Let $f(t)$ be an irreducible polynomial in $k(x_1, \ldots, x_n)[t]$. Show that $f(t)$ is irreducible in $k[x_1,\ldots,x_n][t]$.

I guess it's clear that I am required to apply Gauss' Lemma. The statement in my course notes is:

Let $R$ be a UFD and $F$ its field of fractions. A non-constant polynomial in $R[X]$ is irreducible in $R[X]$ if and only if it is both irreducible in $F[X]$ and primitive in $R[X]$. And a polynomial $P$ in $R[X]$ is primitive if the only elements of $R$ that divide all coefficients of $P$ at once are the invertible elements of $R$.

I first cleared the denominators of the coefficients $a_i \in k(x_1,\ldots,x_n)$ of $f(t)$. But this modified polynomial is not monic anymore, so I ran into trouble showing that $f$ is primitive in $k[x_1,\ldots,x_n][t]$ - or at least, things get very messy.

* * *

EDIT: Additional context added. I am trying to understand Proposition 4.9 in Hartshorne's Algebraic Geometry. The argument given goes as follows:

  • Let $K$ be a finitely generated extension field of $k$. Then $K$ is seperably generated over $k$ since we are working over an algebraically closed field. Hence we can find a transcendence base $x_1,\ldots,x_n \in K$ such that $K$ is a finite seperable extension of $k(x_1,\ldots,x_n)$. By the primitive element therem we can find one further element $y \in K$ such that $K=k(x_1,\ldots,x_n,y)$. Now $y$ is algebraic over $k(x_1,\ldots,x_n)$. So it satisfies a polynomial equation with coefficients which are rational functions in $x_1,\ldots,x_n$. Clearing denominators we get an irreducible polynomial, satisfying $f(x_1,\ldots,x_n,y)=0$

  • My original question aimed to clarify this last step. I assumed he used Gauss' Lemma.

EDIT 2: Reformulation of original question (related to comments on first post).

Let $f(t)$ be an irreducible polynomial in $k(x_1, \ldots, x_n)[t]$. Let $f'(t)$ be the polynomial where we have cleared up the denominators of $f(t)$. Show that $f'(t)$ is irreducible in $k[x_1,\ldots,x_n][t]$.

  • $\begingroup$ If $f$ is monic, then there can be no non-invertible element that divides all coefficients of $f$. $\endgroup$ – Alexander Aug 28 '13 at 19:23
  • $\begingroup$ But what after modifying? Say $f(t)=a_0 t^n + a_1 t^{n-1} + a_n$ with $a_i = f_i/g_i$ and $a_0=1$, then we have after modifying $a_0 = \prod g_i$, $a_1 = f_1 \cdot \prod_{i \neq 1} g_i$, $a_2 = f_2 \cdot \prod_{i \neq 2} g_i$, etc. $\endgroup$ – yannickvda Aug 28 '13 at 19:31
  • $\begingroup$ What are $f_i, g_i$? $\endgroup$ – Alexander Aug 28 '13 at 19:34
  • $\begingroup$ Since $a_i \in k(x_1,\ldots,x_n)$, each $a_i$ can be written as a quotient of polynomials $f_i$ and $g_i$. $\endgroup$ – yannickvda Aug 28 '13 at 19:35
  • $\begingroup$ Are you missing an assumption? Should you have $f \in k[x_1, ..., x_n][t]$? After all, $\frac{t}{x}$ isn't even an element of $k[x][t]$. $\endgroup$ – Alexander Aug 28 '13 at 19:40

From the comments, it seems like you want to ask the following question. Suppose $f(t) \in k(x_1,...,x_n)[t]$ is irreducible (where $k$ is a field). We can assume that the coefficients are reduced so that for each coefficient $\frac{c_i}{d_i}$, the numerator and denominator are relatively prime. Then $k[x_1, ..., x_n]$ is a UFD so we can take the least common multiple $m$ of the denominators of the coefficients of $f(t)$ and gcd $n$ of all the numerators of the coefficients. Let $f'(t)=\frac{m}{n}f(t) \in k[x_1,...,x_n][t]$. We are in an integral domain and $\frac{m}{n} \neq 0$ so the roots of $f'$ are the same as the roots of $f$. If $f$ is irreducible in $k(x_1, ..., x_n)[t]$, is $f'(t)$ irreducible in $k[x_1,...,x_n][t]$?

