5
$\begingroup$

Of course “several” in the title means $n$ strictly greater than 1 and the function is defined on some open subset of $\mathbb{C}^n$.

I tried to use the Weierstrass preparation theorem because it’s the only result on analysis of several variables I know but I couldn’t find a contradiction (I believe the answer is “no”).

$\endgroup$
8
$\begingroup$

The answer is indeed no, which might be surprising since the real analogue of this question has a different outcome ($f(x) = ||x||^2$ is smooth and only has one zero).

The crucial result in this proof is Hartog's Theorem:

Let $n \geq 2$ and $\Omega \subset \mathbb C^n$ be open. Suppose $K \subset \Omega$ is compact and $\Omega \setminus K$ is connected. Then any holomorphic function on $\Omega \setminus K$ can be extended to all of $\Omega$.

Remark: This is not true in one variable: The function $z \mapsto e^{1/z}$ cannot be extended to all of $\mathbb C$ because it has an essential singularity at zero.

We are now going to prove that no holomorphic function of several variables has just one zero.
Toward a contradiction, let $\Omega \subset \mathbb C^n$ be a (connected) domain and let $f: \mathbb{\Omega} \rightarrow \mathbb C$ be holomorphic with $f(z) = 0$ if and only if $z = z_0$. Then the function $1/f$ is holomorphic in $\Omega \setminus \{z_0\}$. By Hartog's Theorem, $1/f$ admits a holomorphic extension to $\Omega$. But then, $f(z_0) \cdot 1/f(z_0) = 1$ by continuity of the product of continuous functions, which is a contradiction to $f(z_0) = 0$.

| cite | improve this answer | |
$\endgroup$
6
$\begingroup$

The answer is no for a function defined on an open subset of the plane. It actually comes down to the following algebra fact :

Let $A$ be a normal ring, then $$A=\cap_{ht(\mathfrak{p})=1} A_{\mathfrak{p}}.$$

where if $\mathfrak{p}$ is a prime ideal of $A$, the height of $\mathfrak{p}$ denoted by $ht(\mathfrak{p})$ is the maximal number of prime ideals contained in $\mathfrak{p}$ that form a strictly increasing chain $$\mathfrak{p_1} \subsetneq\cdots\subsetneq \mathfrak{p_s}\subsetneq \mathfrak{p}.$$

And $A_{\mathfrak{p}}$ is the localization of $A$ with respect to $\mathfrak{p}$.

A proof can be found in Matsumura's "Commutative Algebra".

Now, you may suppose that your open subset of the plane is connected, hence the ring of holomorphic functions on $U$ is normal (this is an exercise). Denote it by $A$ and denote by $K$ its fraction field i.e., the field of meromorphic functions on $U$.

My claim is that if $f$ is holomorphic and has only one point as zero locus, then $1/f$ is holomorphic, which yealds to a contradiction.

But then I only have to prove that $1/f$ is in all the $A_{\mathfrak{p}}$ for all primes of height one.

To do this, I say that the elements of a certain prime of height one of $A$ are exactly the holomorphic functions vanishing on a certain codimension $1$ subset $Z$ of $U$ (this is the analytic version of the Nullstellensatz, again a good exercise).

Therefore $1/f$ is not in any $A_{\mathfrak{p}}$.

| cite | improve this answer | |
$\endgroup$
  • 1
    $\begingroup$ As is, your argument does not work since the result you quote from Matsumura's book (12.4. (i)) requires some noetherianity assumption. $\endgroup$ – Gaussian Nov 21 '19 at 13:12
  • 1
    $\begingroup$ Indeed, there is a step missing in the argument. When localized at any maximal ideal $m$, $A_m$ is noetherian. Then apply the argument to $A_m$ and use the fact that $A=\cap A_m$. $\endgroup$ – Carot Nov 21 '19 at 23:57
  • $\begingroup$ Indeed, it fills the gap. $\endgroup$ – Gaussian Nov 22 '19 at 7:38
2
$\begingroup$

The vanishing set of an holomorphic function $\Bbb{C}^{n}\to\Bbb{C}$ is an union of analytic hypersurfaces (locally biholomorphic to $\Bbb{C}^{n-1}$)

Use the n-polydisk Cauchy integral formula to show holomorphic implies analytic. Let $$f(w,z)=F_w(z)=\sum_n g_n(w) z^n, \qquad w\in \Bbb{C}^{n-1},z\in \Bbb{C}$$ be analytic non-constant and $f(0,0)=0$. The $g_n$ are analytic with a common non-zero radius of convergence.

If $F_0'(0) \ne 0$ it is easy : $F_w'(0)\ne 0$ for $w$ small thus with $r$ small enough from the residue thoerem $$h(w) = \frac1{2i\pi} \int_{|z|=r} \frac{z F_w'(z)}{F_w(z)}dz, \qquad f(w,h(w))=0$$

Otherwise $F_0'(0) =0$, assume $F_0^{(k)}(0)\ne 0$ then from the argument principle the number of zeros of $F_w$ near $z=0$ is the number of times the curve $F_w(r e^{it}),t\in [0,2\pi]$ encloses $0$, for $w$ small it is $\le k$ and it is continuous in $w$.

There may be several hypersurfaces passing through $w=0$ but for $w\ne 0$ small the hypersurfaces separate and replacing $\int_{|z|=r}$ by $\int_\gamma$ we get the equations of our hypersurfaces.

The remaining case is $F_0$ identically zero : thus $f(z,w) = z^m f_2(z,w)$ and if $f_2(0,0)=0$ then $f_2$ satisfies the previous hypothesis.

| cite | improve this answer | |
$\endgroup$
2
$\begingroup$

Suppose for simplicity that $f$ is holomorphic in an open set $U\subset \mathbb C^2$ with $(0,0)\in U.$ Assume that $f$ has an isolated zero at $(0,0).$ Then we can choose $r>0$ such that $f$ is holomorphic in $D(0,2r)^2$ and $f\ne 0$ in $D(0,2r)^2\setminus \{(0,0)\}.$

For $w\in D(0,r)$ define

$$g(w)=\frac{1}{2\pi i}\int_{|z|=r} \frac{D_zf(z,w)}{f(z,w)}\,dz.$$

The argument principle shows that $g(0)$ is a nonzero integer $m,$ while $g(w)= 0$ for $0<|w|<r.$ Note also that $g$ is a continuous function of $w.$ Thus $g(0) = m = \lim_{z\to 0} g(w)=0,$ which is a contradiction.

| cite | improve this answer | |
$\endgroup$

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.