Beruflich Dokumente
Kultur Dokumente
an?112 SELECTED POLYNOMIAL INEQUALITIES and the equality holds if and only if ag = 1—2p/n. In the case q0) shows that F3; = 2((a — b)/n)°. Thus, F; may be zero, positive, or negative. Consider now a general expression Day = A (1) where 2!) (v = 1,...,n—k) are zeros of the k-th derivative P(*) and p< n, Evidently, Dy..2 = Fm. The following result was proved by Aihara [1]; THEOREM 1.3.14. Dm,p = 0 for m= 1,2,... ,p. At the end of this subsection, we mention a result, of Erdés [2]: THEOREM 1.3.15. If P(x) = cg +12 +---+en2", Coch # 0, has integral coefficients only and has all its zeros in (-1,1), then len! > 277. Recently Reddy [1] obtained the following generalization:INEQUALITIES INVOLVING ALGEBRAIC POLYNOMIALS — 113 THEOREM 1.3.16. If P(z) = en(z—21)(t-Z2) ...(2-Z) (n > 1) has all its zeros in [—1, 1] and satisfies, for k = 0,1,...,m(m > 1), [P(n)| 24 >0, me = cos(ke/m), then len| > "Va. Equality holds for P(z) = +aTm(z)"/™ (a > 0) whenever m divides n, where Tm is the m-th degree Chebyshev polynomial of the first kind, defined by T,,(2) = cos(marccos z) {(-1 < z < 1). Notice that for a = 1 and m = 2 this result reduces to Theorem 1.3.15. 2.1.4. Inequalities Involving Zeros of Polynomials f2a=& Kt in, (v = 1,...,n) are the zeros of the polynomial ze P(z)= B a,z”, Ostrowski [3] proved the following inequalities for all real values of z: y (2) >Re P(z)PM(z)— P(x? Qe P(zy , 1)? ae Pilia)? = Plix)PMiz) »D (z) 2 Re P(iz)?? In addition, if m # 0 or & # 0 (v =1,... ,n), then we have n 2 2 " (2) > BRe Piz) fe (z) Sn (2) and “712 P(iz)P"(iz) — P'(ix)? y (G) 2 8R PG , Tespectively. These a are equivalent, respectively to 71 =2aa, sl 2agay — —) >8Re ' =) >8Re———. D(z) 2am, YE) a De Bruijn [1] proved for some special lasses of polynomials the following result:na SELECTED POLYNOMIAL INEQUALITIES THEOREM 1.4.1. Let the polynomial z > P(z) of degree n > 1, have the zeros 21,... ,Zn, and let wy,...,Wn—1 be those of z+ Pz). Then we have ou Ie (1.4.1) A Yiimu,| < = Yo Im... v=1 vl The equality holds if and only if no two zeros of z+ P(z) are sepa- rated by the real azis. In the following we will prove the generalized form of the above theorem (due to de Bruijn and Springer [1]), namely for polynomials with arbitrary real or complex coefficients. For the proof de Bruijn and Springer [1] introduced an auxiliary function z+ P*(z) obtained from P(z) by replacing the zeros of z+ P(z) in the lower half-plane by their complex conjugates. Theorem 1.4.1 follows very easily when all zeros of z + P(z) lie in Imz > 0. Because in that case, by the Gauss-Lucas theorem (see Theorem 1.2.1 in Chapter 3), the same is true for the zeros of z+ P'(z), so that the imaginary parts of 2,... Zn, Wiy-++)Wn-1 all have the same sign. The theorem then follows from the relation Therefore in (1.4.1) the sign of equality holds in this case. The general case is reduced to this one after applying the following lemma of de Bruijn and Springer [1]. Lema 1.4.2. Let ik n P(z)=e]] (2-2) J @-2) (
0 (v =1,...,k), Imz, <0 (v=k+1,...,n). Ifwe sel k n (142) P*()=a][ (2-2) [] @-2) val vektlINEQUALITIES INVOLVING ALGEBRAIC POLYNOMIALS — 115 then IP(@) < IP"@) for all real values of z. The equality holds for all real x if and only if no two zeros of z++ P(z) are separated by the real azis. Proor. Setting n 1 1 y =QtiR, DY 5 2, vokt) ud =S4+iT where z, Q, R, 5, T are real, we obtain P'(2) | (2) | Play (2+ S)+HRET) and Pez) (2 FS) +HR-7). It follows from Imz, 20 (v=1,....k) mz, <0 (v=k+1,...,0) that R>0,T <0. Thus |R+7|<|R—-T|, which implies Fe