Squares in products with terms in an arithmetic progression

Transcription

1 ACTA ARITHMETICA LXXXVI. (998) Squares in products with terms in an arithmetic progression by N. Saradha (Mumbai). Introduction. Let d, k 2, l 2, n, y be integers with gcd(n, d) =. Erdős [4] and Rigge [2] independently proved that a product of two or more consecutive positive integers is never a square. Further Erdős and Selfridge [5] showed that a product of k consecutive integers is never a power, i.e., n(n + )... (n + k ) = y l with integers k 2, l 2, n, y never holds. In [4, Corollary ] the author extended this result by showing that n(n + d)... (n + (k )d) = y l with integers k 3, l 2, n, y never holds for < d 6. In this paper we extend the range of d for the preceding equation with l = 2. () Theorem. The only solution of the equation n(n + d)... (n + (k )d) = y 2 in integers k 3, n, y is (n, d, k) = (8, 7, 3). and < d 22 Theorem is a consequence of the following more general result. Theorem 2. Let k 3, n and < d 22. Then there exists a prime exceeding k which divides n(n + d)... (n + (k )d) to an odd power except when (n, d, k) {(2, 7, 3), (8, 7, 3), (64, 7, 3)}. Equation () implies that every prime exceeding k divides the product n(n + d)... (n + (k )d) to an even power. This contradicts Theorem 2 except when (n, d, k) {(2, 7, 3), (8, 7, 3), (64, 7, 3)}. But in these three cases we find that n(n + d)... (n + (k )d) equals 2 2 2, 20 2, , 99 Mathematics Subject Classification: Primary D6. [27]

2 28 N. Saradha respectively. Hence equation () holds only when (n, d, k) = (8, 7, 3). Thus Theorem follows from Theorem 2. Let P (m) denote the greatest prime factor of m for any integer m > and we write P () =. Then it follows from Theorem 2 that the equation (2) n(n + d)... (n + (k )d) = By 2 in positive integers k 3, n, y, B with P (B) k never holds for < d 22 except when (n, d, k) {(2, 7, 3), (8, 7, 3), (64, 7, 3)}. Marszałek [7] proved that equation (2) with d > and B = implies that (3) k < 2 exp(d(d + ) /2 ). Shorey and Tijdeman [6] proved that equation (2) with d > implies that C/log log d (4) k < d where C is an effectively computable absolute constant. We prove Theorem 3. Equation (2) with d 23 implies that { 4d(log d) 2 if d is odd, (5) k <.3d(log d) 2 if d is even. In Theorem 3 we need to consider only d 23 in view of Theorem 2. The estimate (5) is a considerable improvement of (3). The estimate (4) involves an unspecified constant which turns out to be large. Therefore the estimate (5) is better than (4) for small values of d. Now we exhibit infinitely many solutions in relatively prime integers n and d > of equation (2) with k = 3 and square-free integer B satisfying P (B) 3. We observe that B {, 2, 3, 6}. For B =, the existence of infinitely many solutions follows from a well known result that there are infinitely many triples of relatively prime squares in arithmetic progression. For B >, we prove Theorem 4. Let B {2, 3, 6}. There are infinitely many triples (n, d, y) with integers n, d >, y and gcd(n, d) = satisfying (6) n(n + d)(n + 2d) = By 2. Let d =, k 3 and n(n+)... (n+k ) be divisible by a prime greater than k. Then Erdős and Selfridge [5] proved that there exists a prime p k dividing n(n + )... (n + k ) to an odd power. The author [4] showed that the above assertion is valid with p > k whenever k 4. If d = and k = 3, we prove Theorem 5. There is a prime exceeding 3 which divides n(n + )(n + 2) to an odd power except when n {, 2, 48}.

