3

Consider a real polynomial of the form $p(x) = ax^4 + bx^2+c$. We are given that $0\leq p(x)\leq 1$ in the interval $[0,1]$. Show that $a\leq 4$


Take $x^2=y$. So $p(x)=q(y)$We assumes $a>0$. Now we consider two cases:

  1. $x$ such that $q'(x)=0$ lies outside $[0,1]$ and
  2. $x$ such that $q'(x)=0$ lies inside $[0,1]$. In the first case, notice that the extreme values of $q(x)$ in that interval must lie on the endpoints. Say in the interval $q'(x)>0$ then $a+b+c\leq 1$ and $c\geq 0$. I'm stuck after here and made no reasonable progress after this. Please help
Yoda_2008
  • 75
  • 5
  • You didn't write this, but I'm guessing you're given that $0 \leq p(x) \leq 1$ in the interval $[0, 1]$. As written, the question doesn't make sense because no non-constant polynomial is bounded across $\Bbb R$. Please edit your question to clarify. – Robert Shore Jan 04 '24 at 18:34
  • The question is incorrect as it is stated right now. Maybe $0 \leq p(x) \leq 1$ between some specific values of $x$? – Actually Fritz Jan 04 '24 at 18:34
  • I don't think what you are given is possible. An even powered polynomial may be bounded below (if $a > 0$) or bounded above (if $a < 0$) but not restricted between to values. [The statement I've just stated needs to be proven but a thourough course would have proven it]. Are you stating this right? $0 \le p(x) \le 1$? For which values of $x$? – fleablood Jan 04 '24 at 18:34
  • .... Oh. I suspect you are saying "If $x \in [0,1]$ then $0 \le p(x) \le 1$" and need to prove that this requires $a \le 4$. ... That seems reasonable. But if that is what the question states, then please stat the question correctly. – fleablood Jan 04 '24 at 18:37
  • I'm very sorry for my carelessness while preparing the post. I have made the edit. Thanks for pointing it out – Yoda_2008 Jan 04 '24 at 18:41
  • Please, use descriptive titles. "Prove that $a\le 4$" says nothing about the subject of the question. – jjagmath Jan 04 '24 at 18:54
  • @jjagmath edited a new title, hopefully it's ok now – Yoda_2008 Jan 04 '24 at 18:58
  • That's better but by replacing $p(x)$ with $a x^4 +bx^2+c$ the title will contain the whole description of the question, which would be even better. – jjagmath Jan 04 '24 at 19:02
  • Thank you for editing @Yoda_2008. Interesting exercise. +1 – Mike Jan 04 '24 at 19:40

3 Answers3

2

Since $\,0\leqslant ax^4+bx^2+c\leqslant1\;\;$ for any $\;x\in[0,1]\,,\,$ in particular, for $\,x=0\,,\,$ $x=\dfrac1{\sqrt2}\,,\,$ $x=1\,,\,$ it follows that

$\begin{cases}0\leqslant c\leqslant1\\[3pt]0\leqslant\dfrac14a+\dfrac12b+c\leqslant1\\[3pt]0\leqslant a+b+c\leqslant1\end{cases}$

consequently ,

$\begin{cases}0\leqslant\dfrac12c\leqslant\dfrac12\\[3pt]-1\leqslant-\dfrac14a-\dfrac12b-c\leqslant0\\[3pt]0\leqslant\dfrac12a+\dfrac12b+\dfrac12c\leqslant\dfrac12\end{cases}$

and, by adding the previous inequalities, we get that

$-1\leqslant\dfrac14a\leqslant1\;\;,$

hence ,

$-4\leqslant a\leqslant4\;.$

Angelo
  • 12,328
  • 3
    Hello, nice solution. May I know what was the motivation of taking $x=1/ \sqrt{2}$ – Yoda_2008 Jan 04 '24 at 19:25
  • @Yoda_2008, it permits to get fractions containing small whole numbers and consequently we can get a better constraint for $a$. – Angelo Jan 04 '24 at 19:34
2

One can apply $$ \frac{f(x+h)-2f(x)+f(x-h)}{h^2}=f''(\zeta) $$ (see for example here) to the function $f(x) = ax^2+bx + c$ with $x=1/2$ and $h=1/2$: $$ 2a = f''(\zeta) = \frac{f(1)-2f(1/2)+f(0)}{(1/2)^2} \, . $$ The expression on the right is in the range $[-8, 8]$ and therefore $-4 \le a \le 4$.

The bound is sharp, equality holds for $f(x) = 4(x-1/2)^2$, i.e. for $q(x) = 4(x^2-1/2)^2 = 4x^4 - 4x^2 + 1$.

Martin R
  • 113,040
0

Another solution:

Suppose for contradiction that $a > 4$. The boundary conditions also tell us that $0\le c\le 1$.

Note that for $q(x) = p(\sqrt x)$ to be nonnegative, it must be the case that $b^2 \le 4ac$ (the discriminant) and so $b \ge -2\sqrt{ac}$.

Then, we have by the boundary condition that

$$1 \ge a + b + c \ge a - 2\sqrt{ac} + c = (\sqrt a - \sqrt c)^2.$$

But, if $a > 4$ then $\sqrt a > 2$ while $0\le \sqrt c \le 1$. This would imply that $(\sqrt a - \sqrt c)^2 > (\sqrt a - 1)^2 > 1$, a contradiction.


Edit: didn't notice the constrained nonnegativity constraint. To fix the argument, note that if $a > 4$, then $b < 0$. Then, either $b^2 \le 4ac$ (as above) or the vertex $-\frac{b}{2a} > 1$, or equivalently $b < -2a$. In this latter case, $a + b + c < c - a < -3 < 0$ which is also a contradiction.

  • 2
    The polynomial doesn't need to be non negative in the whole domain, just in $[0,1]$ – Yoda_2008 Jan 04 '24 at 19:36
  • Edited to address your comment! – Misha Ivkov Jan 04 '24 at 19:51
  • @Angelo I mentioned that $b < 0$ in my edit above -- I think that addresses your comment. The quadratic $ax^2 + bx + c$ is minimized at $x = -\frac{b}{2a}$, so if this vertex lies within $[0, 1]$ we must have $b^2 \le 4ac$ and if this vertex does not it must be that $b < -2a$ (since $b < 0$). Also in my first line I assumed by contradiction that $a > 4$. – Misha Ivkov Jan 04 '24 at 20:06