5

I have managed to write the polynomial mentioned in the title as a sum of squares as follows:

$$ z^4 + z^3 + z^2 +2z +3 = \left( z^2 + \frac12 z - \frac18 \right)^2 + \left( z + \frac{17}{16} \right)^2 + \frac{475}{256} $$

Is this the simplest way to prove that it has no real zero, or is there another way that is simpler?

Update. The first comment below, written by Piquito, seems interesting, but I don't understand why the inequality would be true for all real z. Can someone explain it?

Gary
  • 31,845
A A
  • 569

8 Answers8

4

Remove the first two terms with $\left(z^2+\frac12z\right)^2$: $$ z^4+z^3+z^2+2z+3-\left(z^2+\frac12z\right)^2=\frac34z^2+2z+3\tag1 $$ Remove the first two terms of $\frac34z^2+2z+3$ with $\frac34\left(z+\frac43\right)^2$: $$ \frac34z^2+2z+3-\frac34\left(z+\frac43\right)^2=\frac53\tag2 $$ Thus, we get $$ z^4+z^3+z^2+2z+3=\left(z^2+\frac12z\right)^2+\frac34\left(z+\frac43\right)^2+\frac53\tag3 $$

robjohn
  • 345,667
  • Elegant way of writing this particular polynomial as a sum of squares, but what happens if you try to do something similar with a polynomial such as z^4 - 2z^3 + 16? If you remove the first two terms with (z^2 - z)^2 then you are left with -z^2 + 16, and what to do from there? How to get rid of the -z^2 term? – A A Jul 01 '23 at 00:11
  • Sometimes, we can't only use binomials: $z^4-2z^3+16=\left(z^2-z-1\right)^2+(z-1)^2+14$ – robjohn Jul 02 '23 at 06:38
3

Rewrite the polynomial as follows :

$$ \begin{align}P(z):=\left(z^4+z^3+z+1\right)+\left(z^2+z+2\right)\end{align} $$

Then, you have :

$$ \begin{align}z^3\left(z+1\right)+(z+1)+\left(z^2+z+2\right)\end{align} $$

or

$$ \begin{align}(z+1)\left(z^3+1\right)+(z^2+z+2)\end{align} $$

Use the sum of cubes formula $z^3+1=(z+1)(z^2-z+1)$, which leads to :

$$ \begin{align}(z+1)^2\underbrace{\left(z^2-z+1\right)}_{\color{#c00}{\Delta_z<0}}+\underbrace {\left(z^2+z+2\right)}_{\color{#c00}{\Delta_z<0}}>0\end{align} $$

for all $z\in\mathbb R$ . Therefore, $P(z)>0,\thinspace \thinspace \forall z \in\mathbb R\thinspace .$

This completes the answer .


Equivalently if you wish, convert both $\color{#c00}{\Delta_z}$ to completing the square, then you reach the following sum of squares :

$$ \begin{align}\left(z+1\right)^2\left(z-\frac 12\right)^2+\frac 34\left(z+1\right)^2+\left(z+\frac 12\right)^2+\frac 74\end{align} $$

which is equivalent to :

$$ \begin{align}\bbox[5px,border:2px solid #C0A000]{\left(z^2+\frac 12z-\frac 12\right)^2+\frac {1}{28}\left(7z+5\right)^2+\frac {13}{7}\thinspace .}\end{align} $$

lone student
  • 14,709
3

Here is a completely elementary proof (without calculus, complicated calculations or tricky factorizations):

  • For $z\geq 0$, we have $$z^4 + z^3 + z^2 +2z +3\geq 3.$$ Therefore, there is no real zero in the interval $[0,\infty)$.

  • For $-1\leq z\leq 0$, we have (multiplying by $z^2$) $-z^2\leq z^3$. Then, $$z^4 + z^3 + z^2 +2z +3\geq z^4 + (-z^2) + z^2 +2(-1) +3=z^4+1>1.$$ Therefore, there is no real zero in the interval $[-1,0]$.

