Introduction
This paper answers the open questions of Kabbaj and Mimouni concerning the position of the t t t -linkage property in the hierarchy of integral domains and its relationship with almost Krull domains [20 Figure 1 and p. 11] . Their questions bring together two structural viewpoints: linkage expressed through colon ideals and cancellation in multiplicative monoids of fractional ideals. The aim is to identify what full t t t -linkage measures when the local descriptions available for Prüfer v v v -multiplication domains no longer apply.
Our work is motivated by a series of characterizations of domains in terms of linkage properties. Kabbaj and Mimouni showed that finite ordinary linkage characterizes Prüfer domains, whereas ordinary linkage for all nonzero ideals characterizes almost Dedekind domains [19 ] . In Linkage of ideals under star operations , they extended this framework to arbitrary star operations. They proved that finite t t t -linkage and finite v v v -linkage each characterize v v v -domains [20 Theorem 3.1] , while full v v v -linkage characterizes completely integrally closed domains [20 Theorem 3.2] . They also established that finite w w w -linkage characterizes Prüfer v v v -multiplication domains (PvMDs), whereas full w w w -linkage characterizes t t t -almost Dedekind domains [20 Theorem 3.3] . Here and throughout, full refers to their linkage property for all nonzero ideals; the adjective serves only to distinguish this property from its finitely generated version.
Related work appears in the manuscript of Huang and Wang [18 ] , who introduced a w w w -analogue of linkage and studied its finite version in weakly Matlis domains. Recall that a domain is weakly Matlis if each nonzero element belongs to only finitely many maximal t t t -ideals and no two distinct maximal t t t -ideals contain a common nonzero prime ideal. Their Theorem 3.9 characterizes weakly Matlis PvMDs in terms of finite w w w -linkage. In the star-operation framework adopted here, the corresponding characterization of PvMDs obtained by Kabbaj and Mimouni requires no weakly Matlis hypothesis [20 Theorem 3.3] .
Since t = w t=w t = w on every PvMD, these results also characterize full t t t -linkage within this class: a PvMD has full t t t -linkage if and only if it is t t t -almost Dedekind [20 Corollary 3.4] . Outside the class of PvMDs, however, these results do not settle the relationship between full t t t -linkage and the almost Krull property. Following Pirtle, a domain D D D is called almost Krull if D N D_N D N is a Krull domain for every maximal ideal N N N of D D D [28 , 29 , 30 ] .
The present paper investigates this relationship, which is also left unresolved in Figure 1 of [20 ] . More specifically, our investigation is guided by the following questions posed by Kabbaj and Mimouni at the end of their article [20 p. 11, Open Questions] .
Figure 1. The position of the t t t -linkage node from [20 Figure 1] , as determined by Theorem 6.7 . Solid arrows are implications. Within a box, full t t t -linkage and TCL are equivalent, as are the two conditions in each of the t t t -almost Dedekind and CIC boxes. The lower-left box records properties of almost Krull domains, not a converse characterization. No implication connects AK with TCL, or PvMD with TCL.
Question 1.1.
(i) Where should the t t t -linkage property be placed in the diagram?
(ii) What is its precise relationship with almost Krull domains?
(iii) Or is it naturally aligned with another class shown in the figure?
Cancellation provides a natural approach to these questions. Gilmer proved that cancellation by all nonzero finitely generated ideals characterizes Prüfer domains, whereas cancellation by all nonzero ideals characterizes almost Dedekind domains [10 Corollary 1 and Theorem 3] . Anderson and Anderson developed the corresponding residual criteria [1 ] . The parallel with the linkage characterizations suggests that the missing node should be sought in the multiplicative structure of fractional t t t -ideals rather than by replacing w w w with t t t in a local criterion.
The first theorem identifies full ⋆ \star ⋆ -linkage with full ⋆ \star ⋆ -cancellation for every star operation ⋆ \star ⋆ (Theorem 2.3). For ⋆ = t \star=t ⋆ = t , this identifies full linkage with the class TCL of domains whose monoid of all nonzero fractional t t t -ideals, under ( I , J ) ↦ ( I J ) t (I,J)\mapsto(IJ)_t ( I , J ) ↦ ( I J ) t , is cancellative. Writing t t t -AD, AK, and CIC for the t t t -almost Dedekind, almost Krull, and completely integrally closed classes, respectively, we obtain
t -AD ⊊ T C L ⊊ C I C ⊆ v -domain , T C L ∩ P v M D = t -AD . (1) t\text{-AD} \subsetneq\mathrm{TCL} \subsetneq\mathrm{CIC} \subseteq v\text{-domain}, \qquad\mathrm{TCL}\cap\mathrm{PvMD}=t\text{-AD}. \tag*{(1)} t -AD ⊊ TCL ⊊ CIC ⊆ v -domain , TCL ∩ PvMD = t -AD . (1)
The intersection identity recovers Kabbaj–Mimouni’s Corollary 3.4 in cancellation language; the strict containments locate the full t t t -node beyond that previously determined intersection.
Two distinctions are essential. First, Chang’s theorem identifies w w w -cancellation of an ideal with t t t -local principality ; it does not identify it with t t t -invertibility without an additional t t t -finiteness hypothesis [5 ] . Second, full cancellation tests arbitrary fractional ideals in all three positions, whereas the a.b. and e.a.b. conditions require the factor being cancelled to be finitely generated. Although these latter conditions agree for finite-type operations, neither is equivalent to full cancellation [8 , 24 ] . We show that full t t t -cancellation is strictly stronger than the a.b. property of t t t , but does not force t t t -invertibility of finitely generated ideals.
The almost Krull part of Question 1.1 has a particularly concrete interpretation. If every maximal localization of D D D is Krull, then the canonical locally divisorial operation I κ = ⋂ N ∈ Max ( D ) ( I D N ) v N I_\kappa=\bigcap_{N\in\operatorname{Max}(D)}(ID_N)_{v_N} I κ = ⋂ N ∈ Max ( D ) ( I D N ) v N is cancellative and satisfies κ f = t \kappa_f=t κ f = t [6 ] . Thus the question whether almost Krullness implies full t t t -linkage asks whether cancellation survives passage from this operation to its finite-type associate. Equality κ = t \kappa=t κ = t holds exactly on the almost Krull PvMD locus, but equality of operations is stronger than preservation of cancellation. We establish both possibilities when t < κ t<\kappa t < κ .
Reinhart’s cancellative non-t t t -Prüfer monoid, together with the monomial-ideal description of Kim and Park, yields a TCL-domain outside both the PvMD and almost Krull classes [22 , 31 ] ; see Theorem 3.3 . A different finite-patching argument proves that the Heinzer–Ohm essential domain and the Arnold–Matsuda domain both belong to A K ∩ T C L ∖ P v M D \mathrm{AK} \cap\mathrm{TCL} \setminus\mathrm{PvMD} AK ∩ TCL ∖ PvMD . Their local descriptions come from [3 , 15 , 16 ] ; the full cancellation statements are proved here. In the Arnold–Matsuda example, t = κ f < κ = v t = \kappa_f < \kappa= v t = κ f < κ = v , and a height-two non-well-behaved maximal t t t -ideal has t t t -closure smaller than its κ \kappa κ -closure.
For the opposite direction, we introduce infinitely many positive weights tending to zero into a Heinzer–Ohm-type construction. Every maximal localization of the resulting domain is a UFD, but one maximal ideal satisfies
0 ≠ M = M t = ( M 2 ) t ⊊ D , M κ = D . (2) 0 \ne M = M_t = (M^2)_t \subsetneq D,\qquad M_\kappa= D. \tag*{(2)} 0 = M = M t = ( M 2 ) t ⊊ D , M κ = D . (2)
Thus ( M 2 ) t = ( M D ) t (M^2)_t = (MD)_t ( M 2 ) t = ( M D ) t while M t ≠ D t M_t \ne D_t M t = D t . The construction proves that almost Krullness does not imply TCL, even when all maximal localizations are factorial and satisfy their own full t t t -cancellation laws (Theorem 5.3 and Corollary 5.4). Together with the monoid example, it gives the incomparability of AK and TCL, answering the second question in Question 1.1 . It also sharpens the positive examples to
A K ∩ P v M D ⊊ A K ∩ T C L ⊊ A K . (3) \mathrm{AK} \cap\mathrm{PvMD} \subsetneq\mathrm{AK} \cap\mathrm{TCL} \subsetneq\mathrm{AK}. \tag*{(3)} AK ∩ PvMD ⊊ AK ∩ TCL ⊊ AK . (3)
Theorem 6.7 collects the answers to all three questions, and Figure 1 records the resulting position of the missing node.
The paper has six sections. § 2 develops the residual language of linkage and cancellation. § 3 gives the monoid example and the Heinzer–Ohm patching argument. § 4 studies κ \kappa κ and proves full cancellation for the Arnold–Matsuda example. § 5 constructs the almost Krull domain failing TCL. § 6 describes the descent defect and records the final implication diagram. Throughout, D D D is an integral domain with quotient field K K K , and all fractional ideals are nonzero. Fields are included in the t t t -almost Dedekind class; equivalently, the defining DVR condition is imposed on nonzero maximal t t t -ideals. The abbreviation AK always means locally Krull , not an almost-factorial condition.
For any undefined terminology and notation, the reader is referred to [32 ] .
Linkage, cancellation, and the full t t t -node
This section places linkage and cancellation in a common notation and then uses the equivalence to locate full t t t -linkage among familiar Prüfer-type conditions.
Star operations and residual linkage
Let F ( D ) \mathcal{F}(D) F ( D ) be the set of nonzero fractional ideals of D D D , and let f ( D ) \mathfrak{f}(D) f ( D ) be its subset of finitely generated fractional ideals. A star operation on D D D is a map I ↦ I ⋆ I \mapsto I_\star I ↦ I ⋆ on F ( D ) \mathcal{F}(D) F ( D ) such that, for 0 ≠ x ∈ K 0 \ne x \in K 0 = x ∈ K and I , J ∈ F ( D ) I,J \in\mathcal{F}(D) I , J ∈ F ( D ) , ( x I ) ⋆ = x I ⋆ (xI)_\star= xI_\star ( x I ) ⋆ = x I ⋆ , I ⊆ J ⇒ I ⋆ ⊆ J ⋆ I \subseteq J \Rightarrow I_\star\subseteq J_\star I ⊆ J ⇒ I ⋆ ⊆ J ⋆ , I ⊆ I ⋆ I \subseteq I_\star I ⊆ I ⋆ , and I ⋆ ⋆ = I ⋆ I_{\star\star} = I_\star I ⋆⋆ = I ⋆ , and D ⋆ = D D_\star= D D ⋆ = D . It is of finite type if I ⋆ = ⋃ { F ⋆ : F ∈ f ( D ) , F ⊆ I } I_\star= \bigcup\{F_\star: F \in\mathfrak{f}(D),\ F \subseteq I\} I ⋆ = ⋃ { F ⋆ : F ∈ f ( D ) , F ⊆ I } for every I I I . These definitions go back to Krull; standard references are [11 Section 32] and [14 ] . The more flexible semistar formulation was introduced by Okabe and Matsuda [26 ] .
For I ∈ F ( D ) I \in\mathcal{F}(D) I ∈ F ( D ) , put I − 1 = ( D : I ) I^{-1} = (D : I) I − 1 = ( D : I ) , I v = ( I − 1 ) − 1 I_v = (I^{-1})^{-1} I v = ( I − 1 ) − 1 , and I t = ⋃ { F v : F ∈ f ( D ) , F ⊆ I } I_t = \bigcup\{F_v : F \in\mathfrak{f}(D),\ F \subseteq I\} I t = ⋃ { F v : F ∈ f ( D ) , F ⊆ I } . The stable finite-type operation associated to v v v is I w = ⋂ M ∈ t - Max ( D ) I D M I_w = \bigcap_{M \in t\text{-}\operatorname{Max}(D)} ID_M I w = ⋂ M ∈ t - Max ( D ) I D M ; see [7 , 33 ] . Thus w ≤ t ≤ v w \leq t \leq v w ≤ t ≤ v . A fractional ideal is a ⋆ \star ⋆ -ideal if I = I ⋆ I = I_\star I = I ⋆ , and it is ⋆ \star ⋆ -invertible if ( I I − 1 ) ⋆ = D (II^{-1})_\star= D ( I I − 1 ) ⋆ = D . Here t - Max ( D ) t\text{-}\operatorname{Max}(D) t - Max ( D ) denotes the set of ideals maximal among the proper integral t t t -ideals; these ideals are prime. A t t t -ideal is t t t -finite if it equals F t = F v F_t = F_v F t = F v for some finitely generated fractional ideal F F F .
