48
$\begingroup$

Can we prove that every entire one-to-one function is linear?

$\endgroup$
45
$\begingroup$

You can rule out polynomials of degree greater than $1$, because the derivative of such a polynomial will have a zero by the fundamental theorem of algebra, and a holomorphic function is $(n+1)$-to-$1$ near a zero of its derivative of order $n$.

To finish, you need to rule out entire functions that are not polynomials. If $f$ is such a function, then $f(1/z)$ has an essential singularity at $z=0$. To see that this implies that $f$ is not one-to-one, you could apply Picard's theorem as yoyo indicates. Or you could proceed as follows. By Casorati-Weierstrass, $f(\{z:|z|>n\})$ is dense in $\mathbb{C}$ for each positive integer $n$. By the open mapping theorem, the set is open. By Baire's theorem, $D=\bigcap_n f(\{z:|z|>n\})$ is dense in $\mathbb{C}$. In particular, $D$ is not empty, and every element of $D$ has infinitely many preimage points under $f$.


I just realized that there is an easier way to apply Casorati-Weierstrass, with no need for Baire. If $f$ is entire and not a polynomial, then $f(\{z:|z|<1\})$ is open, and $f(\{z:|z|>1\})$ is dense. Therefore these sets have nonempty intersection. Every element of the intersection has at least $2$ preimage points.

$\endgroup$
  • $\begingroup$ Nice application of Baire. Thanks! $\endgroup$ – Digital Gal Mar 29 '11 at 19:24
15
$\begingroup$

By shifting $z$, without loss of generality you can assume $f(0) = 0$. By the open mapping theorem, $f(z)$ maps some open set $U$ containing $0$ to another one, call it $V$. Since $f(z)$ is to be one-to-one, $f(z)$ can't map any $z$ outside of $U$ to $V$. Thus ${1 \over f(z)}$ is bounded outside of $U$. Therefore ${z \over f(z)}$ is an entire function that grows no faster than linearly: $|{z \over f(z)}| < A|z| + B$ for some $A$ and $B$.

It's easy from here to show that $g(z) = {z \over f(z)}$ is linear; for any $z_0$ ${g(z) - g(z_0) \over z - z_0}$ must be bounded and therefore is a constant by Liouville's theorem. So ${z \over f(z)} = c_1z + c_2$ for some $c_1$ and $c_2$. Hence $f(z) = {z\over c_1z + c_2}$. Since $f(z)$ has no poles and is nonconstant, $c_1$ must be zero and $c_2$ nonzero. We conclude that $f(z) = {1 \over c_2} z$.

$\endgroup$
  • $\begingroup$ Why f(z) has simple zero at z=0? $\endgroup$ – nicksohn Dec 30 '15 at 10:33
  • $\begingroup$ Nice proof. In the beginning, you say wlog wma $f(z)=0$. Maybe you mean $f(0)=0$? $\endgroup$ – perlman Oct 12 '17 at 5:55
  • $\begingroup$ @nicksohn That follows from $f(z)$ being one to one $\endgroup$ – Zarrax Oct 12 '17 at 17:40
  • $\begingroup$ @MathewJames yes, thanks, corrected it. $\endgroup$ – Zarrax Oct 12 '17 at 17:40
11
$\begingroup$

Let $f:\mathbb C\to\mathbb C$ entire and injective. Let $U=f(\mathbb C)$. $U$ is an open subset of the plane.

$U$ is simply connected: indeed, to check this it is enough to show that the integral of every analytic function on $U$ along every closed curved in $U$ is zero, and you can do this by "changing variables using $f$".

Next, if $U\subsetneq\mathbb C$, from Riemann's theorem we know that there is an biholomorphic map $U\to D$, with $D$ the unit disc. Composing with $f$, we get a biholomorphic map $\mathbb C\to D$, and this is impossible. We see then that $f$ is in fact bijective and, in fact, an homeomorphism. Composing with a translation, we can assume that $f(0)=0$.

Using this, one can see that the function $1/f(z)$ is bounded at $\infty$ and has a simple pole at $0$, so $g(z)=z/f(z)$ is entire and bounded by a function of the form $cz$ for some constant $c$. Using Cauchy's estimates for the Taylor coefficients of $g$, we see that $g$ is a polynomial of very low degree. Translating this to information about $f$, we can conclude what we want.

