Difference between revisions of "1988 USAMO Problems/Problem 5"

Revision as of 23:40, 24 June 2022

Problem

Let $p(x)$ be the polynomial $(1-x)^a(1-x^2)^b(1-x^3)^c\cdots(1-x^{32})^k$ , where $a, b, \cdots, k$ are integers. When expanded in powers of $x$ , the coefficient of $x^1$ is $-2$ and the coefficients of $x^2$ , $x^3$ , ..., $x^{32}$ are all zero. Find $k$ .

Solutions

Solution 1

First, note that if we reverse the order of the coefficients of each factor, then we will obtain a polynomial whose coefficients are exactly the coefficients of $p(x)$ in reverse order. Therefore, if $p(x)=(1-x)^{a_1}(1-x^2)^{a_2}(1-x^3)^{a_3}\cdots(1-x^{32})^{a_{32}},$ we define the polynomial $q(x)$ to be $q(x)=(x-1)^{a_1}(x^2-1)^{a_2}(x^3-1)^{a_3}\cdots(x^{32}-1)^{a_{32}},$ noting that if the polynomial has degree $n$ , then the coefficient of $x^{n-1}$ is $-2$ , while the coefficients of $x^{n-k}$ for $k=2,3,\dots, 32$ are all $0$ .

Let $P_n$ be the sum of the $n$ th powers of the roots of $q(x)$ . In particular, by Vieta's formulas, we know that $P_1=2$ . Also, by Newton's Sums, as the coefficients of $x^{n-k}$ for $k=2,3,\dots,32$ are all $0$ , we find that $\begin{align*} P_2-2P_1&=0\\ P_3-2P_2&=0\\ P_4-2P_3&=0\\ &\vdots\\ P_{32}-2P_{31}&=0. \end{align*}$ Thus $P_n=2^n$ for $n=1,2,\dots, 32$ . Now we compute $P_{32}$ . Note that the roots of $(x^n-1)^{a_n}$ are all $n$ th roots of unity. If $\omega=e^{2\pi i/n}$ , then the sum of $32$ nd powers of these roots will be $a_n(1+\omega^{32}+\omega^{32\cdot 2}+\cdots+\omega^{32\cdot(n-1)}).$ If $\omega^{32}\ne 1$ , then we can multiply by $(\omega^{32}-1)/(\omega^{32}-1)$ to obtain $\frac{a_n(1-\omega^{32n})}{1-\omega^{32}}.$ But as $\omega^n=1$ , this is just $0$ . Therefore the sum of the $32$ nd powers of the roots of $q(x)$ is the same as the sum of the $32$ nd powers of the roots of $(x-1)^{a_1}(x^2-1)^{a_2}(x^4-1)^{a_4}(x^{8}-1)^{a_4}(x^{16}-1)^{a_{16}}(x^{32}-1)^{a_{32}}.$ The $32$ nd power of each of these roots is just $1$ , hence the sum of the $32$ nd powers of the roots is $P_{32}=2^{32}=a_1+2a_2+4a_4+8a_8+16a_{16}+32a_{32}.\tag{1}$ On the other hand, we can use the same logic to show that $P_{16}=2^{16}=a_1+2a_2+4a_4+8a_8+16a_{16}.\tag{2}$ Subtracting (2) from (1) and dividing by 32, we find $a_{32}=\frac{2^{32}-2^{16}}{2^5}.$ Therefore, $a_{32}=2^{27}-2^{11}$ .

Solution 2

By a limiting process, we can extend the problem to that of finding a sequence $b_1, b_2, \ldots$ of positive integers such that $(1 - z)^{b_1}(1 - z^2)^{b_2}(1 - z^3)^{b_3}\cdots = 1 - 2z.$ (We could make the limiting process more formal by invoking the "usual" ( $(z)$ -adic) topology on the space of formal power series in $z$ , or if we want to avoid the high-powered stuff, we could also stick to the finite situation in the problem and suppress large powers with big-oh notation, then work in a suitably small ball around the origin to ensure convergence of geometric series.)