For fractional ideals A A A and B B B , write ( A : B ) = { x ∈ K : x B ⊆ A } (A : B) = \{x \in K : xB \subseteq A\} ( A : B ) = { x ∈ K : x B ⊆ A } . If A A A and B B B are integral ideals, the integral colon is ( A : D B ) = ( A : B ) ∩ D (A :_D B) = (A : B) \cap D ( A : D B ) = ( A : B ) ∩ D .
Following [20 Definition 2.1] , call two nonzero integral ideals I I I and J J J ⋆ \star ⋆ -linked if there is a nonzero ideal A ⊆ I ∩ J A \subseteq I \cap J A ⊆ I ∩ J such that I ⋆ = ( ( A ⋆ : D J ) ) ⋆ I_{\star}=((A_{\star}:_{D}J))_{\star} I ⋆ = (( A ⋆ : D J ) ) ⋆ and J ⋆ = ( ( A ⋆ : D I ) ) ⋆ J_{\star}=((A_{\star}:_{D}I))_{\star} J ⋆ = (( A ⋆ : D I ) ) ⋆ . The domain has full ⋆ \star ⋆ -linkage if every such pair is ⋆ \star ⋆ -linked, and finite ⋆ \star ⋆ -linkage if this is required only for pairs of nonzero finitely generated ideals. The adjective full is omitted in [20 ] .
The defining ideal can always be replaced by the product. This is [20 Lemma 2.2] ; we include the proof to fix the residual notation.
Lemma 2.1. Two nonzero integral ideals I I I and J J J are ⋆ \star ⋆ -linked if and only if
I ⋆ = ( ( I J ) ⋆ : D J ) ⋆ , J ⋆ = ( ( I J ) ⋆ : D I ) ⋆ . (4) I_{\star} = ((IJ)_{\star}:_{D}J)_{\star}, \qquad J_{\star} = ((IJ)_{\star}:_{D}I)_{\star}. \tag*{(4)} I ⋆ = (( I J ) ⋆ : D J ) ⋆ , J ⋆ = (( I J ) ⋆ : D I ) ⋆ . (4)
Proof. If I I I and J J J are linked through A A A , then I ⊆ ( A ⋆ : D J ) I \subseteq(A_{\star}:_{D}J) I ⊆ ( A ⋆ : D J ) . Indeed, the latter colon is a ⋆ \star ⋆ -ideal: it is the nonzero intersection of D D D and the ⋆ \star ⋆ -ideals j − 1 A ⋆ j^{-1}A_{\star} j − 1 A ⋆ for 0 ≠ j ∈ J 0 \ne j \in J 0 = j ∈ J . Consequently, I J ⊆ A ⋆ IJ \subseteq A_{\star} I J ⊆ A ⋆ and ( I J ) ⋆ ⊆ A ⋆ (IJ)_{\star} \subseteq A_{\star} ( I J ) ⋆ ⊆ A ⋆ . The inclusions I ⊆ ( ( I J ) ⋆ : D J ) ⊆ ( A ⋆ : D J ) I \subseteq((IJ)_{\star}:_{D}J) \subseteq(A_{\star}:_{D}J) I ⊆ (( I J ) ⋆ : D J ) ⊆ ( A ⋆ : D J ) yield the first identity after taking ⋆ \star ⋆ -closures. The second follows by symmetry. Conversely, if the displayed identities hold, choose A = I J A = IJ A = I J , which is a nonzero subideal of I ∩ J I \cap J I ∩ J . □
A fractional ideal A A A is a ⋆ \star ⋆ -cancellation ideal if ( A B ) ⋆ = ( A C ) ⋆ (AB)_{\star}=(AC)_{\star} ( A B ) ⋆ = ( A C ) ⋆ implies B ⋆ = C ⋆ B_{\star}=C_{\star} B ⋆ = C ⋆ for all B , C ∈ F ( D ) B,C \in\mathcal{F}(D) B , C ∈ F ( D ) . The full ⋆ \star ⋆ -cancellation law holds if every member of F ( D ) \mathcal{F}(D) F ( D ) is a ⋆ \star ⋆ -cancellation ideal. This extends the classical notions studied by Gilmer and by Anderson–Anderson [1 , 10 ] .
For the identity operation d d d , Anderson and Anderson call A A A a quasi-cancellation ideal when ( A F : A ) = F (AF:A)=F ( A F : A ) = F for every finitely generated fractional ideal F F F . Their residual and inclusion criteria are the d d d -counterparts of the lemma below [1 Lemma 1] . If A A A is finitely generated, quasi-cancellation, cancellation, and invertibility are equivalent [1 Corollary 1 and Theorem 1] . At the level of the whole domain, every nonzero finitely generated ideal is cancellative exactly when D D D is Prüfer, whereas every nonzero ideal is cancellative exactly when D D D is almost Dedekind [1 Theorem 6] ; compare [10 Corollary 1 and Theorem 3] . By contrast, every nonzero ideal is merely quasi-cancellative exactly when D D D is a completely integrally closed Prüfer domain [1 Theorem 7] . Thus the distinction between finite residual tests and unrestricted cancellation is present before any nontrivial star operation is introduced.
The following lemma is the star-operation residual form of these classical equivalences and converts equality-cancellation into the inclusion form needed later.
Lemma 2.2. For a fixed A ∈ F ( D ) A \in\mathcal{F}(D) A ∈ F ( D ) , the following conditions are equivalent.
(i) If ( A B ) ⋆ = ( A C ) ⋆ (AB)_{\star}=(AC)_{\star} ( A B ) ⋆ = ( A C ) ⋆ , then B ⋆ = C ⋆ B_{\star}=C_{\star} B ⋆ = C ⋆ .
(ii) If ( A B ) ⋆ ⊆ ( A C ) ⋆ (AB)_{\star}\subseteq(AC)_{\star} ( A B ) ⋆ ⊆ ( A C ) ⋆ , then B ⋆ ⊆ C ⋆ B_{\star}\subseteq C_{\star} B ⋆ ⊆ C ⋆ .
(iii) For every B ∈ F ( D ) B \in\mathcal{F}(D) B ∈ F ( D ) , ( ( ( A B ) ⋆ : A ) ) ⋆ = B ⋆ (((AB)_{\star}:A))_{\star}=B_{\star} ((( A B ) ⋆ : A ) ) ⋆ = B ⋆ .
Proof. Assume (i) and ( A B ) ⋆ ⊆ ( A C ) ⋆ (AB)_{\star}\subseteq(AC)_{\star} ( A B ) ⋆ ⊆ ( A C ) ⋆ . Since ( A ( B + C ) ) ⋆ = ( A B + A C ) ⋆ = ( A C ) ⋆ (A(B+C))_{\star}=(AB+AC)_{\star}=(AC)_{\star} ( A ( B + C ) ) ⋆ = ( A B + A C ) ⋆ = ( A C ) ⋆ , cancellation gives ( B + C ) ⋆ = C ⋆ (B+C)_{\star}=C_{\star} ( B + C ) ⋆ = C ⋆ , and hence B ⋆ ⊆ C ⋆ B_{\star}\subseteq C_{\star} B ⋆ ⊆ C ⋆ . Thus (i) implies (ii), and the reverse implication is immediate.
Set H = ( ( A B ) ⋆ : A ) H=((AB)_{\star}:A) H = (( A B ) ⋆ : A ) . This is a nonzero fractional ideal: it contains B B B and is contained in a − 1 ( A B ) ⋆ a^{-1}(AB)_{\star} a − 1 ( A B ) ⋆ for any fixed 0 ≠ a ∈ A 0 \ne a \in A 0 = a ∈ A . The inclusions B ⊆ H B \subseteq H B ⊆ H and A H ⊆ ( A B ) ⋆ AH \subseteq(AB)_{\star} A H ⊆ ( A B ) ⋆ give ( A B ) ⋆ ⊆ ( A H ) ⋆ (AB)_{\star}\subseteq(AH)_{\star} ( A B ) ⋆ ⊆ ( A H ) ⋆ . Condition (ii) therefore yields H ⋆ = B ⋆ H_{\star}=B_{\star} H ⋆ = B ⋆ , proving (iii). Conversely, if ( A B ) ⋆ = ( A C ) ⋆ (AB)_{\star}=(AC)_{\star} ( A B ) ⋆ = ( A C ) ⋆ , applying (iii) to the common product gives B ⋆ = C ⋆ B_{\star}=C_{\star} B ⋆ = C ⋆ . □
The product criterion now identifies the two domain-wide properties.
Theorem 2.3. For a star operation ⋆ \star ⋆ on D D D , the following conditions are equivalent.
(i) D D D has full ⋆ \star ⋆ -linkage.
(ii) The full ⋆ \star ⋆ -cancellation law holds.
(iii) For all A , B , C ∈ F ( D ) A,B,C \in\mathcal{F}(D) A , B , C ∈ F ( D ) , the inclusion ( A B ) ⋆ ⊆ ( A C ) ⋆ (AB)_\star\subseteq(AC)_\star ( A B ) ⋆ ⊆ ( A C ) ⋆ implies B ⋆ ⊆ C ⋆ B_\star\subseteq C_\star B ⋆ ⊆ C ⋆ .
Proof. Assume (i), and first take nonzero integral ideals A , B , C A,B,C A , B , C with ( A B ) ⋆ = ( A C ) ⋆ (AB)_\star=(AC)_\star ( A B ) ⋆ = ( A C ) ⋆ . Applying (4) to ( A , B ) (A,B) ( A , B ) and ( A , C ) (A,C) ( A , C ) gives B ⋆ = ( ( ( A B ) ⋆ : D A ) ) ⋆ = ( ( ( A C ) ⋆ : D A ) ) ⋆ = C ⋆ B_\star=(((AB)_\star:_D A))_\star=(((AC)_\star:_D A))_\star=C_\star B ⋆ = ((( A B ) ⋆ : D A ) ) ⋆ = ((( A C ) ⋆ : D A ) ) ⋆ = C ⋆ . A common scaling of A , B , C A,B,C A , B , C proves (ii) for arbitrary fractional ideals.
Conversely, assume (ii), and let I , J I,J I , J be nonzero integral ideals. Put C = ( ( I J ) ⋆ : D J ) C=((IJ)_\star:_D J) C = (( I J ) ⋆ : D J ) . Since I ⊆ C I\subseteq C I ⊆ C and C J ⊆ ( I J ) ⋆ CJ\subseteq(IJ)_\star C J ⊆ ( I J ) ⋆ , one has ( C J ) ⋆ = ( I J ) ⋆ (CJ)_\star=(IJ)_\star ( C J ) ⋆ = ( I J ) ⋆ . Cancelling J J J gives C ⋆ = I ⋆ C_\star=I_\star C ⋆ = I ⋆ . Interchanging I I I and J J J proves both identities in (4) , and hence (i). Finally, (ii) and (iii) are equivalent by Lemma 2.2.
The unrestricted law also forces the strongest familiar integral-closure condition. Recall that x ∈ K x\in K x ∈ K is almost integral over D D D if there is 0 ≠ d ∈ D 0\ne d\in D 0 = d ∈ D such that d x n ∈ D dx^n\in D d x n ∈ D for every n ≥ 0 n\geq0 n ≥ 0 ; the domain is completely integrally closed if every almost integral element of K K K belongs to D D D . The full v v v -linkage characterization in the following proposition is [20 Theorem 3.2] .
Proposition 2.4. If the full ⋆ \star ⋆ -cancellation law holds for some star operation ⋆ \star ⋆ , then D D D is completely integrally closed. Conversely, complete integral closure is equivalent to the full v v v -cancellation law and to full v v v -linkage.
Proof. Let x ∈ K x\in K x ∈ K be almost integral over D D D . Choose 0 ≠ d ∈ D 0\ne d\in D 0 = d ∈ D with d x n ∈ D dx^n\in D d x n ∈ D for every n ≥ 0 n\geq0 n ≥ 0 , and put A = ∑ n ≥ 0 D x n A=\sum_{n\geq0}Dx^n A = ∑ n ≥ 0 D x n . Then A A A is a fractional ideal and A ( D + x D ) = A = A D A(D+xD)=A=AD A ( D + x D ) = A = A D . Cancellation by A A A gives ( D + x D ) ⋆ = D ⋆ = D (D+xD)_\star=D_\star=D ( D + x D ) ⋆ = D ⋆ = D , so x ∈ D x\in D x ∈ D .
Conversely, suppose that D D D is completely integrally closed. For every fractional ideal J J J , one has ( J : J ) = D (J:J)=D ( J : J ) = D : if u J ⊆ J uJ\subseteq J u J ⊆ J , choose 0 ≠ j ∈ J 0\ne j\in J 0 = j ∈ J and 0 ≠ d ∈ D 0\ne d\in D 0 = d ∈ D with d J ⊆ D dJ\subseteq D dJ ⊆ D ; then d j u n ∈ D dju^n\in D d j u n ∈ D for every n ≥ 0 n\geq0 n ≥ 0 . Thus, for any fractional ideal I I I , ( I I − 1 ) − 1 = ( I v : I v ) = D (II^{-1})^{-1}=(I_v:I_v)=D ( I I − 1 ) − 1 = ( I v : I v ) = D , and consequently ( I I − 1 ) v = D (II^{-1})_v=D ( I I − 1 ) v = D . The fractional divisorial ideals therefore form a group under v v v -multiplication, so full v v v -cancellation holds. Full v v v -linkage is equivalent by Theorem 2.3; compare [11 ] .