  • For $z\leq -1$, we have (multiplying by $z^3$) $z^4\geq-z^3$. Then, $$z^4 + z^3 + z^2 +2z +3\geq(-z^3) + z^3 + z^2 +2z +3= z^2 +2z +3>0.$$ Therefore, there is no real zero in the interval $(-\infty,-1]$.

Pedro
  • 18,817
  • 7
  • 65
  • 127
  • Very nice proof, but maybe the last, strict inequality is not obvious on an elementary level. – A A Jun 30 '23 at 03:22
  • @AA Well, it requires the knowledge of the graph of quadratic function (in this case, concave upward due to positive leading coefficient, without root due to negative discriminant). – Pedro Jun 30 '23 at 04:17
  • @AA Or we can use these estimates (sum of positive terms is positive):

    $z²+2z+3=z(z+1)+(z+3)>0$ for $z$ in $[-3,-1]$;

    $z²+2z+3=z(z+2)+3>0$ for $z$ in $(-\infty,-2]$.

    Thus $>0$ in $(-\infty,-1]$.

    But this is a kind of tricky factorizarion, which I tried to avoid.

    – Pedro Jun 30 '23 at 04:56
2

I'll leave this answer here, if it works for you .


Observe that :

$$ \begin{align}&z^4+z^3+z^2+2z+3\\ =\thinspace\thinspace\thinspace &\color{red}{z^2}(z^2+z+1)+2\color{red}{z}+3\end{align} $$

Then you have :

$$ \begin{align}\Delta_{\color{red}{z}}&=1-3(z^2+z+1)\\ &=-\underbrace {(3\color{blue}{z^2}+3\color{blue}{z}+2)}_{\Delta_{\color{blue}{z}}\thinspace<\thinspace0}\\ &<0\thinspace\thinspace,\forall z\in\mathbb R\thinspace .\end{align} $$

which completes the answer .


$\rm {Explanation:}$

We rewrote the polynomial as a quadratic polynomial respect to the $\color{red}{z}$ . Next, we looked at the general discriminant of the equation with respect to $\color{red}{z}$ . We have shown that, this discriminant is always negative . Thus, you're done .

lone student
  • 14,709
  • Please, explain the downvote, if possible . Because, I am ready to expand any step . Thank you . – lone student Jun 30 '23 at 03:13
  • 1
    (+1) This seems like an odd way to use the discriminant, with the independent variable in the coefficients. However, if there is a root, $z_0$, then definitely $z^2(z_0^2+z_0+1)+2z+3$ would also have $z_0$ as a root. So this method works. – robjohn Jun 30 '23 at 08:04
  • @robjohn Thank you for your time, indeed. I was very curious to know if the method would work. From here we can complete the square with respect to $\color{red}{z}$, am I right? I guess this is a something cleaner ? – lone student Jun 30 '23 at 08:16
  • @robjohn Do you think, is it worth to add these further details to above answer ? – lone student Jul 02 '23 at 01:26
  • The explanation provided is questionable. The polynomial is not rewritten as a quadratic polynomial in $z$, it is reorganized to look like a quadratic polynomial (using colors). Note that when I commented, I used a root $z_0$, different from $z$, to give an actual quadratic in $z$. Then we can derive the result you were after. I have trouble following your completion of squares for the same reason. It would probably be better to write things as an actual quadratic. – robjohn Jul 02 '23 at 03:11
  • @robjohn Thanks for your feedback. Good point , indeed. I need to note that "..it is reorganized to look like a quadratic polynomial (using colors)" .To completely understand the last part of your comment, may I ask if you agree with the following equality/identity? If we omit the term "Completing the square" in the terminological sense from the answer? $$ \begin{align}P(z):=\underbrace{\left(z^2+z+1\right)}{\Delta_z<0}\left(z-\frac {1}{z^2+z+1}\right)^2+\frac{\overbrace{3z^2+3z+2}^{\Delta_z<0}}{\underbrace{\color{black}{z}^2+{z}+1}{\Delta_z<0}}\end{align} $$ Do you agree above identity ? – lone student Jul 02 '23 at 03:47
1