3 Squares in products 29 When n =, 2, 48, we see that n(n + )(n + 2) equals 6, 6 2 2, and the assertion of Theorem 5 is false. For the proof of Theorem 5, it suffices to show that the equation (7) n(n + )(n + 2) = By 2 with B {, 2, 3, 6} has no solution other than B = 6, (n, y) {(, ), (2, 2), (48, 40)}. If B =, the above assertion is a particular case of the result of Erdős and Rigge mentioned at the beginning of this section. If B = 6 and n odd, then the assertion was proved by Meyl [8] whereas Watson [7] and Ljunggren [6] proved the case of n even. The Algorithm in Section 3 was programmed and checkings and computations for the proof of Theorem 2 were carried out using Mathematica. I thank Professor T. N. Shorey for many helpful discussions. I also thank the referee for his valuable comments on an earlier draft of the paper. 2. Lemmas. We suppose throughout this section that n, d > and k 3 with (n, d, k) (2, 7, 3). Then by a result of Shorey and Tijdeman [5], we have (8) P (n(n + d)... (n + (k )d)) > k. Further we suppose that (9) ord p (n(n + d)... (n + (k )d)) 0 (mod 2) for all primes p > k. We write (0) n + id = a i x 2 i, a i square-free, P (a i ) k, x i > 0 for 0 i k. We observe that gcd(a i, d) = for 0 i k since gcd(n, d) =. We denote by {a,..., a t } the set of all the distinct elements from {a 0,..., a k }. By (8), we have () n + (k )d (k + ) 2. Let m be an integer and 2 p (d) < p (d) 2 <... be all the primes which are coprime to d. We define B m,d = {a r P (a r) p (d) m } and g(k, m, d) = B m,d. We observe that (2) g(k, m, d) t where ε (d) i = 0 if p (d) i i m+ ([ k > k and for p (d) i p (d) i ] ) + ε (d) i := g 0 (k, m, d) k, ε (d) i = 0 or according as p (d) i k or not for i m +. We note that g(k, m, d) and g 0 (k, m, d) are the same as g(k, m) and g 0 (k, m) of [4]. Throughout this section we assume without reference that h is a positive integer with h even whenever d is even. Further, let ϱ > 0. Define V h = {α α is a positive integer with αh 2 < ϱ and gcd(α, d) = }. We write

4 30 N. Saradha V h = i V hi such that for every i, positive integers α and β are in V hi if and only if α β ( mod (3) ε h = if 2 d gcd(d, h) d ε h gcd(d,h)) where and ε h = 2 if 2 d gcd(d, h). Further, let δ h = max{ V hi } and δ(d) = h< ϱ δ h. Here we recall that the summation in is taken over even values of h whenever d is even. We note that δ(d) can be computed for every d and ϱ and that the values of δ(d) for 7 d 22 and ϱ = 3 d2 can be found in Table. We begin with the following lemma which gives a lower bound for the number of distinct a i, viz., t. Lemma. Let n (k ) 2 d 2 /(4ϱ). If (9) holds, then t k δ(d). P r o o f. Let b,..., b r,... be the a j s which occur more than once with n + i r d = b r x 2 i r for r and such that x ir is minimal, i.e., if a i = b r with i i r, then x i > x ir. For any b r with r, we say that b r is repeated at the hth place if there exists some j, 0 j k, such that a j = b r, j i r, x j = x ir + h with h. We observe that j is uniquely determined. We set W h = {a j 0 j k, a j = b r, j i r and x j = x ir + h for some r}. In order to get a lower bound for the number of distinct a j s, we need to get an upper bound for h W h. We observe that W h is equal to the number of b r which are repeated at the hth place. We proceed to find an upper bound for this number. Suppose b r is repeated at the hth place. Then by its definition, we obtain for some j, 0 j k, j i r, (4) Thus (k )d (j i r )d = b r (x 2 j x 2 i r ) = b r (2hx ir + h 2 ) > 2hb /2 r (b r x 2 i r ) /2 2hb /2 (5) b r h 2 < ϱ. r n /2 > hb /2 r (k )d ϱ. Hence h < ϱ, i.e., the number of places at which b r can be repeated is at most [ ϱ]. Further, we note from (4), (0) and (5) that h is even whenever d is even and that b r V h. Also we observe that h(2x ir + h) 0 (mod d) from which it follows that x ir c ( d mod ε h gcd(d,h)) where c depends only on h and d with ε h as in (3). Thus n b r c 2 ( mod d ε h gcd(d,h)). Further, we observe that gcd ( d c, ε h gcd(d,h)) = since gcd(n, d) =. If b s b r is such that b s is repeated at the hth place, then by the foregoing argument, we have b s h 2 < ϱ and n b s c ( 2 d mod ε h gcd(d,h)). Thus ( ) ( b r b s mod d ε h gcd(d,h) since gcd c, d ε h gcd(d,h)) =. Hence br, b s belong to

5 Squares in products 3 V hi for some i. Thus the number of b r which are repeated at the hth place is δ h. Since h < ϱ, we have W h δ h = δ(d). h h< ϱ Hence the number of distinct a j s is at least k δ(d). As a consequence of Lemma, we have Corollary. Let k 2(2d 7). If (9) holds, then t k δ(d) where δ(d) is computed with ϱ = 3 d2. P r o o f. By () and k 2(2d 7), we see that n (k + ) 2 (k )d > (k + ) 2 (k )(k + 4) > (k )2. Now the result follows immediately from Lemma. Let = s < s 2 <... be the sequence of all square-free integers and = s < s 2 <... be the sequence of all odd square-free integers. Lemma 2. We have (i) s i (.5)i for i 39. (ii) s i (.6)i for 286 i 570. (iii) s i (2.25)i for i 2. P r o o f. (i) We first check that s i (.5)i for 39 i 70. Further, we check that for 0 r < 36, r S 0 = {0, 4, 8, 9, 2, 6, 8, 20, 24, 27, 28, 32} we can choose an s ir with 39 i r 70 such that s ir r (mod 36). Now we consider any s i with i > 70. Then s i r (mod 36) for some r with 0 r < 36, r S 0. Thus s i s ir (mod 36) with 39 i r 70. Hence (6) s i s ir = 36f for some positive integer f. We know that in any set of 36 consecutive integers, the number of square-free integers is 24. Thus the number of square-free integers 36f is at most 24f. Also we observe from (6) that this number is equal to i i r. Therefore i i r 24f 2 3 (s i s ir ) by (6). Hence s i 3 2 (i i r) + s ir (.5)i since s ir (.5)i r. (ii) The inequality follows by direct checking. (iii) We check that s i (2.25)i for 2 i 35. Also we check for 0 r < 36 with r (mod 2) and r S 0 that we can choose an s i r with 2 i r 35 such that s i r r (mod 36). Further, we observe that the number of odd square-free integers in any set of 36 consecutive integers is 6. Now we repeat the argument in (i) for any s i with i > 35 to obtain (iii).