The localization step used below requires an outer closure. We record the precise statement so that it cannot be confused with unrestricted commutation of t t t and localization; compare [16 Section 2] and [21 Lemma 3.4] .
Lemma 2.5. Let U ⊆ D ∖ { 0 } U\subseteq D\setminus\{0\} U ⊆ D ∖ { 0 } be multiplicatively closed and put T = D U T=D_U T = D U . For every I ∈ F ( D ) I\in\mathcal{F}(D) I ∈ F ( D ) ,
I t T ⊆ ( I T ) t T , ( ( I t ) T ) t T = ( I T ) t T . (5) I_tT\subseteq(IT)_{t_T}, \qquad((I_t)T)_{t_T}=(IT)_{t_T}. \tag*{(5)} I t T ⊆ ( I T ) t T , (( I t ) T ) t T = ( I T ) t T . (5)
If T T T is a valuation domain, then I t T = I T I_tT=IT I t T = I T .
Proof. For a finitely generated fractional ideal F F F , localization commutes with its inverse: ( D : F ) T = ( T : F T ) (D:F)T=(T:FT) ( D : F ) T = ( T : F T ) . Hence F v T ⊆ ( F T ) v T F_vT\subseteq(FT)_{v_T} F v T ⊆ ( F T ) v T . Taking the union over the finitely generated F ⊆ I F\subseteq I F ⊆ I gives I t T ⊆ ( I T ) t T I_tT\subseteq(IT)_{t_T} I t T ⊆ ( I T ) t T . The reverse inclusion after taking v T v_T v T -closures follows from I ⊆ I t I\subseteq I_t I ⊆ I t . On a valuation domain every finitely generated fractional ideal is principal, so t T = d T t_T=d_T t T = d T and the last assertion follows.
The a.b. and e.a.b. levels
The finite-factor cancellation conditions explain why the full law cannot be replaced by a familiar finite-type hypothesis. A finitely generated fractional ideal F F F is ⋆ \star ⋆ -a.b. if ( F G ) ⋆ ⊆ ( F H ) ⋆ (FG)_\star\subseteq(FH)_\star ( F G ) ⋆ ⊆ ( F H ) ⋆ implies G ⋆ ⊆ H ⋆ G_\star\subseteq H_\star G ⋆ ⊆ H ⋆ for all fractional ideals G , H G,H G , H ; it is ⋆ \star ⋆ -e.a.b. if G G G and H H H are also required to be finitely generated. The operation is a.b. or e.a.b. when every finitely generated fractional ideal has the corresponding property. Fontana and Loper prove that e.a.b. and a.b. coincide for finite-type semistar operations [8 Lemma 3 and Proposition 4] . They also give the residual characterization
( ( F H ) ⋆ : F ⋆ ) = H ⋆ (6) ((FH)_\star:F_\star)=H_\star \tag*{(6)} (( F H ) ⋆ : F ⋆ ) = H ⋆ (6)
for a.b. factors; see [8 Lemma 8] . Matsuda’s notation records the stronger unrestricted levels. Let f f f , g g g , and h h h denote finitely generated fractional ideals, arbitrary fractional ideals, and arbitrary nonzero submodules of K K K , respectively. For a finite-type semistar operation, his implications reduce to
h h = h g = h f ⇒ g h ⇒ g g ⇒ g f ⇒ f h = f g = f f . (7) hh = hg = hf \Rightarrow gh \Rightarrow gg \Rightarrow gf \Rightarrow fh = fg = ff. \tag*{(7)} hh = h g = h f ⇒ g h ⇒ g g ⇒ g f ⇒ f h = f g = f f . (7)
Here g g gg g g is full cancellation on fractional ideals, while f h fh f h is the a.b. condition [24 Remarks 2 and 3] . The analogous monoid theory is developed in [23 ] . Thus finite type identifies e.a.b. with a.b., but it does not identify either condition with full cancellation. The 2009 construction intended to separate e.a.b. from a.b. was corrected in [9 ] ; we use that corrected example whenever the strict separation is invoked.
Pseudo-valuation domains as a classical stress test
Pseudo-valuation domains show that the preceding hierarchy is not an artifact of Prüfer domains. A pseudo-valuation domain is a local domain D D D for which there is a valuation overring V V V with Spec ( D ) = Spec ( V ) \operatorname{Spec}(D) = \operatorname{Spec}(V) Spec ( D ) = Spec ( V ) as sets. If M M M is the maximal ideal of D D D , this valuation overring is unique, is called the canonical valuation domain , and is given by V = ( M : M ) V = (M : M) V = ( M : M ) . More precisely, if φ : V → V / M \varphi: V \to V/M φ : V → V / M is the residue map, then D = φ − 1 ( k ) = V × V / M k D = \varphi^{-1}(k) = V \times_{V/M} k D = φ − 1 ( k ) = V × V / M k for a subfield k k k of V / M V/M V / M [25 pp. 3983–3984] .
The ordinary cancellation theory of this class is rigid at the full level but flexible at the finitely tested level. Matsuda and Sugatani prove that a nonzero ideal of a pseudo-valuation domain is a cancellation ideal if and only if it is principal [25 Theorem 3] . Hence a pseudo-valuation domain satisfying the full ordinary cancellation law is a field or a discrete valuation domain; in particular, it cannot be a proper nonvaluation pseudo-valuation domain. If D ≠ V D \ne V D = V and V V V is module-finite over D D D , even every nonzero quasi-cancellation ideal is principal [25 Theorem 4] . Without this finiteness hypothesis, however, they construct proper pseudo-valuation domains having nonprincipal quasi-cancellation ideals that are not cancellation ideals [25 Examples 5 and 7] ; by [25 Proposition 6] , these examples are not flat.
These results concern the identity operation, not the t t t -operation. They illustrate why neither quasi-cancellation nor a change of star operation can be substituted for the unrestricted law. The constructions below verify the full t t t -law directly rather than inferring it from a pseudo-valuation description.
An ideal I I I is t t t -locally principal if I D M ID_M I D M is principal for every M ∈ t M \in t M ∈ t -Max( D ) (D) ( D ) . Chang’s characterization says that this is equivalent to w w w -cancellation. Such an ideal is t t t -invertible precisely when it is also t t t -finite [5 ] ; these are distinct requirements for arbitrary ideals.
The formal position of full t t t -linkage
A domain is a v v v -domain if every finitely generated fractional ideal is v v v -invertible, and it is a Prüfer v v v -multiplication domain , or PvMD, if every finitely generated fractional ideal is t t t -invertible. Equivalently, D M D_M D M is a valuation domain for every M ∈ t M \in t M ∈ t -Max( D ) (D) ( D ) [13 , 21 ] . A domain is t t t -almost Dedekind if D M D_M D M is a discrete valuation domain for every M ∈ t M \in t M ∈ t -Max( D ) (D) ( D ) .
We write TCL for the full t t t -cancellation law: ( A B ) t = ( A C ) t (AB)_t = (AC)_t ( A B ) t = ( A C ) t implies B t = C t B_t = C_t B t = C t for all fractional ideals A , B , C A,B,C A , B , C . By Theorem 2.3 , this is exactly full t t t -linkage.
The next proposition gives the unconditional containments and determines the intersection with the PvMD branch. Its characterization within that branch recovers [20 Corollary 3.4] via Theorem 2.3 ; the full w w w -linkage characterization recorded at the end of the proof is [20 Theorem 3.3] .
Proposition 2.6. For every domain D D D , the following statements hold.
(i) If D D D satisfies TCL, then D D D is completely integrally closed, is a v v v -domain, and the t t t -operation is a.b.
(ii) Every t t t -almost Dedekind domain satisfies TCL.
(iii) If D D D is a PvMD, then D D D satisfies TCL if and only if D D D is t t t -almost Dedekind.
(iv) Consequently, TCL ∩ \cap ∩ PvMD = t t t -AD.
Proof. By Proposition 2.4 , TCL implies complete integral closure. A completely integrally closed domain is a v v v -domain [2 ] , and the inclusion form in Lemma 2.2 gives the a.b. property of t t t .
If D D D is t t t -almost Dedekind, it is a PvMD, so t = w t = w t = w . By Lemma 2.5, a t t t -product equality becomes an ordinary product equality in each DVR D M D_M D M , M ∈ t M \in t M ∈ t -Max( D ) (D) ( D ) . Cancellation there, followed by intersection, gives the full t t t -law.
Conversely, suppose that D D D is a PvMD satisfying TCL. Then t = w t = w t = w , so every nonzero ideal is a w w w -cancellation ideal. Chang’s local-principality characterization implies that D M D_M D M is a principal ideal domain for every M ∈ t M \in t M ∈ t -Max( D ) (D) ( D ) [5 ] . Each nonfield D M D_M D M is therefore a DVR. This proves (iii), and (iv) follows. The same characterization and Theorem 2.3 show, without assuming that D D D is a PvMD, that full w w w -linkage is equivalent to the t t t -almost Dedekind property. □
The inclusions supplied so far are t t t -AD ⊆ \subseteq ⊆ TCL ⊆ \subseteq ⊆ CIC ⊆ \subseteq ⊆ v v v -domain. The examples in the next section prove that the first two are strict and that neither PvMD nor almost Krullness can replace the middle node.
Two boundary constructions
This section constructs a t t t -cancellative monoid domain outside the local Krull world and then proves, by a separate patching argument, that the Heinzer–Ohm almost Krull non-PvMD also satisfies full t t t -cancellation.
A monoid domain outside the PvMD and almost Krull classes
A grading monoid is a cancellative torsion-free additive monoid S S S with quotient group G G G . A nonempty subset I ⊆ G I \subseteq G I ⊆ G is a fractional ideal of S S S if S + I ⊆ I S + I \subseteq I S + I ⊆ I and s + I ⊆ S s + I \subseteq S s + I ⊆ S for some s ∈ S s \in S s ∈ S . Writing ( S : I ) = { g ∈ G : g + I ⊆ S } (S : I) = \{g \in G : g + I \subseteq S\} ( S : I ) = { g ∈ G : g + I ⊆ S } , one defines I v = ( S : ( S : I ) ) I_v = (S : (S : I)) I v = ( S : ( S : I )) and I t = ⋃ { E v : E ⊆ I , E finite } I_t = \bigcup\{E_v : E \subseteq I,\ E\text{ finite}\} I t = ⋃ { E v : E ⊆ I , E finite } . These are the additive counterparts of the domain operations; see [12 ] and [14 ] . For a field k k k , let k [ S ] k[S] k [ S ] be the monoid algebra and, for I ⊆ G I \subseteq G I ⊆ G , let k [ I ] k[I] k [ I ] be the k k k -space with monomial basis { X a : a ∈ I } \{X^a : a \in I\} { X a : a ∈ I } .
The following transfer result turns cancellation in the monoid into cancellation in its monoid algebra.
Proposition 3.1. Let k k k be a field, let G G G be a free abelian group, and let S ⊆ G S \subseteq G S ⊆ G be an integrally closed submonoid with quotient group G G G . If every fractional t t t -ideal of S S S is t t t -cancellative, then k [ S ] k[S] k [ S ] satisfies TCL.
Proof. A free abelian group satisfies the type-(0, 0, . . .) hypothesis of Kim and Park. Their representation theorem first applies to integral t t t -ideals, with scalar in k [ G ] k[G] k [ G ] . Multiplying an arbitrary fractional t t t -ideal by a nonzero element of k [ S ] k[S] k [ S ] makes it integral, so every fractional t t t -ideal has the form α k [ I ] \alpha k[I] α k [ I ] , where 0 ≠ α ∈ Frac ( k [ G ] ) = qf ( k [ S ] ) 0 \ne\alpha\in\operatorname{Frac}(k[G]) = \operatorname{qf}(k[S]) 0 = α ∈ Frac ( k [ G ]) = qf ( k [ S ]) and I I I is a fractional t t t -ideal of S S S . The scalar need not belong to k [ G ] k[G] k [ G ] in the fractional case; see [22 ] , Lemmas 2.1 and 2.4. For every fractional ideal E E E of S S S , one has
( k [ E ] ) − 1 = k [ ( S : E ) ] , ( k [ E ] ) t = k [ E t ] . (8) (k[E])^{-1} = k[(S : E)], \qquad(k[E])_t = k[E_t]. \tag*{(8)} ( k [ E ] ) − 1 = k [( S : E )] , ( k [ E ] ) t = k [ E t ] . (8)
Indeed, multiplying an inverse element by any monomial of k [ E ] k[E] k [ E ] forces that element to lie in k [ G ] k[G] k [ G ] , and comparison of monomial supports gives the inverse formula. Applying it twice gives the ν \nu ν -closure formula. Every finite collection of polynomials in k [ E ] k[E] k [ E ] involves only finitely many monomials, so the t t t -closure formula follows as well.
