4

There's a remark in Infinite Integral Extensions and big Cohen--Macaulay Algebras by Hochster and Huneke which I cannot verify. Let's first define some notation.

Let $I$ be an ideal of a ring $R$ of characteristic $p>0$ prime. Let $I^{[p]}$ be the ideal generated by all $p$-th powers of elements in $I$. Note that it is generated by the $p$-th powers of a set of generators for $I$.

Now let $x_1,\dots,x_k$ be a regular sequence in $R$, and $I = (x_1,\dots,x_k)$, then the following two ideals are equal: $$I^{[p]}:I = I^{[p]} + (x_1\cdots x_k)^{p-1}R.$$ The statement is trivial when $k = 1$, and so the natural idea would be to mod out by $x_1$ (or $x_1^p$ or something) and then try to use induction, but I am getting no luck.

1 Answers1

2

I will prove the following stronger statement by induction on $k$.

Claim: if $R$ is any ring (commutative with $1$), $x_1, \dots, x_k$ is a regular sequence in $R$, and we are given integers $0 \leq d_i \leq e_i$ for all $1 \leq i \leq k$, then we have $$ (x_1^{e_1}, \dots, x_k^{e_k}) : (x_1^{d_1}, \dots, x_k^{d_k}) = (x_1^{e_1}, \dots, x_k^{e_k}, x_1^{e_1 - d_1} \cdots x_k^{e_k - d_k}). $$ (Of course this recovers your claim by taking $R$ to be of characteristic $p$ with all $e_i = p$ and all $d_i = 1$.)

Step 0: Setup

For brevity, we will write the three ideals above as $I, J$, and $K$ respectively; we want to show $I : J = K$. We clearly have $K \subseteq I : J$. For the reverse containment, we induct on $k$; the base case $k=1$ is trivial. Suppose we have an integer $k > 1$ such that the statement is true for regular sequences of length $k-1$ (in any ring), and fix a regular sequence $x_1, \dots, x_k$ in some ring $R$. Recall that then $x_1^{f_1}, \dots, x_k^{f_k}$ is also a regular sequence for any integers $f_i \geq 1$.

Suppose we are given an element $r$ of $I : J$; that is, $r \in R$ such that $r x_i^{d_i} \in I$ for all $i$. We must show that $r \in K$; as a warning, I will repeatedly modify $r$ by elements of $K$ in the course of the proof.

Step 1: first use of inductive hypothesis

Our assumption implies that the image of $r$ in $R/(x_1^{e_1})$ lies in the colon ideal $$ (x_2^{e_2}, \dots, x_k^{e_k}) : (x_2^{d_2}, \dots, x_k^{d_k}), $$ which by the inductive hypothesis equals $$ (x_2^{e_2}, \dots, x_k^{e_k}, x_2^{e_2 - d_2} \cdots x_k^{e_k - d_k}). $$

Lifting back up to $R$, we have $r \in (x_1^{e_1}, x_2^{e_2}, \dots, x_k^{e_k}, x_2^{e_2 - d_2} \cdots x_k^{e_k - d_k})$. Since all but the last generator of this ideal lie in $K$, we may adjust $r$ to assume without loss of generality that $r \in (x_2^{e_2 - d_2} \cdots x_k^{e_k - d_k})$. So set $r = ax_2^{e_2 - d_2} \cdots x_k^{e_k - d_k}$. To show that $r \in K$, it will suffice to show that $a \in (x_1^{e_1 - d_1}, x_2^{d_2}, \dots, x_k^{d_k})$.

