I have problems with this:

I need to prove that in the polynomial ring the radical of an ideal generated by monomials is also generated by monomials.

I found a proof on internet that uses the convex hull of the multidegrees of the monomials, but I want a proof that uses less terminology. For example, can be proved in a simple way the following:

Given a monomial $u=x^a = \prod\limits_{k = 1}^n {x_k ^{\alpha_k}}$ we define $\sqrt u = \prod\limits_{k = 1}^n {x_k }$. How can I prove that if $G(I)$ is a minimal set of generators of $I$ ( I proved that this set it's also a monomial), then the set $\left\{\sqrt u: u \in G\left(I\right) \right\}$ generates the radical of $I$?


Claim: If $m_1,\ldots,m_s$ are monomials in $K[x_1,\ldots,x_n]$, then $$\sqrt{\langle m_1,\ldots,m_s\rangle} = \langle \sqrt{m_1},\ldots,\sqrt{m_s}\rangle.$$

Proof: Put $k_i:=(\text{greatest exponent of any variable in }m_i)\in\mathbb{N}$, i.e. if $m_i=x_{j_1}^{a_1}\cdots x_{j_l}^{a_l}$ then $k_i=\max\{a_1,\ldots,a_l\}$. Now put $k:=k_1+\cdots+k_s-s+1$. We have $\langle \sqrt{m_1},\ldots,\sqrt{m_s}\rangle^k\subseteq \langle m_1,\ldots,m_s\rangle$, because every term of $\prod_{j=1}^k(f_{1,j}\sqrt{m_1}+\cdots+f_{s,j}\sqrt{m_s})$ has the form $f\sqrt{m_1}^{\beta_1}\cdots\sqrt{m_s}^{\beta_s}$ where $b_j\in\mathbb{N}_0$ and $\beta_1+\cdots+\beta_s=k$, which means at least one $\beta_j\geq k_j$. Therefore $$\langle \sqrt{m_1},\ldots,\sqrt{m_s}\rangle^k\subseteq\langle m_1,\ldots,m_s\rangle\subseteq\langle\sqrt{m_1},\ldots,\sqrt{m_s}\rangle/\sqrt{~},$$ $$\sqrt{\langle \sqrt{m_1},\ldots,\sqrt{m_s}\rangle}\subseteq\sqrt{\langle m_1,\ldots,m_s\rangle}\subseteq\sqrt{\langle\sqrt{m_1},\ldots,\sqrt{m_s}\rangle}.$$ Thus it remains to show that $\sqrt{\langle \sqrt{m_1},\ldots,\sqrt{m_s}\rangle}=\langle \sqrt{m_1},\ldots,\sqrt{m_s}\rangle$, i.e. that squarefree monomial ideals are radical.

If $\sqrt{m_1}=x_{j_1}\cdots x_{j_l}$, we have $\langle \sqrt{m_1},\ldots,\sqrt{m_s}\rangle=\bigcap_{r=1}^l\langle x_{j_r},\sqrt{m_2},\ldots,\sqrt{m_s}\rangle$, because by point e) from my post here, $$\langle x_{j_1},\sqrt{m_2},\ldots,\sqrt{m_s}\rangle\cap\langle x_{j_2},\sqrt{m_2},\ldots,\sqrt{m_s}\rangle= \sum\sum\langle\mathrm{lcm}(\ast,\ast)\rangle= \langle x_{j_1}x_{j_2},\sqrt{m_2},\ldots,\sqrt{m_s} \rangle.$$ Next, $\langle \sqrt{m_1},\ldots,\sqrt{m_s}\rangle=\bigcap_r\bigcap_{r'}\langle x_{j_r},x_{j_{r'}},\sqrt{m_3},\ldots,\sqrt{m_s}\rangle$, and so on. Therefore $\langle \sqrt{m_1},\ldots,\sqrt{m_s}\rangle = \bigcap_\lambda\mathfrak{p}_\lambda$ for some ideals $\mathfrak{p}_\lambda$, generated by variables. But $\mathfrak{p}_\lambda$ are prime by b), hence $\sqrt{\langle \sqrt{m_1},\ldots,\sqrt{m_s}\rangle} = \sqrt{\cap_\lambda\mathfrak{p}_\lambda}=\cap_\lambda\sqrt{\mathfrak{p}_\lambda}=\cap_\lambda\mathfrak{p}_\lambda$, since prime ideals are radical. $\blacksquare$