Let A = α k [ I ] A=\alpha k[I] A = α k [ I ] , B = β k [ J ] B=\beta k[J] B = β k [ J ] , and C = γ k [ L ] C=\gamma k[L] C = γ k [ L ] be fractional t t t -ideals of k [ S ] k[S] k [ S ] , and suppose ( A B ) t = ( A C ) t (AB)_t=(AC)_t ( A B ) t = ( A C ) t . By (8) , cancelling the scalar α \alpha α gives
β k [ P ] = γ k [ Q ] , P = ( I + J ) t , Q = ( I + L ) t . (9) \beta k[P]=\gamma k[Q],\qquad P=(I+J)_t,\qquad Q=(I+L)_t. \tag*{(9)} β k [ P ] = γ k [ Q ] , P = ( I + J ) t , Q = ( I + L ) t . (9)
After extension to k [ G ] k[G] k [ G ] , both monomial spaces generate k [ G ] k[G] k [ G ] , so β k [ G ] = γ k [ G ] \beta k[G]=\gamma k[G] β k [ G ] = γ k [ G ] . Since the units of the group algebra of a free abelian group are the scalar monomials, β / γ = u X g \beta/\gamma=uX^g β / γ = u X g for some u ∈ k × u\in k^\times u ∈ k × and g ∈ G g\in G g ∈ G . The monomial basis in (9) then gives Q = g + P Q=g+P Q = g + P . Hence ( I + L ) t = ( I + ( g + J ) ) t (I+L)_t=(I+(g+J))_t ( I + L ) t = ( I + ( g + J ) ) t . Cancellation by I I I in the fractional t t t -ideal monoid of S S S yields L = g + J L=g+J L = g + J , and consequently C = γ k [ L ] = γ X g k [ J ] = u − 1 β k [ J ] = B C=\gamma k[L]=\gamma X^gk[J]=u^{-1}\beta k[J]=B C = γ k [ L ] = γ X g k [ J ] = u − 1 β k [ J ] = B . □
We now use the explicit monoid from Reinhart’s counterexample. Let G = ⨁ i ∈ N 0 Z e i G=\bigoplus_{i\in\mathbb{N}_0}\mathbb{Z}e_i G = ⨁ i ∈ N 0 Z e i , write x = ∑ i x i e i x=\sum_i x_i e_i x = ∑ i x i e i , and set
H = { x ∈ G : x 0 ≥ x i ≥ 0 for every i ∈ N 0 } . (10) H=\{x\in G:x_0\ge x_i\ge0\text{ for every }i\in\mathbb{N}_0\}. \tag*{(10)} H = { x ∈ G : x 0 ≥ x i ≥ 0 for every i ∈ N 0 } . (10)
The quotient group of H H H is G G G , since e 0 , e 0 + e i ∈ H e_0,e_0+e_i\in H e 0 , e 0 + e i ∈ H , and H H H is integrally closed because its defining inequalities are preserved under positive integral division.
Before applying Proposition 3.1 , we record a direct obstruction to almost Krullness. This replaces any reliance on a general semigroup-ring criterion.
Lemma 3.2. Let R = k [ H ] R=k[H] R = k [ H ] and m = ( X h : 0 ≠ h ∈ H ) \mathfrak{m}=(X^h:0\ne h\in H) m = ( X h : 0 = h ∈ H ) . Then R m R_{\mathfrak{m}} R m is not a Krull domain. Consequently, R R R is not almost Krull.
Proof. For i ≥ 1 i\ge1 i ≥ 1 , define λ i : G → Z \lambda_i:G\to\mathbb{Z} λ i : G → Z by λ i ( x ) = x 0 − x i \lambda_i(x)=x_0-x_i λ i ( x ) = x 0 − x i , and put F i = H ∩ ker ( λ i ) F_i=H\cap\ker(\lambda_i) F i = H ∩ ker ( λ i ) and P i = { h ∈ H : λ i ( h ) > 0 } P_i=\{h\in H:\lambda_i(h)>0\} P i = { h ∈ H : λ i ( h ) > 0 } . The complement F i F_i F i is a face of H H H , so P i P_i P i is a prime ideal of the monoid. Hence p i = k [ P i ] \mathfrak{p}_i=k[P_i] p i = k [ P i ] is a prime ideal of R R R .
Write H − F i = { h − f : h ∈ H , f ∈ F i } H-F_i=\{h-f:h\in H,\ f\in F_i\} H − F i = { h − f : h ∈ H , f ∈ F i } . We claim that H − F i = { g ∈ G : λ i ( g ) ≥ 0 } H-F_i=\{g\in G:\lambda_i(g)\ge0\} H − F i = { g ∈ G : λ i ( g ) ≥ 0 } . One inclusion is immediate. For the reverse, let g ∈ G g\in G g ∈ G satisfy λ i ( g ) ≥ 0 \lambda_i(g)\ge0 λ i ( g ) ≥ 0 . Choose N N N large enough that all coordinates used below are nonnegative and that g 0 + N ≥ max { g j , 0 } g_0+N\ge\max\{g_j,0\} g 0 + N ≥ max { g j , 0 } for every j j j . Define f ∈ G f\in G f ∈ G by f 0 = f i = N f_0=f_i=N f 0 = f i = N and f j = max { 0 , − g j } f_j=\max\{0,-g_j\} f j = max { 0 , − g j } for j ∉ { 0 , i } j\notin\{0,i\} j ∈ / { 0 , i } , enlarging N N N so that N ≥ f j N\ge f_j N ≥ f j for all j j j . Then f ∈ F i f\in F_i f ∈ F i and g + f ∈ H g+f\in H g + f ∈ H , proving the claim.
Let L i = ker ( λ i ) L_i=\ker(\lambda_i) L i = ker ( λ i ) . Since λ i ( e 0 ) = 1 \lambda_i(e_0)=1 λ i ( e 0 ) = 1 , the preceding claim gives H − F i = L i + N 0 e 0 H-F_i=L_i+\mathbb{N}_0e_0 H − F i = L i + N 0 e 0 . Inverting first the monomials from F i F_i F i and then the complement of p i \mathfrak{p}_i p i therefore yields
R p i ≅ Frac ( k [ L i ] ) [ T ] ( T ) , T = X e 0 . (11) R_{\mathfrak{p}_i}\cong\operatorname{Frac}(k[L_i])[T]_{(T)},\qquad T=X^{e_0}. \tag*{(11)} R p i ≅ Frac ( k [ L i ]) [ T ] ( T ) , T = X e 0 . (11)
Thus R p i R_{\mathfrak{p}_i} R p i is a discrete valuation domain, and p i \mathfrak{p}_i p i has height one. The primes p i \mathfrak{p}_i p i are distinct: e 0 + e i ∈ F i e_0+e_i\in F_i e 0 + e i ∈ F i but e 0 + e i ∈ P j e_0+e_i\in P_j e 0 + e i ∈ P j when j ≠ i j\ne i j = i . They are all contained in the maximal ideal m \mathfrak{m} m , and the nonzero element X e 0 X^{e_0} X e 0 belongs to every p i \mathfrak{p}_i p i .
If R m R_{\mathfrak{m}} R m were Krull, the element X e 0 X^{e_0} X e 0 would belong to only finitely many height-one prime ideals, by the finite-character property of Krull domains; see [4 ] , Chapter VII, Section 1 or [14 ] . The infinitely many primes p i R m \mathfrak{p}_iR_{\mathfrak{m}} p i R m give a contradiction. Since an almost Krull domain has a Krull localization at every maximal ideal, R R R is not almost Krull. □
The monoid example now gives the first strict boundary point.
Theorem 3.3. For every field k k k , the domain k [ H ] k[H] k [ H ] satisfies full t t t -linkage and TCL, but it is neither a PvMD nor an almost Krull domain. In particular, TCL ⊈ \not\subseteq ⊆ PvMD and TCL ⊈ \not\subseteq ⊆ AK.
Proof. Reinhart proves that every nonempty t t t -ideal of H H H is t t t -cancellative, that t t t -dim ( H ) = 2 \dim(H)=2 dim ( H ) = 2 , and that H H H is not a t t t -Prüfer monoid [31 ] , Example 4.2. Translation in G G G carries a finite family of fractional ideals into integral ideals, so the stated cancellation also holds for fractional t t t -ideals. Thus Proposition 3.1 gives TCL, and Theorem 2.3 gives full t t t -linkage.
If k [ H ] k[H] k [ H ] were a PvMD, every finitely generated monomial ideal would be t t t -invertible. By the inverse and closure identities in (8) , each t t t -finite t t t -ideal of H H H would then be t t t -invertible. This would make H H H a t t t -Prüfer monoid, contrary to Reinhart’s example. The failure of almost Krullness is Lemma 3.2 .
Finite patching and the Heinzer–Ohm domain
The second construction requires a local-to-global argument because the t t t -operation does not generally commute with localization. The next theorem isolates exactly the finite-patching mechanism needed below.
Theorem 3.4. Suppose that Max ( D ) = { M } ∪ { P i : i ≥ 1 } \operatorname{Max}(D)=\{M\}\cup\{P_i:i\geq1\} Max ( D ) = { M } ∪ { P i : i ≥ 1 } . Set R = D M R=D_M R = D M and V i = D P i V_i=D_{P_i} V i = D P i , and let v i v_i v i be the normalized valuation of V i V_i V i . Assume that
(a) R R R is a Krull domain and every V i V_i V i is a discrete valuation domain;
(b) for every 0 ≠ x ∈ D 0\neq x\in D 0 = x ∈ D , the sequence ( v i ( x ) ) i ≥ 1 (v_i(x))_{i\geq1} ( v i ( x ) ) i ≥ 1 is eventually constant.
Then D D D satisfies TCL.
Proof. Replacing each factor by its t t t -closure does not change a t t t -product. After a common scaling, let A A A , B B B , C C C be nonzero integral t t t -ideals with
( A B ) t = ( A C ) t . (12) (AB)_t=(AC)_t. \tag*{(12)} ( A B ) t = ( A C ) t . (12)
For 0 ≠ x ∈ D 0\neq x\in D 0 = x ∈ D , write w ( x ) w(x) w ( x ) for the eventual value of v i ( x ) v_i(x) v i ( x ) . Choose 0 ≠ a 0 ∈ A 0\neq a_0\in A 0 = a 0 ∈ A with w ( a 0 ) w(a_0) w ( a 0 ) minimal among the values w ( a ) w(a) w ( a ) for 0 ≠ a ∈ A 0\neq a\in A 0 = a ∈ A .
Fix 0 ≠ b ∈ B 0\neq b\in B 0 = b ∈ B . Since a 0 b ∈ ( A C ) t a_0b\in(AC)_t a 0 b ∈ ( A C ) t , finite type gives a finitely generated ideal H 0 ⊆ A C H_0\subseteq AC H 0 ⊆ A C with a 0 b ∈ ( H 0 ) v a_0b\in(H_0)_v a 0 b ∈ ( H 0 ) v . Expanding its finitely many generators, choose finitely generated ideals E ⊆ A E\subseteq A E ⊆ A and F ⊆ C F\subseteq C F ⊆ C such that H 0 ⊆ E F H_0\subseteq EF H 0 ⊆ E F . Thus a 0 b ∈ ( E F ) v = ( E F ) t a_0b\in(EF)_v=(EF)_t a 0 b ∈ ( E F ) v = ( E F ) t . By Lemma 2.5, a 0 b ∈ E F V i a_0b\in EFV_i a 0 b ∈ E F V i , and hence v i ( a 0 ) + v i ( b ) ≥ v i ( E ) + v i ( F ) v_i(a_0)+v_i(b)\geq v_i(E)+v_i(F) v i ( a 0 ) + v i ( b ) ≥ v i ( E ) + v i ( F ) for every i i i . Once the finitely many terms have stabilized, minimality of w ( a 0 ) w(a_0) w ( a 0 ) yields w ( b ) ≥ w ( F ) w(b)\geq w(F) w ( b ) ≥ w ( F ) . Therefore
b ∈ F V i for all sufficiently large i . (13) b\in FV_i\qquad\text{for all sufficiently large }i. \tag*{(13)} b ∈ F V i for all sufficiently large i . (13)
Localizing (12) at P i P_i P i and using Lemma 2.5 gives ( A V i ) ( B V i ) = ( A V i ) ( C V i ) (AV_i)(BV_i)=(AV_i)(CV_i) ( A V i ) ( B V i ) = ( A V i ) ( C V i ) . Cancellation in the DVR V i V_i V i yields B V i = C V i BV_i=CV_i B V i = C V i . For each of the finitely many indices omitted in (13) , choose a finite subideal F i ⊆ C F_i\subseteq C F i ⊆ C with b ∈ F i V i b\in F_iV_i b ∈ F i V i . Enlarging F F F by these finitely many subideals gives a finitely generated F ′ ⊆ C F'\subseteq C F ′ ⊆ C such that b ∈ F ′ V i b\in F'V_i b ∈ F ′ V i for every i i i .