Step 2: calculations with $r x_1^{d_1}$

The assumption $r \in I : J$ tells us in particular that $r x_1^{d_1} \in I$; that is, that

$$ \tag{1} a x_1^{d_1} x_2^{e_2 - d_2} \cdots x_k^{e_k - d_k} = b_1 x_1^{e_1} + \cdots + b_k x_k^{e_k} $$

for some $b_i \in R$. Our next goal is to simplify equation (1) slightly by removing its last term, at the cost of modifying $a$ (and thus $r$) as well as $b_1, \dots, b_{k-1}$. To accomplish this, we set \begin{align*} I' & = (x_1^{e_1}, \dots, x_{k-1}^{e_{k-1}}), \\ I_1' & = I' + (x_1^{d_1}), \\ I_2' & = I' + (x_2^{e_2 - d_2}), \\ & \dots \\ I_{k-1}' & = I' + (x_{k-1}^{e_{k-1} - d_{k-1}}), \end{align*} and consider equation (1) separately modulo each of the ideals $I_1', \dots, I_{k-1}'$. Working modulo $I_1'$ gives $$ 0 \equiv 0 + \cdots + 0 + b_k x_k^{e_k} \pmod{I_1'}, $$ which implies that $b_k \in I_1'$. (Reason: $x_k^{e_k}$ is a non-zero-divisor mod $I_1'$, because $x_1^{d_1}, x_2^{e_2}, \dots, x_{k-1}^{e_{k-1}}, x_k^{e_k}$ is a regular sequence--unless one of the exponents is $0$, in which case $I_1'$ is the unit ideal and thus it automatically contains $b_k$.) The analogous calculations modulo $I_2', \dots, I_{k-1}'$ show that $b_k$ lies in each of these ideals.

Now, since $b_k$ lies in each of the ideals $I_1', \dots, I_{k-1}'$, it follows that multiplying it by any of the elements $$ x_1^{e_1 - d_1}, x_2^{d_2}, \dots, x_{k-1}^{d_{k-1}} $$ yields an element of $I'$; that is, $b_k$ lies in the colon ideal $$ I' : (x_1^{e_1 - d_1}, x_2^{d_2}, \dots, x_{k-1}^{d_{k-1}}). $$ Our inductive hypothesis implies that this ideal equals $$ (x_1^{e_1}, \dots, x_{k-1}^{e_{k-1}}, x_1^{d_1} x_2^{e_2 - d_2} \cdots x_{k-1}^{e_{k-1} - d_{k-1}}) = I' + (x_1^{d_1} x_2^{e_2 - d_2} \cdots x_{k-1}^{e_{k-1} - d_{k-1}}). $$ Thus we may write $b_k$ as $$ b_k = c_1 x_1^{e_1} + \cdots + c_{k-1} x_{k-1}^{e_{k-1}} + c x_1^{d_1} x_2^{e_2 - d_2} \cdots x_{k-1}^{e_{k-1} - d_{k-1}}. $$ Plugging this back into equation (1), we may absorb the terms involving $c_1 x_1^{e_1}, \dots, c_{k-1} x_{k-1}^{e_{k-1}}$ into the other terms in the right-hand side of (1), and thereby assume without loss of generality that $$b_k = c x_1^{d_1} x_2^{e_2 - d_2} \cdots x_{k-1}^{e_{k-1} - d_{k-1}}.$$ Then equation (1) says: $$ a x_1^{d_1} x_2^{e_2 - d_2} \cdots x_k^{e_k - d_k} = b_1 x_1^{e_1} + \cdots + b_{k-1} x_{k-1}^{e_{k-1}} + c x_1^{d_1} x_2^{e_2 - d_2} \cdots x_{k-1}^{e_{k-1} - d_{k-1}} x_k^{e_k}, $$ or after moving the last term to the left, $$ (a - c x_k^{d_k}) x_1^{d_1} x_2^{e_2 - d_2} \cdots x_k^{e_k - d_k} = b_1 x_1^{e_1} + \cdots + b_{k-1} x_{k-1}^{e_{k-1}}. $$ We have now succeeded in removing the last term of equation (1)--at the cost of adjusting $b_1, \dots, b_{k-1}$, and adjusting $a$ by $-c x_k^{d_k}$. Since $r = ax_2^{e_2 - d_2} \cdots x_k^{e_k - d_k}$, this has the effect of adjusting $r$ by $$-cx_2^{e_2 - d_2} \cdots x_{k-1}^{e_{k-1} - d_{k-1}} \cdot x_k^{e_k},$$ which lies in $(x_k^{e_k})$ and thus in $K$. So there is no harm in renaming $a - c x_k^{d_k}$ as $a$ and similarly for $r$; we must still show that $r$ lies in $K$.

