On a particular way to compare fields

91 Views Asked by At

I'm aware that the natural analogue of the Cantor–Schröder–Bernstein (CSB) theorem is not true for fields. That means it is possible for a pair of fields $E$ and $K$ to be such that $E$ is isomorphic to a subfield of $K$, $K$ is isomorphic to a subfield of $E$, but $E$ and $K$ are not isomorphic.

This question is about whether the CSB property holds with respect to a stricter way to compare fields.

Let $E$ and $K$ be fields. Let's say "$E\leq K$" iff $E$ is isomorphic to a subfield of $K$ and $K$ is algebraic over this subfield. More clearly, we say "$E\leq K$" if there is an injective field homomorphism $\phi:E\rightarrow K$ such that $K$ is algebraic over $f(E)$. Henceforth, we will refer to such a field homomorphism as an algebraic homomorphism.

First and main question: Is it true that if $E\leq K$ and $K\leq E$, then $E\cong K$?

Related second question: given a field $E$, if $f:E\rightarrow E$ is an algebraic endomorphism, must we have that $E=f(E)$? That is, must $f$ be an automorphism of $E$?

Counterexample: @Martin Brandenburg points out that for any field $k$, a counterexample to this second question is given by the map $f:k(T)\rightarrow k(T)$ that linearly extends $T\mapsto T^2$. $f$ is a field endomorphism, and $k(T)$ is algebraic over $f(k(T))$.

The answer to the second question is yes in the following special case:

Theorem 1: Let $E$ be a field, and let $f:E\rightarrow E$ be an algebraic field endomorphism; that is to say, $E$ is algebraic over $f(E)$.

Let $F=\{x\in E\:|\:f(x)=x\}$. Then $F$ is a subfield of $E$ that is fixed pointwise by $f$, and $F\subseteq f(E)$. Suppose that $E$ is algebraic over $F$.

In this special case, $E=f(E)$.

Proof:

For each $\alpha\in E$, we want to show that $\alpha\in f(E)$.

Let $\min_F(\alpha)$ be the unique monic irreducible polynomial with coefficients in $F$ of which $\alpha$ is a root, and let $d=\text{deg}_F(\alpha)$ be the degree of $\min_F(\alpha)$.

Let $S=\{f^n(\alpha)\}_{0\leq n\leq d}$, where by $f^n(\alpha)$ I mean $f(f(\cdots f(\alpha)\cdots))$.

Since $F$ is fixed pointwise by $f$, every element of $S$ is a root of $\min_F(\alpha)$. But $\min_F(\alpha)$ only has $d$ roots, whereas $S$ has $d+1$ elements. This means that for some $k>0$, $\alpha=f^k(\alpha)\in f(E)$, as desired. $\blacksquare$


How is the second question related to the first?

Suppose $f:E\rightarrow K$ and $g:K\rightarrow E$ are algebraic field homomorphisms. Then $K$ is algebraic over $f(E)$, so $g(K)$ is algebraic over $gf(E)$. Since $E$ is algebraic over $g(K)$, it follows that $E$ is algebraic over $gf(E)$. If the answer to the second question were "yes", then $E=gf(E)$, and so $g$ and $f$ would be a isomorphisms.

Here is a positive answer to the first question in a special case:

Theorem 2: Let $E$ and $K$ be fields, and let $f:E\rightarrow K$ and $g:K\rightarrow E$ be algebraic field homomorphisms. Let $F\subseteq E$ be defined by $F=\{x\in F \:|\:gf(x)=x\}$. Then $F\subseteq E$ is a subfield. If $E$ is algebraic over $F$, then we must have that $f$ and $g$ are actually isomorphisms between $E$ and $K$.

Proof:

Apply Theorem 1 to see that $gf(E)=E$, and thus $g:K\rightarrow E$ is an isomorphism. Since $gf$ and $g$ are isomorphisms, $f$ must also be an isomorphism. $\blacksquare$

This may seem like a rather useless special case, but this gives us the following nice result:

Corollary: Let $E$ and $K$ be fields. Suppose that $E$ and $K$ are both algebraic over their respective prime fields (their unique minimal subfields). If there are algebraic field homomorphisms $f:E\rightarrow K$ and $g:K\rightarrow E$, then $E$ and $K$ are isomorphic.

1

There are 1 best solutions below

0
On

You should be able to obtain a counterexample from any pair of isogenous but non-isomorphic elliptic curves: If $C$ and $D$ are such curves over any fixed field $K$, then an isogeny $f:C\to D$ furnishes an algebraic embedding of their function fields $f^*:K(D)\to K(C)$. Moreover, $f$ always comes with a dual isogeny $g:D\to C$, which gives the desired field embedding $g^*$ in the other direction. Hence, $K(D)$ and $K(C)$ are algebraic extensions of each other, which are not isomorphic over $K$ unless $C$ and $D$ are isomorphic curves. By picking $K=\mathbb{Q}$ (or any other prime field), we can thus ensure that the resulting function fields are not isomorphic overall.

Note that these function fields of curves $K(C)$ are of transcendence degree 1 over $K$ and are thus, by your own observation, in some sense minimal counterexamples. Moreover, there can be no counterexample using "simpler" curves, i.e. those of genus zero: Their function fields are all isomorphic to $K(T)$. By Lüroth's theorem, any field $L$ such that $K\subsetneq L\subseteq K(T)$ is of the form $L=K(f(T))$ for some rational function $f$ and therefore isomorphic to $K(T)$ itself.

Edit: Here is an explicit example taken out of Silverman: We take $K=\mathbb{Q}$ and consider the two curves $$C:y^2=x^3+x \hspace{5mm} \text{and} \hspace{5mm} D:Y^2=X^3-4X.$$ These curves are not isomorphic (since -4 is not a sixth power in $K$, see e.g. here) but are isogenous via the map $f:C\to D$ given by $$f(x,y)=\left(\frac{y^2}{x^2},\frac{y(1-x^2)}{x^2} \right).$$ The function fields ($K(C)=K(x,y)$ with $y^2=x^3-x$ and similarly for $K(D)=K(X,Y)$) together with the embedding $$f^*:K(D)\to K(C), X\mapsto \frac{y^2}{x^2},Y\mapsto \frac{y(1-x^2)}{x^2}$$ thus form a concrete instance of the described phenomenon.