At M M M , Lemma 2.5 gives ( ( L t ) R ) t R = ( L R ) t R ((L_t)R)_{t_R}=(LR)_{t_R} (( L t ) R ) t R = ( L R ) t R for every fractional ideal L L L . Thus (12) gives ( A R B R ) t R = ( A R C R ) t R (ARBR)_{t_R}=(ARCR)_{t_R} ( A R B R ) t R = ( A R C R ) t R . Since R R R is Krull, t R = v R t_R=v_R t R = v R and the divisorial ideal monoid of R R R is a group. Cancelling ( A R ) v R (AR)_{v_R} ( A R ) v R gives ( B R ) v R = ( C R ) v R (BR)_{v_R}=(CR)_{v_R} ( B R ) v R = ( C R ) v R .
Hence b ∈ ( C R ) v R b\in(CR)_{v_R} b ∈ ( C R ) v R . By finite type, there is a finitely generated R R R -submodule L ⊆ C R L\subseteq CR L ⊆ C R with b ∈ L v R b\in L_{v_R} b ∈ L v R . Clearing the finitely many denominators in generators of L L L gives a finitely generated subideal F M ⊆ C F_M\subseteq C F M ⊆ C with L ⊆ F M R L\subseteq F_MR L ⊆ F M R , and therefore b ∈ ( F M R ) v R b\in(F_MR)_{v_R} b ∈ ( F M R ) v R .
Set G = F ′ + F M G=F'+F_M G = F ′ + F M . Then G G G is finitely generated and contained in C C C . If x ∈ G − 1 x\in G^{-1} x ∈ G − 1 , the containment b ∈ G V i b\in GV_i b ∈ G V i gives b x ∈ V i bx\in V_i b x ∈ V i for every i i i . Also x G R ⊆ R xGR\subseteq R x GR ⊆ R , so x ∈ ( R : G R ) x\in(R:GR) x ∈ ( R : GR ) ; since b ∈ ( G R ) v R = ( R : ( R : G R ) ) b\in(GR)_{v_R}=(R:(R:GR)) b ∈ ( GR ) v R = ( R : ( R : GR )) , one has b x ∈ R bx\in R b x ∈ R . Consequently,
b G − 1 ⊆ R ∩ ⋂ i ≥ 1 V i = D . (14) bG^{-1}\subseteq R\cap\bigcap_{i\geq1}V_i=D. \tag*{(14)} b G − 1 ⊆ R ∩ i ≥ 1 ⋂ V i = D . (14)
It follows that b ∈ G v = G t ⊆ C b\in G_v=G_t\subseteq C b ∈ G v = G t ⊆ C . Hence B ⊆ C B\subseteq C B ⊆ C , and the symmetric argument gives C ⊆ B C\subseteq B C ⊆ B .
We apply the criterion to the essential domain of Heinzer and Ohm. Let k k k be a field, let x 1 , x 2 , … , y , z x_{1}, x_{2}, \ldots, y, z x 1 , x 2 , … , y , z be indeterminates, and put
K = k ( x 1 , x 2 , … , y , z ) , R = k ( x 1 , x 2 , … ) [ y , z ] ( y , z ) . (15) K = k(x_{1},x_{2},\ldots,y,z), \qquad R = k(x_{1},x_{2},\ldots)[y,z]_{(y,z)}. \tag*{(15)} K = k ( x 1 , x 2 , … , y , z ) , R = k ( x 1 , x 2 , … ) [ y , z ] ( y , z ) . (15)
For each i ≥ 1 i \ge1 i ≥ 1 , let V i V_i V i be the rank-one discrete valuation domain of K K K defined by the Gauss (minimum-monomial-weight) valuation satisfying v i ( x i ) = v i ( y ) = v i ( z ) = 1 v_i(x_i) = v_i(y) = v_i(z) = 1 v i ( x i ) = v i ( y ) = v i ( z ) = 1 and v i ( x j ) = 0 v_i(x_j) = 0 v i ( x j ) = 0 for j ≠ i j \ne i j = i . Define
D = R ∩ ⋂ i ≥ 1 V i . (16) D = R \cap\bigcap_{i \ge1} V_i. \tag*{(16)} D = R ∩ i ≥ 1 ⋂ V i . (16)
A domain is essential if it is an intersection of valuation overrings that are localizations of the domain. Heinzer and Ohm constructed this domain to show that an essential domain need not be a v v v -multiplication domain [15 ] . Houston, Kabbaj, and Mimouni later isolated the valuation behavior that is needed for patching [16 Example 3.3] .
The cited description gives the following facts in a form suited to Theorem 3.4 .
Proposition 3.5. For the domain in (16) , there are maximal ideals M M M , P 1 P_1 P 1 , P 2 P_2 P 2 , … \ldots … such that
(i) Max ( D ) = { M } ∪ { P i : i ≥ 1 } \operatorname{Max}(D) = \{M\} \cup\{P_i : i \ge1\} Max ( D ) = { M } ∪ { P i : i ≥ 1 } , D M = R D_M = R D M = R , and D P i = V i D_{P_i} = V_i D P i = V i ;
(ii) D D D is almost Krull and is not a PvMD;
(iii) for every 0 ≠ a ∈ D 0 \ne a \in D 0 = a ∈ D , the sequence ( v i ( a ) ) (v_i(a)) ( v i ( a )) is eventually constant;
(iv) J t V i = J V i J_tV_i = JV_i J t V i = J V i for every nonzero ideal J J J of D D D and every i i i .
Proof. Assertions (i), (iii), and (iv) are proved for this construction in [16 Example 3.3] . Since R R R is a two-dimensional regular local ring and each V i V_i V i is a DVR, (i) shows that every localization of D D D at a maximal ideal is Krull; hence D D D is almost Krull. The failure of the PvMD property is the defining conclusion of [15 ] . □
The patching theorem now gives an almost Krull example in which full t t t -cancellation survives without the PvMD property.
Theorem 3.6. The Heinzer–Ohm domain in (16) satisfies full t t t -linkage and TCL, but it is not a PvMD.
Proof. The ring R R R is a UFD and hence Krull, and Proposition 3.5 (i) and (iii) verify all hypotheses of Theorem 3.4 . Thus D D D satisfies TCL; full t t t -linkage follows from Theorem 2.3 , while Proposition 3.5 (ii) gives the failure of the PvMD property. □
The canonical operation and positive descent
This section identifies the operation naturally attached to an almost Krull domain and describes it through well-behaved t t t -primes. The Arnold–Matsuda domain then shows that divisorial height-one primes and a higher-height non-well-behaved maximal t t t -ideal are compatible with full t t t -cancellation. The next section gives the contrasting failure.
The canonical locally divisorial operation
Let v N v_N v N denote the divisorial operation on D N D_N D N . The canonical locally divisorial operation is
I κ = ⋂ N ∈ Max ( D ) ( I D N ) v N ( I ∈ F ( D ) ) . (17) I_{\kappa} = \bigcap_{N \in\operatorname{Max}(D)} (ID_N)_{v_N} \qquad(I \in\mathcal{F}(D)). \tag*{(17)} I κ = N ∈ Max ( D ) ⋂ ( I D N ) v N ( I ∈ F ( D )) . (17)
The intersection in (17) is a star operation. Indeed, a common fractional bound for I I I also bounds every local divisorial closure; scalar compatibility and monotonicity are immediate, and I κ ⊆ ( I D N ) v N I_{\kappa} \subseteq(I D_N)_{v_N} I κ ⊆ ( I D N ) v N gives idempotence after localization and intersection. Finally, D κ = ⋂ N D N = D D_{\kappa} = \bigcap_N D_N = D D κ = ⋂ N D N = D .
Its finite-type operation is defined by I κ f = ⋃ { F κ : F ∈ f ( D ) , F ⊆ I } I_{\kappa_f} = \bigcup\{F_{\kappa}: F \in\mathfrak{f}(D),\ F \subseteq I\} I κ f = ⋃ { F κ : F ∈ f ( D ) , F ⊆ I } . The local theory of this operation is developed in [6 ] .
A nonzero prime t t t -ideal P P P is called well-behaved if P D S P D_S P D S is a t D S t_{D_S} t D S -ideal for every multiplicative set S S S disjoint from P P P ; equivalently, it is enough that P D P P D_P P D P be a t D P t_{D_P} t D P -ideal. Houston and Zafrullah show that maximal well-behaved t t t -ideals induce a spectral star operation and identify that operation on an almost Krull domain [17 Theorem 2.3 and Corollary 4.8] .
The next proposition collects the properties that distinguish κ \kappa κ from its finite-type shadow.
Proposition 4.1. Let D D D be almost Krull and let κ \kappa κ be defined by (17) . Then the following statements hold.
(i) D D D is completely integrally closed.
(ii) The operation κ \kappa κ is cancellative, and D D D has full κ \kappa κ -linkage.
(iii) κ f = t \kappa_f = t κ f = t .
(iv) The following conditions are equivalent: κ = t \kappa= t κ = t ; D D D is a PvMD; and D D D is t t t -almost Dedekind.
Proof. An element almost integral over D D D is almost integral over every Krull domain D N D_N D N , and hence belongs to every D N D_N D N . Since D = ⋂ N ∈ Max ( D ) D N D = \bigcap_{N \in\operatorname{Max}(D)} D_N D = ⋂ N ∈ Max ( D ) D N , this proves (i).
Suppose ( A B ) κ = ( A C ) κ (AB)_{\kappa} = (AC)_{\kappa} ( A B ) κ = ( A C ) κ . For every fractional ideal L L L and maximal ideal N N N , the inclusions L ⊆ L κ ⊆ ( L D N ) v N L \subseteq L_{\kappa} \subseteq(L D_N)_{v_N} L ⊆ L κ ⊆ ( L D N ) v N give ( ( L κ ) D N ) v N = ( L D N ) v N ((L_{\kappa})D_N)_{v_N} = (L D_N)_{v_N} (( L κ ) D N ) v N = ( L D N ) v N . Applying this identity to the common κ \kappa κ -closure of A B AB A B and A C AC A C yields ( A D N B D N ) v N = ( A D N C D N ) v N (A D_N B D_N)_{v_N} = (A D_N C D_N)_{v_N} ( A D N B D N ) v N = ( A D N C D N ) v N . The divisorial fractional ideals of the Krull domain D N D_N D N form a group, so ( B D N ) v N = ( C D N ) v N (B D_N)_{v_N} = (C D_N)_{v_N} ( B D N ) v N = ( C D N ) v N . Intersecting over N N N gives B κ = C κ B_{\kappa} = C_{\kappa} B κ = C κ . Thus κ \kappa κ is cancellative, and full κ \kappa κ -linkage follows from Theorem 2.3 .
For F ∈ f ( D ) F \in\mathfrak{f}(D) F ∈ f ( D ) , we claim that F κ = F v F_{\kappa} = F_v F κ = F v . Since F F F is finitely generated, inverse formation commutes with localization: ( D : F ) D N = ( D N : F D N ) (D : F)D_N = (D_N : F D_N) ( D : F ) D N = ( D N : F D N ) . Hence x ∈ F v x \in F_v x ∈ F v implies x ( D N : F D N ) ⊆ D N x(D_N : F D_N) \subseteq D_N x ( D N : F D N ) ⊆ D N , and therefore x ∈ ( F D N ) v N x \in(F D_N)_{v_N} x ∈ ( F D N ) v N for every N N N ; this proves F v ⊆ F κ F_v \subseteq F_{\kappa} F v ⊆ F κ . Conversely, let x ∈ F κ x \in F_{\kappa} x ∈ F κ and y ∈ F − 1 y \in F^{-1} y ∈ F − 1 . For each N N N , y ∈ ( D N : F D N ) y \in(D_N : F D_N) y ∈ ( D N : F D N ) and x ∈ ( F D N ) v N x \in(F D_N)_{v_N} x ∈ ( F D N ) v N , so x y ∈ D N xy \in D_N x y ∈ D N . Hence x y ∈ D xy \in D x y ∈ D for every y ∈ F − 1 y \in F^{-1} y ∈ F − 1 , and therefore x ∈ F v x \in F_v x ∈ F v . Taking the union over all finitely generated subideals proves κ f = t \kappa_f = t κ f = t .
