Generalized depth and associated primes in the perfect closure R^\infty

\varphi is already known to be a bijection by [NoSh.alpha, Theorem 6.1, \ i)]. \varphi is always continuous, but we claim that \varphi^{-1} : \operatorname{Spec}(R) \rightarrow \operatorname{Spec}(R^\infty) is continuous as well. Let V(J) \subset \operatorname{Spec}(R^\infty) be a closed set for some ideal J \subset R^\infty. We claim \varphi (V(J)) = V(J \cap R), and hence its pre-image under \varphi^{-1} is closed. If J \subseteq P^\infty, then clearly J \cap R \subseteq P^\infty \cap R = P, whence P \in V(J \cap R). Conversely, if J \cap R \subseteq P, fix x^{\frac{1}{q}} \in J. Then x \in (J^{[q]}R^\infty \cap R) \subseteq (J \cap R) \subseteq P = (P^\infty \cap R) \subseteq P^\infty. Hence x^{\frac{1}{q}} \in P^\infty, and J \subseteq P^\infty.

To show J \subseteq \mathfrak{a}, fix r \in J. Then for all e, there exists a q' \geq q such that r^{q'} \in J_e, and r \in \sqrt{J_e} = \mathfrak{a}.

Conversely if r \in \mathfrak{a}, then for all e there exists an n such that r^n \in J_e. Let q' > n, and therefore r^{q'} \in J_e, whereby f^{-e'}(r^{q'}) = r \in J.

For every e, we want to show f^{-1}(J_{e+1}) = J_e. Fix r \in f^{-1}(J_{e+1}), then r^p \in J_{e+1}. Now fix m_e \in F^e(M) \setminus N_e. By remark 2.18, number 2 below, N_e = N_e^F and hence g_{e}(m_e) = m_{e+1} \notin N_{e+1}. Since r^p \in J^{e+1}, we have r^p \circ m_{e+1} = r^\frac{p}{qp}m_{e+1} = r^\frac{1}{q}m_{e+1} \in N_{e+1}. But r \circ m_e = r^\frac{1}{q}m_e, and thus r \circ m_e \in g_{e}^{-1}(r^\frac{1}{q}m_{e+1}) \subseteq N_e. Thus r \in J_e = \operatorname{ann}_R \big( F^e(M)/N_e \big).

Conversely, first note that F^e(M) is generated by elements of the form (1 \otimes_R m)_e for some m \in M, and their images under g_{e} generate F^{e+1}(M).

Now fix r \in J_e, and we want to show r^p \in J_{e+1}. Let (1 \otimes_R m)_{e+1} be a generator of F^{e+1}(M), and consider r^p \circ (1 \otimes_R m)_{e+1} = (r^{\frac{1}{q}} \otimes_R m)_{e+1}. Now, any (r^{\frac{1}{q}} \otimes_R m)_{e} \in g_e^{-1}\big( (r^{\frac{1}{q}} \otimes_R m)_{e+1} \big) will be in N_e since it is equal to r \circ (1 \otimes_R m)_e. Therefore, r^p \circ (1 \otimes_R m)_{e+1} \in N_{e+1}, and since this was an arbitrary generator, we have r^p \circ F^{e+1}(M) \subseteq N_{e+1}, and r^p \in J_{e+1}.

1 \Rightarrow 2: Suppose P \in \bigcup \operatorname{Ass}_R(F^e(M) / N_e) with P = \big( N_e :_R m_e \big). I.e. for all r \in P, we have r \circ m_e = r^{\frac{1}{q}} m_e \in N_e. Equivalently, r^{\frac{1}{q}} m_\infty = 0 in (R^\infty \otimes_R M)/N. Thus P R^\infty \subset P^{[\frac{1}{q}]} R^\infty \subseteq ( N :_{R^\infty} m_\infty).

On the other hand, fix some s^{\frac{1}{q'}} \in ( N :_{R^\infty} m_\infty ). If q' \leq q, then s^{\frac{q}{q'}} \in R and s^{\frac{q}{q'}} \circ m_e = s^{\frac{1}{q'}} m_e \in N_e. Hence s^{\frac{q}{q'}} \in P, and s^{\frac{1}{q'}} \in P^\infty.

