6

I am stuck on this complex analysis problem.

Let $f$ be an entire function and $n$ a positive integer. Show that there exists an entire function $g$ such that $f=g^n$ if and only if the order of each zero of $f$ is divisible by $n$.

I can see that locally around each $z_0$ we can write $f(z)=(z-z_0)^{nk}s(z)$ where $s(z)$ has no zeroes in a neighborhood of $z_0$ and so there exists a logarithm of $s(z)$, say $l(z)$, which means that $g(z)=(z-z_0)^ke^{l(z)/n}$ works, but I don't understand how to get an entire function from this. I know that it suffices to show that this construction gives functions that agree on the overlap of these neighborhoods, but I don't see why they would have to agree.

2 Answers2

1

In order to prove the result, we will first construct an $n^{th}$ root in any open disc $D(r)$ of radius $r$, $r > 0$ around the origin. Then, by taking a sequence of values of $r$ that go off to infinity, we can patch together the roots to get an entire function.

First, it helps to recall a slightly less local logarithm construction. The proof of the following theorem works in almost the exact same way as the usual proof of the local existence of a logarithm in a simply connected domain (i.e. by constructing a primitive $F$ such that $F'(z) = \frac{f'(z)}{f(z)}$).

Theorem: If $f(z)$ is a holomorphic function defined on some domain $\Omega$ such that $f(z)$ is nonvanishing on this domain, then there is a holomorphic function $g(z)$ defined on $\Omega$ such that $e^{g(z)} = f(z)$ for all $z \in \Omega$.

Now, let $r > 0$ be fixed. Since $\overline{D(r)}$ is compact, we know that $f$ can have finitely many zeroes in $\overline{D(r)}$ and hence finitely many zeroes in $D(r)$.

Now, let $z_1, \ldots, z_N$ be the finitely many zeroes of the given entire function $f$ in $D(r)$. Let the order of $z_i$ be $nk_i$. Now, let $$\tilde{f}(z) = \frac{f(z)}{(z - z_1)^{nk_1} \cdots (z - z_N)^{nk_N}}$$

Then $\tilde{f}$ is non-vanishing on $D(r)$ since it is clearly nonvanishing at the $z_i$, and entire since it is a quotient of meromorphic functions with the same order at each $z_i$.

Thus, there is a logarithm $\ell$ of $\tilde{f}$ defined on the simply connected domain $D(r)$. We have that $e^{\ell(z)} = \tilde{f}(z)$, so $\tilde{g}(z) = e^{\frac{\ell(z)}{n}}$ satisfies $\tilde{g}(z)^n = \tilde{f}(z)$. Now, since $f(z) = \tilde{f}(z) (P(z))^n$, where $P(z)$ is the polynomial $\prod_i (z-z_i)^{k_i}$. Thus, letting $g(z) = \tilde{g}(z)P(z)$, we have our desired $n^{th}$ root.

Now, we can "patch together" our $n^{th}$ roots on each $D(N)$, $N \in \mathbb{N}^{+}$. Define $g_1$ to be the $n^{th}$ root constructed above on $D(1)$.

Assume that we've successfully defined $g_k$ such that for all $\ell < k$, $g_k|_{D(\ell)} = g_{\ell}$. Now, let $g'_{k+1}$ be an $n^{th}$ root for $f$ on $D(k+1)$. It may not be the case that $g'_{k+1}|_{D(k)} = g_k$. However, we have that on $D(k)$:

$$ \left(\frac{g'_{k+1}(z)}{g_k(z)}\right)^n = \frac{g'_{k+1}(z)^n}{g_k(z)^n} = \frac{f(z)}{f(z)} = 1 $$

Thus, for all $z \in D(k)$, we have that $h(z) = \frac{g'_{k+1}(z)}{g_k(z)}$ satisfies $h(z)^n = 1$, and thus $h(z) \in \{e^{2\pi i \frac{m}{n}}\}_{m = 1, \ldots, n}$. Since $h(z)$ is holomorphic and thus in particular continuous on $D(k)$, we have that $h(z)$ is constant of value $\zeta \in \{e^{2\pi i \frac{k}{n}}\}_{k = 1, \ldots, n}$ for all $z \in D(k)$. Thus, let $g_{k+1} = \zeta^{-1}g'_{k+1}$. Then, for $z \in D(k)$, we have $g_{k+1}(z) = \zeta^{-1} h(z) g_k(z) = g_k(z)$.

