3

Let $f:\mathbb{R}^n\to\mathbb{R}$ be a nonzero linear map. Let $\alpha \in \mathbb{R}$ e define $A=\{x\in \mathbb{R}^n:f(x)<\alpha\}$. By continuity of $f$, $A$ is open in $\mathbb{R}^n$.

How can I prove (rigorously) that $A$ and $\mathbb{R}^n$ are homeomorphic ?

I have some thoughts but I'm not able to make them rigorous. Intuitively I see that $\{x\in \mathbb{R}^n:f(x)=\alpha\}$ is an hyperplane in $\mathbb{R}^n$ which divides $\mathbb{R}^n$ into to "open rectangles" homeomorphics between them, and one of this "rectangles" is $A$.

For example, if $n=2$, I think I can show that $A$ is homeomorphic with $\mathbb{R}\times(-\infty,\alpha)$ maybe using some rotation, and then I easily can show that $\mathbb{R}\times(-\infty,\alpha)$ is homeomrphic to $\mathbb{R}^2$ since I know that $(-\infty,\alpha)$ is homeomrphic to $\mathbb{R}$. For example if $(a,b)\in \mathbb{R}^2$ is the vector such that $f(x,y)=ax+by$, such homeomorphism could be the linear map $\mathbb{R}\times(-\infty,\alpha)\to A$ represented by the matrix \begin{pmatrix} \cos(b/a)& -\sin(b/a) \\\sin(p/a) & \cos(b/a) \end{pmatrix}

What is a rigorous (and possibly elegant) way to prove that $A$ and $\mathbb{R}^n$ are homeomorphic for general $n$?

Suppose also that $\beta\in \mathbb{R}$ and $\beta<\alpha$. Let $B=\{x\in \mathbb{R}^n:f(x)>\beta\}$.

How can I prove that $A \cap B$ is non empty and homoemorphic to $\mathbb{R}^n$ ?

Minato
  • 1,541

3 Answers3

3

$A$ is a halfspace. Without loss of generality, you can consider $f(x)=x_1$ (with a change of basis, considering a basis completion of $f$, and then looking at the dual basis). Then, it reduces to showing that $(-\infty,a)$ is homeomorphic to $\mathbb{R}$.

Pierre
  • 352
  • If $f$ is any projection, then we have $A$ is a "real rectangle" in $\mathbb{R}^n$, and it is easy as you say. But I'm interested in any linear map and not considering only special cases. In other terms, if we are not allowed to change coordinates and in this way "suppose without loss of generality" that $f$ is a projection, how could one prove the statement? – Minato Mar 06 '19 at 10:50
  • 1
    In other terms you argument skips my "real question": how to show that $A$ is homeomorphic to a subset of $\mathbb{R}^n$ of the form $\mathbb{R}^{n-1}\times I$ with I open interval? – Minato Mar 06 '19 at 10:58
  • @Minato, this answer tells you precisely that. Change of basis is homeomorhpism. – Ennar Mar 06 '19 at 10:59
  • 1
    @Minato The without loss of generality means that the proof for the special case is the same as the proof of the general case. The general case just has more indices to keep track of. – Michael Burr Mar 06 '19 at 11:00
  • @Ennar, Michael Burr: You are right! :) In my preceeding comments I wanted to say "how can I concretely realize the change of coordinates that I need" – Minato Mar 06 '19 at 11:06
  • 1
    @Minato, take a basis ${f,f_2,\ldots,f_n}$ for $(\mathbb R^n)^*$ and take the corresponding dual basis ${b,b_2,\ldots b_n}$ for $\mathbb R^n$. – Ennar Mar 06 '19 at 11:20
  • @Ennar , Michalel Burr , Pierre Thank-you. I didn't know one could realize a change of coordinates using dual bases. Happy to have learned it thanks to you! :) I have been able to conclude with this method and I find it very useful and elegant! – Minato Mar 06 '19 at 12:28
1

Since $f$ is non-zero, $ker(f)$ is an $(n-1)$-dimensional subspace of $\mathbb{R}^n$. Choose any basis $b_1,\dots,b_{n-1}$ of $ker(f)$ and any $b_n$ such that $f(b_n) =1$ (which is possible because $f$ is non-zero linear map). Then $b_1,\dots,b_n$ are linearly independent, thus form a basis of $\mathbb{R}^n$.

The unique linear map $\phi : \mathbb{R}^n \to \mathbb{R}^n$ such that $\phi(e_i) = b_i$, where $e_1,\dots,e_n$ is the standard basis of $\mathbb{R}^n$, is a linear isomorphism and hence a homeomorphism (since linear maps are continuous).

Let $x \in \mathbb{R}$. Write $x = (x',t)$ with $x' \in \mathbb{R}^{n-1}$ and $t \in \mathbb{R}$. Then $\phi(x',t) = \phi(x',0) + \phi(0,t)$, where $\phi(x',0) \in ker(f)$ and $\phi(0,t) = \phi(te_n) = tb_n$. Hence $f(\phi(x',t)) = t$. This shows that for any $J \subset \mathbb{R}$

$$\phi(\mathbb{R}^{n-1} \times J) = \{ x \in \mathbb{R}^n \mid f(x) \in J \} .$$

Therefore $A(\alpha) = \phi(\mathbb{R}^{n-1} \times (\alpha,\infty))$. Similarly, with $A(\alpha,\beta) = \{x \in \mathbb{R}^n \mid \beta < f(x) < \alpha \}$, $A(\alpha,\beta) = \phi(\mathbb{R}^{n-1} \times (\beta,\alpha))$. Since $(\alpha,\infty)$ and $(\beta,\alpha)$ are homeomorphic to $\mathbb{R}$, we are done.

