Problem Statement:-

Show that the roots of the polynomial equation $$x^n+a_1x^{n-1}+a_2x^{n-2}+\ldots+a_n=0$$ cannot be all real if $(n-1){a_1}^{2}-2na_2\lt0$

The inequality was wrongly written before it has been changed now.

My attempt:-

Attempt 1:-

First I tried differentiating the the function $f(x)=x^n+a_1x^{n-1}+a_2x^{n-2}+\ldots+a_n$ to get

$$\begin{aligned} f'(x) &= nx^{n-1}+(n-1)a_1x^{n-2}+(n-2)a_2x^{n-3}+\ldots+a_{n-1}\\ &= \dfrac{n}{x}(f(x))-(a_1x^{n-2}+a_2x^{n-3}+\ldots+na_{n-1})-\dfrac{na_n}{x} \end{aligned}$$

I was thinking that if I could prove that the no of distinct roots of $f'(x)$ are less than $\left \lfloor{\dfrac{n}{2}}\right\rfloor$, then all the roots of $f(x)=0$ are not real. But I couldn't figure out how to manipulate $f'(x)$ so as to arrive at the condition that the question asks to prove or for that matter how to prove that the no. of roots of $f'(x)$ are less than $\left \lfloor{\dfrac{n}{2}}\right\rfloor$.

Attempt 2:-

I thought that why not start messing around with the expression that we are needed to prove $\lt0$, i.e. $(n-1){a_1}^{2}-2na_2$.

$a_1=-\sum{\alpha} \qquad\qquad a_2=\sum{\alpha\beta}$

So, $$(n-1){a_1}^{2}-2na_2=(n-1)\left(\sum\alpha\right)^2-2n\left(\sum{\alpha\beta}\right)=n\sum{\alpha^2}-\left(\sum{\alpha}\right)^2$$

I don't know how to arrive at the required condition from the point that I reached in this attempt.

Can you push me in the right direction by giving a hint as to what should be my line of thought while attempting the problem

  • $\begingroup$ Is each $a_i$ given to be real? $\endgroup$ – Dhanvi Sreenivasan Nov 14 '16 at 9:41
  • $\begingroup$ @DhanviSreenivasan - Though it is not given in the statement of the problem but the author has told to consider all the coefficients as real until told otherwise. $\endgroup$ – user350331 Nov 14 '16 at 9:42

Consider $n$ reals $(\alpha_1,\dots,\alpha_n)$. By Cauchy-Schwarz, $$(1.\alpha_1+\dots+1.\alpha_n)^2\le (1^2+\dots+1^2)(\alpha_1^2+\dots+\alpha_n^2)$$ so $$\left(\sum \alpha_i\right)^2\le n\sum \alpha_i^2$$ What you proved with your second try is that for your polynomial, this inequality doesn't stand. So the hypothesis must be wrong, and not all $\alpha_i$ are real.

You were close :-)

| cite | improve this answer | |

The inequality should be in the other direction, i.e. the roots cannot be all real if $(n - 1)a_1^2 - 2na_2 < 0$. For example, the equation $x^2 - 3x + 2$ has the two real solutions $x = 1$ and $x = 2$, but $1 \cdot 3^2 - 2\cdot2\cdot2 = 1 > 0$.

Your second attempt leads to the solution.

Let $\alpha_1, \ldots, \alpha_n$ denote the roots of the polynomial (according to their multiplicity). Then $a_1 = -\sum \limits_{i} \alpha_i$ and $a_2 = \sum \limits_{i < j} \alpha_i \alpha_j$. Now if all $\alpha_i$ are real, then by Jensen's inequality

$$\left( \sum \limits_{i} \alpha_i\right)^2 = n^2 \left( \sum \limits_i\frac{1}{n} \alpha_i\right)^2 \le n^2 \sum \limits_i \frac{1}{n} \alpha_i^2 = n \sum \limits_{i} \alpha_i^2$$ holds, but this means $$(n - 1)a_1^2 - 2n a_2 = n \sum_i \alpha_i^2 - \left(\sum \limits_i \alpha_i\right)^2 \ge 0.$$

| cite | improve this answer | |

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.