On intervals containing full sets of conjugates of algebraic integers

Size: px

Start display at page:

Download "On intervals containing full sets of conjugates of algebraic integers"

Piers Bell
5 years ago
Views:

1 ACTA ARITHMETICA XCI4 (1999) On intervals containing full sets of conjugates of algebraic integers by Artūras Dubickas (Vilnius) 1 Introduction Let α be an algebraic number with a(x α 1 ) (x α d ) as its minimal polynomial over Z Then α is called totally real if all its conjugates α 1 = α, α,, α d are real Also, α is called an algebraic integer if a = 1 Now, define I(d) as the smallest positive number with the following property: any closed real interval of length at least I(d) contains a full set of conjugates of an algebraic integer of degree d It is clear that I(1) = 1 Theorem 1 We have I() = As we will see from the proof of this simple theorem I(d) can also be computed for all small d The purpose of this paper is to give an upper bound for I(d) for large d In 1918, I Schur [Sc] proved that an interval on the real axis of length smaller than 4 can contain only a finite number of full sets of conjugates of algebraic integers T Zaïmi [Za] gave another proof of Schur s result His approach is based on M Langevin s proof [La] of Favard s conjecture Moreover, in [Za] it is proved that the length of an interval containing a full set of conjugates of an algebraic integer of degree d is greater than 4 ψ 1 (d) with some explicitly given positive function ψ 1 (d) satisfying lim d ψ 1 (d) = 0 Note that a similar result with another explicitly given function ψ (d) also follows from [Sc] On the other hand, R Robinson [Ro] showed that any interval of length greater than 4 contains infinitely many full sets of conjugates of algebraic integers Moreover, V Ennola [En] proved that such an interval contains full sets of conjugates of algebraic integers of degree d for all d sufficiently large Hence lim d I(d) = Mathematics Subject Classification: 11R04, 11R80, 1D10 [379]

2 380 A Dubickas A lower bound for I(d) can be obtained via a Kronecker type theorem In 1857, L Kronecker [Kr] proved that if α is an algebraic integer all of whose conjugates lie in [ ; ] then α = cos(πr) with r rational So if α is a totally real algebraic integer not of the form cos(πr) with r rational, then α = max 1 j d α j > In 1965, A Schinzel and H Zassenhaus [SZ] asked for a lower bound of the house α in terms of the degree d of α They showed that with the same hypotheses, (1) α > + 4 d 3 This lower bound was derived from the lower bound for α, where α is an algebraic integer which is not a root of unity The conjectural inequality α > 1 + c 1 /d (see [SZ]) with an absolute positive constant c 1 is not yet proved This is also the case with D H Lehmer s [Le] more general conjectural inequality d M(α) = a max{1, α j } > 1 + c j=1 where α is an algebraic number which is not a root of unity and c is an absolute positive constant Using the best known lower bound in Lehmer s conjecture [Lo] the author strengthened the inequality (1) We proved [Du] that if α is a totally real algebraic integer of degree d, α cos(πr) with r rational, and d is sufficiently large, then Thus the interval (log )3 α > + 46 d() 4 ( ( ] π (log )3 cos ); + 46 d d() 4 does not contain a full set of conjugates of an algebraic integer of degree d It follows immediately that for all d sufficiently large, () I(d) > (log )3 d() 4 Our main theorem gives an explicit slowly decreasing function, namely 1(log ) /, which cannot replace 9(log ) 3 /d() 4 in () Theorem There is an infinite sequence S of positive integers such that for d S any interval of length greater than or equal to (log ) contains a full set of conjugates of an algebraic integer of degree d

3 Clearly, for d S we have the inequality Conjugates of algebraic integers 381 (log ) I(d) Our proof of Theorem is based on the following statement Lemma Let u, v, w be three fixed positive integers Then there is an infinite sequence S(u, v, w) of positive integers such that every d S(u, v, w) is divisible by where w(vq(d) u ) q(d) q(d)!, [ q(d) = (u + 1) log Here and below [ ] denotes the integral part We will also show that the sequence S in Theorem can be taken to be all sufficiently large elements of S(, 16, ) Now we will prove the Lemma, Theorem and Theorem 1 Proof of the Lemma Put for brevity f(x) = log x log log x For x 16 the function f(x) is increasing Let k be an integer Then the equation (in x) f(x) u + 1 = k has a unique solution which we denote by x k Clearly, x > 5503 and the sequence x k is increasing We now prove that Indeed, if x k+1 x k log x k then x k+1 > x k log x k u + 1 = (u + 1)(k + 1) (u + 1)k = f(x k+1 ) f(x k ) f(x k log x k ) f(x k ) = log(x k log x k ) log log(x k log x k ) log x k log log x k Put y k = log log x k for brevity By the last inequality we have u + 1 y k + e y k log(y k + e y eyk = y k + y ke yk e yk log(y k + e yk ) k ) y k y k log(y k + e y k ) = y k ey k log(1 + y k e y k ) y k log(y k + e y < y k eyk log(1 + y k e yk ) k ) The last expression is less than 1, a contradiction ] y k