However if q' > q, consider the image of s \circ m_e = {s}^{\frac{1}{q}}m_e under the natural map g_{e, e'} : F^e(M) \rightarrow F^{e'}(M):

But {s}^{\frac{1}{q'}}m_{e'} \in N_{e'} = N_{e'}^F (see remark 2.18 fact 2), because s^{\frac{1}{q'}} \in ( N :_{R^\infty} m_\infty ). Therefore g_{e, e'}({s} \ \circ \mathfrak{m}_e) \in N_{e'}, and s \ \circ \ m_e \in N_e because N_e = N_e^F. Thus s \in (N_e :_R m_e ), whereby s \in P and s^{\frac{1}{q'}} \in P^\infty.

We have now shown that PR^\infty \subseteq ( N :_{R^\infty} m_\infty) \subseteq P^\infty. Since \sqrt{P R^\infty} = P^\infty, P^\infty is minimal over P R^\infty. Hence P^\infty is minimal over ( N :_{R^\infty} m_\infty) as well. This fact shows that P^\infty is minimal over the annihilator of an element of (R^\infty \otimes_{R^\infty} M)/N, and therefore P^\infty \in \operatorname{\widetilde{Ass}}_R \big( (R^{\infty} \otimes_R M)/N \big).

2 \Rightarrow 3: Always true, as stated above.

3 \Rightarrow 1: Suppose P^\infty \in \operatorname{sK}_R \big( (R^\infty \otimes_R M)/N \big), then PR^\infty is a finitely generated sub-ideal, and we have PR^\infty \subseteq ( N :_{R^{\infty}} m_\infty ) \subseteq P^\infty for some m_\infty. If P = (r_1, \ldots, r_n), then for each i there exists a q_i such that r_i^{\frac{1}{q_i}}m_{e_i} \in N_{e_i}. Letting q = \text{min} \lbrace q_i \rbrace, we have P \subseteq ( N_e :_{R} m_e ).

On the other hand, fix s \in ( N_e :_{R} m_e ), whereby s \circ m_e = s^\frac{1}{q}m_e \in N_e. Again by definition, s^\frac{1}{q} \in ( N :_{R^{\infty}} m_\infty) \subseteq P^\infty. Thus s \in P^\infty \cap R = P. Therefore ( N_e :_{R} m_e) \subseteq P.

We now have P = ( N_e :_{R} m_e), and P \in \operatorname{Ass}_R \big( F^e(M) / N_e \big).

Let \lbrace N_e \rbrace = \lbrace 0^F_{F^e(M)} \rbrace, and N = 0 = \underrightarrow{\lim} \ 0^F_{F^e(M)}. The result follows from theorem 3.

Letting M = R, \lbrace N_e \rbrace = \lbrace J_e \rbrace, and N = J we see that this corollary is a special case of theorem 3.

Letting J = I R^\infty, and J_e = (I^{[q]})^F for all e, we see this corollary is a special case of corollary 3.3.

Alternately, letting N_e = (I^{[q]})^F and N = IR^\infty, we see the corollary is a special case of theorem 3.

Fix e \in \mathbb{N}. It is easy to see that \mathbf{x} is a left-regular F^e(M)-sequence if and only if \mathbf{x}^{[q]} = x_1^q, \ldots, x_n^q is a left-regular F^e(M)-sequence if and only if \mathbf{x} is a right-regular F^e(M)-sequence.

The first equivalence is true by properties of regular sequences [No-FFR, Theorem 5.1.3], whereby \mathbf{x} is a left-regular F^e(M) if and only if \mathbf{x}^{[q]} is a left-regular F^e(M) sequence.

The second equivalence is true because for any m_e \in F^e(M), the left action by any x_j^q for j = 1, \ldots, n is equivalent to right action by x_j. That is,

By lemma 4.1, \mathbf{x} \subset R is left F^e(M)-regular if and only if it is right F^e(M)-regular. Therefore all maximal such sequences have the same length under right and left action.

