= Solution
A <coarse embedding> $f:X\to Y$ has nondecreasing control functions $\rho_-,\rho_+$ such that
$$
\rho_-(d_X(x,y))\leq d_Y(f(x),f(y))\leq\rho_+(d_X(x,y)),
\qquad \rho_-(t)\longrightarrow\infty.
$$
A family is a <uniform coarse embedding of a family of metric spaces> when all its maps use the same controls.
For $H_n=\{0,1\}^n$ with the <Hamming distance>, the coordinate map into $\ell_1^n$ satisfies
$$
\|x-y\|_1=d_H(x,y),
$$
so the cubes embed isometrically and uniformly into $L^1$. The same coordinate map into $\ell_2^n$ satisfies
$$
\|x-y\|_2=\sqrt{d_H(x,y)}.
$$
Thus they also uniformly coarsely embed into $L^2$, with $\rho_-(t)=\rho_+(t)=\sqrt t$. This is the <Hamming cube as an L1 and L2 metric>.
For a finite graph $G=(V,E)$, define
$$
h(G)=\min_{0<|S|\leq|V|/2}\frac{|\partial_ES|}{|S|}.
$$
An <expander family> is a sequence of graphs with orders tending to infinity, uniformly bounded degrees, and $h(G)$ uniformly bounded below.
Now let $|V|=n$ and let $M$ be a median of $f:V\to\mathbb R$. Put
$$
g=(f-M)_+,
\qquad k=(M-f)_+.
$$
Both supports have at most $n/2$ vertices. The <layer-cake representation> and the definition of $h$ give
$$
\begin{aligned}
\sum_{x,y}a_{xy}|g(x)-g(y)|
&=2\int_0^\infty|\partial_E\{g>t\}|\,dt\\
&\geq2h\int_0^\infty|\{g>t\}|\,dt
=2h\sum_xg(x),
\end{aligned}
$$
and the same holds for $k$. Since their supports are disjoint,
$$
|f(x)-f(y)|=|g(x)-g(y)|+|k(x)-k(y)|.
$$
Adding gives
$$
\boxed{\sum_{x,y}a_{xy}|f(x)-f(y)|
\geq2h\sum_x|f(x)-M|}.
$$
The <triangle inequality> also gives
$$
\sum_{x,y}|f(x)-f(y)|
\leq2n\sum_x|f(x)-M|,
$$
and hence
$$
\boxed{\sum_{x,y}a_{xy}|f(x)-f(y)|
\geq\frac hn\sum_{x,y}|f(x)-f(y)|}.
$$
Represent an $L^1$-valued map as scalar functions on the underlying measure space and integrate this inequality. The result is the <L1 Poincare inequality for an expander graph>.
Suppose next that $G$ is $d$-regular and $F:V\to L^1$ is noncontracting with Lipschitz constant $D$. The Poincare inequality gives
$$
ndD\geq\sum_{x,y}a_{xy}\|F(x)-F(y)\|_1
\geq\frac hn\sum_{x,y}d_G(x,y).
$$
A ball of radius $r$ in a graph of maximum degree $d$ has at most
$$
1+d\sum_{j=0}^{r-1}(d-1)^j\leq e d^r
$$
vertices. Therefore, for
$$
0\leq r\leq\frac{\log(n/2)-1}{\log d},
$$
at least $n/2$ vertices lie outside the radius-$r$ ball about each vertex. Integrating the tail count of the distance gives
$$
\sum_{x,y}d_G(x,y)
\geq\frac{n^2}{2}\frac{\log(n/2)-1}{\log d}.
$$
Consequently every such $F$ has
$$
D\geq\frac{h}{2d\log d}\left(\log\frac n2-1\right),
$$
which proves the <L1 distortion lower bound for an expander graph>.
Finally, suppose an expander family admitted uniform coarse embeddings $F_n$ into $L^1$. Embedded edges have length at most $\rho_+(1)$, whereas the preceding Poincare argument and the fact that at least half the ordered pairs have distance comparable to $\log|V_n|$ imply
$$
d\rho_+(1)\geq\frac h2\rho_-(c\log|V_n|).
$$
This contradicts $\rho_-(t)\to\infty$. Thus expanders do not uniformly coarsely embed into $L^1$. They do not uniformly coarsely embed into $L^2$ either: a Hilbert space embeds isometrically into an $L^1$ space by
$$
v\longmapsto\left(\omega\mapsto
\frac{\langle v,g(\omega)\rangle}{\mathbb E|N(0,1)|}
\right),
$$
so an $L^2$ coarse embedding would produce an $L^1$ one. This is the <expander graph obstruction to uniform coarse embedding>.
Back to article page
