Let $P$ be a polynomial, it has $n$ distinct real roots and all these roots are larger than 1. Let $$ Q(x)=(x^2+1)P(x)P'(x)+x[(P(x))^2+(P'(x))^2]. $$
Prove $Q(x)$ has at least $2n-1$ distinct real roots.

I start to find $Q(x)=0$ between two roots. Assume $P(x)<0$ in $(x_1,x_2)$ and $P(x_1)=P(x_2)=0$. There exists a minimal value at $\xi\in(x_1,x_2)$ s.t. $$ P(\xi)<0, P'(\xi)=0, P''(\xi)>0. $$ Then I get $P(\xi)+P'(\xi)<0$.

It can be deduced by contradiction that not all $x\in (x_1,x_2)$ satisfy $P(x)+P'(x)\leq 0$. Hence there exists $\eta$ s.t. $P(\eta)+P'(\eta)>0$, then there exists $\alpha$ s.t. $$P(\alpha)+P'(\alpha)=0.$$ $Q(\alpha)<2xP(\alpha)P'(\alpha)+x[((P(\alpha))^2+(P'(\alpha))^2]\leq 0$, i.e. $$Q(\alpha)<0.$$ If I have $Q(x_1)>0$ and $Q(x_2)>0$ then by mean value theorem there are two roots in $(x_1,x_2)$.

The problems of this thought are:

(1) Even I have two roots in every interval, only $2n-2$ roots be found;

(2) $Q(x_1)>0$ and $Q(x_2)>0$ may be zero and my process would be useless?

Hope someone give me an answer or a hint.

  • $\begingroup$ @TZakrevskiy Thank you! I've updated. $\endgroup$ – yahoo Sep 27 '17 at 12:52
  • $\begingroup$ Does your assumption implicitly imply that $P$ has degree $n$, i.e. all roots are of multiplicitly one? $\endgroup$ – MooS Sep 27 '17 at 12:54
  • $\begingroup$ @MooS I'd doubt tahat. In fact, any multiple root of $P$ is a common root of $P$ and $P'$, hence also a root of $Q$. $\endgroup$ – Hagen von Eitzen Sep 27 '17 at 12:58
  • $\begingroup$ Neither are mentioned in the book. Hence no degree and multipilicity restriction. @MooS $\endgroup$ – yahoo Sep 27 '17 at 12:59
  • 1
    $\begingroup$ if $h(x)=f(x)g(x)$ then $deg[h(x)]=deg[f(x)]+deg[g(x)]$ and if $deg[f(x)]=n $ then $deg[f'(x)]=n-1$ since f is a polynomial? Correct? does this help at all? $\endgroup$ – AmateurMathPirate Sep 27 '17 at 13:29

Ad (1): Note that $P(x)$ and $P'(x)$ have opposite signs for $x\ll 0$, hence $Q(x)<0$ for $x\ll 0$. But at the first real root $x_0$ of $P$, we have $Q(x_0)=x_0P'(x_0)^2\ge0$ because $x_0\ge1$. Thus there is an additional root of $Q$ in $(-\infty,x_0]$.

Ad (2): If $P(x_i)=0$, then $Q(x_i)=x_iP'(x_i)^2\ge 0$. Hence if both $x_i$ are only simple roots, then we obtain two distinct root insinde $(x_1,x_2)$. Whereas if $P(x_i)$ is a $k$-fold root, $k\ge 2$, then $x_i$ is a precisely $(2k-2)$-fold root of $Q$. So first of all, the $Q(x_i)$ count as distinct roots by themselves (but that''s too few). But at least we know that $Q$ has no sign change at $x_i$. Does that help?


I claim that $$H(x) := \frac{Q}{xP(x)P'(x)} = \frac{x^2+1}{x}+\frac{P^2+P'^2}{PP'}$$ has at least $2n-2$ roots in its range of definition. Then in particular $Q$ has at least $2n-2$ roots.

As you have proposed, we fix two roots $x_1 < x_2$ with no more roots in between them and some intermediate value $a$ with $P'(a)=0$ (Rolle's Theorem).

Let us assume $P(a) > 0$, the other case is dealt with the same way.

Then $P'<0$ in $(a,x_2)$.

Note that $|P(a)| > |P'(a)|$ but $|P(x_2)| \leq |P'(x_2)|$, hence at some point in $(a,x_2)$ we have $|P|=|P'|$. At this point we have $|\frac{P^2+P'^2}{PP'}| =2$ and thus $\frac{P^2+P'^2}{PP'}=-2$. We also know $\frac{x^2+1}{x}>2$ in $(a,x_2)$, because $x>1$.

Hence $H>0$ at this point and clearly $\lim\limits_{x \to a, x>a} H = -\infty = \lim\limits_{x \to x_2, x<x_2} H$, thus we obtain two roots of $H$ in $(a,x_2)$ by the intermediate value theorem.

Doing this for any two adjacent roots of $P$, this addds up to $2n-2$ roots, i.e $Q$ has at least $2n-2$ roots.

Here comes the punchline: $Q$ is a polynomial of odd degree, hence the number of real roots is odd. Thus having at least $2n-2$ real roots actually ensures having at least $2n-1$ roots.

  • $\begingroup$ I agree that $\bigg|\frac{P^2+P'^2}{PP'}\bigg| =2$ , but how do you see that $\frac{P^2+P'^2}{PP'}=-2$ and not $+2$ ? $\endgroup$ – Ewan Delanoy Sep 27 '17 at 13:37
  • $\begingroup$ $P$ is positive and $P'$ is negative in $(a,x_2)$, hence the fraction is negative. $\endgroup$ – MooS Sep 27 '17 at 13:38
  • $\begingroup$ I see now, thanks. $\endgroup$ – Ewan Delanoy Sep 27 '17 at 13:40

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.