3
$\begingroup$

Let $X$ be a compact metric space. A $k$ mean on $X$ is a continuous map $f:X^{k}\to X$ which is identity on the diagonal and is invariant under all $k$-permutations. For details, See the following paper http://www.sciencedirect.com/science/article/pii/S0723086904800161

Assume that $X$ is a compact metric space which admit a $k$ mean for every $k$. Can we equip $X$ to a family of $k$-means $f_{k}:X^{k} \to X$ with the following uniform continuity property:

For every $\varepsilon$ there is a $\delta$ such that for every two finite subset $A$ and $B$ of $X$ with cardinality $n,m$,resp. and $d_{H}(A,B)<\delta$ we have $d(f_{n}(A), f_{m}(B))< \varepsilon$.

Here $d_{H}$ is the Hausdorff distance.

Then the next question is that: for any such metric space, is it true that $X$ is a retract of $2^{X}$, the hyperspace.(Since every compact subset can be approximated by a finite set)

$\endgroup$
0

1 Answer 1

2
$\begingroup$

The uniform continuity property that you have mentioned does guarantee that $X$ is a retract of the hyperspace $H(X)$. Let $[X]^{<\omega}$ denote the collection of all finite subsets of $X$. Then $[X]^{<\omega}$ is a dense subspace of $H(X)$. Suppose that for each $\epsilon>0$ there is a $\delta>0$ such that for each $A,B\subseteq X$ with $d_{H}(A,B)<\delta$, we have $d(f_{n}(A),f_{m}(B))<\epsilon$. Define a mapping $j:[X]^{<\omega}\rightarrow X$ by letting $j(A)=f_{n}(A)$ whenever $|A|=n$. Then $j$ is uniformly continuous.

It is well known that if $(X,\mathcal{U}),(Y,\mathcal{V})$ are uniform spaces and $(Y,\mathcal{V})$ is complete, then every uniformly continuous mapping $f:X\rightarrow Y$ extends to a uniformly continuous mapping $\overline{f}:\overline{X}\rightarrow Y$ where $\overline{X}$ is the completion of $\overline{X}$. Therefore, the mapping $j:[X]^{<\omega}\rightarrow $ extends to a uniformly continuous map $k:H(X)\rightarrow X$ and the mapping $k$ is clearly a retraction.

However, I must mention that it is often the case that there does not exist a retraction from $H(X)$ to $X$ since if $X$ is a Peano continuum and there is a retraction from $H(X)$ to $X$, then $X$ is contractible.

$\mathbf{Proposition}$ If $X$ is a Peano continuum and there is a retraction from $H(X)$ to $X$, then there is a deformation retract from $X$ to any point $x_{0}\in X$.

$\mathbf{Proof}$ By this answer, the hyperspace of any Peano continuum (locally connected continuum) is homeomorphic to the Hilbert cube. Since $H(X)$ is homeomorphic to the Hilbert cube, there is a deformation retract $T:H(X)\times[0,1]\rightarrow H(X)$ from $H(X)$ to the point $x_{0}$. In other words, $T$ is a continuous mapping with $T(x_{0},t)=x_{0}$ and $T(C,0)=x,T(C,1)=x_{0}$ for each $C\in H(X)$. Let $j:H(X)\rightarrow X$ be a retraction. Define a mapping $V:X\times[0,1]\rightarrow X$ by $V(x,t)=j T(x,t)$. Then $V$ is clearly a deformation retract from $X$ to the point $x_{0}$. $\mathbf{QED}$

$\textbf{There are sometimes retracts from spaces of measures}$ For this section, we shall suppose that $X$ is a compact metric space.

Using a mean to obtain a retraction from the hyperspace $H(X)$ to $X$ seems a little bit unnatural and the uniform continuity condition that you have mentioned is a bit unnatural as well (I must mention that I can imagine that there are contexts in which this uniform continuity condition would appear). We shall therefore now take a retraction from the space $\mathcal{M}(X)$ of all regular Borel probability measures on $X$ to $X$. The space $\mathcal{M}(X)$ shall be endowed with the weak$^{*}$-topology.

Let $\delta_{x}$ be the measure where $\int fd\delta_{x}=f(x)$ for each continuous $f$. By associating each point $x\in X$ with the measure $\delta_{x}$, we shall regard the space $X$ as a closed subspace of $\mathcal{M}(X)$. Take note that since $\mathcal{M}(X)$ is a convex subset of the space of all signed measures on $X$, there for each $x_{0}\in X$, there is a deformation retract of $\mathcal{M}(X)$ to any point measure $\delta_{x_{0}}$. Therefore, if there is a retract from $\mathcal{M}(X)$ to $X$, then for each $x_{0}\in X$, there is a deformation retract from $X$ to $x_{0}$.