The equivalence κ = t ⟺ D \kappa= t \Longleftrightarrow D κ = t ⟺ D is t t t -almost Dedekind is [6 Propositions 1.6 and 1.7] . Every t t t -almost Dedekind domain is a PvMD. Conversely, if D D D is almost Krull and a PvMD, then for P ∈ t -Max ( D ) P \in t\text{-Max}(D) P ∈ t -Max ( D ) and a maximal ideal N ⊇ P N \supseteq P N ⊇ P , the localization D P = ( D N ) P D N D_P = (D_N)_{P D_N} D P = ( D N ) P D N is both Krull and a valuation domain; hence it is a DVR unless it is a field. Thus D D D is t t t -almost Dedekind. □
The well-behaved-prime description makes the local support of κ \kappa κ explicit and will be used in the second patching construction.
Proposition 4.2. Let D D D be an almost Krull domain, and let κ \kappa κ be defined by (17) . Then the following statements hold.
(i) A nonzero prime t t t -ideal P P P is well-behaved if and only if ht P = 1 \operatorname{ht} P = 1 ht P = 1 .
(ii) For every fractional ideal I I I ,
I κ = ⋂ ht P = 1 I D P . (18) I_{\kappa} = \bigcap_{\operatorname{ht} P=1} I D_P. \tag*{(18)} I κ = ht P = 1 ⋂ I D P . (18) Moreover, I κ D P = I D P I_{\kappa}D_P = I D_P I κ D P = I D P for every height-one prime P P P .
(iii) The equality κ = v \kappa= v κ = v holds if and only if every height-one prime of D D D is divisorial.
Proof. If P P P is nonzero and well-behaved, the maximal ideal P D P PD_P P D P is a proper t D P t_{D_P} t D P -ideal of the Krull domain D P D_P D P . Such a maximal ideal has height one, so ht P = 1 \operatorname{ht} P=1 ht P = 1 . Conversely, if P P P has height one, then D P D_P D P is a DVR and Lemma 2.5 gives P t D P = P D P P_tD_P=PD_P P t D P = P D P . Since P t ⊆ D P_t\subseteq D P t ⊆ D , contraction shows that P t = P P_t=P P t = P ; also P D P PD_P P D P is a t D P t_{D_P} t D P -ideal, proving (i).
Represent each divisorial ideal ( I D N ) v N (ID_N)_{v_N} ( I D N ) v N as the intersection of its extensions to the height-one localizations of the Krull domain D N D_N D N . Primes contained in N N N have the same height before and after localization. Taking the intersections over N N N gives (18) . For a fixed height-one P P P , the inclusions I ⊆ I κ ⊆ I D P I\subseteq I_\kappa\subseteq ID_P I ⊆ I κ ⊆ I D P immediately give I κ D P = I D P I_\kappa D_P=ID_P I κ D P = I D P . For (iii), suppose first that κ = v \kappa=v κ = v . Since P κ ⊆ P D P ∩ D = P P_\kappa\subseteq PD_P\cap D=P P κ ⊆ P D P ∩ D = P for a height-one prime P P P , one has P v = P P_v=P P v = P .
Conversely, assume that every height-one prime P P P is divisorial. Fix such a P P P and let ν P \nu_P ν P be the normalized valuation of D P D_P D P . By complete integral closure and the characterization in the proof of Proposition 2.4 , ( P P − 1 ) v = D (PP^{-1})_v=D ( P P − 1 ) v = D . Since P = P v ≠ D P=P_v\ne D P = P v = D , there exist a ∈ P a\in P a ∈ P and b ∈ P − 1 b\in P^{-1} b ∈ P − 1 with a b ∉ P ab\notin P ab ∈ / P . The equality P D P = { z ∈ D P : ν P ( z ) > 0 } PD_P=\{z\in D_P:\nu_P(z)>0\} P D P = { z ∈ D P : ν P ( z ) > 0 } and b P ⊆ D bP\subseteq D b P ⊆ D then force ν P ( a ) = 1 \nu_P(a)=1 ν P ( a ) = 1 and ν P ( b ) = − 1 \nu_P(b)=-1 ν P ( b ) = − 1 .
We claim that if I ⊆ D I\subseteq D I ⊆ D is nonzero and ν P ( I ) ≥ n \nu_P(I)\ge n ν P ( I ) ≥ n for an integer n ≥ 0 n\ge0 n ≥ 0 , then ν P ( I v ) ≥ n \nu_P(I_v)\ge n ν P ( I v ) ≥ n . For n = 0 n=0 n = 0 this follows from I v ⊆ D I_v\subseteq D I v ⊆ D ; for n = 1 n=1 n = 1 it follows from I ⊆ P = P v I\subseteq P=P_v I ⊆ P = P v . If n ≥ 2 n\ge2 n ≥ 2 , then I ⊆ P I\subseteq P I ⊆ P and b I ⊆ D bI\subseteq D b I ⊆ D , with ν P ( b I ) ≥ n − 1 \nu_P(bI)\ge n-1 ν P ( b I ) ≥ n − 1 . Induction and ( b I ) v = b I v (bI)_v=bI_v ( b I ) v = b I v give the claim. The values of a nonzero integral ideal in the DVR form a subset of N 0 \mathbb{N}_0 N 0 with a minimum. Applying the claim at this minimum yields I v D P = I D P I_vD_P=ID_P I v D P = I D P . Scaling proves the same equality for fractional ideals. Intersecting over height-one primes gives I v ⊆ I κ I_v\subseteq I_\kappa I v ⊆ I κ . The reverse inclusion holds because every star operation is bounded by v v v , proving κ = v \kappa=v κ = v . This also proves directly the characterization in [17 Corollary 4.10] . □
Fontana–Loper theory identifies the canonical finite-type representative of κ \kappa κ , but it does not by itself prove full cancellation after passage to finite type.
Proposition 4.3. Extend κ \kappa κ to a semistar operation on all nonzero D D D -submodules E E E of K K K by retaining (17) on fractional ideals and setting E κ = K E_\kappa=K E κ = K when E E E is not fractional. Then
t = κ f = κ K r = b ( κ ) = κ a . (19) t=\kappa_f=\kappa_{\mathrm{Kr}}=b(\kappa)=\kappa_a. \tag*{(19)} t = κ f = κ Kr = b ( κ ) = κ a . (19)
In this formula, t t t is extended to arbitrary nonzero D D D -submodules by E t = ⋃ F ∈ f ( D ) , F ⊆ E F v E_t=\bigcup_{F\in f(D),\,F\subseteq E}F_v E t = ⋃ F ∈ f ( D ) , F ⊆ E F v .
A semistar operation satisfies the closure and scalar axioms of a star operation on all nonzero D D D -submodules of K K K , without requiring D ⋆ = D D_\star=D D ⋆ = D . The notation in (19) is that of [8 ] . Explicitly, let V ( κ ) \mathcal{V}(\kappa) V ( κ ) be the valuation overrings V V V such that F κ ⊆ F V F_\kappa\subseteq FV F κ ⊆ F V for every F ∈ f ( D ) F\in f(D) F ∈ f ( D ) ; then E b ( κ ) = ⋂ V ∈ V ( κ ) E V E_{b(\kappa)}=\bigcap_{V\in\mathcal{V}(\kappa)}EV E b ( κ ) = ⋂ V ∈ V ( κ ) E V . If X X X is an indeterminate and c ( f ) c(f) c ( f ) denotes the coefficient ideal of a polynomial f f f , the e.a.b. Kronecker function ring is
Kr ( D , κ ) = { 0 } ∪ { f / g : 0 ≠ f , g ∈ D [ X ] , c ( f ) κ ⊆ c ( g ) κ } , \operatorname{Kr}(D,\kappa)=\{0\}\cup\{f/g:0\ne f,g\in D[X],\ c(f)_\kappa\subseteq c(g)_\kappa\}, Kr ( D , κ ) = { 0 } ∪ { f / g : 0 = f , g ∈ D [ X ] , c ( f ) κ ⊆ c ( g ) κ } ,
and E κ K r = E Kr ( D , κ ) ∩ K E_{\kappa_{\mathrm{Kr}}}=E\operatorname{Kr}(D,\kappa)\cap K E κ Kr = E Kr ( D , κ ) ∩ K . Finally, κ a \kappa_a κ a is the finite-type operation determined on finitely generated F F F by F κ a = ⋃ H ∈ f ( D ) ( ( F H ) κ : H κ ) F_{\kappa_a}=\bigcup_{H\in f(D)}((FH)_\kappa:H_\kappa) F κ a = ⋃ H ∈ f ( D ) (( F H ) κ : H κ ) .
Proof. Full κ \kappa κ -cancellation implies the e.a.b. property, since all three ideals occurring in the e.a.b. test are fractional. Fontana and Loper prove that every e.a.b. semistar operation ⋆ \star ⋆ satisfies ⋆ f = ⋆ K r = b ( ⋆ ) = ⋆ a \star_f=\star_{\mathrm{Kr}}=b(\star)=\star_a ⋆ f = ⋆ Kr = b ( ⋆ ) = ⋆ a [8 Proposition 9] . Now apply Proposition 4.1(iii). □
The Heinzer–Ohm example shows that the equality of operations in Proposition 4.1(iv) is not required for cancellation to descend.
Corollary 4.4. For the Heinzer–Ohm domain, κ f = t < κ \kappa_f=t<\kappa κ f = t < κ , while both t t t and κ \kappa κ are cancellative. In particular, A K ∩ P v M D ⊊ A K ∩ T C L \mathrm{AK}\cap\mathrm{PvMD}\subsetneq\mathrm{AK}\cap\mathrm{TCL} AK ∩ PvMD ⊊ AK ∩ TCL .
Proof. By Proposition 3.5 and Theorem 3.6 , the domain is almost Krull, satisfies TCL, and is not a PvMD. Hence Proposition 4.1(iv) gives t < κ t < \kappa t < κ , while Proposition 4.1(ii)–(iii) gives the two cancellation statements. Every almost Krull PvMD is t t t -almost Dedekind and therefore belongs to TCL by Proposition 2.6 ; strictness follows from the same example.
A second finite-patching model with divisorial height-one primes
The next example tests descent under the strongest height-one divisoriality available in an almost Krull domain. It shows that κ = v \kappa= v κ = v can coexist with strict and fully cancellative passage to t = κ f t = \kappa_f t = κ f .
Let p 1 , p 2 , … p_1,p_2,\ldots p 1 , p 2 , … be the positive rational primes, let x , y x,y x , y be indeterminates, and set
D A M = Z [ x p i , y p i : i ≥ 1 ] ⊆ Q [ x , y ] . (20) D_{\mathrm{AM}} = \mathbb{Z}\left[\frac{x}{p_i},\frac{y}{p_i}:i \ge1\right] \subseteq\mathbb{Q}[x,y]. \tag*{(20)} D AM = Z [ p i x , p i y : i ≥ 1 ] ⊆ Q [ x , y ] . (20)
Put T = Q [ x , y ] T = \mathbb{Q}[x,y] T = Q [ x , y ] . Arnold and Matsuda determine the height-one valuation representation of this domain and use it to disprove Pirtle’s conjecture on divisorial height-one primes.
Proposition 4.5. For the domain D A M D_{\mathrm{AM}} D AM in (20) , the following statements hold.
(i) D A M D_{\mathrm{AM}} D AM is almost Krull but not Krull.
(ii) Its height-one primes are the ideals P i = p i D A M P_i = p_iD_{\mathrm{AM}} P i = p i D AM and the contractions Q f = f T ∩ D A M Q_f = fT \cap D_{\mathrm{AM}} Q f = f T ∩ D AM , where f f f ranges over the irreducible polynomials of T T T up to associates.
(iii) Every height-one prime is divisorial, but D A M D_{\mathrm{AM}} D AM is not a PvMD.
(iv) For the canonical operation κ \kappa κ ,
t = κ f < κ = v . (21) t = \kappa_f < \kappa= v. \tag*{(21)} t = κ f < κ = v . (21)
Proof. We give details of the construction in [3 ] . Write D = D A M D = D_{\mathrm{AM}} D = D AM . For each rational prime p i p_i p i , localizing at the integers not divisible by p i p_i p i gives
D Z ∖ p i Z = Z ( p i ) [ x / p i , y / p i ] , D Z ∖ { 0 } = T . (22) D_{\mathbb{Z}\setminus p_i\mathbb{Z}} = \mathbb{Z}_{(p_i)}[x/p_i,y/p_i], \qquad D_{\mathbb{Z}\setminus\{0\}} = T. \tag*{(22)} D Z ∖ p i Z = Z ( p i ) [ x / p i , y / p i ] , D Z ∖ { 0 } = T . (22)
Both rings are UFDs. Every prime of D D D contracts to either ( 0 ) (0) ( 0 ) or p i Z p_i\mathbb{Z} p i Z in Z \mathbb{Z} Z , so every maximal localization of D D D is a localization of one of these UFDs. Thus D D D is locally factorial and almost Krull.