Piz) P(z) If z is real then |P(z)| = |P*(2)|, and therefore IP(2) < lP"@)I- The equality holds for all rea] z only if either R = 0 or T = 0, in other words if all zeros of z+ P(z) lie either inImz 0. O Applying the above lemma, the proof of Theorem 1.4.1 follows. Proor of THEOREM 1.4.1. Let the polynomial z 4 P*(z) be defined by (1.4.2). Because of the fact all zeros of z+ P*(z) lie in Imz > 0, it follows that n~1 1 nor hee 1.4.3 Imw}] = — (1.43) Y leat alma,née SELECTED POLYNOMIAL INEQUALITIES where wj,..., w%_, denote the zeros of z++ P*’(z). Also, A A aa ff roglPteide sf teele elds (A> 0) -A -A or nol n-1 (1.4.5) ¥ [steele eS [lew We have A (1.4.6) il log |z — al dz = 2(Alog A — A) + x[Ima]+ O(1/A). -A From (1.4.5) and (1.4.6), and taking the limit as A ++ +00 we obtain n-1 n-1 (1.4.7) Yitmu,| < Ym uy]. tat oa From (1.4.3) and (1.4.7) we obtain aol LS |Imw, [<1 hima) ws ls vt vs1 which proves (1.4.1). The equality in (1.4.7) holds if and only if there is equality in (1.4.4), in other words if |P(z)| = |P*'(z)| for all real values of z. Then using Lemma 1.4.2 one has that all zeros of z ++ P(z) lie either in Imz > O or in Imz < 0. This completes the proof of Theorem 14.1. 0 Remark 1.4.1. In geometric terms Theorem 1.4.1 means that the zeros of z+ P"(z) lie, in the mean, closer to the real axis, than the zeros of z+ P(z). The same applies about any line when we replace [Im z] by |Im(az + A)|, where o and f are complex numbers. This can be proved if one applies Theorem 1.4.1 to the function z ~ P((z — B)/a). In an analogous manner de Bruijn and Springer [1] proved the following theorems.INEQUALITIES INVOLVING ALGEBRAIC POLYNOMIALS M7 THEOREM 1.4.3. Let 2,...,2n be the zeros of the polynomial 2 P(2), wy... ,Un-1 the zeros of P'(z), and let Im 2 < Im z < oS Imzy. If z+ Y(z) is a conver real-valued function for x € [Im 2, Imz,], and if 12 1 oad DWP) = = vm 2.) - —> Yo W(Imw,), 1 val then D(y,P) > 0, with equality only if is a linear function for x € [Imz,,Imz,]. This implies the case Im 2, = --- = Im 2p- One obtains a special case when #(z) = |z|" (r > 1), for which the following theorem holds: THEOREM 1.4.4. Using the notation of Theorem 1.4.1, for every r>1, we have 1 12 . ao Dla $y dlimat" = vel The equality holds in the following two cases: a) if r = 1 and all zeros of P lie in the same half-plane Imz > 0 or Imz <0, and b) ifr > 1 and Imz =---=Imzy. THEOREM 1.4.5. Under the hypotheses of Theorem 1.4.1 we have, forr>1, 1 n-1 1 n (1.48) ales. Dial a m The equality holds in the following two cases: a) if r = 1 and all zeros of P lie in the same half-line with endpoint 0, and b) if 21 = --- = Zn. Proor. The distance of the point z € C to the line through the point 0 making an angle a with the positive rea] axis, is [cos alm z ~ sina Rez|. By Theorem 1.4.4 we have n-1 n i, 1 > \cosalm z—sinaRez|" < = > |cosa Im z— sina Rez|", v=l v1 n-1us SELECTED POLYNOMIAL INEQUALITIES wherefrom, by integrating we obtain 12 an al | cosa Im z — sin a Re z|"da nah ilo PS pr sip | cosa Im z — sin a Re 2|"da, ny=i/0 or (1.4.8). O Some similar results can be found in the papers of Bray [1], Erdés and Niven [1], and de Bruijn and Springer (2]. Let a, b, and ¢ be three points in the complex plane, and let A(a, b,c) denote the area of the triangle determined by those points. Kramer [1] proved the following result: THEOREM 1.4.6. Let z ++ P(z) be a polynomial of degree n with Z€TOS 21, -.