22
$\begingroup$

Let $$f(x)=(x-1)^2(x-2)^2(x-3)^2\cdots(x-2013)^2+2014\tag{1}$$

Prove or disprove: $f(x)$ is reducible on the field of rational numbers $Q$.

my try: since I know this $$g(x)=(x-1)(x-2)(x-3)\cdots(x-n)+1 \qquad(n\in N^{+},n\neq 4)$$is reducible on the field of rational numbers $\mathbb Q$

because when $n=4$ we have $$(x-1)(x-2)(x-3)(x-4)+1=(x^2-5x+4)(x^2-5x+6)+1=(x^2-5x+5)^2$$

this simaler problem: $[(x-a_1)(x-a_2) \cdots (x-a_n)]^2 +1$ is irreducible over $\mathbb Q$

$(1)$ problem is from Beijing university mathematics postgraduate entrance exams today,and this constant is $2014$ follow I can't works,

$\endgroup$
6
  • 1
    $\begingroup$ $f$ seems to be symmetric around $x = 1007$. $\endgroup$
    – TMM
    Jan 5, 2014 at 17:37
  • 1
    $\begingroup$ @ziangchen,yes. $\endgroup$
    – user94270
    Jan 5, 2014 at 17:39
  • 3
    $\begingroup$ math.stackexchange.com/questions/317622/x-a-1x-a-2-cdots-x-a-n2-1-is-irreducible-over-mathbb-q?rq=1 $\endgroup$
    – medicu
    Jan 5, 2014 at 19:29
  • 2
    $\begingroup$ @ziangchen : see this $\endgroup$
    – user87543
    Jan 6, 2014 at 2:16
  • 2
    $\begingroup$ are the Beijing entrance problems available online? $\endgroup$
    – zyx
    Jan 11, 2014 at 5:37

2 Answers 2

14
+50
$\begingroup$

Since $f(x)$ has integer coefficients and the gcd of these coefficients is $1$, by Gauss's lemma it suffices to show that $f(x)$ is irreducible in $\mathbf{Z}[x]$. Suppose that $f(x)=g(x)h(x)$ where $g,h\in\mathbf{Z}[x]$ are nonconstant. For $1\le c\le 2013$, we have $2014=f(c)=g(c)\cdot h(c)$, so both $g(c)$ and $h(c)$ are (positive or negative) divisors of $2014$. Since $2014=2\cdot 19\cdot 53$, we see that $2014$ is divisible by exactly $16$ integers, namely $\pm 2^i 19^j 53^k$ with $i,j,k\in\{0,1\}$. Thus there must be some divisor $d$ of $2014$ which occurs as $g(c)$ for at least $\lceil 2014/16\rceil=126$ values $c\in\{1,2,\dots,2013\}$.

I claim that $g(x)-d$ is divisible by $(x-1)(x-2)\dots (x-2013)$ in $\mathbf{Z}[x]$. To show this, let $S$ be the set of all $c\in\{1,2,\dots,2013\}$ for which $g(c)=d$. Since $p(x):=\prod_{s\in S} (x-s)$ is monic, we can write $g(x)-d=p(x)q(x)+r(x)$ with $q,r\in\mathbf{Z}[x]$ and $\deg(r)<\deg(p)$. But every element of $S$ is a root of both $p(x)$ and $g(x)-d$, and hence is a root of $r(x)$, so the number of roots of $r(x)$ is at least $\#S=\deg(p)$, and hence is bigger than $\deg(r)$, so we must have $r(x)=0$. Now suppose that $S\ne\{1,2,\dots,2013\}$, and let $c$ be an integer with $1\le c\le 2013$ and $c\notin S$. Since $g(c)-d=p(c)q(c)$ where $d$ and $g(c)$ are distinct divisors of $2014$ (and where $p(c)$ and $q(c)$ are integers), it follows that $\lvert p(c)\rvert\le 4018$. But $p(c)=\prod_{s\in S}(c-s)$ is the product of at least $126$ distinct nonzero integers, hence has absolute value at least $(63!)^2$, which is bigger than $4018$, a contradiction. Therefore $S=\{1,2,\dots,2013\}$, so that $p(x)=(x-1)(x-2)\dots(x-2013)$ and $q(x)\in\mathbf{Z}[x]$ satisfy $g(x)-d=p(x)q(x)$.