(This avoids Picard but uses Riemann... :( )

$\endgroup$
  • $\begingroup$ How do we know that the pole at 0 is of order 1, from the fact that f(0)=0? What if 0 was zero of order n>1 for f(z)...thanks, @Mariano Suárez-Alvarez. $\endgroup$ – User001 Dec 18 '14 at 6:18
9
$\begingroup$

I'll give the "usual" proof.

Note that by Little Picard, $f$ misses at most one point; but it is a homeomorphism onto its image, and the plane minus a point is not simply connected. Thus $f$ is onto $\mathbb{C}$, and hence bijective. Then $f$ has a holomorphic inverse, which is enough to imply $f$ is proper, that is, the pre-image of a compact set is compact. This in turn implies $$ \lim_{z\rightarrow\infty} f(z)=\infty,$$ and thus if we define $f(\infty)=\infty$, $f$ becomes a Möbius transformation of the Riemann sphere. So $f$ has the form $f(z) = \frac{az+b}{cz+d},$ and it is easy to see that if $f$ is entire on $\mathbb{C}$, then $c=0$.

$\endgroup$
  • $\begingroup$ I should also add that I believe there is a proof using just Little Picard and the Cauchy-Riemann equations. I just forget all the details :-). Basically you show $f$ is a Euclidean similarity on $\mathbb{R}^2$, so it has the form $kAz+b$, where $A$ is $2\times2$ orthogonal. Since $f$ is holomorphic, $A\in SO(2)$. Using the well known form of such matrices, you then show $A$ is rotation by some angle $\theta$, so $f(z)=ke^{i\theta}z+b$. $\endgroup$ – user641 Mar 29 '11 at 20:36
  • $\begingroup$ Why is it that $c=0$ if $f$ is entire? $\endgroup$ – user225477 Feb 16 '17 at 18:33
  • $\begingroup$ @Zermelo's_Choice, I think otherwise $f$ would have a pole at the point $z$ such that $cz+d=0$. $\endgroup$ – user135520 Apr 27 '18 at 16:43
5
$\begingroup$

This proof uses only the Open Mapping Theorem and Cauchy's Estimates.