-, Zn, and let wy, ..., Wa be the zeros of the derivative P!. Then we have ar > Alstps Wyte) $ > Al Zi, 25s 2k). Agp 2, : [AP oa Papo [Poel Pe (14 oles) ek 2 vel where equality occurs if and only if \ = k = 2 or, forX =2 ifk-2 of the left-hand terms are equal to 1. In the book of Ostrowski [1, p. 32] the following result is mentioned as a problem: THEOREM 1.4.8. If 2),... ,2n are zeros of the complex polynomial R z+ P(z) = } a,2” (a, € C; a, #0) then v0 (1.4.9) Shall” < £ len] Te + 2-4). v=0 We give the proof of Kalajdaié [1]. Proor or THEOREM 1.4.8. Let Qe) = Soh = Jol [Te +a) (bn = lanl) vat and let o,(z1,-..,2n) be the elementary symmetric function of 71, s+ ®n (cf. Subsection 1.3.1). Since —|z,| (v = 1,... ,n) are zeros of the polynomial Q, according to Viete’s formulae we conclude that (-1ytev(lah- be) = (D8,120 SELECTED POLYNOMIAL INEQUALITIES baw = lanlor(I2il,--- »Lznl) 2 lenllo.(215--- zn) On-y Gy, = {onl = lan] (v=... yn). Therefore, for v = 0,1,... ,n we have 6, — |a,| > 0, which implies the inequality " YL. - leet’ 20. v=0 This inequality is equivalent to (1.4.9), foreveryz€C. O At the end of this subsection we mention a result of Colucci [1]: Tueorem 1.4.9. If the moduli of each of the zeros of a complez polynomial z ++ P(z) = bo az” (a, € C; a, # 0) do not exceed a positive number M, then’ [PM(2)| < u(t 1) lel +m)", for every k = 0,1,...,n. 2.1.5. Inequalities for Non-negative Polynomials For polynomials with non-negative coefficients, Milovanovié [2] proved the following result: THEOREM 1.5.1. Let P(z) = S az” with a, >0(0 0 the inequality, (15.1) 2(P(2) ~ Ple)P"(2)) < P@)P(2) holds. Proor. Using the Cauchy-Schwarz-Buniakowsky inequality | Yew Ps (o I.) (5 Iwl?),INEQUALITIES INVOLVING ALGEBRAIC POLYNOMIALS 121 for z, = af/*z4/? and y, = val/*z"/? (2 > 0) we obtain (Deen) < (La)(Sa-’) which is equivalent to (1.5.1). O For the Lorentz’s class of non-negative polynomials on [—1, I], i-e., for polynomials of the form (1.5.2) Q(t)= ya —yr(en’, by 20 (OS 0) we obtain Pz) P"(z) < 1 Piz) Pz) ~ 2° Of course, for z > 0 the following inequalities PHY z) _ PIED ( 2) PUD (2) PUN(z) (15.4) st usesn) also hold. Since P("+))(z) = 0, using (1.5.4) we find that (z>0), which was proved by Soble [1]. The case P(x) = 2” shows that this is the best possible result. If, moreover, the constant term in P(z) is zero, then Piz) P(z) The case P(x) = z shows that this is the best possible result. Soble [1] also proved the following three theorems: 1 <> (e>0). ‘Tneonem 1.5.3. Let P(z) = Ss eyz"-", with c, >0. If0 1(v=1,...,n), then i P\(z) P(z) (2 -¢/2? a(z—cy ls n (z >).INEQUALITIES INVOLVING ALGEBRAIC POLYNOMIALS — 123 n THeoreM 1.5.5. Let P(z) = } cya"-” (c, > 0). Ife, > eva =0 (v=1,...,n), then Pz P(z ntl sz (0 e). 2.1.6. Inequalities for the Coefficients of Polynomials In 1945 Visser [1] proved: Theorem 1.6.1. Ifz++ P(z) = 3> a,2” is a polynomial of degree w=0 n and |P(z)| <1 for |2| <1, then leo] + len] < 1. There are several results for self-inversive polynomials (see Chapter 1, Subsection 1.2.1). Using the previous theorem, Visser [1] stated the following result: n THEOREM 1.6.2. If z++ P(z) = ¥> a,z” is a self-inversive poly- =0 nomial of degree n and |P(2)| <1 for |z| <1, then mie leo] < An improvement of Theorem 1.6.2 was obtained and proved by Malik [2] in the following theorem: n THEOREM 1.6.3. Let z +> P(z) = ¥ a,2" be a self-inversive =0 polynomial of degree n and |P(2)| <1 for [zl < 1, then a9 (1.6.1) tat + (7) lac ( 0 we have $(n,0) = 1. Suppose now that k < 1. mS O62) sone S7()() = (1) +0 (62) +72") so Feo V()-C0 10 “Ee(S Ci) = S(n-1,k-1). Iterating this identity we prove Lemma 1.6.4. O In order to prove Lemma 1.6.5, Malik [2] used the Visser’s theorem 1.6.1, the