxq−1 . i i ! i=0(cid:18) (cid:19) X (cid:2) (cid:3) HencebythechoiceofdandbytheStaudt-Clausentheorem,wehaved q B ∈ i i Z and q dB ∈ Z for k = 1,2,..., q−1. If d is odd, then necessarily q and 2k 2k 2 (cid:0)i(cid:1) q must be even for k = 1,2,..., q−1. Write q = 2µr where µ ≥ 1 and r is 2k (cid:0) (cid:1) 2 (cid:0) (cid:1) odd. Then q is odd, giving a contradiction unless r = 1. So (cid:0) (cid:1) 2µ (cid:0) (cid:1) d is odd⇐⇒ q = 2µ for some µ ≥1. If q 6= 2µ for any µ ≥ 1 then (2.3) d≡ 2 (mod 4). 6 GO¨KHANSOYDAN We distinguish three cases: I. Suppose q = 2µ for some µ ≥ 1, so that d is odd. We first prove (i) so we may assume that µ ≥ 3. Considering (2.2) modulo 4, we have q−2 q−2 q 2 q (2.4) dP(x) ≡ d (lx+1)q−iB −d B xq−2i (mod 4). i 2i i 2i i=0(cid:18) (cid:19) i=0(cid:18) (cid:19) X X Firstly, let l ≡ 0 (mod 4). Then we obtain q−2 q−2 q q−2 q 2 q (2.5) d (lx+1)q−iB ≡d B ≡ d B (mod 4). i i 2i i i 2i i=0(cid:18) (cid:19) i=0(cid:18) (cid:19) i=0(cid:18) (cid:19) X X X q It is easy to see that q B = 0. Hence we get q−i q−i i=1 P(cid:0) (cid:1) q−2 2 q (2.6) B = −B −qB . 2i 0 1 2i i=1(cid:18) (cid:19) X By using (2.5) and (2.6), one gets q−2 q−2 q q 2 q d (lx+1)q−iB ≡ d B + B ≡ 0 (mod 4). i 0 2i i 0 2i i=0(cid:18) (cid:19) (cid:18)(cid:18) (cid:19) i=1(cid:18) (cid:19) (cid:19) X X Then we deduce by (2.4) the following: q−2 2 q (2.7) dP(x) ≡ −d B xq−2i (mod 4). 2i 2i i=0(cid:18) (cid:19) X Secondly, let l ≡ 2 (mod 4). Then we obtain q−2 q−2 q q (2.8) d (lx+1)q−iB ≡ d (2x+1)q−iB (mod 4). i i i i i=0(cid:18) (cid:19) i=0(cid:18) (cid:19) X X Then the RHS of (2.8) becomes q−2 q d (2x+1)q−iB = d(B .(2x+1)q +qB .(2x+1)q−1 i 0 1 i i=0(cid:18) (cid:19) (2.9) X q−2 2 q + (2x+1)q−2iB . 2i 2i i=0(cid:18) (cid:19) X ON THE DIOPHANTINE EQUATION (x+1)k+(x+2)k+...+(lx)k =yn 7 Since 2x+1 is odd and q = 2µ,µ ≥ 3, is even, considering (2.9) modulo 4 and using (2.6), (2.8) becomes q−2 q d (lx+1)q−iB ≡ 0 (mod 4). i i i=0(cid:18) (cid:19) X So in all cases (2.4) reduces to (2.7). Note that q is divisible by 8 unless 2i is divisible by 2µ−2. We have 2i therefore for some odd d′, writing t = 1q (cid:0) (cid:1) 4 (2.10) dP(x) ≡ d′x4t+2x3t +dx2t+2xt (mod 4). Write dP(x) = R2(x)S(x) where R(x),S(x) ∈Z[x] and S contains each factor of odd multiplicity of P in Z[x] exactly once. Assumethat degS(x) ≤ 2. Since R2(x)S(x) ≡ x4t +x2t ≡ x2t(x2t +1) (mod 2), R2(x) must be divisible by x2t−2 (mod 2). So R(x) = xt−1R (x)+2R (x), 1 2 R2(x) = x2t−2R2(x)+4R (x), 1 3 for certain R ,R ,R ∈ Z[x]. If q > 8, then t > 2 so the last identity is 1 2 3 incompatible with (2.10) because of the term 2xt. Hence degS(x) ≥ 3, which proves (i). If q = 8, then by (2.7) dP(x) ≡ 3x8+2x6+x4+2x2 (mod 4). From