Suppose $f$ is entire and injective. Consider the injective entire function $$\widetilde{f} \colon z \mapsto f(z) - f(0).$$ We see that $\widetilde{f}(0) = 0$. Let $r > 0$, and consider the ball $B(0,r)$ of radius $r$ centered at the origin. By the Open Mapping Theorem, $\widetilde{f}(B(0,r))$ is an open set containing the origin. By openness, there is some $\epsilon > 0$ such that $B(0,\epsilon) \subset \widetilde{f}(B(0,r))$. Since $\widetilde{f}$ is injective, no point of $B(0,r)^c$ may be sent to any point of $B(0,\epsilon)$. In other words, $$|z| \geq r \Rightarrow |\widetilde{f}(z)| \geq \epsilon.$$ Now $\widetilde{f}$ has a zero at the origin, so let's say this zero has order $k \geq 1$. This means there is some entire function $g \colon \Bbb C \rightarrow \Bbb C$ such that $g(0) \neq 0$ and $$\widetilde{f}(z) = z^k g(z) \ \forall \ z \in \Bbb C.$$ Since $\widetilde{f}$ is injective, $\widetilde{f}$ has no other zeroes apart from its zero at the origin. So $g$ has no zeroes. This means the function $F \colon \Bbb C \rightarrow \Bbb C$ defined by $$F(z) = \frac{1}{g(z)} \ \forall \ z \in \Bbb C$$ is entire and also has no zeroes. Write the Taylor expansion of $F$ at the origin as $$F(z) = \sum_{n=0}^\infty \frac{F^{(n)}(0)}{n!}z^n.$$ Given any $R > r$, we can apply Cauchy's Estimates on the circle of radius $R$ to obtain bounds on the derivatives of $F$ at the origin. Indeed, $$F^{(n)}(0) \leq \max_{|z| = R}|F(z)|\frac{n! }{R^n}.$$ But we can apply the fact that $|\widetilde{f}(z)| \geq \epsilon \ \forall \ |z| \geq r$ to notice that $$\max_{|z| = R} |F(z)| = \max_{|z| = R} \frac{1}{|g(z)|} = \max_{|z| = R} \frac{|z|^k}{|\widetilde{f}(z)|} \leq \frac{R^k}{\epsilon}.$$ So if $n > k$, we have $$F^{(n)}(0) \leq \max_{|z| = R}|F(z)|\frac{n!}{R^n} \leq \frac{R^k n!}{\epsilon R^n} = \frac{n!}{\epsilon R^{n-k}} \xrightarrow{R \to \infty} 0.$$ So we conclude from the Taylor expansion of $F$ that $$F(z) = \sum_{n=0}^k \frac{F^{(n)}(0)}{n!} z^n$$ is a polynomial of degree at most $k$. But as previously noted, $F$ has no zeroes. So $F \equiv c$ for some $c \in\Bbb C \backslash\{ 0\}$! This means $g \equiv c^{-1}$, and furthermore, $$\widetilde{f}(z) = c^{-1}z^k \ \forall \ z \in \Bbb C.$$ Supposing $k \geq 2$, the polynomial $z^k - 1 \in \Bbb C[z]$ has at least two roots $\xi_1, \xi_2 \in \Bbb C$ which are certainly non-zero on account of the fact that $0^k \neq 1$. The multiplicity of the root $\xi_1$ is precisely $1$ because $$\frac{d}{dz}\Big\vert_{z=\xi_1} z^k - 1 = k \xi_1^{k-1} \neq 0.$$ This allows us to conclude that $\xi_1 \neq \xi_2$, and that $$\widetilde{f}(\xi_1) = c^{-1} \xi_1^k = c^{-1} \cdot 1 = c^{-1} \xi_2^k = \widetilde{f}(\xi_2).$$ This result contradicts the fact that $\widetilde{f}$ is injective. So $k=1$, and our final conclusion is that $$f(z) = \widetilde{f}(z) + f(0) = c^{-1} z + f(0) \ \forall \ z \in \Bbb C,$$ where for completeness's sake I recall that $c$ was necessarily non-zero.

$\endgroup$
  • 2
    $\begingroup$ Very nice! And you can even add that $c=1/(f(1)-f(0))$. $\endgroup$ – Unit Mar 3 '16 at 21:15
4
$\begingroup$

Here is a (longer) proof using very little of complex analysis. Assume that $f:\mathbb{C}\to\mathbb{C}$ is holomorphic and injective. The function $f$ extends to a holomorphic map of the Riemann sphere to itself, $\mathbb{CP}^1\to\mathbb{CP}^1$. Indeed, choose any $z_0$ such that $f'(z_0)\neq 0$ (if it doesn't exist then $f$ is constant (of course from injectivity we know that actually $f'\neq0$ everywhere)). Then a small neighbourhood $U\ni z_0$ is mapped bijectively to a small neighbourhood $V\ni f(z_0)$. The function $1/(f(z)-f(z_0))$ is therefore bounded in $\mathbb{C}-U$, hence by Riemann removable singularity theorem it extends to a holomorphic function on $\mathbb{CP}^1-U$, therefore $f$ extends to a holomorphic map $\mathbb{CP}^1\to\mathbb{CP}^1$.

As an application of Liouville's theorem, any holomorphic map $F:\mathbb{CP}^1\to\mathbb{CP}^1$ such $F(z)\neq\infty$ for $z\neq\infty$ is a polynomial. If we wish, we can also avoid Liuoville theorem and use some topology. If the order of pole of $f$ at $\infty$ is $>1$ then $f$ is not injective in the neighbourhood of $\infty$. Hence there is $a\in\mathbb{C}$ such that $f-az$ is holomorphic $\mathbb{CP}^1\to\mathbb{C}$, hence it's bounded (being a map from a compact space), hence it's a constant: if $f-az$ is not constant then it is a map $\mathbb{CP}^1\to\mathbb{CP}^1$ which is of positive degree but which is not surjective (as it avoids $\infty$).

$\endgroup$

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.