6 32 N. Saradha It can be checked that s i (.5) 63 (63)!, s i (.6) 286 (286)!, 5 By Lemma 2 and from an induction argument we derive Corollary 2. We have s i (2.25) 5 (5)!. (i) ν s i (.5) ν ν! for ν 63. (ii) ν s i (.6) ν ν! for 286 ν 570. (iii) ν s i (2.25)ν ν! for ν 5. The inequality in (i) of the above corollary has already appeared in [5]. Lemma 3. Let 7 d 22. Suppose (9) holds. Then k k 0 (d) := k 0 where k 0 is as given in Table. P r o o f. Suppose k > k 0. Then k > 2(2d 7). Hence Corollary is valid. Thus t k δ(d) where δ(d) is computed with ϱ = 3 d2. We note from Table that δ(d) 20. Thus t k 20. From now onwards we shall assume that k 83. Since a i for i t are square-free integers, we use Corollary 2(i) to obtain (7) a i t k 20 a i k 20 On the other hand, by (0), we have s i (.5) k 20 (k 20)!. (8) a... a t (k )! p k p. We put g q = ord q (a... a t ) and h q = ord q ((k )! p k p) for any prime q k. Then it follows from Marszałek [7, p. 22] that g q k q + + log k log q + and h q k q log k log q. Thus (9) g q h q 2k q 2 + q q + 2 log k log q. Further, from (8) we get ( )( (20) a... a t (k )! ) p q g q h q p k q 9 where in the product signs p, q run over primes. Now by (20) and (9), we have ( (2) a... a t (k )! p )k 6( q q/(q ))( ) k q 2/(q2 ). p k q 9 q 9

7 Squares in products 33 We find that q q/(q ) , q 2/(q2 ) 2.889, q 9 q 9 (22) p (2.78) k (see [3, p. 7]). p k Using (22) in (2) and comparing with (7), we get (.5549) k ( )(.5) 20 k 35. This inequality is not valid for k 570. Thus we obtain k < 570. Now let k 485. We use Corollary 2(ii) to get a i (.6) k 20 (k 20)!. t Comparing this lower bound with the upper bound in (2), we get (.6586) k ( )(.6) 20 k 35. This inequality is not valid for k 485. Thus we conclude that k < 485. We shall bring down the value of k to k 0 in all cases except d = 9 by a counting argument which will be presented in the next paragraph. When d = 9 the counting argument fails. But a refinement of the above argument itself enables us to bring k < 35. When d = 9 we observe that g 9 = 0 and we rewrite (20) as ( )( a... a t (k )! p q g q h q )(9) h q. p k q 7 On the other hand, by Corollary 2(ii), we have for 485 > k 303, a... a t (.6) k 7 (k 7)! since δ(d) = 7. Now we combine the preceding estimates for a... a t to conclude that k < 35 whenever d = 9. Let 7 d 22, d 9 and k < 485. Since a i for i t are distinct and square-free we have ( ) ( ) m m (23) g(k, m, d) = 2 m. m Thus if g 0 (k, m, d) 2 m +, we get a contradiction by (2). Since t k δ(d), we replace t in (2) by k δ(d) and using Table for the values of δ(d) we check that g 0 (k, m, d) 2 m + for a proper choice of m whenever k 0 < k < 485. For example, when d = 3 we observe from p (d) 6 = 7 and the definition of g 0 (k, m, d) that g 0 (k, 5, 3) 33 for 20 k < 485. The other cases are checked similarly. See Table for the choices of m when different values of d and k are considered. This completes the proof.

