Maths Olympiad Prep

Library / /5 of 8

, 2022

Algebra Difficulty 8.8 Shortlist Prove it China

(1) Prove that on the complex plane, the convex hull of all complex roots of the equation
z20+63z+22=0 z^{20} + 63z + 22 = 0
is larger than π\pi.

(2) Let nn be a positive integer, and 1k1<k2<<kn1 \le k_1 < k_2 < \dots < k_n be nn odd integers. Prove that for any nn complex numbers a1,a2,,ana_1, a_2, \dots, a_n with sum 1 and any complex number ww with absolute value at least 1, the equation
a1zk1+a2zk2++anzkn=w a_1z^{k_1} + a_2z^{k_2} + \dots + a_nz^{k_n} = w
has at least one complex root with norm less than or equal to 3nw3n|w|.

Solution

(1) We first prove the following.

Lemma: [Gauss-Lucas Theorem] All zeros of the polynomial f(z)f'(z) lie in the closure formed by zeros of f(z)f(z).

Proof of Lemma: By the fundamental theorem of algebra, we know that f(z)f(z) admits a decomposition:
f(z)=A(zz1)α1(zz2)α2(zzn)αn. f(z) = A(z - z_1)^{\alpha_1} (z - z_2)^{\alpha_2} \cdots (z - z_n)^{\alpha_n}.
Thus, z1,z2,,znz_1, z_2, \dots, z_n are zeros of f(z)f'(z) with multiplicities α11,α21,,αn1\alpha_1 - 1, \alpha_2 - 1, \dots, \alpha_n - 1, respectively. Say ZZ is any other zero of f(z)f'(z), then
f(Z)f(Z)=α1Zz1+α2Zz2++αnZzn=0. \frac{f'(Z)}{f(Z)} = \frac{\alpha_1}{Z - z_1} + \frac{\alpha_2}{Z - z_2} + \dots + \frac{\alpha_n}{Z - z_n} = 0.
That is,
α1Zz12(Zˉzˉ1)+α2Zz22(Zˉzˉ2)++αnZzn2(Zˉzˉn)=0. \frac{\alpha_1}{|Z - z_1|^2} (\bar{Z} - \bar{z}_1) + \frac{\alpha_2}{|Z - z_2|^2} (\bar{Z} - \bar{z}_2) + \dots + \frac{\alpha_n}{|Z - z_n|^2} (\bar{Z} - \bar{z}_n) = 0.
After taking complex conjugations, the equality becomes
Z=α1Zz12z1+α2Zz22z2++αnZzn2znα1Zz12+α2Zz22++αnZzn2. Z = \frac{\frac{\alpha_1}{|Z-z_1|^2} z_1 + \frac{\alpha_2}{|Z-z_2|^2} z_2 + \dots + \frac{\alpha_n}{|Z-z_n|^2} z_n}{\frac{\alpha_1}{|Z-z_1|^2} + \frac{\alpha_2}{|Z-z_2|^2} + \dots + \frac{\alpha_n}{|Z-z_n|^2}}.
Hence ZZ is a convex linear combination of z1,z2,,znz_1, z_2, \dots, z_n. This proves the lemma.

Back to the original problem. Gauss-Lucas Lemma implies that the convex hull of the zeros of the given equation contains the convex hull of the zeros of the following equation
20z1963=0. 20z^{19} - 63 = 0.
This is a regular 19-gon inscribed in a circle of radius (6320)1/19\left(\frac{63}{20}\right)^{1/19}.

The inscribed circle of this regular 19-gon has a radius
(6320)1/19cosπ19>(6320)1/19(112(π19)2)>(6320)1/19(112(16)2)=7172(6320)1/19. \left(\frac{63}{20}\right)^{1/19} \cos \frac{\pi}{19} > \left(\frac{63}{20}\right)^{1/19} \left(1 - \frac{1}{2} \left(\frac{\pi}{19}\right)^2\right) > \left(\frac{63}{20}\right)^{1/19} \left(1 - \frac{1}{2} \left(\frac{1}{6}\right)^2\right) = \frac{71}{72} \left(\frac{63}{20}\right)^{1/19}.
On the other hand, since
(7271)19=(1+171)19<1+1971+18C1921722<3<6320, \left(\frac{72}{71}\right)^{19} = \left(1 + \frac{1}{71}\right)^{19} < 1 + \frac{19}{71} + 18 \cdot C_{19}^2 \frac{1}{72^2} < 3 < \frac{63}{20},
the radius of the inscribed circle is larger than 1. This proves (1).

(2) We need to prove the following.

(A) If a1,a2,,ana_1, a_2, \dots, a_n and ww satisfy the assumptions, then the minimal norm of the roots of
a1zk1+a2zk2++anzkn=w() a_1 z^{k_1} + a_2 z^{k_2} + \dots + a_n z^{k_n} = w \quad (*)
is less than or equal to 3mn3mn.
This is equivalent to the following.

(B) If a1,a2,,ana_1, a_2, \dots, a_n and ww satisfy the assumptions, then the maximal norm of the roots of
wzkna1zknk1a2zknk2an1zknkn1an=0,() wz^{k_n} - a_1z^{k_n-k_1} - a_2z^{k_n-k_2} - \dots - a_{n-1}z^{k_n-k_{n-1}} - a_n = 0, \quad (**)
denoted by M0M_0, is greater than or equal to 1/(3mn)1/(3mn).

