Norming vector for Holder inequality (source code)

= Norming vector for Holder inequality

Let $1<q<\infty$, let $p=q/(q-1)$, and take nonzero $y\in\ell^q$. For the complex pairing $\sum_nx_n\overline{y_n}$, put
$$
x_n=\frac{\operatorname{phase}(y_n)|y_n|^{q-1}}{\|y\|_q^{q-1}},
\qquad
\operatorname{phase}(z)=\begin{cases}z/|z|,&z\ne0,\\0,&z=0.\end{cases}
$$
Since $(q-1)p=q$, direct summation gives $\|x\|_p=1$ and $\sum_nx_n\overline{y_n}=\|y\|_q$. This <vector> attains equality in the <Holder inequality>. For the bilinear pairing $\sum_nx_ny_n$, conjugate the displayed phase instead.

At $q=1$, the norming <vector> lies in $\ell^\infty$ and is simply the phase of $y$, with the same pairing convention. At $q=\infty$, an $\ell^1$ norming <vector> need not exist. For $y_n=1-1/(n+1)$, the supremum <norm> is1 but no coordinate has modulus1. Every $x\in\ell^1$ of <norm>1 then satisfies $|\sum_nx_ny_n|<1$, though coordinate unit <vectors> approach the bound. Endpoint duality therefore cannot be handled by blindly inserting $q=\infty$ into the finite-exponent formula.