8 34 N. Saradha Table d δ(d) Range for k m k 0 d δ(d) Range for k m k We see from Table that δ(d) 20 for 7 d 22. In the following lemma we give an upper bound for δ(d) whenever d 23. Lemma 4. For d 23 and ϱ = 3 d2, we have { 4d log d + (.8323)d if d is odd, δ(d) 3d log d + (.8)d if d is even. P r o o f. By the definition of δ(d), we obtain (24) δ(d) ([ ] ) d 2 3h 2 εh gcd(d, h) + d h<d/ 3 where the sum is taken over even values of h whenever d is even. We observe from (3) that ε h 2. Thus from (24) we get (25) δ(d) d 3 h + d 2 if d is even. 3 h<d/(2 3) Let d be odd. Then by (3), ε h =. Further, gcd(d, h) h/2 whenever h is even. Hence from (24) we get δ(d) d 3 h + d 3 2h + d 3 d 3 h<d/ 3, h odd h<d/ 3 h d 6 h<d/ 3, h even h<d/ 3, h even h + d 3

9 Thus d 3 d 4 (26) δ(d) d 4 h<d/ 3 h<d/(2 3) h<d/(2 3) Squares in products 35 h d 2 h + d 3 h<d/(2 3) h + d 3 d/(2 3)<h<d/ 3 h + d 3 + d h + d. 3 if d is odd. We use h<x /h < log x + γ + /x where x > and γ is Euler s constant whose value is <.5773 (see [, p. 55] and [3, pp ]) and d 23 in the estimates (25) and (26) to prove the assertion of the lemma. Lemma 5. Let k = 3. Suppose (9) holds. Then (a 0, a, a 2 ) S where S = S S 2 S 3 S 4 with S = {(,, )}, S 2 = {(2,, 2)}, S 3 = {(, 2, 3), (2, 3, ), (3,, 2), (6,, 2), (,, 2)} and S 4 = {(, 3, 2), (2,, 3), (3, 2, ), (2,, 6), (2,, )}. Further, we have 0 (mod 8) if (a 0, a, a 2 ) S ; ± (mod 8) if (a d 0, a, a 2 ) S 2 ; (mod 8) if (a 0, a, a 2 ) S 3 ; (mod 8) if (a 0, a, a 2 ) S 4. P r o o f. From (9), we see that (0) holds and therefore {a 0, a, a 2 } {, 2, 3, 6}. Also gcd(n, n+d) = gcd(n+d, n+2d) = and gcd(n, n+2d) = or 2 since gcd(n, d) =. Thus we find that there are 20 possible values for the triple (a 0, a, a 2 ), viz., (,, ), (,, 2), (,, 3), (,, 6), (, 2, ), (, 2, 3), (, 3, ), (, 3, 2), (, 6, ), (2,, ), (2,, 2), (2,, 3), (2,, 6), (2, 3, ), (2, 3, 2), (3,, ), (3,, 2), (3, 2, ), (6,, ), (6,, 2). We use without mentioning that x 0, x, x 2 are pairwise coprime and a 0 x a 2 x 2 2 = 2a x 2. We exclude the possibilities (,, 3) and (,, 6) since 2x 2 x (mod 3); (, 3, ), (, 6, ), (2, 3, 2) since x x (mod 3); (3,, ) and (6,, ) since 2x 2 x (mod 3); (, 2, ) since x x (mod 4). The remaining 2 possibilities are given by S. We take (a 0, a, a 2 ) = (,, ) S. Then n = x 2 0, n+d = x 2, n+2d = x 2 2, implying x 0 and x 2 are odd since n and n + 2d are both odd or both even and gcd(n, n + 2d) = or 2. Thus 2d = x 2 2 x (mod 8), yielding d 0, 4 (mod 8). If d 4 (mod 8), then x 2 = n + d 5 (mod 8), which is not possible. Thus d 0 (mod 8). Next we consider (a 0, a, a 2 ) = (2,, 2) S 2. Then x is odd and n 0 or 2 (mod 8) according as x 0 is even or odd. Hence x 2 = n + d d or d + 2 (mod 8), which implies that d or 7 (mod 8).