Suppose now that $(X,d)$ is a compact metric space. The we shall call a family of functions $\mathcal{A}\subseteq\bigcup_{n\in\omega}C(X^{n},X)$ equicontinuous in the mean if for each $\epsilon>0$ there is some $\delta>0$ such that if $f\in\mathcal{A},f:X^{n}\rightarrow X$ and if $\frac{1}{n}[d(x_{1},y_{1})+...+d(x_{n},y_{n})]<\delta$, then $d(f(x_{1},...,x_{n}),f(y_{1},...,y_{n}))<\epsilon$.

We say that a continuous binary operation $*$ on $X$ is a mean if it satisfies the following identities

  1. $x*x=x$

  2. $x*y=y*x$

  3. $(w*x)*(y*z)=(w*y)*(x*z)$.

Suppose that $*$ is a mean. Then define operations $m_{n}^{*}:X^{2^{n}}\rightarrow X$ (the operations $m_{n}^{*}$ were mentioned by Eric Wofsey in this answer) for all $n>0$ by induction:

$m^{*}_{1}(x,y)=x*y$

$m^{*}_{n+1}(x_{1},...,x_{2^{n+1}})=m^{*}_{n}(x_{1},...,x_{2^{n}})*m^{*}_{n}(x_{1},...,x_{2^{n}})$.

$\mathbf{Example}$ Suppose that $*$ is a mean on a compact metric space $(X,d)$ such that $d(x_{1}*x_{2},y_{1}*y_{2})<\frac{1}{2}(d(x_{1},y_{1})+d(x_{2},y_{2}))$. Then the family of mappings $\{m_{n}^{*}|n\in\mathbb{N}\}$ is equicontinuous in the mean.

Suppose now that $*$ is a mean on a compact metric space $(X,d)$.

Let $\mathcal{A}\subseteq\mathcal{M}(X)$ be the set of all measures of the form $\frac{1}{2^{n}}(\delta_{x_{1}}+...+\delta_{x_{2^{n}}})$. Since $\mathcal{M}(X)$ is a compact space in the weak$^{*}$-topology, the space $\mathcal{M}(X)$ has a unique uniformity.

Suppose that $*$ is a mean on a compact metric space $(X,d)$. Define a mapping $j_{*}:\mathcal{A}\rightarrow X$ by letting $j_{*}(\frac{1}{2^{n}}(\delta_{x_{1}}+...+\delta_{x_{2^{n}}}))=m^{*}(x_{1},...,x_{2^{n}})$.

$\mathbf{Theorem}$ Suppose that $*$ is a mean on a compact metric space such that $\{m^{*}_{n})|n\in\mathbb{N}\}$ is equicontiuous in the mean. Then the mapping $j_{*}$ is uniformly continuous. In particular, the mapping $j_{*}$ extends to a unique mapping $k_{*}:\mathcal{M}(X)\rightarrow X$ (and the mapping $k_{*}$ is a retract).

$\mathbf{Proof}$ For this theorem, let $D=\max_{x,y\in X}d(x,y)$

Suppose that $\epsilon>0$. Then since $\{m_{n}^{*}|n\in\mathbb{N}\}$ is equicontinuous in the mean, there is some $\delta>0$ such that if $n>0$ and $\frac{1}{2^{n}}(d(x_{1},y_{1})+...+d(x_{2^{n}},y_{2^{n}}))<\delta$, then $d(m_{n}^{*}(x_{1},...,x_{2^{n}}),m_{n}^{*}(y_{1},...,y_{2^{n}})<\epsilon$

Now let $\beta>0,\alpha>0$. Then there are $f_{1},...,f_{k}:X\rightarrow[0,1]$ such that whenever $1\leq i\leq k$ the support $\textrm{supp}(f_{i})$ has diameter less than $\alpha$ and where $\{(f_{i}^{-1}[\{1\}])^{\circ}|1\leq i\leq k\}$ covers $X$. Let $U_{i}=f^{-1}_{i}[\{1\}]^{\circ}$ whenever $1\leq i\leq k$.

If $A\subseteq\{1,...,k\}$, then let $f_{A}=\max_{a\in A}f_{a}$.

Now suppose that $\mu,\nu\in\mathcal{A}$ and $$|\int f_{A}d\mu-\int f_{A}d\nu|<\beta$$ for each $A\subseteq\{1,...,k\}$.