4 38 A Dubickas Set N k = { [x k ] + 1, [x k ] +,, [x k log x k ] } Clearly, for n N k, [ ] f(n) q(n) = = k u + 1 Note that for all n sufficiently large the expression is less than w(vq(n) u ) q(n) q(n) q(n) exp so that for all n N k, r(n) = w(vq(n) u ) q(n) q(n)! ( log w + + log n log log n log < exp(log n) = n, r(n) = w(vk u ) k k! [x k ] log v log n (u + 1) log log n ( log n (u + 1) log log n Hence, at least one of the integers [x k ] + 1, [x k ] +,, [x k ] is divisible by w(vk u ) k k! Since [x k ] [x k log x k ], at least one element of N k belongs to S(u, v, w) The Lemma is proved 3 Proof of Theorem Let d be a sufficiently large positive integer from S(, 16, ) Suppose [A; B] is a real interval such that B A (log ) Let also [ ] q = 3 log We take two integers p 1 and p in the intervals [ ( ) ) (( ) ] 4 log 4 log Aq; A + q and B q; Bq respectively Then [p 1 /q; p /q] [A; B] and p q p 1 q B A 8 log )) > (log ) log We will show that the interval [p 1 /q; p /q] contains a full set of conjugates of an algebraic integer of degree d Define ϱ = p 1 + p q, λ = p p 1 4q

5 Conjugates of algebraic integers 383 Following [Ro] and [En] an irreducible monic polynomial of degree d with all d roots in the interval [p 1 /q; p /q] = [ϱ λ; ϱ + λ] can be constructed by means of the Chebyshev polynomials [m/] T m (x) = x m + ( 1) j m j4 j j=1 In [ 1; 1] these are also given by the formula Set We write ( m j 1 j 1 T m (x) = 1 m cos(m arccos x) ( ) x ϱ P m (x) = (λ) m T m λ P d (x) = x d + d c d,j x d j j=1 ) x m j The denominators of the rational numbers ϱ and λ are both at most q Hence the coefficients c d,1, c d,,, c d,q are all even integers if d is divisible by q!4 q (q) q This is exactly the case, since d S(, 16, ) (see the Lemma) All the polynomials P m (x) are monic, except for P 0 (x) = So in [ 1; 1) there are numbers b q+1, b q+,, b d such that the polynomial Q d (x) = P d (x) + d j=1+q d 1 b j P d j (x) = x d + a j x j has all coefficients a k integral and even, a 0 not being divisible by 4 Therefore, Q d (x) is irreducible by Eisenstein s criterion In the interval [ϱ λ; ϱ + λ] the maximum of the absolute value of P m (x) equals λ m Consequently, in this interval we bound Q d (x) P d (x) d j=1+q λ d j = λd q 1 λ 1 j=0 < λ d λ q (λ 1) Since q > 3 log 1 and λ > (log ) log, for large d we have ( q log λ > q(λ 1) 1 λ 1 ) ( ) 3(log ) 3(log ) > log 1 )(1 > log