Step 3: finishing up

At this point, the modified equation (1) says that $$ a x_1^{d_1} x_2^{e_2 - d_2} \cdots x_k^{e_k - d_k} \in I' = (x_1^{e_1}, \dots, x_{k-1}^{e_{k-1}}). $$ Since $x_k^{e_k-d_k}$ is a non-zero-divisor mod $I'$, we may cancel it from the product on the left-hand side: $$ a x_1^{d_1} x_2^{e_2 - d_2} \cdots x_{k-1}^{e_{k-1} - d_{k-1}} \in I'. $$ We can simplify this further using some degenerate cases of our inductive hypothesis: the element $$ a x_2^{e_2 - d_2} \cdots x_{k-1}^{e_{k-1} - d_{k-1}} $$ lies in the colon ideal $$ I' : (x_1^{d_1}) = I' : (x_1^{d_1}, x_2^{e_2}, \dots, x_{k-1}^{e_{k-1}}) = I' : I_1', $$ which by the inductive hypothesis equals $$ I' + (x_1^{e_1 - d_1}) = (x_1^{e_1 - d_1}, x_2^{e_2}, \dots, x_{k-1}^{e_{k-1}}). $$ Repeating this argument to eliminate powers of $x_2, \dots, x_{k-1}$ in turn ultimately implies that $$ a \in (x_1^{e_1 - d_1}, x_2^{d_2}, \dots, x_{k-1}^{d_{k-1}}), $$ and thus $r = ax_2^{e_2 - d_2} \cdots x_k^{e_k - d_k} \in K$, as claimed.

  • This answer got a little longer than I was hoping, but I tried to make it fairly readable. Please ask if anything needs to be clarified. Where the argument came from: I managed to prove the inductive step for $k = 2$ directly, but for $k = 3$ I needed to use a stronger inductive hypothesis, allowing powers other than $p$ and $1$. So I started over and proved the stronger statement from the beginning. (I think this is called inductive loading.) As far as I can tell, the characteristic of $R$ only comes up in proving the equivalence you mention between definitions of $I^{[p]}$. – Ravi Fernando Jan 13 '23 at 05:45
  • Hi, thanks for the response! I read through it and it seemed good to me, but someone also told me they found a counterexample so I’m now even more confused. Specifically, taking $R = k[x,y,z]; x(y-1),y,z(y-1)$. Then the element $x^{p-1}y^{p-1}(z(y-1))^{p-1}$ appears to be in the colon ideal but not the right ideal. This came up because I said I can prove it if I can rearrange the regular series, so I was able to prove it if the ring was Noetherian local, and this is an example of a regular sequence you can’t rearrange. Do you have any thoughts? – Alex Scheffelin Jan 15 '23 at 03:14
  • @AlexScheffelin I think the element you mention is in the right-hand ideal. The right-hand ideal contains $y^p$ and also $x^{p-1}y^{p-1}z^{p-1}(y-1)^{2p-2}$, which is congruent mod $y^p$ to $(xyz)^{p-1}$, by expanding $(y-1)^{2p-2}$ and dropping all positive-degree terms. Similarly your element is congruent to $(-xyz)^{p-1}$. – Ravi Fernando Jan 15 '23 at 06:00
  • Oh wow you’re completely right! This is great, your proof was incredibly easy to follow. I’ve accepted your answer. – Alex Scheffelin Jan 15 '23 at 07:30