0

Given a non-constant analytic function $f(x)$ on a domain $D \subseteq \mathbb{R}^n$. I want to prove that

$\mathcal{Z} = \{x \in \mathcal{D} |f(x) = 0 \}$

is closed nowhere dense.

I originally wanted to prove this by contradiction as follows using the identity theorem.

"Assume that $\mathcal{Z}$ is not a closed nowhere dense set, then $\mathcal{Z}$ must contain a limit point", which shows that $f(x)$ is constant zero by the identity theorem and thus contradicts that $f(x)$ is non-constant.

However, I am not sure if the following statement is true:

"Assume that $\mathcal{Z}$ is not a closed nowhere dense set, then $\mathcal{Z}$ must contain a limit point"

  1. If it is true, could you provide one reference?

  2. If not, is $\mathcal{Z}$ closed nowhere dense, and how can I prove it?

chirico
  • 1,339
Shi James
  • 411
  • You should say connected domain. Otherwise, your function could be identically zero on some component of $D$ but not on others. The identity theorem you cite is for domains in $\mathbb C$, not in $\mathbb R^n$; so you will need a different reference. – GEdgar Mar 19 '21 at 12:22
  • Here "domain" refers to a domain in $\mathbb{R}^n$ and is open and connected by definition. – Shi James Mar 19 '21 at 13:07
  • The existence of a limit point of $Z$ is not going to get you very far if $n>1.$ – zhw. Mar 22 '21 at 18:49
  • For example, consider $f(x,y)=x$ on $\mathbb R^2.$ – zhw. Mar 22 '21 at 20:20
  • $\mathcal{Z}$ is closed as it is the preimage of a closed set under a continuous function $(f^{-1}({0}))$, so it's equal to its closure. If it had nonempty interior it would contain an open set, implying $f=0$ by the identity theorem, hence it(s closure) has empty interior, giving nowhere denseness by definition. – pabk Dec 17 '24 at 11:00

2 Answers2

1

We know $Z$is closed. So we want to show $Z$ is nowhere dense. Assume, to reach a contradiction, that the interior of $Z,$ denoted by $\text {int } Z,$ is nonempty.

Let $x\in \text {int } Z.$ Let $y\in D.$ Because $D$ is open and connected, $D$ is path connected. So there is a continuous path $\gamma:[0,1]\to D$ with $\gamma(0)=x, \gamma(1)=y.$

Let $E$ be the set of $t\in [0,1]$ such that $f=0$ in an open ball $B_{\gamma(t)}$ centered at $\gamma(t).$ Clearly $0\in E.$ By continuity of $\gamma,$ $\gamma^{-1}(B_{\gamma(t)})$ is an open neighborhood of $t$ in $[0,1].$ And if $s$ is in this neighborhood, $\gamma(s)\in B_{\gamma(t)},$ hence there is an open ball centered at $\gamma(s)$ where $f=0.$ It follows that $E$ is a nonempty open subset of $FE.$

Furthermore, since $f$ is real analytic in $D,$ $E$ is precisely the set of $t\in [0,1]$ where all derivatives of $f$ vanish at $\gamma(t).$ Thus $E$ is closed in $[0,1].$ It follows that $E$ is nonempty and clopen in $[0,1].$ Since $[0,1]$ is connected, $E=[0,1].$

This implies $f(y)=0.$ Since $y\in D$ was arbitrary, we have $f=0$ in $D.$ This is a contradiction, since $f$ was assumed to be nonconstant in $D.$ Therefore $\text {int } Z$ is empty, and $Z$ is nowhere dense as desired.

zhw.
  • 107,943
0

Indeed, more is true. It is enough to assume that $\mathcal Z$ is not nowehere dense to show that it has a limit point.

To see it, recall that a set is nowhere dense, by definition, if and only if its closure has empty interior. Then if $\mathcal Z$ is not nowhere dense, its closure has nonempty interior. Any of these points is a limit point of $\mathcal Z$.

EDIT: As Conrad points out in reply to Shi James' comment, to complete the proof for the given $\mathcal Z$, which is a zero set of an analytic (hence continuous) function, we need to notice first that $\mathcal Z$ is necessarily closed.

Giulio R
  • 3,267
  • I think for the contradiction proof, If we assume that it does not hold, then it should mean that " $\mathcal{Z}$ is not a closed set or $\mathcal{Z}$ is not a nowhere dense set", right? So I think we should also consider the case where $\mathcal{Z}$ is not closed but still nowhere dense. – Shi James Mar 19 '21 at 11:32
  • 2
    Zero set of continuous function is closed always ( pretty much by definition) – Conrad Mar 19 '21 at 11:37
  • Maybe more relevant to your question and comment, the statement "Assume that $\mathcal Z$ is not a closed nowhere dense set, then $\mathcal Z$ must contain a limit point" is not correct in itself I think. Indeed $\mathcal Z={1/n:\ n$ positive integer$}\subset\mathbb R$ is not a closed nowhere dense set (it is not closed), and it does not contain a limit point. – Giulio R Mar 19 '21 at 14:47
  • Merely having a limit point is not sufficient for the identity theorem for functions on $\mathbb{R}^n$, see e.g. this discussion – pabk Dec 17 '24 at 10:55