here, we follow the proof in the corrigendum paper [17]. This fact can also be reduced from (2.7). So, the proof of (i) is completed where q = 2µ, µ ≥ 3. To prove (ii), let p be an odd prime and write P(x) = (R(x))pS(x) where R,S ∈ Z[x] and all the roots of multiplicity divisibly by p are incorporated in (R(x))p. We have, writing δ = 1q, by (2.10) 2 dP(x) ≡ (R(x))pS(x) ≡ xδ(xδ +1) ≡ xδ(x+1)δ (mod 2). Since δ is prime to p, S has at least two different zeros, proving (ii) in case I. II. Suppose q is even and q 6= 2µ for any µ. Then d≡ 2 (mod 4) and hence considering (2.2) in modulo 2, we get q q dP(x) ≡ d (1−xq−i)B (mod 2). i i i=0(cid:18) (cid:19) X 8 GO¨KHANSOYDAN Since B d q ≡ q (mod 2) for i= 1,2,3,...,q, we have i i i (cid:0) (cid:1) q(cid:0)−2(cid:1) 2 q q−1 q dP(x) ≡ x2k = xt ≡ (x+1)q −xq −1 (mod 2). 2k t k=1(cid:18) (cid:19) t=1(cid:18) (cid:19) X X Write q = 2µr, where r > 1 is odd. Then dP(x) ≡ (x+1)q −xq −1≡ ((x+1)r −xr −1)2µ (mod 2). Since r > 1 is odd,(x+1)r−xr−1has x and x+1as simplefactors (mod 2). Thus dP(x) ≡x2µ(x+1)2µK(x) (mod 2) where K(x) is neither divisible by x nor by (x + 1) (mod 2). As in the preceding case, P(x) must have two roots of multiplicity prime to p. This proves part (ii) of the lemma. In order to prove part (i), first we consider the case q = 6. In this case dP(x) ≡ (2l6+2)x6+(2l5 +2)x5+(l4+3)x4+(3l2+1)x2 (mod 4). Since l is even, we can write dP(x) ≡ 2x6+2x5+3x4+x2 (mod 4). So, P(x) has at least three simple roots. To prove our claim, suppose dP can be written as (2.11) dP(x) ≡ S(x)R2(x) (mod 4) with degS ≤ 2. If degS = 0, then clearly S is an odd constant, so R2(x) ≡ x4+x2 (mod 2). Hence R(x) ≡ x2+x (mod 2) and R2(x) ≡ x4+2x3+x2 (mod 4), which is a contradiction. If degS = 1, then either S(x) ≡x or S(x) ≡ x+1 (mod 2). In both cases, the quotient of P and S can not be written as a square (mod 2). If degS = 2, theneither S(x) ≡ x2 or S(x) ≡ x2+x or S(x) ≡ x2+1 (mod 2), since x2 +x+1 does not divide P (mod 2). In the first case R(x) ≡ x+1 (mod 2), hence R2(x) ≡ x2+2x+1 (mod 4) which is a contradiction. In the secondcase,thequotientofP andS isnotevenasquare (mod 2). Inthethird case R(x) ≡ x (mod 2), hence R2(x) ≡ x2 (mod 4) which is a contradiction. We conclude that dP cannot bewritten in form (2.11) with degS < 3, proving our claim. Secondly, as q = 2 and 4 are the exceptional cases, q = 6 case is treated in this section and finally the case q = 8 was treated in Section I,we may assume ON THE DIOPHANTINE EQUATION (x+1)k+(x+2)k+...