Suppose $p$ is a prime factor dividing each coefficient $\frac{m}{d_i} \cdot \frac{c_i}{n}$ of $f'$. Then for each $i$, $p|\frac{m}{d_i}$ or $p|\frac{c_i}{n}$. If in all cases the latter is true, then $p| c_i$ for each $i$ so this contradicts the definition of $n$. Thus, $p|\frac{m}{d_i}$ for some $i$, and in particular, $p|m$. Thus, if $d_i$ contains the largest power of $p$ of all the denominators, then $p$ does not divide $\frac{m}{d_i}$ so must divide $\frac{c_i}{n}$. But then $p|c_i$ which contradicts $p|d_i$ (and we assumed $\frac{c_i}{d_i}$ is reduced). Thus, $f'$ is primitive and so if we can show it is irreducible in $k(x_1, ..., x_n)[t]$, then it is irreducible in $k[x_1, ..., x_n][t]$ by Gauss's Lemma.

But in particular, if $gh=f'$, then by Gauss's Lemma, we can assume $g, h \in k[x_1, ..., x_n][t]$. Then we have $(ng)h=mf$. Each prime factor of $m$ must divide $ng$ or $h$, and so canceling prime factors, we end up with a factorization of $f$.

I haven't checked this too carefully. Does this seem right?


Your assertion is false already for $n=1$:

The polynomial $f(t)=x+xt\in k(x)[t]$ is irreducible in $k(x)[t]$, like all degree one polynomials over a field (which here is $k(x)$).
However it is reducible in $k[x][t]=k[x,t]$ since it factors as $f(t)=x+xt=x(1+t)$.
The apparent paradox is eexplained by the fact that $x$ is invertible in $k(x)[t]$ and so doesn't count as a factor of a factorization, whereas it is not invertible $k[x][t]$ and thus counts as a bona fide factor in a factorization.

This post answers the original question, namely:
"Let $f(t)$ be an irreducible polynomial in $k(x_1, \ldots, x_n)[t]$. Show that $f(t)$ is also irreducible in $k[x_1,\ldots,x_n][t]$. "


For clarity I'll post a new answer in order to explain Hartshorne's reasoning.

We have a monic irreducible polynomial $g(x_1,x_2,\cdots,x_n,Y)\in k(x_1,\cdots,x_n)[Y]$ with $g(x_1,x_2,\cdots,x_n,y)=0$.
We can write $g(x_1,x_2,\cdots,x_n,Y)=Y^s+a_1Y^{s-1}+\cdots a_s$ with $a_i=a_i(x_1,x_2,\cdots,x_n)\in k(x_1,\cdots,x_n)$.
Then we write $a_i=\frac {b_i}{d}$ where $b_i=b_i(x_1,x_2,\cdots,x_n),d=d(x_1,x_2,\cdots,x_n)\in k[x_1,\cdots,x_n]$ are polynomials.
In other words we have considered a common denominator $d$ for all our rational functions $a_i$.

The equation $g(x_1,x_2,\cdots,x_n,y)=y^s+a_1y^{s-1}+\cdots a_s=0$ after multiplication by $d$ then becomes : $$ dy^s+b_1y^{s-1}+\cdots b_s=0 $$ Dividing $d,b_1,\cdots b_s$ by their gcd $g$ in $k[x_1,\cdots,x_n]$, we get $\delta y^s+\beta _1 y^{s-1}+\cdots \beta _s=0$, with $\delta=\frac {d}{g}, \beta_i=\frac {b_i}{g}\in k[x_1,\cdots,x_n]$.
Hartshorne's polynomial is then $f(x_1,\cdots,x_n,Y)=\delta Y^s+\beta _1 Y^{s-1}+\cdots \beta _s\in k[x_1,\cdots,x_n][Y]$.

This polynomial $ f(x_1,\cdots,x_n,Y)$ is indeed irreducible in $k[x_1,\cdots,x_n][Y]$ by the criterion in your course note since it is primitive in $k[x_1,\cdots,x_n][Y]$ and irreducible in $k(x_1,\cdots,x_n)[Y]$.
(Recall that $\frac {\beta_i}{\delta}=\frac {b_i}{d}=a_i$ , so that $f=g \delta$ is indeed irreducible in $k(x_1,\cdots,x_n)[Y]$)


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.