| cite | improve this answer | |
  • 1
    $\begingroup$ Where did the chain of inclusions $\sqrt{\langle \sqrt{m_1},\ldots,\sqrt{m_s}\rangle}\subseteq\sqrt{\langle m_1,\ldots,m_s\rangle}\subseteq\sqrt{\langle\sqrt{m_1},\ldots,\sqrt{m_s}\rangle}.$ come from? (why did the $k$ disappear? $\endgroup$ – lego323 Mar 1 '17 at 15:03

Let $J=\oplus J_i$ be a $\mathbb{Z}$ or $\mathbb{N}$ graded module over a graded ring $R=\oplus R_i$ (this question is such a case). Denote by $J^\ast$ the ideal generated by homogenous elements of $J$.

Whenever we write something in its unique grade decomposition, we will write $x=\sum x_i$ where $x_i$ is homogeneous of grade $i$.

Key idea: By definition, an element $x\in J$ is in $J^\ast$ iff $x_i\in J$ for all $i$.

Proposition: $(\sqrt{J})^\ast=\sqrt{J^\ast}$.

Lemma: If $z$ is a homogenous element of $J$, then $z\in \sqrt{J}\iff z\in\sqrt{J^\ast}$.

Proof of lemma: $z\in \sqrt{J}\iff \exists n, z^n\in J\iff \exists n, z^n\in J^\ast\iff z\in \sqrt{J^\ast} $ QED.

Proof of proposition: By the lemma, any homogenous element of $\sqrt{J}$ is in $\sqrt{J^\ast}$, and so $(\sqrt{J})^\ast\subseteq\sqrt{J^\ast}$.

We now proceed to show equality via a contradiction. Suppose there exists an $x=\sum x_i\in \sqrt{J^\ast}\setminus(\sqrt{J})^\ast $. If one of the $x_i$ is in $\sqrt{J}$, then $x-x_i$ is still in $\sqrt{J^\ast}\setminus(\sqrt{J})^\ast $. WLOG then, replace $x$ with what is left after subtracting all of the $x_i$ that are already in $\sqrt{J}$.

Suppose momentarily that $x$ has a strictly positive grade term, and let $x_{\max}$ be the term of highest positive grade. By definition, there exists an $n$ such that $x^n\in J^\ast$, so every homogenous term of $x^n$ is in $J$. But the largest grade term in $x^n$ is $x_{\max}^n$, so $x_{\max}^n\in J$, and consequently $x_{\max}\in\sqrt{J}$. This is a contradiction since $x_{max}$ is not in $\sqrt{J}$. Therefore $x$ has no strictly positive grade term.

If the grading is By a similar argument, $x$ has no strictly negative grade term. Thus, $x$ is homogeneous of grade zero, and hence is in $(\sqrt{J})^\ast$: a contradiction!

| cite | improve this answer | |
  • $\begingroup$ I could not think of a way to simplify the non-obvious direction! Can someone explain how/if it can be dramatically simplified? $\endgroup$ – rschwieb Jun 5 '12 at 2:02
  • 3
    $\begingroup$ Here it's only proved that the radical of a homogeneous ideal is homogeneous. $\endgroup$ – user26857 Apr 14 '13 at 8:06

You can do it as follows:

Given $F \in K[X_1,\ldots,X_n] \setminus \{0\}$ one defines its support, $\mathrm{supp}(F)$ as the set of monomials of its distributive form. It is useful (not only for this exercise) to prove that:

If $I$ is a non-zero ideal of $K[X_1,\ldots,X_k]$, then $I$ is monomial $\iff \forall F \in I$, $\mathrm{supp}(F) \subseteq I$.

In this case, you use this characterization and perform induction in the Newton diagram of polynomial (non-null powers of polynomial), $N(F)$.

The case $N(F) = 1$ is direct. The case $N(F) > 1$ you use the induction hypothesis by removing the leading term of your polynomial.

| cite | improve this answer | |

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.