By the lemma in (1), the maximal norm of roots of ()(**) cannot be less than the maximal norm of roots of
wknzkn1a1(knk1)zknk11an1(knkn1)zknkn11=0.(1) w k_n z^{k_n-1} - a_1(k_n - k_1)z^{k_n-k_1-1} - \dots - a_{n-1}(k_n - k_{n-1})z^{k_n-k_{n-1}-1} = 0. \quad (1)
That is, the maximal norm, say M1M_1, of the roots of
wknzkn1a1(knk1)zknk11a2(kn1k2)zknk21an1(knkn1)=0.(1) w k_n z^{k_n-1} - a_1(k_n - k_1)z^{k_n-k_1-1} - a_2(k_{n-1} - k_2)z^{k_n-k_2-1} - \dots - a_{n-1}(k_n - k_{n-1}) = 0. \quad (1')
Beginning with (1)(1'), we repeat the same process for nsn-s times. It is known that the maximal norm of the roots decreases each time. Therefore,
M0M1Mns, M_0 \ge M_1 \ge \dots \ge M_{n-s},
where MnsM_{n-s} is the maximal norm of the roots of the equation
wknkn1ks+1zksa1(knk1)(kn1k1)(ks+1k1)zksk1as(knks)(ks+1ks)=0.((ns)) w k_n k_{n-1} \cdots k_{s+1} z^{k_s} - a_1(k_n - k_1)(k_{n-1} - k_1) \cdots (k_{s+1} - k_1) z^{k_s - k_1} \\ - \cdots - a_s(k_n - k_s) \cdots (k_{s+1} - k_s) = 0. \quad ((n-s)')

By Vieta's theorem, we have
M0M1Mns[(knks)(ks+1ks)knkn1ks+1asw]1/ks. M_0 \ge M_1 \ge \dots \ge M_{n-s} \ge \left[ \frac{(k_n - k_s) \cdots (k_{s+1} - k_s)}{k_n k_{n-1} \cdots k_{s+1}} \cdot \left| \frac{a_s}{w} \right| \right]^{1/k_s}.
Hence it suffices to show that there exists s{1,2,,n}s \in \{1, 2, \dots, n\} such that
[knkn1ks+1(knks)(kn1ks)(ks+1ks)]1/ks[was]1/ks3mn. \left[ \frac{k_n k_{n-1} \cdots k_{s+1}}{(k_n - k_s)(k_{n-1} - k_s) \cdots (k_{s+1} - k_s)} \right]^{1/k_s} \cdot \left[ \left| \frac{w}{a_s} \right| \right]^{1/k_s} \le 3mn.
Note that
kpkpks=1+kskpks(1+1kpks)ks. \frac{k_p}{k_p - k_s} = 1 + \frac{k_s}{k_p - k_s} \le \left(1 + \frac{1}{k_p - k_s}\right)^{k_s}.
So it remains to show that there exists some ss such that
[p=s+1n(1+1kpks)](mas)1/ks3mn.2 \left[ \prod_{p=s+1}^{n} \left( 1 + \frac{1}{k_p - k_s} \right) \right] \cdot \left( \frac{m}{|a_s|} \right)^{1/k_s} \le 3mn. \quad 2
We also make the following observations.

* Since all kjk_j's are odd numbers, kpks2(ps)k_p - k_s \ge 2(p-s). Therefore,
1+1kpks1+12(ps)<ps+1ps; 1 + \frac{1}{k_p - k_s} \le 1 + \frac{1}{2(p-s)} < \frac{p-s+1}{p-s};
* Since j=1naj=1\sum_{j=1}^n a_j = 1, we must have j=1naj1\sum_{j=1}^n |a_j| \ge 1. Thus there must be ss satisfying as12s|a_s| \ge \frac{1}{2^s}, or equivalently, 1as2s\frac{1}{|a_s|} \le 2^s.

We separate the discussion into the following two cases.

(1) If there is ss (allowing s=1s=1), such that ks3k_s \ge 3 and as12s|a_s| \ge \frac{1}{2^s}, then (note that s2s1kss \le 2s-1 \le k_s)
[p=s+1n(1+1kpks)](mas)1/ks<[p=s+1nps+1ps]m1/ks2s/ks<nm32<3mn. \left[ \prod_{p=s+1}^{n} \left( 1 + \frac{1}{k_p - k_s} \right) \right] \cdot \left( \frac{m}{|a_s|} \right)^{1/k_s} < \left[ \prod_{p=s+1}^{n} \frac{p-s+1}{p-s} \right] \cdot m^{1/k_s} \cdot 2^{s/k_s} < n \cdot \sqrt[3]{m} \cdot 2 < 3mn.

(2) Otherwise, we must have k1=1k_1 = 1 and a112|a_1| \ge \frac{1}{2}. Hence
[p=2n(1+1kpk1)]ma1[p=2n2(p1)+12(p1)]m2<k=12n2k+1k632=2n1m2<3mn. \left[ \prod_{p=2}^{n} \left( 1 + \frac{1}{k_p - k_1} \right) \right] \cdot \frac{m}{|a_1|} \le \left[ \prod_{p=2}^{n} \frac{2(p-1)+1}{2(p-1)} \right] \cdot m \cdot 2 < \sqrt{\prod_{k=1}^{2n-2} \frac{k+1}{k}} \cdot 63 \cdot 2 = \sqrt{2n-1} \cdot m \cdot 2 < 3mn.

This completes the proof.

Want a route through all this instead of 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.