23
$\begingroup$

Consider the set of polynomials with real coefficients as a vector space with the following inner-product: $\langle f, g \rangle = \int_{a}^{b} f(x)g(x) dx$.

Hilbert showed, in a paper from 1894, that the norm (with respect to this inner-product) of a non-zero polynomial in $\mathbb{Z}[X]$ can get arbitrarily small when $b-a < 4$. In other words, $\min_{0 \neq p \in \mathbb{Z}[X]} \int_{a}^{b} p^2(x) dx = 0$.

My questions are:

  1. What happens when $b-a \ge 4$? Can the norm get arbitrarily small in this case? Or is there some (positive) lower bound for $\min_{0 \neq p \in \mathbb{Z}[X]} \int_{a}^{b} p^2(x) dx$?

  2. When $b-a < 4$, is there an explicit construction of a sequence $p_n \in \mathbb{Z}[X],n\ge 1$, with norm tending to 0?

I posted similar questions in MSE and got no responses.

(Hilbert's proof was as follows: Minimizing the norm for (non-zero) polynomials of degree less than $n$ is equivalent to minimizing a certain positive-definite quadratic form. The corresponding matrix $A_n$ has entries $a_{i,j} = \langle x^i, x^j \rangle = \frac{b^{i+j+1} - a^{i+j+1}}{i+j+1}, 0 \le i,j \le n-1$. A calculation using an orthonormal basis for our inner-product space shows that $\det A_n = (\frac{b-a}{4})^{n^2}n^{-1/4} (2 \pi)^n c_n$ where $c_n$ converges to a positive constant. A result by Minkowski shows that in general, the minimal value of a positive quadratic form $\langle v, Av \rangle$ in $n$ variables is at most $n (\det A)^{1/n}$. Since $\lim_{n} n (\det A_n)^{1/n} = 0$ for $b-a<4$, the result follows.)

$\endgroup$
  • $\begingroup$ This problem might be answered with potential theory methods .note that logarithmic capacity of a segment is a quarter of its length. $\endgroup$ – BigM Mar 20 '14 at 18:24
  • $\begingroup$ Dear @user48438, can you please add some details? Thank you very much. $\endgroup$ – Ricardo Andrade Mar 23 '14 at 20:02
  • $\begingroup$ Honestly I am not aware how to relate the posted problem to potential theory machinery. however I have the feeling that it can be done via potential theory machinery. you can take a look at books.google.com/books/about/… $\endgroup$ – BigM Mar 31 '14 at 20:42
18
$\begingroup$

For Question 1: Once $b-a \geq 4$, the norm cannot get arbitrarily small.

Suppose $p(x)$ has leading term $c_n x^n$. The minimum of $\int_a^b p^2(x) \, dx$ over all polynomials $p(x) = c_n x^n + O(x^{n-1})$ with real coefficients is a multiple of $P_n(l(x))$, where $P_n$ is the $n$-th Legendre polynomial and $l$ is the affine-linear transformation $l(x) = (2x-(a+b))/(b-a)$ that takes $(a,b)$ to $(-1,1)$. Now $P(x/2)$ has leading coefficient $$ \lambda_n := \frac12 \frac34 \frac56 \cdots \frac{2n-1}{2n} \sim \frac1{\sqrt{\pi n}} $$ and norm $\int_{-2}^2 P(x/2)^2 \, dx = 4/(2n+1)$. Thus in general $$ \int_a^b p^2(x) \, dx \geq \frac4{(2n+1)\lambda_n^2} \left(\frac{b-a}{4}\right)^{\!2n+1} c_n^2, $$ with the factor $r_n := 4/(2n+1)\lambda_n^2$ approaching $2\pi$ as $n \rightarrow \infty$. Since $|c_n| \geq 1$, we deduce a lower bound on $\int_a^b p^2(x) \, dx$ that approaches $2\pi$ when $b-a=4$, and increases exponentially with $n$ when $b-a > 4$. [Added later, in connection with the recent MO Question 188807: since $\{ r_n \}_{n \geq 0}$ is an increasing sequence and the $n=0$ bound $b-a$ is attained by the integer polynomial $p(x) = 1$ (or $p(x) = -1$), it also follows that $b-a$ is the minimum over all nonzero $p \in {\bf Z}[X]$ once $b-a \geq 4$.]

It follows also that if $b-a>4$ then for any $M$ there are only finitely many $p \in {\bf Z}[X]$ for which $\int_a^b p^2(x) \, dx < M$. When $b-a = 4$ this is not true when $a,b$ are integers; I do not know what happens otherwise. [For the integer case: it is enough to prove it for $(a,b) = (-2,+2)$. let $T_n$ be the $n$-th Chebyshev polynomial; then $p(x) = 2T_n(x/2)$ has integer coefficients and $$ \int_{-2}^2 p(x)^2 \, dx \leq \int_{-2}^2 2^2 \, dx = 16. $$ (In fact for large $n$ the integral approaches $8$.)]

| cite | improve this answer | |
$\endgroup$

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.