Now for any $\ell \leq k$, we have that $g_{k+1}|_{D(\ell)} = (g_{k+1}|_{D(k)})|_{D(\ell)} = g_k|_{D(\ell)} = g_\ell$ by the induction hypothesis. Thus, we construct by induction $g_N$ for all $N \in \mathbb{N}^+$. Finally, we may define $g(z)$ by: $g(z) = g_N(z)$ for any $N$ such that $z \in D(N)$. By the above proof, we know that these are all equal, so this is well-defined. In addition, if $z \in D(N)$, then some small neighborhood $U$ of $z$ is in $D(N)$ as well. Thus, $g(z)|_U = g_N(z)|_U$, and this is holomorphic on $U$. Thus, $g(z)$ is entire. Since $g_N(z)^n = f(z)$ for $z \in D(N)$, we have that $g$ satisfies $g(z)^n = f(z)$ for all $z \in \mathbb{C}$.

The converse is easy to see by a local argument. Let $f = g^n$ where $f, g$ are entire. If $z_0$ is a zero of $f$, then in some neighborhood $U$ of $z_0$, $f(z) = (z-z_0)^m s(z)$ where $s(z)$ is nonvanishing in $U$. Since $g$ must vanish at $z_0$ but not anywhere else on $U$, we must have $g(z) = (z-z_0)^k t(z)$ for some $k \geq 1$ and $t(z)$ nonvanishing on $U$. Thus, $f(z) = (z-z_0)^{nk} t(z)^n$. Since $t(z)^n$ is nonvanishing on $U$, we must have that $t(z)^n = s(z)$ and thus $nk = m$, so the order $m$ of the zero $z_0$ is divisible by $n$.

Dorebell
  • 4,154
1

The one direction is clear, if $f = g^n$, and $z_0$ is a zero of $f$, then $z_0$ is a zero of $g$, say of order $k$, so in a neighbourhood of $z_0$ we have $g(z) = (z-z_0)^k\cdot h(z)$ with a holomorphic nonzero $h$. Then in that neighbourhood $f(z) = (z-z_0)^{nk}\cdot h(z)^n$, and of course $h(z)^n$ is nonzero there. So indeed $n$ divides the order of the zero of $f$ at $z_0$.

Conversely, let all zeros of $f$ have orders divisible by $n$. The case $f \equiv 0$ is trivial, so in the following we assume $f\not\equiv 0$. Consider the entire meromorphic function

$$h_0(z) = \frac{f'(z)}{f(z)}.$$

Since $f$ is entire, the poles of $g$ are exactly the zeros of $f$. Let $P$ be the pole set of $g$ and $U = \mathbb{C}\setminus P$.

All poles of $h_0$ are simple, with the residue of $h_0$ in $z_0$ being the order of the zero of $f$ at $z_0$: If $f(z) = (z-z_0)^m\cdot k(z)$ with a holomorphic nonzero $k$ in a neighbourhood of $z_0$, then

$$\frac{f'(z)}{f(z)} = \frac{m(z-z_0)^{m-1}\cdot k(z) + (z-z_0)^m\cdot k'(z)}{(z-z_0)^m\cdot k(z)} = \frac{m}{z-z_0} + \frac{k'(z)}{k(z)}$$

in that neighbourhood, and $\frac{k'}{k}$ is holomorphic there since $k$ is nonzero.

