Let $P(x)\in \mathbb{R}[x]$. If $P(x)=n$ has at least one rational zero for $\forall n \in \mathbb{N}$, show that $P(x)=ax+b$ where $a$ and $b$ are rational.

I totally do not know how I should solve this.

When I set $g(x)=P(x)-n$, the derivation is $g'(x)=P'(x)$. ... That's what I found. In addition, how can I use the condition that $P(x)=n$ has a rational zero? Could you please give me some key points to the problem I'm suffering from? Thanks.

  • $\begingroup$ $g(\xi)=0$ does not imply $g'(\xi)=0$. $\endgroup$ – Gae. S. Feb 9 at 14:12
  • 1
    $\begingroup$ Looks interesting. Where is that problem from ? Do you know for sure that it has a simple solution ? It would also help to know a little about your level of knowledge. $\endgroup$ – Ewan Delanoy Feb 9 at 14:15
  • $\begingroup$ @Gae.S. I corrected that part. Thanks for your advice! $\endgroup$ – ToBY Feb 9 at 14:16
  • $\begingroup$ @EwanDelanoy Some Iranian competition? $\endgroup$ – Aqua Feb 9 at 14:22

Let $P(x)=\alpha_nx^n+\ldots +\alpha_0$ with $\alpha_i\in\Bbb R$ and $\alpha_n\ne 0$. For $k\in\Bbb N$, let $q_k$ be a rational root of $P(x)-k$. Then the $\deg P+1$ points $(q_k,k)$, $k=1,2,\ldots, \deg P+1$ determine $P$ uniquely and we can find the coefficients of $P$ from them by solving a system of linear equations with all coefficients rational. It follows that all $\alpha_i$ are rational. Then for suitable $M\in\Bbb N$, all $a_i:=M\alpha_i$ are integer. By the rational root theorem, any rational root $q_k$ of $MP(x)-kM $, i.e., of $$a_nx^2n+a_{n-1}x^{n-1}+\ldots +(a_0-kM) $$ (at least when $kM\ne a_0$) is of the form $q_k=\frac {u_k}{v_k}$ where $u_k\mid a_0-kM$ and $v_k\mid a_n$. In particular, $r_k:=a_nq_k\in \Bbb Z$ for all $k$. (If $kM=a_0$, we can divide out a power of $x$ and at least still obtain $q_k\in\frac1{a_n}\Bbb Z$, so still $r_k\in\Bbb Z$). So with $Q(x):=MP(a_n x)\in\Bbb Z[x]$ we are given that $Q(x)-kM$ has an integer root $r_k$ for all $k\in\Bbb N$.

By the Mean Value Theorem, there exists $\xi_k$ between $r_k$ and $r_{k+1}$ with $$ \left|Q'(\xi_k)\right|=\left|\frac{Q(r_{k+1})-Q(r_k)}{r_{k+1}-r_k}\right|=\frac{M}{|r_{k+1}-r_k|}\le M.$$ Similarly, there exists $\eta_k$ between $r_k$ and $r_{k+2}$ with $$ \left|Q'(\eta_k)\right|=\left|\frac{Q(r_{k+2})-Q(r_k)}{r_{k+2}-r_k}\right|=\frac{2M}{|r_{k+2}-r_k|}\le 2M.$$

Assume $Q'$ is not constant. Then there exists $L$ such that $|Q(x)|>2M$ for all $x$ with $|x|>L$. Therefore $|\xi_k|\le L$ and $|\eta_k|\le L$ for all $k$. Since the $r_k$ are all distinct and only finitely many integers are between $-L$ and $L$, there exists $k$ such that $|r_k|, |r_{k+1}|, |r_{k+2}|$ are all $>L$. As among $r_k,r_{k+1},r_{k+2}$, two numbers must have the same sign, it follows that one of $|\xi_k|, |\xi_{k+1}\|, |\eta_k|$ is $>L$, contradiction. We conclude that $Q'$ is constant.

Then $Q$, as well as $P$ is at most linear. It can clearly not be constant, hence $$ P(x)=ax+b$$ with $a,b\in\Bbb Q$.


WLOG, suppose that the leading coefficient of $P$ is positive. By the Lagrange interpolation formula, $P$ must have rational coefficients.

Let $Q(x) = m P(x)$, where $m$ is a positive integer chosen so that $Q$ has integer coefficients. Let the leading coefficient of $Q$ be $c$, and let its constant term be $d$. From the given hypothesis, for all sufficiently large primes $p$, we can find some positive rational number $r$ such that $Q(r) = p + d$.

By the rational root theorem, all rational roots of the polynomial $Q(x) - (p-d)$ must be of the form $\dfrac{\pm p}{s}$, where $s$ is a divisor of $c$. Hence, $r \geq \dfrac{p}{c}$. On the other hand, if $\deg Q > 1$, then

$$0 = Q(r) - (p-d) \geq Q(\frac{p}{c}) - (p-d) \geq O(p^2),$$

which is clearly impossible. Hence, the degree of $P$ must be less than or equal to 1.

  • $\begingroup$ How can you suppose that the leading coefficient is positive? Thank you $\endgroup$ – Giuseppe Negro Feb 9 at 18:52
  • $\begingroup$ @GiuseppeNegro, if it's not positive, you can let $Q(x)=-mP(x)$. $\endgroup$ – LHF Feb 9 at 18:54
  • $\begingroup$ And will $Q$ satisfy that $Q(x)=n$ has a rational zero for all $n\in\mathbb N$? Is this obvious? Sorry if it is a silly question $\endgroup$ – Giuseppe Negro Feb 9 at 18:56
  • $\begingroup$ @GiuseppeNegro, It's very good you ask :) $Q(x)$ does not satisfy $Q(x) = n$, that's $P(x)$. $Q(x)$ is the 'common-denominator' version of $P(x)$ with a positive leading coefficient. $\endgroup$ – LHF Feb 9 at 19:09

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.