Let $P_n(t) = p_0 + p_1 t + \cdots + p_n t^n$ be a polynomial (with real coefficients) of degree $n$ in the variable $t$. I am interested in the quantity $$\Phi_n = \min_{\sum_{i=1}^n p_i^2 = 1} \int_0^1 P_n^2(t).$$ In particular: how fast does $\Phi_n$ approach zero as $n \rightarrow \infty$? Can anything be said about which polynomials achieve the minimum, or which achieve something close to the minimal $\Phi_n$?

Note that this can be reformulated as a question about the smallest eigenvalue of a certain matrix.

  • 2
    $\begingroup$ I suggest that you fix your notation, since it is very confusing to have $p_n$ be a coefficient of the polynomial $p_n(t)$. Probably best to define $\|P(t)\|_2$ to be the $L^2$-norm of the vector formed from the coefficients of $P$, then write $\Phi_n$ as $\min_{\deg P=n,\,\|P\|=1}\int_0^1 P(t)^2\,dt$. $\endgroup$ – Joe Silverman Aug 8 '15 at 18:10
  • 10
    $\begingroup$ As you already pointed out yourself, you are equivalently asking about the smallest eigenvalue of the Hilbert matrix $H_{jk}=1/(j+k+1)$. Googling this will produce a lot of information, including asymptotic formulae. $\endgroup$ – Christian Remling Aug 8 '15 at 18:11
  • 7
    $\begingroup$ More specifically, the link below seems to give that $\Phi_n\sim (1+\sqrt{2})^{-4n}\sqrt{n}$: link.springer.com/article/10.1023%2FA%3A1004180718725#page-1 $\endgroup$ – Christian Remling Aug 8 '15 at 18:13
  • 2
    $\begingroup$ @ChristianRemling: maybe you should elevate your comments into an answer so that this question attains "answered" status? $\endgroup$ – Suvrit Aug 8 '15 at 22:11
  • 2
    $\begingroup$ Googling around, it seems that another reference for the decay of $\Phi_n$ is jstor.org/stable/2035162 ( published in the 60s) $\endgroup$ – Bertram Q. Aug 8 '15 at 22:42

This summarizes various comments from above. Write $p=(p_0,p_1,\ldots, p_n)$. Since $$ \Phi_n = \min_{\|p\|_2 =1} \sum_{j,k=0}^n \frac{p_jp_k}{j+k+1} , $$ we can also interpret $\Phi_n$ as the smallest eigenvalue of the (positive definite) Hilbert matrix $H_{jk}=1/(j+k+1)$, $j,k=0,1, \ldots , n$.

We find the following asymptotic formula in the literature: $$ \Phi_n = K\sqrt{n}\left( 1+\sqrt{2}\right)^{-4n}(1+o(1)), \quad\quad K=\frac{8\cdot 2^{3/4}\pi^{3/2}}{(1+\sqrt{2})^4} $$ See here and here.

| cite | improve this answer | |

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.