The same argument shows that there is a divisor $D$ of $2014$ for which $h(x)-D=p(x)Q(x)$ with $Q(x)\in\mathbf{Z}[x]$ and $p(x)=(x-1)(x-2)\dots (x-2013)$ as above. In particular, since we assumed that both $g(x)$ and $h(x)$ are nonconstant, it follows that both of them have degree at least $2013$. But their product has degree $4026$, so we must have $\deg(g)=\deg(h)=2013$, whence both $q(x)$ and $Q(x)$ are nonzero constants (which we will call $q$ and $Q$). Hence we have $$ p(x)^2 + 2014 = f(x) = g(x)\cdot h(x) = (d+p(x)\cdot q)\cdot (D+p(x)\cdot Q). $$ It follows that $y^2+2014=(d+yq)(D+yQ)$ (since this is an equality of degree-$2$ polynomials which is valid for more than two values $y$, namely for any value $y=p(x)$). But this is impossible, since $\sqrt{-2014}$ is irrational. This contradiction shows that $f(x)$ is irreducible in $\mathbf{Z}[x]$, and hence (by Gauss's lemma) in $\mathbf{Q}[x]$.

$\endgroup$
0
6
$\begingroup$

As Michael Zieve said my previous answer had a wrong quote, which I am terribly sorry for. However, the following is true: if $ax^2+bx+c$ is irreducible $a,b,c \in \mathbb{Z}$ then $Q(x)=aP(x)^2 + bP(x) + c$ is irreducible where $P(x)=(x-a_1)...(x-a_n), n>2\tau(c)(2+[\log_2|c|]), a_1, ... , a_n \in \mathbb{Z}$ see theorem 7.1 in K. Gyoury,L.Hajdu and R. Tijdeman ( http://www.math.unideb.hu/~hajdul/ght20.pdf‎)

Below is the previous version of the answer. Dorwart and Ore proved the result only for $c=1$.

There is a result by Dorwart and Ore (see http://www.jstor.org/discover/10.2307/1968341?uid=3739232&uid=2&uid=4&sid=21103238485287) stating that if $ax^2+bx+c$ is irreducible $a,b,c \in \mathbb{Z}$ then $Q(x)=aP(x)^2 + bP(x) + c$ is irreducible where $P(x)=(x-a_1)...(x-a_n), n \geq 5, a_1, ... , a_n \in \mathbb{Z}$ This answers both your questions

$\endgroup$
4
  • 2
    $\begingroup$ Nice reference. Do you really need to write an Annals paper to pass the Peking university entrance exam? $\endgroup$
    – Igor Rivin
    Jan 8, 2014 at 18:15
  • 2
    $\begingroup$ I think you misquoted the Dorwart-Ore result. The closest result I can see in their paper is Theorem IV on page 86, which is the same as what you wrote except with the added condition that $c=1$. So this answers the "similar problem", but does not answer the original question from the entrance exam. $\endgroup$ Jan 9, 2014 at 3:38
  • 1
    $\begingroup$ Thanks for editing your answer, and adding the Gyory-Hadju-Tijdeman reference. It looks like my proof is essentially the same as theirs, and the key ideas for both proofs appear already in the Dorwart-Ore paper from 1933. In response to Igor Rivin's comment: many Annals papers from 1933 (including this one) are indeed at the level of entrance exam questions. $\endgroup$ Jan 9, 2014 at 6:47
  • $\begingroup$ Also, I think that the result itslef is more intresting than the proof. General idea that composing something completely reducible with something irreducible must lead to something irreducible is intresting. $\endgroup$
    – user68061
    Jan 9, 2014 at 9:51

You must log in to answer this question.