We know that if $F$ is a field and $f(X)$ a non-zero polynomial in $F[X]$, then for every polynomial $g(X)$ we can find $q(X),r(X)$ such that $$g(X)=f(X)\cdot q(X)+r(X)$$ with $r(X)$ the zero polynomial or $\deg r(X)<\deg f(X)$. My question is: the same algorithm holds for many variables? So can i write $$g(X_1,\ldots,X_n)=f(X_1,\ldots,X_n)\cdot q(X_1,\ldots,X_n)+r(X_1,\ldots,X_n)$$


2 Answers 2


No, the polynomial division algorithm does not immediately generalize to multivariate rings. Here is a simple proof. Just as for $\rm\:\Bbb Z,\:$ a domain having an algorithm for division with smaller remainder, also enjoys Euclid's algorithm for gcds, which, in extended form, yields Bezout's identity. Therefore gcds have linear representation $\rm\:gcd(a,b) = r a + s b\:$ (i.e. Euclidean domains are Bezout). But this fails in multivariate polynomial rings $\rm\:F[x_1,\ldots, x_n],\ n \ge 2,\:$ since $\rm\:gcd(x_1,x_2) = 1\:$ but there is no Bezout equation $\rm\: 1 = x_1 f + x_2 g\ $ (evaluating at $\rm\:x_1 = 0 = x_2\,$ $\Rightarrow$ $\rm\,1 = 0\:$ in $\rm\,F,\,$ contra field axioms).

But all Euclidean is not lost, since one can generalize the polynomial division algorithm in a way that recovers many of the important properties. For this, look up the Grobner basis algorithm, which may be viewed as a nonlinear generalization of the Euclidean/division algorithm and Gaussian elimination.

  • $\begingroup$ Just to point out what I was missing at first: You can try to do polynomial division the way you are used to but then $deg\ r(X) < deg\ f(X)$ will not necessarily hold. $\endgroup$
    – Sarien
    Feb 29, 2016 at 14:21

If you can do Euclidean division in a reasonable sense in a domain $R$, then you can prove that $R$ is a principal ideal domain. This is not the case for the $F[x_{1}, \dots, x_{n}]$ for $n > 1$.

Please see the Wikipedia page for a clear and precise definition.

  • $\begingroup$ so,suppose $f(b_1,\ldots,b_n)=0$. Can i say that $X_i-b_i$ divides $f$ for every $i=1,\ldots,n$? $\endgroup$ Feb 28, 2013 at 10:35
  • 1
    $\begingroup$ $f(x_{1}, x_{2}) = (x_{1} - 1)^{2} + (x_{2} - 1)^{2} - 1$ has the root $(b_{1}, b_{2}) = (0, 1)$, but surely $x_{1}$ does not divide $f$, because the locus of $f$ does not contain the line $x_{1} = 0$. More generally, if $x-b_{1}$ divides $f(x_{1}, \dots, x_{n})$, then $f(b_{1}, a_{2}, \dots, a_{n}) = 0$ for all $a_{2}, \dots, a_{n}$, so the locus of $f$ contains the affine hyperplane $x_{1} = b_{1}$. $\endgroup$ Feb 28, 2013 at 10:45
  • 1
    $\begingroup$ @Federica But there are extremely useful multivariate generalizations of the division algorithm, such as the Grobner basis algorithm - see my answer. $\endgroup$
    – Math Gems
    Feb 28, 2013 at 16:47

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.