Then there is some $n$ and $x_{1},...,x_{2^{n}},y_{1},...,y_{2^{n}}$ such that $$\mu=\frac{1}{2^{n}}(\delta_{x_{1}}+...+\delta_{x_{2^{n}}}),\nu=\frac{1}{2^{n}}(\delta_{y_{1}}+...+\delta_{y_{2^{n}}})$$. Then since $$\int f_{A}d\mu=\frac{1}{2^{n}}(f_{A}(x_{1})+...+f_{A}(x_{2^{n}})),\int f_{A}d\nu=\frac{1}{2^{n}}(f_{A}(y_{1})+...+f_{A}(y_{2^{n}})),$$ we have $$|\frac{1}{2^{n}}(f_{A}(x_{1})+...+f_{A}(x_{2^{n}}))-\frac{1}{2^{n}}(f_{A}(y_{1})+...+f_{A}(x_{2^{n}})|<\beta$$ for each $A\subseteq\{1,...,k\}$, so $|(f_{A}(x_{1})+...+f_{A}(x_{2^{n}}))-(f_{A}(y_{1})+...+f_{A}(y_{2^{n}}))|<2^{n}\beta$.

Let $r=\lfloor 2^{n}\cdot\beta\rfloor$.

If $1\leq i\leq 2^{n}$, then let $M(i)=\{j\in\{1,...,2^{n}\}|d(x_{i},y_{j})<\alpha\}$ and if $R\subseteq\{1,...,2^{n}\}$, then let $M(R)=\bigcup_{r\in R}M(r)$.

Suppose now that $R\subseteq\{1,...,2^{n}\}$. Let $a_{r}\in\{1,...,k\}$ be such that $x_{r}\in U_{a_{r}}$. Let $A=\{a_{r}|r\in R\}$.

Then $f_{A}(x_{r})=1$ for each $r\in R$, so

$f_{A}(x_{1})+...+f_{A}(x_{2^{n}})\geq |R|$. Therefore, $f_{A}(y_{1})+...+f_{A}(y_{2^{n}})>|R|-2^{n}\beta$.

Take note that there must be at least $|R|-r$ indices $i$ such that $f_{A}(y_{i})\neq 0$.

However, if $f_{A}(y_{i})>0$, the $y_{i}\in\mathrm{supp}(f_{a_{r}})$ for some $r\in R$, so since $\mathbf{supp}(f_{a_{r}})$ has diameter less than $\alpha$, we conclude that $y_{i}$ is within $\alpha$ distance of $x_{r}$. We therefore conclude that $i\in M(r)$, so $i\in M(R)$. Therefore, we conclude that $|M(R)|\geq|R|-r$.

If $1\leq i\leq 2^{n}$, then let $N(i)=\{j:1\leq j\leq 2^{n},d(x_{i},y_{i})<\alpha\}\cup\{2^{n}+1,...,2^{n}+r\}$ and if $2^{n}<i\leq 2^{n}+r$, then let $N(i)=\{1,...,2^{n}+r\}$. If $A\subseteq\{1,...,2^{n}+r\}$, then define $N(A)=\bigcup_{a\in A}N(a)$. Then $|N(A)|\geq|A|$ for each $A\subseteq\{1,...,2^{n}+r\}$. Therefore, by Hall's marriage theorem, there exists a bijection $f:\{1,...,2^{n}+r\}\rightarrow\{1,...,2^{n}+r\}$ such that $f(i)\in N(i)$ for $1\leq i\leq 2^{n}+r$. Therefore, by restricting $f$ to a function whose domain and range are subsets of $\{1,...,2^{n}\}$, we conclude that there are $A,B\subseteq\{1,...,2^{n}\}$ such that $|A|\geq 2^{n}-r,|B|\geq 2^{n}-r$ and a bijection $h:A\rightarrow B$ such that $h(i)\in N(i)$ for $i\in A$. Extend the bijection $h$ to a bijection $g:\{1,...,2^{k}\}\rightarrow\{1,...,2^{k}\}$. Then $$\frac{1}{2^{n}}(d(x_{1},y_{g(1)})+...+d(x_{2^{n}},y_{g(2^{n})}))$$ $$=\frac{1}{2^{n}}[\sum_{a\in A}d(x_{a},y_{g(a)})+\sum_{b\in\{1,...,2^{n}\}\setminus A}d(x_{b},y_{g(b)})]$$

$$<\frac{1}{2^{n}}(2^{n}\cdot\alpha+r D)=\alpha+D\frac{r}{2^{n}}\leq \alpha+D\beta.$$

Therefore, choose $\alpha,\beta$ such that $\alpha+D\beta<\delta$. Then we have

$$\frac{1}{2^{n}}(d(x_{1},y_{g(1)})+...+d(x_{2^{n}},y_{g(2^{n})}))<\delta.$$

Therefore, we have

$$d(j_{*}(\mu),j_{*}(\nu))=d(m^{*}_{n}(x_{1},...,x_{2^{n}}),m^{*}_{n}(y_{1},...,y_{2^{n}})))$$

$$=d(m^{*}_{n}(x_{1},...,x_{2^{n}}),m^{*}_{n}(y_{g(1)},...,y_{g(2^{n})}))<\epsilon.$$

We conclude that the mapping $j^{*}$ is uniformly continuous.

$\mathbf{QED}$

$\endgroup$
0

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.