Assume that $Q$ is a degree $k$ polynomial and $I$ an interval of finite length $c$. Can you please give me some hints on how to show that

$$ \left( \sum_{m=0}^k \frac{c^{m}}{m!} |Q^{(m)} (t_0) | \right)^2 \leq \frac{C(k)^{2c}}{c} \int_{I} Q^{2}(t) dt, $$

for some positive constant $C(k)$ depending only on the degree $k$ and any $t_0 \in I$. The statement is very puzzling and I have had a hard time establishing it. All hints are therefore greatly appreciated.

EDIT: If this is wrong, as it looks like, can it be repaired under some additional assumptions? Thank you.

  • $\begingroup$ Chebyshev polynomials certainly give a minimum possible supremum norm, and I believe I recall that these give rise to nontrivial and optimal lower bounds on integrals of polynomials. I'll search for those. $\endgroup$ – Robert Wolfe Nov 16 '18 at 15:33
  • $\begingroup$ I've found this: ac.els-cdn.com/0022247X85900654/… but I recall a somewhat more recent paper. Can't recall if they were lecture notes or an actual article. $\endgroup$ – Robert Wolfe Nov 16 '18 at 15:50

I think the statement is false.

For $I \subset \mathbb R$, $k\in \mathbb N$ and $C>0$ let $P(I,C,k)$ be the proposition : $$\forall Q\in \mathbb R_k[x],\quad\quad \max_{t_0\in I}\left(\sum_{m=0}^{k} \frac{l(I)^m}{m!}|Q^{(m)}(t_0)|\right)^2 \leq \frac{C^{2l(I)}}{l(I)}\int_I Q^2(t)\mathrm d t$$ where $l(I)$ is the length of $I$.

You can notice that for all $a\in \mathbb R, P(I,k,C)\Leftrightarrow P(I+a,k,C)$. Then, notice that the change of variable $t= \lambda s$ transforms $Q(X)$ into $Q(\lambda X)$ which is of same degree and $(Q(\lambda X))^{(m)}=\lambda^m Q^{(m)}(\lambda X)$. Furthermore, this change of variable leaves invariant the integral over $l(I)$ on the right hand side. Therefore, $$\forall \lambda,C>0, \forall k\in \mathbb N, \quad\quad P(I,C,k) \Leftrightarrow P(\lambda I, C^{1/\lambda},k).$$ If the statement is true for some constant $k$ and some constant $C(k)$ then $C$ can be chosen equal to $1$ : $$\forall k\in \mathbb N, \exists C>0, \forall I\subset \mathbb R, P(I,C,k) $$ $$\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad \Rightarrow \forall k\in \mathbb N, \forall I\subset \mathbb R, P(I,1,k).$$

We thus showed : $$\forall k\in \mathbb N, \exists C>0, \forall I\subset \mathbb R, P(I,C,k) \quad \quad\Leftrightarrow \quad\quad \forall k\in \mathbb N, P([0,1],1,k) $$

Finally, lets test this on $Q=X^k$, we have : $$ \max_{t_0\in [0,1]} \left(\sum_{m=0}^k \frac{k!}{(k-m)!m!} t_0^{k-m}\right)^2 = 2^k$$ and $$ \int_{0}^1 X^k = \frac{1}{k+1}$$ thus contradicting $P([0,1],1,k)$ for $k\geq 1$.

  • $\begingroup$ You make a good point. Thanks for your input. $\endgroup$ – JohnK Nov 7 '18 at 7:00

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.