previous lemma and the following result of de Bruijn [2] (see also Chapter 3, Subsection 3.1.3):INEQUALITIES INVOLVING ALGEBRAIC POLYNOMIALS 125 THEOREM 1.6.6. Let z++ P(z) be a polynomial of degree n such that |P(z)| < 1 for |z| < 1, and let z ++ Pe(z) be @ polynomial of degree n — 1, defined by 1 Pez) = E- z)P"(z) + P(z). Then, |Pe(z)| < 1 for |z| < 1 and |€| < 1. Proor oF THEOREM 1.6.3. Since P is given to be a self-inversive polynomial of degree n, we have an_, = a, (k = 0,1,...,n). Then Lemma 1.6.5 implies -1 n 1 6. <= (1.6.2) |eo+a(?) a |< 5 for |A| = 1. Choose arg A = — arg(ax/ao) a priori, then (1.6.2) implies (1.6.1). O An immediate corollary of Theorem 1.6.3 is the following result: Coroutary 1.6.7. Let z > P(z) = s a,z” be a self-inversive =0 polynomial of degree n, such that |P(2)|'< 1 for |z| < 1 and |ao| = 1/2. Then P(z) = Fast +&) for |al = 1. Using the method of Visser [1], van der Corput and Visser [1] proved the following result: n THEOREM 1.6.8. Let z + P(z) = > a,z” be a polynomial of =0 degree n such that |P(z)| < M for |2| <1. If ap, a, (p< 9) ore two coefficients such that for no other coefficient a, # 0 we have 5 =pmod(q-— p), then (1.6.3) lopl+ lagl < Mf. Also On (1.6.4) lol + lol < 3 if [P(e a.126 SELECTED POLYNOMIAL INEQUALITIES Also, they considered real trigonometric polynomials of the form . (1.6.5) T(0)= Y> ae*, ven where a, = @,, and proved that 2 1 Br fol + 2a] < maxI7(O), lool + Floal <¢ f”17(0108 for k > n/2. The constant 1/4 in the last inequality was improved by Boas [1] to (3 + V2)/(6r) = 0.23418. Later Boas [2] found the best constant C, for any positive y and k > n/2 in the following inequality 2m lal +2rlae] < Cy f” Ire). , Namely, if y is the smallest positive root of the equation : 1 sing = gue — 2y), then C, = (21 — 4y)-!. For 7 = 1/3 the corresponding constant becomes 0.213653... . Also, Boas [2] proved that 1 7 2m lal < jos [ ercoi, where p = 2[(n — k)/(2k)] + 1 is the largest odd integer with pk n. If on the other hand, z ++ Q(z) has all its zeros in the closed interior of the unit disk, then lao! + lon! < [bol men; max{[ao|, |x|} < lbm| — bol, m > 15 and (1.6.6) hold.128 SELECTED POLYNOMIAL INEQUALITIES THEOREM 1.6.11. If z 4 P(z) = x a,z” is a polynomial of =0 degree n and ay, ay (p < q) are two coefficients such that for no other coefficient a, # 0 we have s = pmod(q — p), then for every r>1, eres lal + lol < 208s (= [Veter a0)", where T(ir+l 062) Mt=c,= 2% = HD) Rs erdy BT (Rr + 3) The result is best possible. Remark 1.6.1. The constant Cy given by (1.6.7) appears in several extremal problems with polynomials (cf. Theorem 1.7.13 in Chapter 5). Rahman [1] also proved: n THEOREM 1.6.12. If z ++ P(z) = }> a,2” is a polynomial of =0 degree n andO 1, <2c(y( 2 [ |rre*yra0)” lol + log < 20) (5— f“1Pte ras) where if" = (r-1)/r -(1 ja r/(r-1) cte)= (ge f° Joos Hanya? a)" The result is best possible. THEOREM 1.6.13. If z+ Plz) = 35 a,2” is @ polynomial of v=0 degree n such that |P(z)| < M for |z| < 1, then 5M la,| + lag] < TrINEQUALITIES INVOLVING ALGEBRAIC POLYNOMIALS 129 if either 0 <2p
