Let $P(x)$ be a polynomial with degree $\leq n$ and $|P(x)|\leq\frac{1}{\sqrt{x}}$ for $x\in(0,1]$. Prove that $|P(0)|\leq 2n+1$. The idea should be that if $|P(0)|$ is too large, then the polynomial cannot change values fast enough to avoid intersecting the curve $ \pm 1/\sqrt{x}$, but I don't see how to formalize this.

  • $\begingroup$ Perhaps you could say that because $P(x)$ is continuous? $\endgroup$ – Chad Shin Feb 1 '16 at 7:04
  • 4
    $\begingroup$ @ChadShin Continuity is not enough. One can easily draw a continuous function $f$ on $[0,1]$ so that $|f(x)|\le \frac{1}{\sqrt x}$ and $f(0)$ is arbitrarily large. Indeed, one can also choose a polynomial to do that, but the degree might be very large. $\endgroup$ – user99914 Feb 1 '16 at 7:26
  • 1
    $\begingroup$ What's the source of this problem, please? $\endgroup$ – Gerry Myerson Feb 4 '16 at 10:30
  • 9
    $\begingroup$ komal.hu/verseny/feladat.cgi?a=feladat&f=A654&l=hu $\endgroup$ – user141614 Feb 4 '16 at 14:10
  • $\begingroup$ It would be great if someone can translate the solution in the above link and post as an answer below. $\endgroup$ – user99914 Feb 6 '16 at 3:09

Since $-\frac{1}{\sqrt{x}}\le P(x)\le\frac{1}{\sqrt{x}}$ for $0<x\le 1$, $\;\;\;-1\le \sqrt{x}P(x)\le1\;$ for $0\le x\le1$.

If we let $t=\sqrt{x}$ and $h(t)=tP(t^2)$, then $h$ is a polynomial of degree $2n+1$ with $\big|h(t)\big|\le1$ for $|t|\le1$.

By Bernstein's inequality for polynomials, $\;\;\displaystyle\big|h^{\prime}(t)\big|\le\frac{2n+1}{\sqrt{1-t^2}}\;$ for $|t|<1$; so $\;\big|P(0)\big|=\big|h^{\prime}(0)\big|\le 2n+1$.

(See (4.2) in http://www.damtp.cam.ac.uk/user/na/people/Alexei/papers/markov.pdf or

Th. 2.8 in http://www.emis.de/journals/HOA/JIA/Volume3_4/156027.pdf)


What follows is more or less the translation of the solution given here.

Part A. First, we show that a polynomial $F(x)$ exists for which the conditions are satisfied and equality in the estimate holds, i.e. $F(0)=2n+1$. Let us denote by $U_{2n}$ the $2n$-th Chebyshev polynomial of the second kind: $$U_{2n}(\cos t)=\frac{\sin(2n + 1)}{\sin t}.$$ This polynomial has degree $2n$. It is known that $U_{2n}(x)$ is even: the reason is that it contains only terms having even degree. This fact can be obtained from the recursive formula for the Chebyshev polynomial of the second kind $U_n(\cos t)=\frac{\sin(n+1)t}{\sin t}$, which is $U_{n+1}(x) = 2x U_n(x)-U_{n-1}(x)$, starting from $U_0(x)=1$, $U_1(x)=2x$.

So we can write $U_{2n}(x)=u(x^2)$ for some polynomial $u$. Now let $$F(x) = u(1-x) = U_{2n}(\sqrt{1-x})$$ and write any $x\in (0,1]$ as $x=\sin^2 t$ (with $t\in(0,\frac{\pi}{2}]$). We get $$F(x) = u(1-\sin^2 t) = u(\cos^2 t) = U_{2n}(\cos t) = \frac{\sin (2n + 1)t}{\sin t} = \frac{\sin (2n + 1)t}{\sqrt{x}}.$$ So for $x\in(0,1]$ we have $|F(x)|\le\frac{1}{\sqrt{x}}$, with equality occurring when $\sin (2n + 1)t = \pm 1$, i.e. at $x_k:=\sin^2\frac{(k + \frac{1}{2})\pi}{2n+1}$ (for $k = 0,1,\dots,n$).
Moreover, $F(0)$ can be found by taking the limit as $t\to 0$: $$F(0) = U_{2n}(1) = \lim_{t\to 0}U_{2n}(\cos t) = \lim_{t\to 0}\frac{\sin (2n + 1)t}{\sin t} = 2n+1.$$

Part B. Now we show the inequality $|P(0)|\le 2n+1$ for a generic polynomial $P$ satisfying the hypotheses. Suppose by contradiction that $|P(0)|>2n+1$; wlog we can assume $P(0)>2n+1$. Choose $c>1$ such that $P(0)>c(2n+1)$ and consider the polynomial $$Q(x) := P(x)-cF(x).$$ The polynomial $Q$ takes alternatively positive and negative values at the points $x_{-1}:=0<x_0<x_1<\dots<x_n$: indeed $Q(x_{-1})=Q(0)>0$ and, if $k$ is even, $Q(x_k)\le\frac{1}{\sqrt{x}}-\frac{c}{\sqrt{x}}<0$; similarly, if $k\ge 1$ is odd, $Q(x_k)>0$. So it possesses at least $n+1$ roots, precisely at least one in each of the intervals $(x_{-1}, x_0), (x_0, x_1), \dots, (x_{n-1}, x_n)$. But this is impossible as $Q$ is not identically zero and has degree $\le n$, and so it cannot have more than $n$ roots.

  • $\begingroup$ did you find the "optimal" polynomial of degree $\le n$ maximizing $P(0)$ and being bounded by $1/x$ on $]0;1]$ ? isn't it of the form $1 + C (1-x)^n$ ? $\endgroup$ – reuns Feb 7 '16 at 2:06
  • $\begingroup$ Yes, but the condition is $|P(x)|\le\frac{1}{\sqrt{x}}$. I think $F(x)$ does not have an explicit formula as the one you wrote. $\endgroup$ – Mizar Feb 7 '16 at 10:59

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.