10 36 N. Saradha Now we take (a 0, a, a 2 ) = (, 2, 3) S 3. Then x 0, x 2 are odd and d = 2x 2 x 2 0 or (mod 8) according as x is even or odd. If d (mod 8), then d = 3x 2 2 2x (mod 8), a contradiction. Thus d (mod 8). Similarly we prove for other possibilities in S 3 that d (mod 8). Lastly, we consider (a 0, a, a 2 ) = (, 3, 2) S 4. Then x 0 is even and hence x, x 2 are odd. Thus d = 2x 2 2 3x 2 7 (mod 8). Likewise we prove for other possibilities in S 4 that d 7 (mod 8). Lemma 6. Let k = 3. Suppose (9) holds. Then one of the following possibilities holds: (i) d =, (ii) d 23, (iii) (n, d) {(2, 7), (8, 7), (64, 7)}. P r o o f. Let < d < 23. We shall show that (iii) holds. By Lemma 5, we need to consider (a 0, a, a 2 ) S with d = 8, 6; (a 0, a, a 2 ) S 2 S 3 with d = 9, 7 and (a 0, a, a 2 ) S 2 S 4 with d = 7, 5. Let (a 0, a, a 2 ) S with d = 8, 6. Since x 2 x 2 0 = d we find that x 0 =, x = 3 and x 0 = 3, x = 5 and hence x 2 2 = 7 and 4, respectively. This is not possible. Let (a 0, a, a 2 ) S 2 with d = 7, 9, 5, 7. Then x 2 2 x 2 0 = d implies that (n, d) = (8, 7). Let (a 0, a, a 2 ) S 3 with d = 9, 7. Let d = 9. In the first 4 possibilities in S 3 we observe that 3 divides one of n, n + d, n + 2d. Hence 3 n, which is a contradiction since gcd(n, d) =. Let (a 0, a, a 2 ) = (,, 2). Then x 2 x 2 0 = 9 gives x 0 = 4 and hence n + 2d = 2x 2 2 = 34, which is impossible. Let d = 7 and (a 0, a, a 2 ) = (, 2, 3). Then n , 3 (mod 9), implying x 2 0 = n 2, 5 (mod 9), a contradiction. The next three possibilities are excluded similarly. Let (a 0, a, a 2 ) = (,, 2). Then x 2 x 2 0 = 7 implies that (x 0, x, x 2 ) = (8, 9, 7). Thus (n, d) = (64, 7). Let (a 0, a, a 2 ) S 4 with d = 7, 5. Let d = 7 and (a 0, a, a 2 ) = (, 3, 2). Then n + 7 0, 3 (mod 9). Hence x 2 0 = n 2, 5 (mod 9), a contradiction. The next three possibilities are excluded similarly. Let (a 0, a, a 2 ) = (2,, ). Then x 2 2 x 2 = 7 gives (x 0, x, x 2 ) = (, 3, 4). Thus (n, d) = (2, 7). Let d = 5. The first four possibilities in S 4 are excluded since 3 divides one of n, n + d, n + 2d. Let (a 0, a, a 2 ) = (2,, ). Then x 2 2 x 2 = 5 implies that x = 7 or, giving 2x 2 0 = n = 34 or 4, which are impossible. Lemma 7. Let 7 d 22 and (n, d) {(2, 7), (8, 7), (64, 7)}. Assume that t = k. Then (9) does not hold. P r o o f. Suppose (9) holds. Then by Lemmas 3, 6 and Table, we have 4 k k We observe that (0) holds and a 0,..., a k are all distinct since t = k. We often use these facts and the property that

11 Squares in products 37 gcd(a i, d) = for 0 i < k without any reference. We check that g 0 (k,, d) 3 for 4 k 8 if 2 or 3 divides d; g 0 (k, 2, 7) 5 for k = 7, 8; g (27) 0 (k, 2, d) 5 for 9 k 22; g 0 (k, 3, d) 9 for 23 k 78; g 0 (k, 4, d) 7 for 79 k 276; g 0 (k, 5, d) 33 for 277 k 34. But (27) contradicts (23), by (2). Thus we may assume that 4 k 6 if d = 7 and 4 k 8 if d {, 3, 7, 9}. Let k = 4 and d {7,, 3, 7, 9}. We know that P (a i ) 3 and hence a i {, 2, 3, 6}. Thus n(n + d)(n + 2d)(n + 3d) is a square. But this is impossible by a well known result of Euler (see Dickson [3, p. 635] and Mordell [9, p. 2, Corollary]). We also use this fact without reference when we deal with other values of k. Let k = 5. Since a i s are distinct, we need only consider the case when 5 divides one and only one of n, n + d, n + 2d, n + 3d, n + 4d and hence at most one a i. The values of the other a i s belong to {, 2, 3, 6}. We may assume that 5 divides one of n + d, n + 2d, n + 3d. Suppose 5 n + d. Then {n, n + 2d, n + 3d, n + 4d} {y 2, 2y 2 2, 3y 2 3, 6y 2 4} for some positive integers y, y 2, y 3, y 4. We explain the case d = 7. Then n 3 (mod 5). Hence n = 2y 2 2 or 3y 2 3. Let n = 2y 2 2. Then n+4 = 3y 2 3, n+2 = y 2 and hence n+28 = 6y 2 4, which gives 3 4, a contradiction. When n = 3y 2 3, we get n + 4 = 2y 2 2, n + 2 = y 2 and hence n + 28 = 6y 2 4, implying 3 28, a contradiction. As another example, we take d =. Then n 4 (mod 5). We find that n = 6y 2 4, n + 22 = y 2, n + 33 = 3y 2 3, n + 44 = 2y 2 2. Here we observe that y is even, y 2, y 3, y 4 are odd. Hence n 6 (mod 8) and n (mod 8). But n + 33 = 3y (mod 8), a contradiction. By a similar argument, we exclude all the cases 5 n + d, 5 n + 2d, 5 n + 3d for d {7,, 3, 7, 9}. Thus k 5. Let k = 6. Then P (a i ) 5 and we may assume that 5 n. Hence 5 divides only one of {n + d, n + 2d, n + 3d, n + 4d}. Therefore five of the a i s belong to {, 2, 3, 6}. This is not possible since a i s are all distinct. Thus k 6. Let k = 7 and d {, 3, 7, 9}. Then P (a i ) 7 and we may assume that there exist distinct i, i 2 and i 3 between 0 and 6 such that 7 n + i d, 5 n+i 2 d, 5 n+i 3 d since otherwise g 0 (k, 2, d) 5 leading to a contradiction. There are 8 possibilities for (i, i 2, i 3 ) for each d. We check the case 7 n + d, 5 n, 5 n + 5d for d = 7. Then n + 2d = 6y 2 4, n + 3d = y 2, n + 4d = 2y 2 2 and hence n + 6d = 3y 2 3, which implies 3 4d, a contradiction. The other cases are excluded similarly. Finally, let k = 8 and d {, 3, 7, 9}. Then P (a i ) 7 and we may assume that 7 n, 7 n + 7d, 5 n + d, 5 n + 6d for otherwise g 0 (k, 2, d) 5,