It is not difficult to see that $z^3(z+1)=z^2z(z+1)\geq -1$ for all real numbers $z$. So, $$f(z)=z^4+z^3+z^2+2z+3=z^3(z+1)+(z+1)^2+2\geq -1+0+2=1.$$

Bob Dobbs
  • 10,988
0

A solution(maybe more difficult but easier to think)

Let $f(z)=z^4+z^3+z^2+2z+3$, then $f'(z)=4z^3+3z^2+2z+2$.

It is easily to see that $f'(z)$ have only one real zero.

Let it be $z_0$, so $4z_0^3+3z_0^2+2z_0+2=0$.

And $f'(0)=2>0$, so $z_0<0$

(This is because f is a monotonically increasing function)

And $f(z)\geq f(z_0)=z_0^4+z_0^3+z_0^2+2z_0+3=z_0^4+z_0^3+z_0^2+1-(4z_0^3+3z_0^2)=z_0^4-3z_0^2-2z_0^2+1=z_0^2(z_0^2-3z_0+2)+1=z_0^2(z_0-1)(z_0-2)+1>1$

Done!

Itoz Darbien
  • 466
  • 16
0

Alternative approach:

Let $P(x)$ a polynomial of even degree, (simple roots,) positive leading coefficient, such that its derivative $P'(x)$ has exactly one real root $\xi$ ( for instance, if $P''(x) \ge 0$ for all $x \in \mathbb{R}$). The minimum value of $P$ is $P(\xi)$, and, since the other roots are conjugate in pairs, it will have the sign of

$$\prod_{x_k, P'(x_k) = 0} P(x_k)$$

that is, the resultant of $P$, and $P'$, or, using the discriminant

$$(-1)^{\binom{n}{2}} \operatorname{Discr}(P(x))$$

Therefore, if the above is $>0$ then $P(x)>0$ on $\mathbb{R}$ ( this is a generalization of the case $P(x)$ of degree $2$).

In our case $n=4$, $P''(x)= 12 x^2 + 6 x + 2 >0$ on $\mathbb{R}$, $\binom{4}{2} = 6$ even, and the discriminant $$\operatorname{Discr}(P(x)) = 4345>0$$ so $P(x)> 0$ on $\mathbb{R}$.

orangeskid
  • 53,909
-1

First note $x=0$ is not a root. The idea is to split it into two parts and control the $\Delta=b^2-4ac<0$ for each part. For example,

$$f(z)=z^4 + \color{red}1\cdot z^3 + \alpha z^2+\beta z^2 +\color{blue}2\cdot z +3$$

where $\alpha+\beta=1$ and the coefficients of odd terms, which are $\color{red}1$ and $\color{blue}2$ will serve as the $b$ in our $\Delta$ for each part, namely $$\color{red}1^2-4\alpha<0~\land~\color{blue}2^2-12\beta<0~\land~\alpha+\beta=1$$ hence

$$\boxed{\alpha>\frac14~\land~\beta>\frac13~\land~\alpha+\beta=1}$$

So we have plenty of space to choose $\alpha$ and $\beta$, for example, $\alpha=\beta=\frac12$, then we have

$$\begin{align} f(z)&=z^4 + \color{red}1\cdot z^3 + \frac12z^2+\frac12z^2 +\color{blue}2\cdot z +3\\ \\ &=z^2\underbrace{\left(z^2+\color{red}1\cdot z+\frac12\right)}_{\Delta=\color{red}1^2-4\cdot\frac12<0}+\underbrace{\left(\frac12z^2 +\color{blue}2\cdot z +3\right)}_{\Delta=\color{blue}2^2-4\cdot\frac12\cdot3<0} \end{align}$$

Therefore, $f(z)>0, \forall z\in \mathbb R$

MathFail
  • 21,128