I want to understand the proof (or under which conditions a proof holds) of the following statement:
Let $f$ be a function in $L^p$ and let $q$ be such that $\frac{1}{p}+\frac{1}{q}=1$. Then we can find a function $g$ in $L^q$ such that:
- $\| g \|_{q}=1$
- $fg \in L^1$
- $\|fg\|_{1} = \|f\|_{p}$
For context: This is used in Kenneth Hoffman "Banach Spaces and Analytic Functions" to prove that if $f \in L^p(-\pi, \pi)$ then the Césaro means of the Fourier series for $f$ converge to $f$ in the $L^p$ norm. In the book, the author merely states that this is a consequence of the Hahn-Banach theorem.
I know the Hahn-Banach version for extending linear functionals and for separating convex sets and I can't wrap my head around on how to use it here.
In addition to it, I am not entirely sure if what I stated is entirely correct, maybe we need the additional assumption of finite measure spaces like $(-\pi, \pi)$ as used in the book.
All in all, this seems like a very cool trick for going back and forth between $L^p$-spaces to establish inequalities. In addition to an explanation of this result, I would also like to know about other related tricks and applications of the Hahn-Banach to this type of questions.
Recall that the dual of $L^p$ is $L^q$, which is valid for arbitrary measure spaces when $1 < p < \infty$. (It is also valid for $p=1$ when the space is $\sigma$-finite, but for $p=1$ the statement at hand here is trivial; simply take $g=1$.)
Suppose without loss of generality $\|f\|_p = 1$. Consider the 1-dimensional subspace of $L^p$ spanned by $f$, and define a linear functional $\ell_0$ on this space by $\ell_0(f)=1$. Clearly $\|\ell_0\| = 1$. Use Hahn-Banach to extend $\ell_0$ to a bounded linear functional $\ell$ on all of $L^p$, whose norm is still 1. Since the dual of $L^p$ is $L^q$, there exists $g \in L^q$ such that $\int gh = \ell(h)$ for all $h \in L^p$; in particular, $\int fg = \ell(f) = 1$. Moreover, $\|g\|_q = \|\ell\| = 1$, so (1) and (2) are established.
For (3), by the triangle inequality and Hölder's inequality we have $$1 = \left|\int fg \right| \le \|fg\|_1 \le \|f\|_p \|g\|_q = 1$$ so that $\|fg\|_1 = 1$ as desired.