12 38 N. Saradha which is a contradiction. Then (n + 2d)(n + 3d)(n + 4d)(n + 5d) is a square, which is impossible. The following lemma deals with the integral solutions of certain Diophantine equations. Lemma 8. (i) There are infinitely many integral solutions in x and y of the equation x 2 2y 2 = with x odd and of the equation x 2 3y 2 = with x odd as well as with x even. (ii) All solutions of the equation 3x 2 + y 2 = z 2 in integers x, y, and z are given by x = ϱ 0 us, y = 2 ϱ 0(αu 2 βs 2 ), z = 2 ϱ 0(αu 2 + βs 2 ) where αβ = 3, u and s are positive integers with gcd(u, s) = and ϱ 0 is any integer when u and s are odd but ϱ 0 is even when one of u and s is even and the other is odd. (iii) The only solutions in non-zero integers of x 4 + y 4 = 2z 2 with gcd(x, y) = are x 2 =, y 2 = and z 2 =. There is no solution in non-zero integers of the equation x 4 y 4 = 2z 2 with gcd(x, y) =. Lemma 8(i) is a well known result in continued fraction theory. We refer to [0, Theorem 7.25, pp ] from where the result in Lemma 8(i) can be derived easily using the facts that 2 =, 2 and 3 =,, 2. Lemma 8(ii) can be found in [2, pp. 40 4]. The first assertion in Lemma 8(iii) is proved in [, p. 38]. It also follows from A4.4 of [, p. 7]. In fact, the statement given therein should be corrected as: If m 0 and x 4 +y 4 = 2 m z 2 with gcd(x, y) =, then m = and x 2 = y 2 = z 2 =. The second assertion in Lemma 8(iii) follows from A4.5 of [, p. 72]. 3. An algorithm. In this section, we modify the algorithm given in [4, 4]. Algorithm. Let d and k 4 be given. Also let µ > 0. Step. Find all primes q,..., q θ, q θ+,..., q θ+η which are coprime to d and such that q <... < q θ k < q θ+ <... < q θ+η and qi 2 < k2 d 2 /µ for i θ + η. Step 2. Set D = {q α... qα θ θ q2β θ+... q2β η θ+η qα... qα θ θ q2β θ+... q2β η θ+η < k 2 d 2 /µ for non-negative integers α i, β j, i θ, j η and β,..., β η not all zero}. Step 3. For every q D, find the smallest j 0 such that d < q/(k j 0 ). Then find some j = j(q) with j 0 j k such that P (q + jd) and P (q (k j)d) are > q θ+η. In our application, it is always possible to find j 0 in Step 3 because d 22 and q q 2 θ+ 25. Also q (k j)d is positive since d < q/(k j 0) <

