1
$\begingroup$

Theorem (Eisenstein Criterion). Let $P$ be a prime ideal of the integral domain $R$ and let $f(x) = x^n +a_{n-1}x^n+\cdots+a_1 x +a_0$ be a polynomial in $R [x]$ (here $n \geq 1$). Suppose $a_{n-1},\dots, a_1, a_0$ are all elements of $P$ and suppose $a_0$ is not an element of $P^2$. Then $f(x)$ is irreducible in $R[x]$.

Proof. Suppose $f(x)$ were reducible, say $f(x) = a(x)b(x)$ in $R[x]$, where $a(x)$ and $b(x)$ are nonconstant polynomials. Reducing this equation modulo $P$ and using the assumptions on the coefficients of $f(x)$ we obtain the equation $x^n = \overline{a(x)b(x)} $ in $(R/ P)[x]$, where the bar denotes the polynomials with coefficients reduced mod $P$. Since $P$ is a prime ideal, $R/P$ is an integral domain, and it follows that both $\overline{a(x)}$ and $\overline{b(x)}$ have $0$ constant term, i.e., the constant terms of both $a (x)$ and $b(x)$ are elements of $P$ . But then the constant term $a_0$ of $f(x)$ as the product of these two would be an element of $P^2$, a contradiction. $\Box$


This proof is from Abstract Algebra by Dummit & Foote. The puzzling point is bold-italic in the proof. Why it is not possible that only one constant of $\overline{a(x)}, \overline{b(x)}$ is $0$ and the other is not? I don't think it violates the rule of integral domain in this case.

Could any one give me some idea ? Thank you in advance!

$\endgroup$
1
  • $\begingroup$ The proof as given is either incomplete or incorrect. This is often a stumbling point for students first learning the proof, so it is pedagogically inexcusable to omit justification of this crucial point. $\endgroup$ – Bill Dubuque Nov 3 '20 at 12:26
1
$\begingroup$

Match coefficients. You obtain a system of equations where if $\deg a = i$ and $\deg b = j$ then $i+j = n$, and, denoting their coefficients by $\alpha_s$ and $\beta_t$, you have $\alpha_i \beta_j = 1$ and then intermediary sums of cross terms, all of which must be $0$. Then the proof as in Dummit and Foote resumes - if these are zero, they are elements of $P$. Ideal arithmetic says that such products are elements of $P^2$, obtaining the contradiction.

Edit: I did not see your more refined question at the end. Suppose that $a(x)$ has $0$ constant term, and it's $b(x)$ that doesn't. Then $a \mod P$ would have to be identically $0$. If not, you could look at its lowest nonzero coefficient and this would eventually be multiplied by the constant term in $b \mod P$. The result is a nonzero monomial with $\deg < n$ in the product $ab \mod P$, since $P$ is a prime ideal. But that means that the whole polynomial $a(x)$ is in $P$, and therefore the whole polynomial is in $P$ since ideals are closed under multiplication. But $f$ is monic, so this is impossible.

$\endgroup$
5
  • $\begingroup$ By matching coefficients I get $$ \begin{cases} \alpha _0\beta _0=0,\\ \alpha _1\beta _0+\alpha _0\beta _1=0,\\ \alpha _2\beta _0+\alpha _1\beta _1+\alpha _0\beta _2=0,\\ \cdots\\ \alpha_i \beta_j=0\\ \end{cases} $$ The coefficients of cross sums are zero indeed, but how to make sure those $\alpha_s$ and $\beta_t$ when $\alpha<i$ and $\beta<j$ are all zero? $\endgroup$ – atlantic0cean Nov 3 '20 at 3:10
  • $\begingroup$ I've answered what your question actually is now, I think. Sorry about that.. $\endgroup$ – Alfred Yerger Nov 3 '20 at 3:39
  • $\begingroup$ @Algred Yerger Thank you very much for your contribution! I think the proof by Dummit & Foote is actually deficient while your proof works. $\endgroup$ – atlantic0cean Nov 3 '20 at 4:47
  • $\begingroup$ Certainly they could have elaborated some more on this point. $\endgroup$ – Alfred Yerger Nov 3 '20 at 5:19
  • $\begingroup$ @atlantic0cean I added a proof which greatly clarifies the inferences you ask about above. $\endgroup$ – Bill Dubuque Nov 3 '20 at 11:14
2
$\begingroup$

That proof of Eisenstein's criterion is clearer using unique factorization of prime products.

$\rm\color{#0a0}{Assume}$ that $f$ is reducible $\,f = g\,h,\,$ $\,i = \deg g,\ j = \deg h,\ \color{#c00}{i,j\ge 1}$, wlog $\,g,h\,$ monic.

This factorization maps to $\,x^n = \bar g\,\bar h\,$ in $\,\bar R := R/P,\,$ a domain, so $\,x\,$ is prime in $\bar R[x]$.

Since it is a prime power: $\,x^n = x^k\, x^{n-k}\,$ are the only possible monic factorizations.

Thus $\,\bar g = x^i,\ \bar h = x^j,\,$ so $\,\color{#c00}{i,j\ge 1}\Rightarrow\,\bar g(0)\!=\!0\!=\!\bar h(0),\,$ i.e. $\,g(0),h(0)\in P$.

Therefore $\,f(0) = g(0)h(0)\in P^2,\,$ contra $\rm\color{#0a0}{hypothesis},\,$ hence $f$ is irreducible.

Alternative direct inductive proof that $\,\bar g(0)\!=\!0\!=\!\bar h(0),\,$ i.e. $\,x\mid \bar g,\bar h.\,$ If not, wlog $\, x\nmid \bar g,\,$ then $\,x^n = \bar g\,\bar h\,$ $\Rightarrow\,x\mid \bar g\,\bar h,\,$ so $\,x\nmid \bar g\Rightarrow x\mid \bar h,\,$ by $\,x\,$ prime. Repeating this shows all $\,n\,$ factors of $\,x\,$ must divide into $\,\bar h,\,$ so $\,n = \deg \bar h = \deg h,\,$ so $\,\color{#c00}i = \deg g \color{#c00}{= 0},\,$ contra hypothesis.

$\endgroup$
1
  • $\begingroup$ Thanks for your explicit answer! BTW how do you change the color of text? $\endgroup$ – atlantic0cean Nov 3 '20 at 12:49

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.