MHB Are There a Finite Number of Solutions to $f(x) = y$ in a Bounded Region?

  • Thread starter Thread starter Euge
  • Start date Start date
Click For Summary
The discussion centers on proving that for a differentiable function \( f : \mathbb{R}^n \to \mathbb{R}^m \) with a constant rank derivative \( Df \), there are finitely many solutions to the equation \( f(x) = y \) within a bounded region \( \Omega \subset \mathbb{R}^n \) for any \( y \in \mathbb{R}^m \). GJA has successfully demonstrated this for smooth functions, while another participant offers a solution applicable to general differentiable functions. The key focus is on the implications of the constant rank condition of the derivative. The discussion highlights the mathematical reasoning behind the finiteness of solutions in the specified context. Overall, the thread explores important aspects of differential topology and the behavior of functions in bounded regions.
Euge
Gold Member
MHB
POTW Director
Messages
2,072
Reaction score
245
Here is this week's problem!

-----
Suppose $f : \Bbb R^n \to \Bbb R^m$ is a differentiable function such that the derivative $Df$ has constant rank $n$. If $\Omega \subset \Bbb R^n$ is bounded, prove that to every $y\in \Bbb R^m$ there corresponds only finitely many solutions to the equation $f(x) = y$ in $\Omega$.
-----

 
Physics news on Phys.org
GJA has provided a correct proof in the case $f$ is smooth. You can read his solution below.

Suppose there are infinitely many solutions to $f(x)=y$ in $\Omega$. From this collection of infinite solutions, select a sequence $\{x_{n}\}$ of distinct points, noting that $x_{n}\in\Omega$ and $f(x_{n}) = y$ for all $n$.

Since $\Omega$ is a bounded subset of $\mathbb{R}^{n}$, $\text{cl}(\Omega)$ is compact by the Heine-Borel theorem. According to the Bolzano-Weierstrass theorem, $\text{cl}(\Omega)$ is sequentially compact. Hence, $\{x_{n}\}$ admits a convergent subsequence whose limit, say $x$, is in $\text{cl}(\Omega)$. For ease of notation, we denote the convergent subsequence by $\{x_{n}\}$, as the original sequence is no longer needed.

Since $Df$ has constant rank at $x$, the Constant Rank theorem says that there is a diffeomorphism $G$ defined on a neighborhood, say $U$, of $x$ sending $x$ to the origin of $\mathbb{R}^{n}$, and a diffeomorphism $F$ defined on a neighborhood of $f(x) = y$ sending $y$ to the origin of $\mathbb{R}^{m}$ such that $$(F\circ f\circ G^{-1})(x^{1},\ldots, x^{n}) = (x^{1},\ldots, x^{n}, \underbrace{0,\ldots, 0}_{m-n})\qquad (*)$$ Since the sequence of points we chose from the outset are distinct, since $U$ is a neighborhood of $x$, and since $x_{n}\rightarrow x$, there exists $N$ such that $x_{N}\in U$ and $x_{N}\neq x$. Since $x_{N}\neq x$ and since the diffeomorphism $G$ maps $x$ to the origin, we know $G(x_{N})\neq (\underbrace{0,\ldots, 0}_{n})$. Let $(p^{1}, \ldots, p^{n})$ be such that $x_{N} = G^{-1}(p^{1},\ldots, p^{n})$ and note that $(p^{1},\ldots, p^{n})\neq (\underbrace{0,\ldots, 0}_{n})$, as the previous sentence implies.

According to $(*)$, $$(F\circ f\circ G^{-1})(p^{1},\ldots, p^{n}) = (p^{1},\ldots, p^{n}, \underbrace{0,\ldots, 0}_{m-n})\neq (\underbrace{0,\ldots, 0}_{m})\qquad(**),$$ where the $\neq$ follows because $(p^{1},\ldots, p^{n})\neq (\underbrace{0,\ldots, 0}_{n})$.

However, we now have a contradiction because we know $f(x_{N})=y$ and that $F(y) = 0$. Thus, $$(F\circ f\circ G^{-1})(p^{1},\ldots, p^{n}) = (F\circ f)(x_{N}) = F(y) = (\underbrace{0,\ldots, 0}_{m}),$$ contradicting $(**)$.

For general differentiable $f$, you can read my solution below.

Fix $y\in \mathbb{R}^m$, and set $\Sigma = \{x\in \mathbb{R}^n : f(x) = y\}$. Given $a\in \Sigma$, $Df(a)$ is an injective linear map (since $Df(a)$ has rank $n$); thus, there is a linear map $L : \mathbb{R}^m\to \mathbb{R}^n$ such that $L\circ Df(a) = \text{id}$. Since $f$ is differentiable at $a$, there is an open neighborhood $U$ of $a$ such that for all $x\in U$, $|f(x) - y - Df(x)(x - a)| < \frac{1}{2\|L\|}|x - a|$. Hence if $x\in U$, $$|f(x) - y| \ge |Df(x)(x - a)| - |f(x) - y - Df(x)(x - a)| > \frac{1}{\|L\|}|x - a| - \frac{1}{2\|L\|}|x - a| = \frac{1}{2\|L\|}|x - a|$$ In particular, $U \cap \Sigma = \{a\}$. Since $a$ was arbitrary, $\Sigma$ is discrete.

Let $X = \Sigma \cap \overline{\Omega}$. Continuity of $f$ implies $\Sigma$ is closed; thus, $X$ is closed. Boundedness of $\Omega$ forces boundedness of $X$.. By the Heine-Borel theorem, $X$ is compact. As compact discrete sets are finite, $X$ is finite. Hence $\Sigma \cap \Omega$ is finite, as desired.