All residues of $h_0$ are multiples of $n$, hence the entire meromorphic function $h = \frac{1}{n}\cdot h_0$ has integer residues.

Now choose an $a \in U$, and for every $z\in U$ choose a - piecewise continuously differentiable - path $\gamma_z \colon [0,1] \to U$ with $\gamma_z(0) = a$ and $\gamma_z(1) = z$. Then define $\lambda \colon U \to \mathbb{C}$ by

$$\lambda(z) := \int_{\gamma_z} h(\zeta)\,d\zeta.$$

The function $\lambda$ depends on the choice of the path $\gamma_z$, but if we choose a different path $\delta_z$ from $a$ to $z$, then the composition $\gamma_z\delta_z^{-1}$ is a closed path in $U$, and $n(\gamma_z\delta_z^{-1},\omega) \neq 0$ for only finitely many $\omega\in P$. Hence

\begin{align} \int_{\gamma_z} h(\zeta)\,d\zeta - \int_{\delta_z} h(\zeta)\,d\zeta &= \int_{\gamma_z\delta_z^{-1}} h(\zeta)\,d\zeta\\ &= 2\pi i \sum_{\omega\in P} n(\gamma_z\delta_z^{-1},\omega)\operatorname{Res}(h;\omega)\\ &\in 2\pi i\cdot \mathbb{Z}, \end{align}

so a different choice of paths alters $\lambda(z)$ by an integer multiple of $2\pi i$, whence $g_0 \colon U\to \mathbb{C}$ given by

$$g_0(z) = \exp(\lambda(z))$$

is independent of the choice of paths. The independence of the choice of paths shows that $g_0$ is holomorphic: For $w\in U$ and $r > 0$ such that $D_r(w) \subset U$, we can define

$$\tilde{\lambda}(z) = \lambda(w) + \int_w^z h(\zeta)\,d\zeta$$

on $D_r(w)$, which is holomorphic, and we have

$$\exp(\tilde{\lambda}(z)) \equiv \exp(\lambda(z))$$

on $D_r(w)$ by the independence of $g_0$ from the choice of paths. Also, this shows that $g_0'(z) = \frac{f'(z)}{nf(z)}\cdot g_0(z)$.

Now we see that $f\cdot g_0^{-n}$ is constant on $U$ by differentiation:

$$(f\cdot g_0^{-n})'(z) = f'(z)\cdot g_0^n(z) - nf(z)\cdot g_0^{n-1}(z)\cdot g_0'(z) = f(z)g_0^n(z)\biggl( \frac{f'(z)}{f(z)} - n\frac{g_0'(z)}{g_0(z)}\biggr) = 0.$$

Hence, for a suitable $c\in \mathbb{C}\setminus \{0\}$, $g(z) := c\cdot g_0(z)$ is a holomorphic $n$-th root of $f$ on $U$.

But for $\omega\in P$ we have

$$\lim_{z\to \omega} g(z) = 0,$$

hence these are all removable singularities and $g$ extends to an entire function.

Daniel Fischer
  • 206,697
  • I am thinking about to use Hadamard’s factorization theorem/Weierstrass Factorization Theorem to prove the second part, which might be shorter proof. I am not sure whether it is correct way to do that. – Ariel So Dec 27 '19 at 03:22
  • 1
    @ArielSO Of course one can do that (Weierstraß in any case, Hadamard if there is a function of finite order with the prescribed zeros) if one has the factorisation theorems available. And the proof will (can) be shorter since much of the argument went into the proof of the factorisation theorem. By Weierstraß there is an entire function $g_1$ such that $g_1^n$ has the same zeros (including multiplicities) as $f$, then $f/g_1^n$ is entire without zeros, hence has an $n^{\text{th}}$ root $g_2$, thus $g = g_1g_2$ is entire with $g^n = f$. – Daniel Fischer Dec 27 '19 at 10:32
  • Thank you!! It looks much simpler now. – Ariel So Dec 27 '19 at 19:41