13 Squares in products 39 q/(k j) as j j 0. We derive from the above Algorithm the following result. Lemma 9. Let d, k and µ > 0 be given such that n + (k )d < k 2 d 2 /µ. If (9) and Step 3 hold, then (8) does not hold. P r o o f. Since n + (k )d < k 2 d 2 /µ, every term n + id for 0 i k is of the form q D or q with P (q) q θ. Now we follow the proof of [4, Lemma ] to obtain the assertion of the lemma. 4. Proof of Theorem 2. Let < d 22 and (n, d, k) {(2, 7, 3), (8, 7, 3), (64, 7, 3)}. We assume that (9) holds and arrive at a contradiction. We apply Lemma 6 to get k 4. Next we use Theorem of [4] to derive that d 7. Then by Lemma 3, we may assume that k k 0 where k 0 is as given in Table. Suppose n (k ) 2 d 2 /4. Then we take ϱ = in Lemma and observe that δ(d) = 0. Thus from Lemma, we derive that t = k and hence it follows from Lemma 7 that (9) does not hold. Thus our supposition n (k ) 2 d 2 /4 is false. We assume from now onwards that n < (k ) 2 d 2 /4 and hence (28) n + (k )d < k2 d 2 for 4 k k 0 and 7 d Suppose k 27. Then from Table we have d 9. Assume that n (k ) 2 d 2 /36. Then we take ϱ = 9 in Lemma. Thus h 2. Suppose d = 9. Then V = {, 2, 4, 5, 7, 8}, V 2 = {, 2}, ε =, ε 2 =, δ = and δ 2 =. Hence δ(d) 2. Similarly, for other values of d we find that δ(d) 2. Hence by Lemma, t k 2. We check that g(k, 3, d) 9 for 27 k 66. This contradicts (23). Thus we derive that (29) n + (k )d < (k )2 d 2 + (k )d < k2 d for 27 k 66 and 9 d 22. Let k > 66. Then we see from Table that d {3, 7, 9, 2, 22} and the corresponding upper bound for k, viz., k 0 is large. We use the idea in the preceding paragraph in order to get a good upper bound for n+(k )d in different ranges of k. First we assume that n+(k )d k 2 d 2 /µ for some positive integer µ. Then we find a lower bound for t, say t 0, and a range of k, say R k R 2, in which by (2), we check that g(k, m, d) 2 m + for a suitable choice of m. Since this is a contradiction we derive that (30) n + (k )d < k2 d 2 for R k R 2. µ In Table 2, we tabulate the choice of µ, values of t 0, R, R 2 and m when d {3, 7, 9, 2, 22}.

14 40 N. Saradha Table 2 d µ t 0 R R 2 m 3 92 k k k , , 5 84 k , , k k k , , 5 85 k k k For a given d, k, we use (28) (30) with Table 2 and construct the set D mentioned in Step 2 of the Algorithm in Section 3. Next we proceed to check that Step 3 holds for the given d, k and q D. This would contradict (8) by Lemma 9. The verification of Step 3 is done as follows. First, we delete from D all the integers q for which both P (q + jd) and P (q (k j)d) exceed q θ+η with j = j 0. We denote the set of remaining integers of D by D. Secondly if D, we delete from D those integers for which both P (q + jd) and P (q (k j)d) exceed q θ+η with j = j 0 +. The remaining set of integers from D is denoted by D 2. The above process is continued till we reach j = k or until D i becomes an empty set for some integer i. For the values of d and k under consideration, we find that we need only take j with j 0 j min(k, 25). There are triples (d, k, q) for which the Algorithm fails, i.e., we are unable to find some j with j 0 j < k such that both P (q + jd) and P (q (k j)d) exceed q θ+η. In all, we find 207 triples which are not covered by the Algorithm. For each d, we give below a few examples of such triples. For a given d, we have chosen as examples those triples for which either k or q is maximum among all the triples (d, k, q): (7, 4, 25), (8, 5, 49), (9, 2, 69), (0, 6, 49), (, 2, 69), (2, 4, 49), (3, 3, 058), (4, 5, 363), (4, 3, 36), (5, 5, 578), (6, 7, 36), (6, 2, 69), (7, 7, 058), (7, 25, 96), (8, 5, 289), (8, 9, 2), (9, 9, 68), (9, 25, 96), (20, 0, 867), (20, 6, 289), (2, 5, 36), (2, 6, 289), (22, 6, 637), (22, 6, 289). We observe that in all the 207 cases k and q are not very large. For all these triples (d, k, q), we factorize the product n(n + d)... (n + (k )d) directly with n {q, q d,..., q (k )d} to find a prime exceeding k which divides the product to an odd power. This completes the proof of Theorem Proof of Theorem 3. Suppose that d 23 and k 4d(log d) 2 if d is odd and k (.3)d(log d) 2 if d is even. Then k > 2(2d 7). Also

