Let W:[0,1]2→[−1,1]. Then
∣∣W∣∣2,2≤(1)4⋅∣∣W∣∣∞→1≤(2)16⋅∣∣W∣∣□
ie, cut norm convergence implies convergence in L2
Proof of (2)
We first show inequality (2). Using the definition of the cut norm, we have
\lvert \lvert W \rvert \rvert_{\square} &= \sup_{S,T \subseteq [0,1]} \left\lvert \int \int _{S \times T} W(u,v) \, du \, dv \right\rvert \\
&= \sup_{f,g: [0,1]\to[0,1]} \left\lvert \int_{0}^1 \int_{0}^1 W(u,v) f(u) g(v) \, du \, dv \right\rvert \\
(*) &= \sup_{f,g: [0,1] \to [0,1]}\lvert \langle T_{W}f, g \rangle \rvert \\
\end{aligned}$$
Since we are taking the supremum, it is equivalent to taking the supremum over all $L_{\infty}$ functions $f$ and $g$.
To get $(*)$, we notice that $\int _{0}^1 W(u,v) f(u) \, du$ is an [[Concept Wiki/integral linear operator]] with kernel $W$. We denote this with $T_{W}$ so that
$$T_{W}f\,\,(v) =\int _{0}^1 W(u,v) f(u) \, du$$
- or equivalently that $T_{W}f = \int _{0}^1 W(u,\cdot) f(u) \, du$
We can then see that the definition becomes the absolute value of the inner product of two $L_{\infty}$ functions.
Note that $\lvert \lvert W \rvert \rvert_{\infty \to 1}$ is an induced [[Concept Wiki/operator norm]] of the [[Concept Wiki/graphon]] mapping functions from $L_{\infty}$ to $L_{1}$.
$$\begin{aligned}
\lvert \lvert W \rvert \rvert_{\infty \to 1} &= \sup_{-1 \leq g \leq 1} \lvert \lvert T_{W}g \rvert \rvert_{1} \\
&= \sup_{-1 \leq g \leq 1} \int \lvert T_{W}g(x) \rvert \, dx \\
&= \sup_{-1 ≤f,g≤1} \langle T_{W}g, f \rangle \\
\end{aligned}$$
- we omit the denominator in our usual definition of the [[Concept Wiki/operator norm]] since $g$ is bounded by $-1$ and $1$ and $g \in L_{\infty}$
- We can rewrite this using $f$ to match in the definition of the $L_{1}$ norm since we are taking the supremum
- For more about $L_{p}$ spaces see [wikipedia](https://en.wikipedia.org/wiki/Lp_space#Lp_spaces_and_Lebesgue_integrals) (hand wavy/cursory for this class on functional analysis details)
We can then rewrite this as
$$\begin{aligned}\lvert \lvert W \rvert \rvert_{\infty \to 1} &= \sup_{-1 \leq f, f' \leq 1, -1 \leq g, g' \leq 1} \langle T_{W}(g-g'), f-f' \rangle \\
(**) &\leq \sup_{-1 \leq g, f \leq 1} \langle T_{W}g, f \rangle - \sup_{-1 \leq g,f' \leq 1} \langle T_{W}g, f' \rangle + \sup_{-1 \leq g',f\leq 1} \langle T_{W}g', f \rangle - \sup_{-1 ≤g', f' ≤1} \langle T_{W}g', f' \rangle \\
({\dagger}) &\leq 4 \lvert \lvert W \rvert \rvert_{\square}
\end{aligned}$$
Where we get $(**)$ by the [[Concept Wiki/triangle inequality]] and $({\dagger})$ from the definition of $\lvert \lvert \cdot \rvert \rvert_{\square}$.
- Note that this is a loose bound!
And this gives us the desired result for (2).
^proof-2
Proof of (1)
Proving (1) is more involved and uses more functional analysis.
Use the Riesz-Thorin interpolation theorem for complex Lp spaces:
∣∣W∣∣p→q≤∣∣W∣∣p1→q01−θ∣∣W∣∣p1→q1θ
Where
θ=min(1−p1,q1) and
p0,q0∈[1,∞)
with p1=p01−θ and
1−q1=(1−θ)(1−q01) and
p1=∞,q1=1.
Define operator norm∣∣W∣∣□,C=supf,g:[0,1]∣∣f∣∣∞,∣∣g∣∣∞→C≤1∫01∫01W(u,v)f(u)g(v)dudv
So for complex functions, we can see that
\text{for complex functions } \lvert \lvert W \rvert \rvert_{\infty \to 1} &= \lvert \lvert W \rvert \rvert_{\square, \mathbb{C}} \\
&≤ 2\lvert \lvert W \rvert \rvert_{\infty \to 1} \text{ for real functions}
\end{aligned}$$
We then have
$$\lvert \lvert W \rvert \rvert _{p_{0} \to q_{0}} \leq \lvert \lvert W \rvert \rvert _{1 \to \infty} \leq \lvert \lvert W \rvert \rvert_{\infty} \leq 1$$
Since $W \leq 1$, and this gives the desired result.