We proceed by induction on |\mathbf{x}|. Let \mathbf{x} = x be a singleton. Then M / xM \neq 0 by Nakayama’s lemma since x \in \mathfrak{m} and M \neq 0. Furthermore, x clearly annihilates no element of M \subset N.

Now assume the statment is true for |\mathbf{x}| = n - 1. Let |\mathbf{x}| = n, and let \mathbf{x}' = x_1, \ldots, x_{n-1} denote the first n-1 elements of \mathbf{x}. Again, M / \mathbf{x}' M \neq 0 by Nakayama’s lemma since \mathbf{x}' \subset \mathfrak{m} and M \neq 0.

Additionally, M / \mathbf{x}' M embeds into N / \mathbf{x}' N. In order to see this fact, recall that M / \mathbf{x}' M \cong M \otimes_R R / (\mathbf{x}'), and N / \mathbf{x}' N \cong N \otimes_R R / (\mathbf{x}'), and since M is a pure sub N-module we have the embedding M \otimes_R R / (\mathbf{x}') \hookrightarrow N \otimes_R R / (\mathbf{x}').

Thus since x_n annihilates no element of N / \mathbf{x}' N, it will therefore annihilate no element of the submodule M / \mathbf{x}' M. Therefore, \mathbf{x} is an M-regular sequence.

M \subseteq F(M) is a pure submodule. Hence by 4.3, if \mathbf{x} is regular over F(M), it is also regular over M.

If \operatorname{depth}_R \big( F(M) \big) = d and x_1, \ldots, x_d is a maximal F(M)-sequence, then by lemma 4.4 x_1, \ldots, x_d is M-regular, and \operatorname{depth}_R (M) \geq d.

If \mathbf{x} is a regular (R^\infty \otimes_R M)-sequence, then by 4.3, \mathbf{x} is a regular right F^e(M)-sequence since F^e(M) \hookrightarrow (R^\infty \otimes_R M) is a pure submodule by F-purity of R, and \mathbf{x} is left-regular if and only it is right-regular.

Conversely, we proceed by induction on |\mathbf{x}|. If \mathbf{x} = x is a singleton, suppose x is not (R^\infty \otimes_R M)-regular, with m_\infty x = 0. Then 0 = m_{e} x in some F^{e}(M), and 0 \neq m_{e}, since it was not zero in (R^\infty \otimes_R M) = \underrightarrow{Lim} \ F^e(M).

Now suppose \mathbf{x} = x_1, \ldots, x_n \subset R is F^e(M)-regular for all e, in particular right-regular, and let \mathbf{x}' = x_1, \ldots, x_{n-1}, which is by hypothesis regular over (R^\infty \otimes_R M). We must show that x_n is regular over

We then apply the identical argument to the regular element x_n over the module F^e(M / \mathbf{x}'M), which is therefore is also regular over the module (R^\infty \otimes_R M / \mathbf{x}'M).

Let \bf x' = x_1^{\frac{1}{q_1}}, \ldots, x_n^{\frac{1}{q_n}} \subset R^\infty be an (R^\infty \otimes_R M)-regular sequence. Then by properties of regular sequences [No-FFR, Theorem 5.1.3], \bf x := x_1, \ldots, x_n \subset R \subseteq R^\infty is (R^\infty \otimes_R M)-regular as well. By lemma 4.7, \bf x is F^e(M)-regular for all e. Hence any maximal (R^\infty \otimes_R M)-sequence can have at most the length of \operatorname{s \ depth}_R (M) = \operatorname{depth}_R \big( F^e(M) \big) for e \gg 0.

We know that \operatorname{k \ depth}_{R^\infty} (R^\infty \otimes_R M) = sup\lbrace \operatorname{k \ gr}_{R^\infty} \big(I, (R^\infty \otimes_R M) \big) \rbrace, where I \subset R^\infty is any finitely generated sub-ideal. We therefore must show that the ideals \mathfrak{m}^{[\frac{1}{q}]}R^\infty determine the Koszul depth.

Recall for ideals I \subseteq J \subset R^\infty, we have \operatorname{k \ gr}_{R^\infty} (I,M) \leq \operatorname{k \ gr}_{R^\infty} (J,M).

