Let $P$ be a polynomial of degree $n>0$ with coefficients in $\mathbb{C}$. By the Fundamental Theorem of Algebra, for any complex $y$, there exist $n$ complex numbers $z$ such that $P(z)=y$ (possibly with multiplicity). So, we can consider $P$ to be a sort of "$n$-injective" function, in that, for each output, there exist $n$ inputs that give that output.


We can assign each complex $c$ with a "degree" $\mu(z)\leq n$ in accordance with the multiplicity of the root $z=c$ in the polynomial $P(z)-P(c)$.

Does there exist a (almost) partition of $\mathbb{C}$ into $n$ Borel sets such that:the polynomial $P$ is bijective from each set to $\mathbb{C}$ and each complex number $z$ appears in exactly $\mu(z)$ of the sets? This would almost be a partition since, for almost all $z$, $\mu(z)=1$ (in fact, $\mu(z)\neq 1$ only for a finite number of $z$ as that would require $P'(z)=0$).

I've chosen Borel as the qualifier as, as far as I can tell, it's the best way to define a "nice set" - obviously if the quantifier is taken away the problem is true - but I'm not sure if I'm using it properly.


The statement is trivially true if $n=1$ (all non-constant linear polynomials are bijective, so one can simply take the set to be $\mathbb{C}$). If $n=2$, the polynomial can be completed into $P(z)=a\left((z-b)^2+c\right)$ for some complex $a,b,c$, so we can choose our sets to be, if $b=x+iy$ where $x,y\in\mathbb{R}$,

  1. The half-plane above $\Im(z)=y$ and the ray $\Re(z)\geq x, \Im(z)=y$

  2. The half-plane below $\Im(z)=-y$ and the ray $\Re(z)\leq x, \Im(z)=y$.

However, this is all based on the fact that, as functions, all quadratic polynomials act like $z^2$. So this sort of thing is, as far as I can see, not readily generalizable to $n>2$.

Is this a known problem (and if so, where can I read more about it)? Am I missing something simple?


$P$ defines an $n$-to-$1$ map $X\to Y$ where $X=\mathbb C\setminus \{z\mid P'(z)=0\}$ and $Y=\mathbb C\setminus \{P(z)\mid P'(z)=0\}.$ Consider a simply-connected set $Y'\subseteq Y,$ a point $y\in Y',$ and a choice $x_j\in P^{-1}(\{y\}).$ The function $P$ can be inverted locally at $y$ so that $y$ is sent to $x_j,$ and this extends to an inverse on $Y'.$ Call the image $X_j.$ The images $X_1,\dots,X_n$ cannot intersect so they will be disjoint open sets in $X.$ If the closure of $Y'$ is $Y,$ then the closures of $X_1,\dots,X_n$ will cover $X.$

Restricting to a simply connected set $Y'$ ensures the extension of $P^{-1}$ is well-defined - it avoids the monodromy around the branch points $\{P(z)\mid P'(z)=0\}$.

Assuming all the branch points ($P(z)$ with $P'(z)=0$) have different imaginary part, we could define $Y'$ by branch cuts going in the direction of $+\infty,$ i.e. $Y'=\mathbb C\setminus \{P(z)+r \mid P'(z)=0, r\geq 0\}.$ Then $X$ gets carved up along the preimages of these branch cuts. To get the boundary partitioned we could specify that $X_j$ includes the boundary points coming from above the branch cuts in $Y'$ - the limits coming from a continuous choice of $P^{-1}(P(z)+r+i\epsilon)$ as $\epsilon\to 0$. In the simple case $P(z)=z^n,$ each set $X_j$ would then correspond to $\left\{z\in\mathbb C\mid z\neq 0, \arg(z)\in \left[\tfrac{2\pi (j-1)}n,\tfrac{2\pi j}n\right)\right\}.$

Alternatively we can consider the uniformization of $Y$: a map of Riemann surfaces $\pi:D\to Y$ and a group $G$ acting on the universal cover $D$ such that $Y$ is the quotient $D/G.$ For example in the case $Y=\mathbb C\setminus \{0\}$ we can think of $D$ as being $\mathbb C$, with $\pi=\exp$ and with $G$ acting by translations $z\mapsto z+2\pi i.$ In this context what we are looking for is a nice fundamental domain and this problem has several nice solutions, though the boundary is usually dealt with differently. The surface $D$ comes with a Riemannian metric and after choosing a point $z_0\in D$ we can define the Dirichlet polygon

$$F=\{z\in D :d(z,z_{0})<d(z,gz_{0})\;\;\forall g\in G,g\neq 1\}$$

whose image $\pi(F)$ serves as the set $Y'$ above. The geometry of $F$ is quite tame. See Alan F. Beardon's "The Geometry of Discrete Groups" for more. I'm sure the boundary given by these sets could also be partitioned with a bit more care.

  • $\begingroup$ Thanks! This seems like a great answer, although I can't quite understand exactly what's going on as I'm not well-versed in this terminology. One question: when you say "this extends to an inverse on $Y'$" in your first paragraph, what do you mean? Do you mean that every value in $Y'$ is reached in the neighborhood of $x_j$ (and if so, why is that?) or do you mean something else entirely? $\endgroup$ – Carl Schildkraut Oct 19 '17 at 0:25
  • $\begingroup$ @CarlSchildkraut: $P^{-1}$ can be defined as a (single-valued) function near $y$ by inverting $P$ in a neighbourhood of $x_j.$ It can then be extended to the whole of $Y',$ e.g. by analytic continuation, which is unique by the monodromy theorem. Actually this is just a purely topological fact about covering maps. $\endgroup$ – Dap Oct 19 '17 at 6:47

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.