15 Squares in products 4 by equation (2), we may assume that (9) holds. Thus we conclude from Corollary that t k δ(d) where δ(d) is computed with ϱ = 3 d2. By Lemma 4, δ(d) (.52)d log d for d odd, δ(d) (.38)d log d for d even and hence k δ(d) 63. Hence by Corollary 2(i), we have (3) a i t k δ(d) a i We use (3), (2) and (22) to get k δ(d) s i (.5) k δ(d) (k δ(d))!. (32).5549 ( ) /k (.5) δ(d)/k k (5+δ(d))/k. Let d be even. Then we turn to sharpening (32). Since gcd(a i, d) = for i t we find that a,..., a t are odd. By Corollary 2(iii), we have t ( ) k δ(d) 9 a i (k δ(d))!. 4 We note that g 2 = 0 and we use (20) with (9) for q 3, g 2 h 2 k+ log k log 2 to estimate t a i ( )(.7657) k k 5 (k )!. Finally, we combine the upper and lower estimates for t that a i (33) ( ) /k ( 9 4) δ(d)/k k (4+δ(d))/k for d even. to conclude We observe that the right hand sides of the inequalities (32) and (33) are decreasing functions of k. Therefore, we put k = (3.8)d(log d) 2 in (32) when d is odd and k = (.3)d(log d) 2 in (33) when d is even. By Lemma 4, we may replace δ(d) in (32) by 4d log d + (.8323)d and δ(d) in (33) by d log d + (.8)d. The resulting inequalities do not hold for d Proofs of Theorems 4 and 5 Proof of Theorem 4. We choose (2, x 2 0 2, 2x 0 y 0 ) where x 2 0 2y0 2 = with x 0 odd, (n, d, y) = (2, x 2 0 2, 2x 0 y 0 ) where x 2 0 3y0 2 = with x 0 odd, (, x 2 0/2, x 0 y 0 /2) where x 2 0 3y0 2 = with x 0 even. Then we observe that (n, d, y) is a solution of equation (6) with B {2, 3, 6}. Now by Lemma 8(i), there are infinitely many such triples (n, d, y) satisfying equation (6) with B {2, 3, 6}. This proves Theorem 4.

16 42 N. Saradha Proof of Theorem 5. Let n {, 2, 48}. By the remarks following Theorem 5 in Section, we need to consider equation (7) with B = 2, 3. Thus we may assume that (9) holds and we shall arrive at a contradiction. We apply Lemma 5 to assume that (a 0, a, a 2 ) = (,, 2) if B = 2 and (a 0, a, a 2 ) = (6,, 2) if B = 3. The first case implies x 2 x 2 0 =, which is not possible. In the second case we have 3x x 2 2 = x 2 with x 0 even, and x 0, x, x 2 pairwise coprime. Hence by Lemma 8(ii), we have ϱ 0 = 2 and = x 2 6x 2 0 = α 2 u 4 + β 2 s 4 8u 2 s 2 with αβ = 3. It is no loss of generality to assume that α =, β = 3 while dealing with this equation. Thus we consider u 4 + 9s 4 8u 2 s 2 =. This implies that + 8u 4 = c 2 for some integer c. This immediately reduces to the equation u 4 2u 4 2 = ± for some integers u and u 2 with u u 2 = ±u. We apply Lemma 8(iii) to observe that (u, u 2 ) = (±, 0) or (±, ±). Hence u = 0 or ±. Thus we have 9s 4 = or s 2 (9s 2 8) = 0. The former is impossible while the latter gives s = 0, implying that x 0 = 0 by Lemma 8(ii) and this is not possible. References [] T. M. Apostol, Introduction to Analytic Number Theory, Springer, 976. [2] L. E. Dickson, Introduction to the Theory of Numbers, Univ. of Chicago Press, 946. [3], History of the Theory of Numbers, Vol. II, Washington, 99; reprint: Chelsea, New York, 97. [4] P. E r d ő s, Note on products of consecutive integers, J. London Math. Soc. 4 (939), [5] P. Erdős and J. L. Selfridge, The product of consecutive integers is never a power, Illinois J. Math. 9 (975), [6] W. L j u n g g r e n, New solution of a problem proposed by E. Lucas, Norsk Mat. Tidsskrift 34 (952), [7] R. Marszałek, On the product of consecutive elements of an arithmetic progression, Monatsh. Math. 00 (985), [8] A. Meyl, Question 94, Nouv. Ann. Math. (2) 7 (878), [9] L. J. Mordell, Diophantine Equations, Academic Press, 969. [0] I. Niven and H. S. Zuckerman, An Introduction to the Theory of Numbers, Wiley, 972. [] P. Ribenboim, Catalan s Conjecture, Academic Press, 994. [2] O. Rigge, Über ein diophantisches Problem, in: 9th Congress Math. Scand., Helsingfors, 938, Mercator, [3] J. B. Rosser and L. Schoenfeld, Approximate formulas for some functions of prime numbers, Illinois J. Math. 6 (962), [4] N. Saradha, On perfect powers in products with terms from arithmetic progressions, Acta Arith. 82 (997), [5] T. N. Shorey and R. Tijdeman, On the greatest prime factor of an arithmetical progression, in: A Tribute to Paul Erdős, A. Baker, B. Bollobás and A. Hajnal (eds.), Cambridge Univ. Press, 990,

17 Squares in products 43 [6] T. N. Shorey and R. Tijdeman, Perfect powers in products of terms in an arithmetical progression, Compositio Math. 75 (990), [7] G. N. Watson, The problem of the square pyramid, Messenger Math. 48 (99), 22. School of Mathematics Tata Institute of Fundamental Research Homi Bhabha Road Mumbai , India Received on and in revised form on (325)