The ideal p i D p_iD p i D is prime. To see this directly, if j ≠ i j \ne i j = i , Bezout’s identity gives x / ( p i p j ) ∈ D x/(p_ip_j) \in D x / ( p i p j ) ∈ D and y / ( p i p j ) ∈ D y/(p_ip_j) \in D y / ( p i p j ) ∈ D . Hence x / p j x/p_j x / p j and y / p j y/p_j y / p j vanish modulo p i D p_iD p i D . The remaining generators x / p i , y / p i x/p_i,y/p_i x / p i , y / p i satisfy no polynomial relation over F p i \mathbb{F}_{p_i} F p i , as is seen by localizing the first ring in (22) . Therefore D / p i D ≅ F p i [ x / p i , y / p i ] D/p_iD \cong\mathbb{F}_{p_i}[x/p_i,y/p_i] D / p i D ≅ F p i [ x / p i , y / p i ] . The localization at p i D p_iD p i D is a DVR, whose normalized valuation is
ν i ( ∑ c a b x a y b ) = min c a b ≠ 0 { v p i ( c a b ) + a + b } . (23) \nu_i\left(\sum c_{ab}x^a y^b\right) = \min_{c_{ab}\ne0}\{v_{p_i}(c_{ab}) + a + b\}. \tag*{(23)} ν i ( ∑ c ab x a y b ) = c ab = 0 min { v p i ( c ab ) + a + b } . (23)
The height-one primes contracting to p i Z p_i\mathbb{Z} p i Z consist just of p i D p_iD p i D , by (22) . Those contracting to zero are precisely the Q f Q_f Q f , by the second localization in that formula. This proves (ii). The nonzero element x x x belongs to all the distinct height-one primes p i D p_iD p i D , so D D D is not Krull.
By almost Krullness and (ii), D = T ∩ ⋂ i V i D = T \cap\bigcap_i V_i D = T ∩ ⋂ i V i , where V i = D p i D V_i = D_{p_iD} V i = D p i D . The primes p i D p_iD p i D are principal, hence divisorial. Fix an irreducible f ∈ T f \in T f ∈ T and multiply it by a rational scalar so that f ∈ Z [ x , y ] ⊆ D f \in\mathbb{Z}[x,y] \subseteq D f ∈ Z [ x , y ] ⊆ D . Put d = ord ( x , y ) ( f ) d = \operatorname{ord}_{(x,y)}(f) d = ord ( x , y ) ( f ) , the least total degree occurring in f f f . Formula (23) gives ν i ( f ) = d \nu_i(f) = d ν i ( f ) = d for all but finitely many i i i . Choose h ∈ { x , y } h \in\{x,y\} h ∈ { x , y } not divisible by f f f and an integer n ≥ max { 1 , d } n \ge\max\{1,d\} n ≥ max { 1 , d } . Multiplying h n h^n h n by a suitable positive integer produces s ∈ D s \in D s ∈ D with ν i ( s ) ≥ ν i ( f ) \nu_i(s) \ge\nu_i(f) ν i ( s ) ≥ ν i ( f ) for every i i i and f ∤ s f \nmid s f ∤ s in T T T . The intersection representation then gives
Q f = ( f D : D s ) = ( f / s ) D ∩ D . (24) Q_f = (fD :_D s) = (f/s)D \cap D. \tag*{(24)} Q f = ( f D : D s ) = ( f / s ) D ∩ D . (24)
Indeed, for z ∈ D z \in D z ∈ D the element s z / f sz/f sz / f automatically belongs to every V i V_i V i , and it belongs to T T T exactly when f f f divides z z z in T T T . Thus every Q f Q_f Q f is divisorial as well.
For completeness, put m = ( x , y ) T ∩ D \mathfrak{m}=(x,y)T\cap D m = ( x , y ) T ∩ D . Every finite set of nonzero elements of m \mathfrak{m} m has eventual ν i \nu_i ν i -values at least one. If E ⊆ m E\subseteq\mathfrak{m} E ⊆ m is finitely generated and 0 ≠ z ∈ E v 0\ne z\in E_v 0 = z ∈ E v , then Lemma 2.5 gives z ∈ E V i z\in EV_i z ∈ E V i for all i i i . The eventual value of z ∈ D ⊆ T z\in D\subseteq T z ∈ D ⊆ T is therefore positive, so z ∈ m z\in\mathfrak{m} z ∈ m . Hence m t = m \mathfrak{m}_t=\mathfrak{m} m t = m . Let F = ( x , y ) D F=(x,y)D F = ( x , y ) D . Coprimality of x , y x,y x , y in the UFD T T T implies F − 1 ⊆ T F^{-1}\subseteq T F − 1 ⊆ T . Consequently, F F − 1 ⊆ ( x , y ) T ∩ D = m FF^{-1}\subseteq(x,y)T\cap D=\mathfrak{m} F F − 1 ⊆ ( x , y ) T ∩ D = m , so F F F is not t t t -invertible. This proves the remaining assertion in (iii). Finally, (iv) follows from Proposition 4.2(iii) and Proposition 4.1(iii)–(iv).
We now prove that the strict inequality in (21) does not destroy full cancellation. This is a new consequence of the valuation structure recorded by Arnold and Matsuda.
Theorem 4.6. The Arnold–Matsuda domain D A M D_{\mathrm{AM}} D AM satisfies full t t t -linkage and TCL.
Proof. For each i i i , put V i = ( D A M ) P i V_i=(D_{\mathrm{AM}})_{P_i} V i = ( D AM ) P i and let ν i \nu_i ν i be its normalized valuation. By (23) , V i V_i V i is the valuation domain associated with the Gauss extension of the p i p_i p i -adic valuation satisfying ν i ( p i ) = ν i ( x ) = ν i ( y ) = 1 \nu_i(p_i)=\nu_i(x)=\nu_i(y)=1 ν i ( p i ) = ν i ( x ) = ν i ( y ) = 1 . Since D A M D_{\mathrm{AM}} D AM is almost Krull and its height-one primes are those in Proposition 4.5 (ii),
D A M = T ∩ ⋂ i ≥ 1 V i . (25) D_{\mathrm{AM}}=T\cap\bigcap_{i\geq1}V_i. \tag*{(25)} D AM = T ∩ i ≥ 1 ⋂ V i . (25)
Indeed, ( D A M ) Q f = T f T (D_{\mathrm{AM}})_{Q_f}=T_fT ( D AM ) Q f = T f T and the intersection of these localizations over all irreducible f f f is the UFD T T T .
For 0 ≠ h = ∑ c a b x a y b ∈ T 0\ne h=\sum c_{ab}x^ay^b\in T 0 = h = ∑ c ab x a y b ∈ T , the monomial valuation is
ν i ( h ) = min a , b { v p i ( c a b ) + a + b } . (26) \nu_i(h)=\min_{a,b}\{v_{p_i}(c_{ab})+a+b\}. \tag*{(26)} ν i ( h ) = a , b min { v p i ( c ab ) + a + b } . (26)
Consequently, if 0 ≠ ξ = f / g ∈ Q ( x , y ) 0\ne\xi=f/g\in\mathbb{Q}(x,y) 0 = ξ = f / g ∈ Q ( x , y ) , then the sequence ( ν i ( ξ ) ) i ≥ 1 (\nu_i(\xi))_{i\geq1} ( ν i ( ξ ) ) i ≥ 1 is eventually constant. Its eventual value is
ω ( ξ ) = ord ( x , y ) ( f ) − ord ( x , y ) ( g ) , (27) \omega(\xi)=\operatorname{ord}_{(x,y)}(f)-\operatorname{ord}_{(x,y)}(g), \tag*{(27)} ω ( ξ ) = ord ( x , y ) ( f ) − ord ( x , y ) ( g ) , (27)
where ord ( x , y ) ( h ) \operatorname{ord}_{(x,y)}(h) ord ( x , y ) ( h ) is the least total degree of a monomial occurring in h h h . Only the finitely many rational primes occurring in the nonzero coefficients of f f f and g g g can be exceptional. In particular, ω ( a ) ≥ 0 \omega(a)\geq0 ω ( a ) ≥ 0 for every 0 ≠ a ∈ D A M 0\ne a\in D_{\mathrm{AM}} 0 = a ∈ D AM .
Let A , B , C A,B,C A , B , C be fractional t t t -ideals with ( A B ) t = ( A C ) t (AB)_t=(AC)_t ( A B ) t = ( A C ) t . After a common scaling, assume that they are integral. Choose 0 ≠ a 0 ∈ A 0\ne a_0\in A 0 = a 0 ∈ A such that ω ( a 0 ) \omega(a_0) ω ( a 0 ) is minimal among the eventual values of the nonzero elements of A A A .
Fix 0 ≠ b ∈ B 0\ne b\in B 0 = b ∈ B . Since a 0 b ∈ ( A C ) t a_0b\in(AC)_t a 0 b ∈ ( A C ) t , there are finitely generated ideals E ⊆ A E\subseteq A E ⊆ A and F ⊆ C F\subseteq C F ⊆ C such that a 0 b ∈ ( E F ) v a_0b\in(EF)_v a 0 b ∈ ( E F ) v . For every i i i , extension to the valuation domain V i V_i V i gives
ν i ( a 0 ) + ν i ( b ) ≥ ν i ( E ) + ν i ( F ) , (28) \nu_i(a_0)+\nu_i(b)\geq\nu_i(E)+\nu_i(F), \tag*{(28)} ν i ( a 0 ) + ν i ( b ) ≥ ν i ( E ) + ν i ( F ) , (28)
where the value of a finitely generated ideal is the minimum of the values of a finite generating set. For all sufficiently large i i i , every term in (28) has stabilized. Since E ⊆ A E\subseteq A E ⊆ A , minimality of ω ( a 0 ) \omega(a_0) ω ( a 0 ) gives ω ( E ) ≥ ω ( a 0 ) \omega(E)\geq\omega(a_0) ω ( E ) ≥ ω ( a 0 ) , and therefore ω ( b ) ≥ ω ( F ) \omega(b)\geq\omega(F) ω ( b ) ≥ ω ( F ) . Hence
b ∈ F V i for all sufficiently large i . (29) b\in FV_i\quad\text{for all sufficiently large }i. \tag*{(29)} b ∈ F V i for all sufficiently large i . (29)
For each exceptional index, Proposition 4.2(ii) and t ≤ κ t\leq\kappa t ≤ κ give L t V i = L V i L_tV_i=LV_i L t V i = L V i for every fractional ideal L L L . Localizing ( A B ) t = ( A C ) t (AB)_t=(AC)_t ( A B ) t = ( A C ) t at P i P_i P i and cancelling in the DVR V i V_i V i therefore gives B V i = C V i BV_i=CV_i B V i = C V i . We may enlarge F F F by finitely many elements of C C C so that b ∈ F V i b\in FV_i b ∈ F V i for every i i i .
It remains to patch the height-one primes contracted from T T T . Since T = ( Z ∖ { 0 } ) − 1 D A M T=(\mathbb{Z}\setminus\{0\})^{-1}D_{\mathrm{AM}} T = ( Z ∖ { 0 } ) − 1 D AM , Lemma 2.5 gives
( ( L t ) T ) ν T = ( L T ) ν T . (30) ((L_t)T)_{\nu_T}=(LT)_{\nu_T}. \tag*{(30)} (( L t ) T ) ν T = ( L T ) ν T . (30)
for every fractional ideal L L L ; here t T = v T t_T = v_T t T = v T because T T T is a UFD. Thus ( A B ) t = ( A C ) t (AB)_t = (AC)_t ( A B ) t = ( A C ) t implies ( A T B T ) v T = ( A T C T ) v T (A_TB_T)_{v_T} = (A_TC_T)_{v_T} ( A T B T ) v T = ( A T C T ) v T . Cancellation in the divisorial ideal group of the Krull domain T T T gives ( B T ) v T = ( C T ) v T (B_T)_{v_T} = (C_T)_{v_T} ( B T ) v T = ( C T ) v T . Since T T T is Noetherian, choose a finitely generated subideal F 0 ⊆ C F_0 \subseteq C F 0 ⊆ C with C T = F 0 T CT = F_0T C T = F 0 T . Then b ∈ ( F 0 T ) v T b \in(F_0T)_{v_T} b ∈ ( F 0 T ) v T .