Paul Frost
  • 87,968
  • Thank-you! This is precisely what I meant with "rigorous and elegant" :) Even thought, I have learned something from all the answers! Thank you all very much! – Minato Mar 06 '19 at 16:18
  • Elegant answer, will appreciate it if you could help with some details in my homeomorphism problem too :) https://math.stackexchange.com/questions/3124772/show-t-is-a-homeomorphism-from-e-onto-te – Homaniac Mar 06 '19 at 21:22
  • @Homaniac It seems that mechanodroid has already done this. – Paul Frost Mar 07 '19 at 10:04
0

Let $f$ be represented by $a = (a_1, \ldots, a_n) \in \mathbb{R}^n$, $a \ne 0$ via the Riesz representation theorem. WLOG we can assume $a_1 \ne 0$ and by considering $\frac{\alpha}{\|a\|}$, we can assume $\|a\| = 1$.

Let $U : \mathbb{R}^n \to \mathbb{R}^n$ be a unitary map such that $Ua = e_1$. Then $$A = \{x \in \mathbb{R}^n : \langle x,a\rangle < \alpha\} = \{x \in \mathbb{R}^n : \langle Ux,Ua\rangle < \alpha\} = \{x \in \mathbb{R}^n : \langle Ux,e_1\rangle < \alpha\}$$

Consider $\phi : \mathbb{R}^n \to A$ given by $$\phi(x_1, \ldots, x_n) = U^*(\alpha - e^{\langle x, a\rangle}, x_2, \ldots, x_n) \in A$$ Clearly $\phi$ is continuous and $\phi^{-1}$ is given by $$\phi^{-1}(y) = \left(\frac1{a_1}\left(\ln(\alpha - \langle Uy, e_1\rangle) - \sum_{j=2}^n a_j\langle Uy, e_j\rangle\right), \langle Uy, e_2\rangle, \ldots, \langle Uy, e_n\rangle\right)$$ which is also continuous.

If you're unsure about this, explicit calculation gives:

$$\phi^{-1}(\phi(x)) = \phi^{-1}(U^*(\alpha - e^{\langle x, a\rangle}, x_2, \ldots, x_n)) = \left(\frac1{a_1}\left(\langle x, a_1\rangle - \sum_{j=2}^n a_jx_j\right), x_2, \ldots, x_n\right) = x$$ \begin{align} \phi(\phi^{-1}(y)) &= \phi\left(\frac1{a_1}\left(\ln(\alpha - \langle Uy, e_1\rangle) - \sum_{j=2}^n a_j\langle Uy, e_j\rangle\right), \langle Uy, e_2\rangle, \ldots, \langle Uy, e_n\rangle\right) \\ &= U^*\left(\alpha-e^\left(a_1\cdot\frac1{a_1}\left(\ln(\alpha - \langle Uy, e_1\rangle) - \sum_{j=2}^n a_j\langle Uy, e_j\rangle\right) + \sum_{j=2}^na_j\langle Uy, e_n\rangle\right), \langle Uy, e_2\rangle, \ldots, \langle Uy, e_n\rangle\right)\\ &=U^*\left(\alpha - e^{\ln(\alpha- \langle Uy, e_1\rangle)},\langle Uy, e_2\rangle, \ldots, \langle Uy, e_n\rangle\right)\\ &=U^*\left(\langle Uy, e_1\rangle,\langle Uy, e_2\rangle, \ldots, \langle Uy, e_n\rangle\right)\\ &= U^*Uy\\ &= y \end{align}


In the case with $\beta < \alpha$, we have that $A \cap B = \{x \in \mathbb{R}^n : \langle x, a\rangle \in (\alpha, \beta)\}$. This is nonempty because if $\langle x, a\rangle \ne 0$, then $$\frac{\alpha + \beta}2 \frac{x}{\langle x,a\rangle} \in A \cap B$$ We know that $(\alpha, \beta) \cong \mathbb{R}$. To construct $\phi$ in the preceding proof I used the homeomorphism $x\mapsto \alpha - e^x$ between $\mathbb{R}$ and $(-\infty, \alpha)$.

In this case pick a homeomorphism $g : \mathbb{R} \to (\alpha, \beta)$ and define $\phi : \mathbb{R}^n \to A \cap B$ as $$\phi(x) = U^*(g(\langle x, a\rangle), x_2, \ldots, x_n)$$ and $\phi^{-1}$ similarly.

mechanodroid
  • 47,570
  • Thank you! I will take some time to look carefully at it, but thank you in advance for your great work! :) – Minato Mar 06 '19 at 11:45
  • 1
    Like your attention to details, will appreciate it if you could help with some details in my homeomorphism problem too :) https://math.stackexchange.com/questions/3124772/show-t-is-a-homeomorphism-from-e-onto-te – Homaniac Mar 06 '19 at 21:24
  • @Homaniac Sure, I posted an answer. Let me know if you need clarification. – mechanodroid Mar 06 '19 at 23:19