Let I = (y_1^{\frac{1}{q_1}}, \ldots, y_n^{\frac{1}{q_n}}) \subset R^\infty be a finitely generated ideal. Then I \subseteq \mathfrak{m}^\infty since (R^\infty,\mathfrak{m}^\infty) is a local ring. Letting q \geq \text{max} \lbrace q_i \rbrace, we see that I \subseteq \mathfrak{m}^{[\frac{1}{q}]}R^\infty. Hence for any finitely generated I \subset R^\infty, we have \operatorname{k \ gr}_{R^\infty} (I,M) \leq \operatorname{k \ gr}_{R^\infty} (\mathfrak{m}^{[\frac{1}{q}]}R^\infty,M) \big) for q \gg 0, since it is a sub-ideal. Therefore, ideals of the form \mathfrak{m}^{[\frac{1}{q}]}R^\infty achieve sup\lbrace \operatorname{k \ gr}_{R^\infty} \big(I, (R^\infty \otimes_R M) \big) \rbrace.

To show \operatorname{k \ depth}_{R^\infty} (R^\infty \otimes_R M) \geq \operatorname{s \ depth}_R (M), let \mathbf{x} = x_1, \ldots, x_n be minimal system of generators for \mathfrak{m}. We claim that we have the comparison \operatorname{k \ gr}_{R^\infty} \big((\mathbf{x})R^\infty, (R^\infty \otimes_R M) \big) \geq \operatorname{s \ depth}_R (M), which will prove the inequality, since \operatorname{k \ depth}_{R^\infty} (R^\infty \otimes_R M) = \operatorname{k \ gr}_{R^\infty} \big(\mathfrak{m}^\infty, (R^\infty \otimes_R M) \big) \geq \operatorname{k \ gr}_{R^\infty} \big((\mathbf{x})R^\infty, (R^\infty \otimes_R M) \big).

To prove this claim, let \operatorname{depth}_R \big( F^e(M) \big) \geq d for all e, and let j > n - d. Fix e. Recall that since R is Noetherian, \operatorname{k \ depth}_R (F^e(M)) = \operatorname{depth}_R (F^e(M)). Thus H_j \big( \mathbf{x}; F^e (M) \big) = 0, and we must also show H_j \big( \mathbf{x}; (R^\infty \otimes_R M) \big) = 0. Fix z_\infty \in \text{ker}(d_j^{\infty}). Then z_e \in \text{ker}(d_j^{q}). By exactness at this j-th homology group, there exists a non-zero y_e \in \big( F^e(M) \big)^{n \choose j+1} such that d_{j+1}^q(y_e) = z_e. By F-purity of R, we have an injection F^e(M)^{n \choose j+1} \hookrightarrow (R^\infty \otimes_R M)^{n \choose j+1}, hence the image y_\infty in (R^\infty \otimes_R M)^{n \choose j+1} is non-zero. But now d_{j+1}^{\infty}(y_\infty) = z_{\infty}.

Since z_\infty \in \text{ker}(d_j^{\infty}) was chosen arbitrarily, therefore \text{ker}(d_j^{\infty}) = \text{image}(d_{j+1}^{\infty}), and H_j \big( \mathbf{x}; (R^\infty \otimes_R M) \big) = 0.

To show the reverse inequality, again let \mathbf{x} = x_1, \ldots, x_n be minimal system of generators for \mathfrak{m}, and recall by lemma 4.9, \operatorname{k \ depth}_{R^\infty} (R^\infty \otimes_R M) = \sup \lbrace \operatorname{k \ gr}_{R^\infty} \big((\mathbf{x})^{[\frac{1}{q}]}R^\infty, (R^\infty \otimes_R M) \big) \rbrace. We claim that for any e, \operatorname{k \ gr}_{R^\infty} \big((\mathbf{x})^{[\frac{1}{q}]}R^\infty, (R^\infty \otimes_R M) \big) \leq \operatorname{depth}_R \big( F^e(M) \big). Hence we know \operatorname{k \ depth}_{R^\infty} (R^\infty \otimes_R M) is finite, and since the statement is true for e \gg 0, the inequality is true.

