Are the restrictions of constructible functions on the level set of smooth map still constructible functions?

38 Views Asked by At

This question is in relation with the definition of Euler-Integral transforms, page 147 of Michael Robinson book "Topological Signal Processing".

Definition: Euler characteristic and Constructible function

The euler characteristic $\chi$ is a valuation from a cell complex $X_f$ to $\mathbb{Z}$ defined by: $$ \chi(X_f) = \sum_{c \in X_f}^{}{-1^{dim(c)}}$$

${\bf Remark:} $The definition of $\chi$ makes it possible to define it for an arbitraty collection of cell. For instance, a sub-collection $X_f'$ of cells in $X_f$ is not always a cell complex, but we will admit that $\chi(X_f')$ is well-defined. Even though the author doesn't make it clear in the book, this assumption seems to be largely used.

An integered valued constructible function is a function on a topological space $f:X \mapsto \mathbb{Z}$ such that there exists a non-unique cell-complex $X_f$ for which:

a) There is an homeomorphism $h : X \mapsto X_f$

b) The function $f \circ h$ is constant on each cell of $X_f$

This can be seen as a generalization of piecewise constant function on topological spaces having a cell-complex skeleton.


Problem statement

Given a topological space $X$ and a smooth map $P:X \mapsto \mathbb{R}$. We assume that $X$ is such that there is a homeomorphism $h$ between $X$ and a cell complex $X_f$ . We define the level sets of $P$: $$ X_c =\{ x \in X, P_x(x,\bullet) = s \}$$

We define a constructible function f with the same underlying cell-structure (i.e, $f\circ h$ is piecewise constant on the cells of $X_f$).

The question is: Is $f_{\vert X_c}$ still constructible for the homeomorphism $h$?

The author seems to consider so. But I am disturbed by the fact that the level set themselves can very well fail to be sub-cell complex, or even collections of cells.


For instance, if $Y$ is the ball centered in O in $\mathbb{R}^2$ with the cell-complex structure made of one $0$-cell, one $1$-cell and one $2$-cell. Let's have $P : y \mapsto \| y \|$, and $f$ is the indicatrice function of the ball.

The level set of $P$ for $s$ are the intersection between a circle of radius s and $Y$. Three cases can occur:

  1. For $s=0$, the level set is the single point (0,0).
  2. For $0<s<{max}$ Y_s is a closed circle in the interior of $Y$.
  3. For $s = s_{max}$ the level set is the frontier made of the $1$-cell attached to the $0$-cell.

It seems that apart for the third case, the level set are neither cells or a collection of cells. Which results in my inability to understand how can the author use the function $f_{\vert Y_s}$ as a constructible function (computing its euler integral for instance).

I am wondering if I missed something involved by $h$ being a homeomorphism, or by $P$ being smooth, as I don't seem to use any of these assumptions. I would appreciate some clues about what the author is really trying to do (for those who read the book) or about the constructability of $f_{\vert X_c}$.