+(lx)k =yn 9 that q ≥ 10. Considering (2.2) modulo 4 where d≡ 2 (mod 4), we have (2.12) q−2 q−2 q 2 q dP(x) ≡ d (lx+1)q−iB −dqB xq−1−d B xq−2i (mod 4). i 1 2i i 2i i=0(cid:18) (cid:19) i=0(cid:18) (cid:19) X X Firstly, let l ≡ 0 (mod 4). Then we obtain q−2 q−2 q 2 q (2.13) d (lx+1)q−iB ≡ dqB +d B (mod 4). i 1 2i i 2i i=0(cid:18) (cid:19) i=0(cid:18) (cid:19) X X q We know that q B = 0. By (2.6) and (2.13), one gets q−i q−i i=1 P(cid:0) (cid:1) q−2 q−2 q 2 q d (lx+1)q−iB ≡ d qB + B ≡ 0 (mod 4). i 1 2i i 2i i=0(cid:18) (cid:19) (cid:18) i=0(cid:18) (cid:19) (cid:19) X X Then we deduce by (2.12) the following: q−2 2 q (2.14) dP(x) ≡ −dqB xq−1−d B xq−2i (mod 4). 1 2i 2i i=0(cid:18) (cid:19) X Secondly,letl ≡ 2 (mod 4). Thenwehave(2.8)andtheRHSof (2.8)becomes q−2 q d (2x+1)q−iB = d(B .(2x+1)q +qB .(2x+1)q−1 i 0 1 i i=0(cid:18) (cid:19) (2.15) X q−2 2 q + (2x+1)q−2iB . 2i 2i i=1(cid:18) (cid:19) X Since2x+1isoddandq 6= 2µ iseven(q ≥ 10)anddq ≡ 0 (mod 4),considering (2.15) modulo 4 and using (2.6), (2.15) becomes q−2 q d (lx+1)q−iB ≡ 0 (mod 4). i i i=0(cid:18) (cid:19) X So in all cases (2.12) reduces to (2.14). Then by (2.14) we have 1 q q (2.16) dP(x) ≡ 2xq −qxq−1+ d xq−2+...+dB x2 (mod 4). q−2 6 2 2 (cid:18) (cid:19) (cid:18) (cid:19) Write dP(x) ≡ R2(x)S(x), where R,S ∈ Z[x] and S(x) only contains each factor of odd multiplicity of P once. Then degS(x) ≥ 3. The assertion easily follows by repeating the corresponding part of the proof of Lemma 5. Thus, the proof is completed for the case II. 10 GO¨KHANSOYDAN III. Let q ≥ 3 be odd. Then d ≡ 2 (mod 4) and for i = 1,2,4,...,q −1, q q d B ≡ (mod 2). i i i (cid:18) (cid:19) (cid:18) (cid:19) Now considering (2.2) modulo 2, we have q q dP(x) ≡ d (1−xq−i)B (mod 2). i i i=0(cid:18) (cid:19) X q−2 2 Since q = 2q−1−1 ≡ 1 (mod 2), we have 2λ λ=1 P (cid:0) (cid:1) q−1 2 q (2.17) dP(x) ≡ xq−1+ xq−2λ (mod 2). 2λ λ=1(cid:18) (cid:19) X From (2.2), we get q q (2.18) dP′(x) = d( [(lx+1)q−i −xq−i]B )′−dq(q−1)xq−2 i i i=0(cid:18) (cid:19) X and then q−1 2 q (2.19) xdP′(x) ≡ (q−2λ)xq−2λ (mod 2). 2λ λ=1(cid:18) (cid:19) X Hence by using (2.17) and (2.19) d(P(x)+xP′(x)) ≡ xq−1 (mod 2). Any common factor of dP(x) and dP′(x) must therefore be congruent to a power of x (mod 2). Considering (2.18) modulo 2, dP′(x) ≡ q = q ≡ q−1 1 (mod 2). Since dP′(0) ≡ 1 (mod 2), we find that dP(x) and dP′(x) are (cid:0) (cid:1) relatively prime (mod 2). So any common divisor of dP(x) and dP′(x) in Z[x] is of the shape 2R(x) + 1. Write dP(x) = Q(x)S(x) where Q(x) = Q (x)ki ∈ Z[x] contains the multiple factors of dP and S ∈ Z[x] contains i i Y its simple factors where k denotes the multiplicity of the polynomial factor i Q (x). Then Q(x) is of the shape 2R(x)+1 with R ∈ Z[x], so i S(x) ≡ dP(x) ≡xq−1+... (mod 2). Thus the degree of S(x) is at least q−1, proving case III whence q > 3. If q = 3, then (2.20) dP(x) = (l−1)x(2(l2 +l+1)x2+3(l+1)x+1).