To prove this claim, fix e. Let \operatorname{k \ gr}_{R^\infty} \big((\mathbf{x})^{[\frac{1}{q}]}R^\infty, (R^\infty \otimes_R M) \big) = d, and j > n - d, whereby H_j \big( \mathbf{x}^{[\frac{1}{q}]}; (R^\infty \otimes_R M) \big) = 0. We must show that H_j \big( \mathbf{x}; F^e(M) \big) = 0 as well, where \mathbf{x} acts on the left, which is equivalent to right action by \mathbf{x}^{\frac{1}{q}} \in R^{\frac{1}{q}}. Therefore, we can equivalently ask whether H_j \big( \mathbf{x}^{[\frac{1}{q}]}; F^e(M) \big) = 0 via right action. Fix z_e \in \text{ker}(d_j^q), whereby z_\infty \in \text{ker}(d_j^\infty). Then \overline{z_\infty} = 0 \in \text{coker}(d_{j+1}^\infty) by exactness of H_{\bullet} \big( \mathbf{x}^{\frac{1}{q}}; (R^\infty \otimes_R M) \big) at the j-th position. But \text{coker}(d_{j+1}^\infty) = \text{coker}(d_{j+1}^q \otimes_R 1_{R^\infty}) = \text{coker}(d_{j+1}^q) \otimes_R R^\infty, and since R is F-pure we have an injection \text{coker}(d_{j+1}^q) \hookrightarrow \text{coker}(d_{j+1}^q) \otimes_R R^\infty. Since \overline{z_{\infty}} = 0 in the image of this map, therefore its preimage \overline{z_e} = 0 as well.

Since z_e \in \text{ker}(d_j^q) was chosen arbitrarily, H_{\bullet} \big( \mathbf{x}; F^e (M) \big) is exact at the j-th position.

We have already shown that \operatorname{s \ depth}_R (M) \geq \operatorname{c \ depth}_{R^\infty} (R^\infty \otimes_R M).

Conversely, suppose that \operatorname{s \ depth}_R (M) > \operatorname{c \ depth}_{R^\infty} (R^\infty \otimes_R M) = d. Let \mathbf{y} = y_1, \ldots, y_d \subset R be a maximal (R^\infty \otimes_R M) sequence. Then for all e, \operatorname{depth}_R \big( F^e(M) / F^e(M)\mathbf{y} \big) > 0 = \operatorname{c \ depth}_{R^\infty} \big( (R^\infty \otimes_R M) / (R^\infty \otimes_R M) \mathbf{y} \big). We claim that \mathfrak{m} = \displaystyle \bigcup \bigcup_{e} \operatorname{Ass}_R \big( F^e(M) / F^e(M) \mathbf{y} \big). Note that this union consists of at most countably many prime ideals since it is a countable union of finite sets.

Clearly the union is contained in \mathfrak{m}, since (R, \mathfrak{m}) is local. Conversely, suppose that there exists some z \in \mathfrak{m} \setminus \displaystyle \bigcup \bigcup_{e} \operatorname{Ass}_R \big( F^e(M) / F^e(M) \mathbf{y} \big). Then z is not contained in any associated prime of F^e(M) / F^e(M) \mathbf{y} for any e. Hence z is regular over F^e(M) / F^e(M) \mathbf{y} for all e. Then \lbrace \mathbf{y},z \rbrace is a right regular F^e(M)-sequence for all e, and hence it is (R^\infty \otimes_R M)-regular by lemma 4.7. But this fact contradicts the maximality of \mathbf{y}.

Now suppose either condition of the theorem holds. Since \mathfrak{m} is contained in this union of prime ideals, then \mathfrak{m} \subseteq P for some P \in \operatorname{Ass}_R \big( F^e(M) / F^e(M) \mathbf{y} \big) and for some e. But since \mathfrak{m} is maximal, \mathfrak{m} = P. We now have the maximal ideal of R shown to be an associated prime of F^e(M) / F^e(M) \mathbf{y}, and hence \operatorname{depth}_R \big( F^e(M) / F^e(M) \mathbf{y} \big) = 0. We have contradicted our assumption that \operatorname{depth}_R \big( F^e(M) / F^e(M)\mathbf{y} \big) > 0.