Let $X = [0,1]^d$ be the unit cube in the $d$-dimensional Euclidean space. For every $x \in X$ and every coordinate $i=1,2,\ldots,d$ denote by $x_{-i} := (x_j)_{j \neq i}$. Given a set $Y \subseteq X$ and a vector $x_{-i} \in [0,1]^{d-1}$ denote by $Y_{x_{-i}} := \{(x_i,x_{-i}) \in Y \colon x_i \in [0,1]\}$ the $x_{-i}$-section of $Y$. Let $Y \subseteq X$ be a compact set that satisfies the following condition: for every $x \in X$ and every coordinate $i=1,2,\ldots,d$, the $x_{-i}$-section $Y_{x_{-i}}$ is nonempty and convex. Is it true that the set $Y$ is contractible?


TL;DR: Yes, it is contractible, and it is enough for that condition to hold for a single $i \in \{1,...,d\}$. To see this, we will construct a strong deformation retraction of $[0,1]^{d}$ to $Y$.

Firstly, a little of notation. Fix one $i \in \{1,...,d\}$, without loss of generality $i=d$. Write $\pi:[0,1]^d \rightarrow [0,1]^{d-1}$ for the map $x \mapsto x_{-d}$, write $\sigma:[0,1]^d \rightarrow [0,1]$, for the map $x \mapsto x_{d}$, and for every $x \in [0,1]^{d-1}$ write \begin{equation} Y_{x} := Y \cap \pi^{-1}(\{x\}) = \{y \in Y \vert \pi(y)=x\} \end{equation}

Over every $x \in [0,1]^{d-1}$, your assumption that $Y_{x}$ be non-empty and convex forces it to be an interval $Y_{x} = I_x\times \{x\}$; because $Y$ is compact, furthermore, it must be closed, and hence all intervals $I_x$ are also closed. Let $m_x,M_x$ be the minimum/maximum of $I_x$ respectively, so that $I_x = [m_x,M_x]$.

We could try to define a strong deformation retraction $f:[0,1]^d \times [0,1] \rightarrow [0,1]^d$ of $[0,1]^d$ to $Y$ as follows: \begin{equation} f(z,t) := \begin{cases} \big(\pi(z),(1-t)\sigma(z)+t M_{\pi(z)}\big) &\text{ if } \sigma(z) \in [M_{\pi(z)},1]\\ z &\text{ if } \sigma(z) \in [m_{\pi(z)},M_{\pi(z)}]\\ \big(\pi(z),(1-t)\sigma(z)+t m_{\pi(z)}\big) &\text{ if } \sigma(z) \in [0,m_{\pi(z)}]\\ \end{cases} \end{equation} It is easy to check that $f(z,0) = z$ for all $z \in [0,1]^d$, that $f(z,1) \in Y$ and that $f(z,t) = z$ for all $z \in Y$. Unfortunately, $f$ is not continuous: there is in general no guarantee that $m_{x_n} \rightarrow m_x$ and $M_{x_n} \rightarrow M_x$ whenever $x_n \rightarrow x$. Instead, we work with a ``regularised'' version of minima and maxima.

Let $(\psi^{(t)})_{t \in [0,1]}$ be a continuous family of continuous probability density functions on $[0,1]$, supported on $[0,1-t]$ and converging to the Diract delta distribution at $0$ as $t \rightarrow 1$. For $x \in [0,1]^{d-1}$, define a function $ms_x:[0,1] \rightarrow [0,1]$ by letting $ms_x(r)$ be the minimum of $m_y$ over all $y$ in the closed ball of radius $r$ around $x$; similarly, define $Ms_x:[0,1] \rightarrow [0,1]$ by letting $Ms_x(r)$ be the maximum of $M_y$ over all $y$ in the closed ball of radius $r$ around $x$. We have $ms_x(0) = m_x$ and $Ms_x(0) = M_x$.

The functions $ms_x$ and $Ms_x$ are monotonically decreasing/increasing respectively, and hence we can use the probability density functions to define the following continuous functions $[0,1]^{d-1}\times [0,1] \rightarrow [0,1]$: \begin{equation} \bar{m}(x,t) := \int_0^1 ms_x(r) \psi^{(t)}(r) dr \hspace{2cm} \bar{M}(x,t) := \int_0^1 Ms_x(r) \psi^{(t)}(r) dr \end{equation} In particular, $\bar{m}(x,1) = m_x$ and $\bar{M}(x,1) = M_x$. Then we can re-define the retraction $f:[0,1]^d \times [0,1] \rightarrow [0,1]^d$ as follows, this time as a continuous function: \begin{equation} f(z,t) := \begin{cases} \big(\pi(z),(1-t)\sigma(z)+t \bar{M}(\pi(z),t)\big) &\text{ if } \sigma(z) \in [M_{\pi(z)},1]\\ z &\text{ if } \sigma(z) \in [m_{\pi(z)},M_{\pi(z)}]\\ \big(\pi(z),(1-t)\sigma(z)+t \bar{m}(\pi(z),t)\big) &\text{ if } \sigma(z) \in [0,m_{\pi(z)}]\\ \end{cases} \end{equation} This is our desired strong deformation retraction from $[0,1]^d$ to $Y$.

| cite | improve this answer | |
  • $\begingroup$ This answer is likely related, but I couldn't determine whether all necessary conditions for the theorems mentioned therein are necessarily satisfied in this case. $\endgroup$ – Stefano Gogioso Sep 7 '18 at 1:02

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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