6
$\begingroup$

I'm wondering if anyone has come across the following ring before. Let $R$ be the ring of complex power series $f=\sum_{n \ge 0} a_n t^n$ such that

  1. $a_n \in \mathbb{Q} \: \: \forall \: n$
  2. The function $f: \mathbb{C} \rightarrow \mathbb{C}$ is entire

Any information about it would be welcome but in order to keep the question specific, I'll list a few things I would like to know in particular.

  1. Is $R$ a unique factorisation domain?
  2. Are there any elements of $R$ which are known to be prime/irreducible but are not associates of polynomials?
  3. If an element $f \in R$ is prime/irreducible, must all its zeroes (in $\mathbb{C}$) be simple?
  4. Is every $\alpha \in \mathbb{C}$ a root of some nonzero $f \in R$ ?
$\endgroup$

2 Answers 2

12
$\begingroup$

Relevant here is the paper of Hurwitz:

Über beständig convergirende Potenzreihen mit rationalen Zahlencoefficienten und vorgeschriebenen Nullstellen. Acta Math. 14 (1890), no. 1, 211–215.

Since the results of that paper are simple and short, I will explain them here. Let $f$ be any entire function. Hurwitz proves that there exists another entire function $g$ such that $f \cdot e^{g}$ (which is entire) has coefficients in $\mathbf{Q}(i)$. The proof is quite easy. Assume (for convenience, one can reduce to this case) case that $f(0) \ne 0$. Then $\log(f(z))$ is holomorphic in a neighbourhood of zero, and so one can write:

$$\log f(z) = b_0 + b_1 z + b_2 z^2 + \ldots $$

Since $\mathbf{Q}(i)$ is dense in $\mathbf{C}$, there exists a function $h(z) = a_0 + a_1 z + a_2 z^2 + \ldots $ so that:

  1. $a_0 = 0$,
  2. $g(z) = h(z) - \log f(z)$ is entire,
  3. $h(z)$ has coefficients in $\mathbf{Q}(i)$.

For example, one can take $a_k$ for $k \ge 1$ to be $1/k!$ times the nearest Gaussian integer to $k! \cdot b_k \in \mathbf{C}$, so $|a_k - b_k| < 1/k!$ is decreasing fast enough to ensure that $g(z)$ is entire. It follows that

$$f \cdot e^g = e^{h(z)}.$$

By construction, the function $h(z)$ has trivial constant term and coefficients in $\mathbf{Q}(i)$. It follows easily that so does $f \cdot e^g$. Now we also observe that $f \cdot e^g$ has the same zeros as $f$. So the properties of ideals in the ring of entire functions with coefficients in $\mathbf{Q}(i)$ is very similar to the properties of ideals in the ring of entire functions. Similarly, the ring of entire functions with coefficients in $\mathbf{Q}$ is very similar to the properties of ideals in the ring of entire functions with coefficients in $\mathbf{R}$ (equivalently, entire functions with $\overline{f(z)} =f(\overline{z})$).

Now recall:

  1. If $S \subset \mathbf{C}$ is any subset of $\mathbf{C}$ which is closed and discrete, there exists an entire function with zeros exactly at $S$ to any prescribed multiplicity. (One can simply write down such a function as an infinite product.)

  2. If $S \subset \mathbf{C}$ is any subset as above such that $\overline{S} = S$, then there exists an entire function with zeros exactly at $S$ to any prescribed multiplicity with coefficients in $\mathbf{R}$, as long as the multiplicity at $s$ is the same as at $\overline{s}$. By Part I, there exists a complex entire function $f = \sum a_n z^n$. The conjugate $\overline{f}(z) = \sum \overline{a_n} z^n$ is a second function. It follows that $$\sqrt{f(z) \cdot \overline{f}(z)}$$ is the required function. (This squareroot exists as an entire function because the order of every zero of $f(z) \cdot \overline{f}(z)$ is even.)

In particular, Hurwitz proves that:
1. If $\alpha$ is any real number, there exists an entire function with rational coefficients that has a simple zero at $\alpha$ and nowhere else.
2. If $\beta$ is any complex number, there exists an entire function with rational coefficients that has a simple zero at $\beta$ and $\overline{\beta}$ and nowhere else.

The construction of these functions can even be said to be quite explicit.

Functions of type 1. and 2. are both prime/irreducible, and consist of all the irreducible elements up to units. (Clearly if $f(z)$ is not a unit, it either has a real zero or a pair of conjugate complex zeros, and so is divisible by such a function.)

We deduce:

  1. $R$ is not a UFD, because $\sin(z)$ does not factor into irreducibles (any factorization of $\sin(z)$ into finitely many factors has to contain at least one function with infinitely many zeros, which is not irreducible.)

  2. Yes; any function of type 1 or 2 where $\alpha$ is not rational or $\beta$ is not a quadratic imaginary will do.

  3. Yes, by the above, if $f(z)$ is irreducible, not only are the zeros of $f(z)$ simple, but there are at most two of them! (In particular, most "irreducible polynomials" are not irreducible in $R$.)

  4. Yes, there is even such a function with zeros only at $\alpha$ and $\overline{\alpha}$.

Some extra facts which are easy to deduce from this construction: $R$ is a Bezout domain: any finitely generated ideal is principal.

$\endgroup$
4
$\begingroup$

A start, but perhaps better than nothing:

  1. If $r$ is a rational number, then $\sin(r t)$ and $\cos(r t)$ are elements of your ring. There is an identity

$$\sin(r t) = 2 \sin(r/2 t) \cos(r/2 t).$$

Neither element on the right is a unit. This implies that the sequence of ideals:

$$ (\sin(t)) \subsetneq (\sin(t/2)) \subsetneq (\sin(t/4)) \subsetneq \cdots $$

is an infinite ascending chain of distinct ideals. So $R$ is not Noetherian (and hence definitely not a PID).

$\endgroup$
0

You must log in to answer this question.

Start asking to get answers

Find the answer to your question by asking.

Ask question

Explore related questions

See similar questions with these tags.