Bruns and Herzog in their book Cohen-Macaulay Rings, page 120 write: "The Hilbert-Burch theorem 1.4.17 identifies perfect ideals of grade $2$ as the ideals of maximal minors of certain matrices".

For the convenience of the community I restate the Hilbert-Burch theorem as given by Bruns and Herzog:

Hilbert-Burch Theorem [1.4.17 in B&H]: Let $R$ be a Noetherian ring and $I$ an ideal with a free resolution $F_{\bullet}:0 \rightarrow R^n \stackrel{\phi}{\rightarrow} R^{n+1} \rightarrow I \rightarrow 0$. Then there exists an $R$-regular element $a$ such that $I = a I_n(\phi)$, ($I_n(\phi)$ is the $n$-th Fitting invariant of $\phi$). If $I$ is projective, then $I=(a)$, and if $\operatorname{projdim}I=1$, then $I$ is perfect of grade $2$. Conversely, if $\phi:R^n \rightarrow R^{n+1}$ is an $R$-linear map such that $\operatorname{grade} I_n(\phi) \ge 2$, then $I=I_n(\phi)$ has the free resolution $F_{\bullet}$.

My question: Now let $I$ be a perfect ideal of grade equal to $2$. Since $I$ is perfect we must have that $\operatorname{projdim} R/I = 2$. How does it follow then from the Hilbert-Burch theorem that $I$ is $I_n(\phi)$ for some $\phi$ and some $n$? How do we construct this $\phi$?


It follows from the first part of the statement when the ring is local. Consider the following minimal free resolution for $R/I$: $$ 0 \to R^m \to R^n \to R \to R/I \to 0. $$ Since rank is additive, we see that $m + 1 = n$. Then the map $\phi: R^m \to R^n$ can be written as $m \times n$ matrix. The Theorem says $I = aI_n(\phi)$ where $a$ is an regular element. If $a$ is not a unit, then height of $I$ is at most $1$. Therefore, $a$ is a unit and $I = I_n(\phi)$.

  • $\begingroup$ Dear Youngsu, could you please expand the details of your answer? In particular i) how do we see that the rank of $R/I$ is zero? ii) why have you taken $R$ as the first free module in the free resolution, i.e. why did you not take a free resolution $0 \rightarrow R^m \rightarrow R^n \rightarrow R^k \rightarrow R/I$? iii) why if $a$ is not a unit then $I$ has height at most $1$ and why this can not be true? $\endgroup$ – Manos Mar 20 '14 at 17:53
  • $\begingroup$ @Manos: 1) Since the grade of $I$ is $2$, $I$ contains a regular element. Therefore, $Quot(R) \otimes I \cong Quot(R)$. 2) I guess you could, but then how are you going to apply the Theorem? I wanted to make the kernel of the map $R^k \to R/I$ to be $I$. 3) $aI_n(\phi) \subseteq (a)$ and the height of $(a)$ is at most one if $a$ is not a unit. $\endgroup$ – Youngsu Mar 20 '14 at 19:44
  • $\begingroup$ I am good with 1) and 3). Regarding 2) my question is how do you know that if you choose a projective resolution of $R/I$ to start with $R$, then this projective resolution will have length equal to $2$? Because the projective dimension of $R/I$ is equal to $2$ that means that there exists some projective resolution of length $2$, but not any projective resolution needs to have length equal to $2$. Right? $\endgroup$ – Manos Mar 20 '14 at 22:04
  • $\begingroup$ @Manos: I see. I believe what you mentioned is delicate. In fact I should change my answer since I do not know how to construct such a resolution. My answer only works when the ring is local. Then the minimal free resolution can be used. However, I do not think one can patch them in general. I'm sorry for confusion. $\endgroup$ – Youngsu Mar 20 '14 at 23:43
  • $\begingroup$ Your answer has still some interesting parts, so i will upvote it. But maybe you can add that it is valid for local rings. In fact B&H may have the local case in their mind even though they do not say so explicitely. $\endgroup$ – Manos Mar 21 '14 at 0:14

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.