Q such that 4c An Soa where n € N. It is a theorem of Rahman [1] that 2 T (1.6.9) An(P,9) $ Pa where 0 < p (m—2)(q— p). Inequality (1.6.9) cannot be sharp in every case, since A 4 elim, An(p.9) = for p,q fixed, which contradicts (1.6.8). To find the value of A,(p,q) in the cases where (1.6.9) is not sharp seems to be a more difficult problem. Brickman, Rahman and Ruscheweyh [1] asked the following interesting question: Narrow the gap in (1.6.10) and (1.6.11) and to decide the question of whether or not alim,, B,, ezists. At the end of this subsection we mention some inequalities of Mabler [1] and Duncan [1]. n Let z+ P(z) = } a,2” be an arbitrary polynomial with complex =0 coefficients. Set H(P) = max{ ao], |ail,--- ,Jan]}, Le(P) = (Jaal” + lal! +--+ laal")"”" (7 > 0), and m(P)=exn( [oe Pte*)),INEQUALITIES INVOLVING ALGEBRAIC POLYNOMIALS 131 with M(P) = 0, when P(z) = 0. Then, for any two polynomials P and Q, we have the following inequalities for their lengths Ty(PQ)S L(P)Li(Q), — n(P #Q) < Li(P) + L1(Q). The height H, the length £,, and the measure M of P are connected by the inequalities (see Mahler (1]) = (era) H(P) < M(P) < H(P)VaFT, 2 Ly(P) < M(P) & Li(P)- If both P and Q are at most of degree n, then M(P¥Q) $ Li(P ¥Q) < La(P) + Li(Q) < 2"(M(P) + M(Q)). Duncan [1] proved that the factor 2” in the inequality M(P+Q) < 2"(M(P) + M(Q)) can be replaced by (°*)'/”. He also proved the following inequalities ~1f2 (*) LAP) < M(P) < Li(P) and Lx(PYLAQ) < y C2) CR) La PQ), where dg P = n and dgQ =m. Remark 1.6.2. These inequalities have applications in the theory of transcendental numbers. Some extremal problems for coefficients of polynomials will be treated in Chapter 5.132 SELECTED POLYNOMIAL INEQUALITIES 2.1.7. Integral Inequalities Lukacs [1] proved the following results: THEOREM 1.7.1. If P is an algebraic polynomial of degree at most 2n~1, which is non-negative on [a,b], then (1.7.1) P(a)< met) if ” pa)de and b (1.7.2) P(t) < wet i P(2) dz with equality in (1.7.1) if and only if P(z) = C(b- 2) {r (a 4 74) y i. and in (1.7.2) if and only if _ 2 bta\)* rerecte-afra(sta- Hs) ), where C is a positive constant and P,, is the Legendre polynomial of degree n defined by") Le Pal) = Tow a (-1)". Under the same conditions the following inequality n n b P(2)< ee f P(z)dz (a2) with only real zeros, and such that P(-1) = P(1) = 0, P(z) # 0 for -l 2) having only real zeros, and let P(z) > 0 for z € (-1,1). Then -I “a max P(z). If, in addition, P(—1) = P(1) = 0, then io P(z)dz > 7 Hes +— a)” amas P(z). Dimovski and Cakalov i" provided the following generalization: TueoreM 1.7.7. Let P and Q be real polynomials of degree at most m and n, respectively having no zeros in the open disk |z| <1, and let P(x) > 0 and Q(z) > 0 for z € (-1,1). Then i. 2 men \ ie [Pees ( 5 ) _pm, PCE)_ max, Oe), with equality if and only if P(z) = a(1 tz)" and Q(z) = B(1 Fz)", where a and 3 are arbitrary positive constants. They also proved:INEQUALITIES WITH TRIGONOMETRIC POLYNOMIALS 135, THEOREM 1.7.8. Let P be a real polynomial of degree m having no zeros in the open disk |2| <1 and let_max P(z)=1. IQ is a -1 PE ("t ar 2.2, INEQUALITIES WITH TRIGONOMETRIC POLYNOMIALS 2.2.1. Inequalities for Trigonometric Sums ‘We start this subsection with some simple estimates of trigono- metric sums. Let m and m + 1 be natural numbers and 0 < @ < 2z. Since | by itt |= le torn Ea | [=i sin(o/2) we obtain fa a simple inequality ntm (2.1.1) | yet? 1 naa sin(0/2) Taking a monotonically decreasing sequence {r,}#°¢ such that (0<@< 2). lim r,=0 k+00 and applying the Abel’s inequality (cf. Mitrinovic [1, pp. 32-33]) to inequality (2.1.1) we find ntm ike Tntl Tn | Do ne | sin(6/2) ~ sin(0/2)° kant Finally, letting m — +00, we obtain an inequality of Blankfield and Zeitlin [1], namely (s ryei*® |< —™ (0<@<2n). Ree sin(8/2) Using a geometric method Kuz’min (1] proved the following result:136 SELECTED POLYNOMIAL INEQUALITIES THEOREM 2.2.1. Let n be a natural number > 2 and let ax (k = 1,...,) be real numbers such that 0 (bie 7) Me Assuming that (2.1.2) 2x (lends bal) =1