6 384 A Dubickas Therefore, λ q (λ 1) > (log ) (λ 1) > > 1 e 3 Hence, in the interval ϱ λ x ϱ + λ we have (3) Q d (x) P d (x) < λ d Suppose ξ 1 < < ξ d+1 are the points in [ϱ λ; ϱ + λ] such that P d (ξ j ) = λ d By our choice, d is even So P d (ξ j ) is positive for odd j and negative for even j From (3) we see that at each of the points ξ 1,, ξ d+1 the signs of the values of Q d (x) and P d (x) coincide Hence in each of the d intervals (ξ i ; ξ i+1 ), where i = 1,, d, there is a zero of Q d (x) This proves Theorem 4 Proof of Theorem 1 We first prove that I() It suffices to show that no interval [A; B] with A > 1, B < (3 + 5)/ contains both conjugates of an algebraic integer of degree two Indeed, suppose that p p 4q (4) > 1, p + p 4q < (5) with integers p, q such that p 4q is a positive integer which is not a perfect square Then p 4q < = < 31, so that p 4q 9 Since p 4q modulo 4 is zero or one and p 4q is not a perfect square, it equals 5 or 8 In the case p 4q = 5 inequality (5) implies that p Also, p is odd, hence p 1 Then p p 4q 1 5 < 1, which contradicts (4) If p 4q = 8, then inequality (4) implies that p > Hence p 4, since p is even Then p + p 4q which contradicts (5) > 3 + 5,

7 To prove the inequality Conjugates of algebraic integers 385 I() we show that any closed interval [A; B] of length (1 + 5)/ + contains both conjugates of an algebraic integer of degree two Without loss of generality, we assume that A ( 1; 0], since the intervals [A; B] and [A + z; B + z] with z an integer either both contain or both do not contain any set of conjugates of an algebraic integer There are three possibilities: A ( 1; (1 5)/], A ((1 5)/; 1 ] and A (1 ; 0] In the first case we have B > = + and [A; B] contains both roots of x x 1 In the second case, B > and [A; B] contains both roots of x x 1 Finally, in the third case, B > = = and [A; B] contains both roots of x 3x + 1 Theorem 1 is proved We now show how to compute I(d) for small d As we already noticed, there is no loss of generality to assume that the left endpoints of the intervals lie in ( 1; 0] So the right endpoinds are bounded above, say, by 5 However the interval ( 1; 5) contains only a finite number of sets of conjugates of algebraic integers of degree d Suppose there are M such sets Clearly, M 1, since [0; 4] contains such a set Let also β 1,, β M be the smallest conjugates in these sets and let γ 1,, γ M be the largest ones Since for d we have I(d) > 3, the intersection of the intervals [β i ; γ i ] and [β j ; γ j ] is nonempty So (6) I(d) = max{γ j β i }, where the maximum is taken over all i, j, 1 i, j M, such that for each s, 1 s M, either β s < β i or γ s > γ j If d is small, then M is not very large Having all M polynomials of degree d with all roots in ( 1; 5) one can apply formula (6) in order to find I(d) explicitly From (6) it is also clear that we can replace the word closed by half-closed (interval) in the definition of I(d)

8 386 A Dubickas This research was partially supported by a grant from Lithuanian Foundation of Studies and Science References [Du] A Dubickas, On the maximal conjugate of a totally real algebraic integer, Lithuanian Math J 37 (1997), [En] V Ennola, Conjugate algebraic integers in an interval, Proc Amer Math Soc 53 (1975), [Kr] L K r o n e c k e r, Zwei Sätze über Gleichungen mit ganzzahligen Coefficienten, J Reine Angew Math 53 (1857), [La] M Langevin, Solution des problèmes de Favard, Ann Inst Fourier (Grenoble) 38 (1988), no, 1 10 [Le] D H Lehmer, Factorization of certain cyclotomic functions, Ann of Math 34 (1933), [Lo] R Louboutin, Sur la mesure de Mahler d un nombre algébrique, C R Acad Sci Paris 96 (1983), [Ro] R M Robinson, Intervals containing infinitely many sets of conjugate algebraic integers, in: Studies in Mathematical Analysis and Related Topics, Stanford Univ Press, 196, [SZ] A Schinzel and H Zassenhaus, A refinement of two theorems of Kronecker, Michigan Math J 1 (1965), [Sc] I Schur, Über die Verteilung der Wurzeln bei gewissen algebraischen Gleichungen mit ganzzahligen Koeffizienten, Math Z 1 (1918), [Za] T Zaïmi, Minoration du diamètre d un entier algébrique totalement réel, C R Acad Sci Paris 319 (1994), Department of Mathematics and Informatics Vilnius University Naugarduko Vilnius, Lithuania arturasdubickas@mafvult Received on and in revised form on (357)

TWO VARIATIONS OF A THEOREM OF KRONECKER

TWO VARIATIONS OF A THEOREM OF KRONECKER ARTŪRAS DUBICKAS AND CHRIS SMYTH ABSTRACT. We present two variations of Kronecker s classical result that every nonzero algebraic integer that lies with its conjugates