If we take logarithmic derivatives on both sides, we get $\sum_{n = 1}^{\infty}\frac{b_n\cdot (-nz^{n - 1})}{1 - z^n} = \frac{-2}{1 - 2z},$ and upon multiplying both sides by $-z$ , this gives us the somewhat simple form $\sum_{n = 1}^{\infty} nb_n\cdot\frac{z^n}{1 - z^n} = \frac{2z}{1 - 2z}.$ Expanding all the fractions as geometric series, we get $\sum_{n = 1}^{\infty} nb_n\sum_{k = 1}^{\infty} z^{nk} = \sum_{m = 1}^{\infty} 2^mz^m.$ Comparing coefficients, we get $\sum_{d\mid n} db_d = 2^n$ for all positive integers $n$ . In particular, as in Solution 1, we get $\begin{array}{ll} b_1 + 2b_2 + 4b_4 + 8b_8 + 16b_{16} + 32b_{32} &= 2^{32}, \\ b_1 + 2b_2 + 4b_4 + 8b_8 + 16b_{16}\phantom{ + 32b_{32}} &= 2^{16}, \end{array}$ from which the answer $b_{32} = 2^{27} - 2^{11}$ follows.

Solution 3

From the starting point of Solution 2, $(1 - z)^{b_1}(1 - z^2)^{b_2}(1 - z^3)^{b_3}\cdots = 1 - 2z,$ taking reciprocals and expanding with geometric series gives us $\prod_{n = 1}^{\infty}\left(\sum_{k = 0}^{\infty} z^{kn}\right)^{b_n} = \sum_{n = 0}^{\infty} 2^nz^n.$ On the right, we have the generating function for the number of monic polynomials of degree $n$ over the field $\mathbb{F}_2$ of two elements, and on the left, we have the factorisation of this generating function that considers the breakdown of any given monic polynomial into monic irreducible factors. As such, we have the interpretation $b_n = \text{number of monic irreducible polynomials of degree }n\text{ over }\mathbb{F}_2.$ From here, to determine $b_n$ , we analyse the elements of $\mathbb{F}_{2^n}$ , of which there are $2^{n}$ in total. Given $\alpha\in\mathbb{F}_{2^n}$ , if the minimal polynomial $f_{\alpha}$ of $\alpha$ has degree $d$ , then $d\mid n$ and all other roots of $f_{\alpha}$ appear in $\mathbb{F}_{2^n}$ . Moreover, if $d\mid n$ and $f$ is an irreducible polynomial of degree $d$ , then all roots of $f$ appear in $\mathbb{F}_{2^n}$ . (These statements are all well-known in the theory of finite fields.) As such, for each $d\mid n$ , there are precisely $db_d$ elements of $\mathbb{F}_{2^n}$ of degree $d$ , and we obtain the same equation as in Solution 2, $\sum_{d\mid n} db_d = 2^n.$ The rest is as before.

@@ Line 38: / Line 38: @@
 By a limiting process, we can extend the problem to that of finding a sequence <math>b_1, b_2, \ldots</math> of positive integers such that
 <cmath> (1 - z)^{b_1}(1 - z^2)^{b_2}(1 - z^3)^{b_3}\cdots = 1 - 2z. </cmath>
-(We could make the limiting process more formal by invoking the "usual" (<math>(z)</math>-adic) topology on the space of formal power series in <math>z</math>, or if we want to avoid the high-powered stuff, we could also stick to the finite situation in the problem and suppress large powers with big-oh notation. We'll use the power series approach here in anticipation of Solution 3 (and because power series appear anyway).)
+(We could make the limiting process more formal by invoking the "usual" (<math>(z)</math>-adic) topology on the space of formal power series in <math>z</math>, or if we want to avoid the high-powered stuff, we could also stick to the finite situation in the problem and suppress large powers with big-oh notation, then work in a suitably small ball around the origin to ensure convergence of geometric series.)
 If we take logarithmic derivatives on both sides, we get
@@ Line 54: / Line 54: @@
 \end{array}</cmath>
 from which the answer <math>b_{32} = 2^{27} - 2^{11}</math> follows.
 ===Solution 3===

1988 USAMO (Problems • Resources)
Preceded by Problem 4	Followed by Last Question
1 • 2 • 3 • 4 • 5
All USAMO Problems and Solutions

Page

Toolbox

Search