Replace F F F by G = F + F 0 G = F + F_0 G = F + F 0 . Then G ⊆ C G \subseteq C G ⊆ C is finitely generated, b ∈ G V i b \in GV_i b ∈ G V i for every i i i , and b ∈ ( G T ) v T b \in(GT)_{v_T} b ∈ ( GT ) v T . If z ∈ G − 1 z \in G^{-1} z ∈ G − 1 , then z G ⊆ D A M zG \subseteq D_{\mathrm{AM}} z G ⊆ D AM . The first family of containments gives b z ∈ V i bz \in V_i b z ∈ V i for every i i i , while z ∈ ( T : G T ) z \in(T : GT) z ∈ ( T : GT ) and b ∈ ( G T ) v T b \in(GT)_{v_T} b ∈ ( GT ) v T give b z ∈ T bz \in T b z ∈ T . By (25) , b z ∈ D A M bz \in D_{\mathrm{AM}} b z ∈ D AM . Hence b G − 1 ⊆ D A M bG^{-1} \subseteq D_{\mathrm{AM}} b G − 1 ⊆ D AM , so b ∈ G v = G t ⊆ C t = C b \in G_v = G_t \subseteq C_t = C b ∈ G v = G t ⊆ C t = C .
We have proved B ⊆ C B \subseteq C B ⊆ C . Interchanging B B B and C C C gives equality, so D A M D_{\mathrm{AM}} D AM satisfies TCL. Full t t t -linkage follows from Theorem 2.3 . □
The example also contains a concrete higher-height locus at which t t t and κ \kappa κ differ. Thus the possible support singled out by well-behaved t t t -primes can be present even when cancellation descends.
Proposition 4.7. Let m = ( x , y ) T ∩ D A M \mathfrak{m} = (x,y)T \cap D_{\mathrm{AM}} m = ( x , y ) T ∩ D AM . Then:
(i) m \mathfrak{m} m is a height-two prime and ( D A M ) m = T ( x , y ) (D_{\mathrm{AM}})_{\mathfrak{m}} = T_{(x,y)} ( D AM ) m = T ( x , y ) ;
(ii) m t = m \mathfrak{m}_t = \mathfrak{m} m t = m , and m \mathfrak{m} m is a maximal t t t -ideal;
(iii) m \mathfrak{m} m is not well-behaved;
(iv) m κ = m v = D A M \mathfrak{m}_{\kappa} = \mathfrak{m}_v = D_{\mathrm{AM}} m κ = m v = D AM .
Proof. Evaluation at x = y = 0 x = y = 0 x = y = 0 gives D A M / m ≅ Z D_{\mathrm{AM}}/\mathfrak{m} \cong\mathbb{Z} D AM / m ≅ Z , so m \mathfrak{m} m is prime. Since every nonzero integer lies outside m \mathfrak{m} m , localization gives ( D A M ) m = T ( x , y ) (D_{\mathrm{AM}})_{\mathfrak{m}} = T_{(x,y)} ( D AM ) m = T ( x , y ) ; hence ht m = 2 \operatorname{ht}\mathfrak{m} = 2 ht m = 2 .
Let F ⊆ m F \subseteq\mathfrak{m} F ⊆ m be finitely generated. Every nonzero generator of F F F has positive ω \omega ω -value. If u ∈ F v u \in F_v u ∈ F v , then u ∈ F V i u \in FV_i u ∈ F V i for every i i i , and so ω ( u ) ≥ ω ( F ) ≥ 1 \omega(u) \geq\omega(F) \geq1 ω ( u ) ≥ ω ( F ) ≥ 1 . Since F v ⊆ D A M F_v \subseteq D_{\mathrm{AM}} F v ⊆ D AM , this means that u u u has zero constant term and belongs to m \mathfrak{m} m . Thus F v ⊆ m F_v \subseteq\mathfrak{m} F v ⊆ m for every such F F F , and m t = m \mathfrak{m}_t = \mathfrak{m} m t = m .
To see maximality among proper t t t -ideals, let J = J t J = J_t J = J t properly contain m \mathfrak{m} m . Modulo m \mathfrak{m} m , the ideal J J J contains a nonzero integer n n n . If n n n is a unit, then J = D A M J = D_{\mathrm{AM}} J = D AM . Otherwise, let p i 1 , … , p i s p_{i_1}, \ldots, p_{i_s} p i 1 , … , p i s be its distinct prime divisors and set
H = ( n , x p i 1 , … , x p i s ) ⊆ J . (31) H = \left(n,\frac{x}{p_{i_1}},\ldots,\frac{x}{p_{i_s}}\right) \subseteq J. \tag*{(31)} H = ( n , p i 1 x , … , p i s x ) ⊆ J . (31)
At P i j P_{i_j} P i j , the element x / p i j x/p_{i_j} x / p i j is a unit; at every other P i P_i P i , the integer n n n is a unit; and at every Q f Q_f Q f , the nonzero constant n n n is a unit. Hence H D P = D P HD_P = D_P H D P = D P for every height-one prime P P P . By Proposition 4.5 (iv) and (18) , H v = H κ = D A M H_v = H_{\kappa} = D_{\mathrm{AM}} H v = H κ = D AM . Therefore J = J t = D A M J = J_t = D_{\mathrm{AM}} J = J t = D AM , and m \mathfrak{m} m is maximal among proper t t t -ideals.
Since m \mathfrak{m} m is a height-two prime t t t -ideal, Proposition 4.2(i) shows that it is not well-behaved. Finally, m D P i = D P i \mathfrak{m}D_{P_i} = D_{P_i} m D P i = D P i because x / p i ∈ m x/p_i \in\mathfrak{m} x / p i ∈ m is a unit in D P i D_{P_i} D P i . If P = Q f P = Q_f P = Q f , then the irreducible polynomial f f f cannot divide both x x x and y y y in T T T , so one of x , y x,y x , y belongs to m ∖ Q f \mathfrak{m} \setminus Q_f m ∖ Q f and is a unit in D P D_P D P . Thus m D P = D P \mathfrak{m}D_P = D_P m D P = D P at every height-one prime. Equation (18) gives m κ = D A M \mathfrak{m}_{\kappa} = D_{\mathrm{AM}} m κ = D AM , and Proposition 4.5 (iv) gives m v = m κ \mathfrak{m}_v = \mathfrak{m}_{\kappa} m v = m κ . □
Corollary 4.8. For the Arnold–Matsuda domain,
D A M ∈ A K ∩ T C L ∖ P v M D , t = κ f < κ = v , (32) D_{\mathrm{AM}} \in\mathrm{AK} \cap\mathrm{TCL} \setminus\mathrm{PvMD}, \qquad t = \kappa_f < \kappa= v, \tag*{(32)} D AM ∈ AK ∩ TCL ∖ PvMD , t = κ f < κ = v , (32) and both t t t and κ \kappa κ are cancellative. This remains true despite the maximal t t t -ideal in Proposition 4.7 , which is non-well-behaved and is closed for t t t but not for κ \kappa κ .
Proof. Combine Proposition 4.5 , Theorem 4.6 , and Proposition 4.7 with Proposition 4.1(ii)–Proposition 4.1(iii). □
An almost Krull domain without full cancellation
We now give the opposite outcome for finite-type descent. The construction has only factorial maximal localizations, but its global t t t -ideal monoid contains a nonidentity idempotent. The proof verifies the maximal spectrum and the t t t -idempotence independently.
The weighted construction
The intersection below is a weighted, infinite-variable modification of the construction of Heinzer and Ohm [15 ] . The finite-patching behavior of their original example was analyzed by Houston, Kabbaj, and Mimouni [16 Example 3.3] . All properties of the modified construction needed here are proved directly; in particular, its maximal spectrum is not assumed to be inherited from the original example.
Let k k k be any field, and let x 1 , x 2 , … , y 1 , y 2 , … x_1,x_2,\ldots,y_1,y_2,\ldots x 1 , x 2 , … , y 1 , y 2 , … be algebraically independent over k k k . Set
L = k ( x 1 , x 2 , … ) , K = L ( y 1 , y 2 , … ) , L = k(x_1,x_2,\ldots), \qquad K = L(y_1,y_2,\ldots), L = k ( x 1 , x 2 , … ) , K = L ( y 1 , y 2 , … ) ,
S = L [ y 1 , y 2 , … ] , m = ( y 1 , y 2 , … ) S , R = S m . S = L[y_1,y_2,\ldots], \qquad\mathfrak{m} = (y_1,y_2,\ldots)S, \qquad R = S_{\mathfrak{m}}. S = L [ y 1 , y 2 , … ] , m = ( y 1 , y 2 , … ) S , R = S m .
The ring S S S is a UFD: factorization of any polynomial takes place in a finite-variable polynomial ring, and its prime factors remain prime after adjoining further indeterminates. Hence R R R is a local UFD, and therefore a Krull domain. No Noetherian hypothesis is imposed on R R R .
For i , j ≥ 1 i,j \ge1 i , j ≥ 1 , define e i j = 2 max { i − j , 0 } e_{ij} = 2^{\max\{i-j,0\}} e ij = 2 m a x { i − j , 0 } . Let v i v_i v i be the weighted Gauss valuation of K K K determined by
v i ( x i ) = 1 , v i ( x h ) = 0 ( h ≠ i ) , v i ( y j ) = e i j . (33) v_i(x_i) = 1, \qquad v_i(x_h) = 0 \ (h \ne i), \qquad v_i(y_j) = e_{ij}. \tag*{(33)} v i ( x i ) = 1 , v i ( x h ) = 0 ( h = i ) , v i ( y j ) = e ij . (33)
Here weighted Gauss means that the value of a nonzero polynomial is the minimum of the weights of its nonzero monomials, and the value of a quotient is the difference of the values of numerator and denominator. Let V i = { 0 } ∪ { a ∈ K × : v i ( a ) ≥ 0 } V_i = \{0\} \cup\{a \in K^\times: v_i(a) \ge0\} V i = { 0 } ∪ { a ∈ K × : v i ( a ) ≥ 0 } . Each V i V_i V i is a DVR, with uniformizer x i x_i x i . More explicitly, put
z i j = y j / x i e i j , E i = k ( x h ( h ≠ i ) , z i 1 , z i 2 , … ) . z_{ij} = y_j/x_i^{e_{ij}}, \qquad E_i = k(x_h\ (h \ne i),z_{i1},z_{i2},\ldots). z ij = y j / x i e ij , E i = k ( x h ( h = i ) , z i 1 , z i 2 , … ) .
Then K = E i ( x i ) K = E_i(x_i) K = E i ( x i ) , V i = E i [ x i ] ( x i ) V_i = E_i[x_i]_{(x_i)} V i = E i [ x i ] ( x i ) , and the residue field of V i V_i V i is E i E_i E i .
Define
D = R ∩ ⋂ i ≥ 1 V i , M = m R ∩ D , P i = x i V i ∩ D . (34) D = R \cap\bigcap_{i \ge1} V_i, \qquad M = \mathfrak{m}R \cap D, \qquad P_i = x_iV_i \cap D. \tag*{(34)} D = R ∩ i ≥ 1 ⋂ V i , M = m R ∩ D , P i = x i V i ∩ D . (34)
The polynomial ring k [ x 1 , x 2 , … , y 1 , y 2 , … ] k[x_1,x_2,\ldots,y_1,y_2,\ldots] k [ x 1 , x 2 , … , y 1 , y 2 , … ] is contained in D D D , so the quotient field of D D D is K K K .
An auxiliary valuation records the asymptotic behavior of the v i v_i v i . It is denoted by ω \omega ω to distinguish it from the usual w w w -operation.
Lemma 5.1. Let ω \omega ω be the weighted Gauss valuation of K K K that is trivial on L L L and satisfies ω ( y j ) = 2 − j \omega(y_j) = 2^{-j} ω ( y j ) = 2 − j . For every 0 ≠ a ∈ K 0 \ne a \in K 0 = a ∈ K ,
v i ( a ) = 2 i ω ( a ) for all sufficiently large i . (35) v_i(a) = 2^i\omega(a) \qquad\text{for all sufficiently large } i. \tag*{(35)} v i ( a ) = 2 i ω ( a ) for all sufficiently large i . (35) Moreover, ω ( a ) ≥ 0 \omega(a) \ge0 ω ( a ) ≥ 0 for every 0 ≠ a ∈ R 0 \ne a \in R 0 = a ∈ R , and ω ( a ) > 0 \omega(a) > 0 ω ( a ) > 0 holds precisely when a ∈ m R a \in\mathfrak{m}R a ∈ m R .
Proof. A rational function a a a involves only finitely many of the x h x_h x h and y j y_j y j . For i i i larger than all their indices, v i v_i v i is trivial on every coefficient occurring in a polynomial presentation of a a a , and v i ( y j ) = 2 i ω ( y j ) v_i(y_j) = 2^i\omega(y_j) v i ( y j ) = 2 i ω ( y j ) for every variable occurring there. Taking minimum monomial weights, and then differences for a quotient, proves (35) .
An element of R R R is a quotient f / g f/g f / g with f , g ∈ S f,g \in S f , g ∈ S and g ∉ m g \notin\mathfrak{m} g ∈ / m . The polynomial g g g has a nonzero constant term, so ω ( g ) = 0 \omega(g) = 0 ω ( g ) = 0 . All variable weights are positive, and each polynomial has only finitely many monomials. Thus ω ( f ) ≥ 0 \omega(f) \ge0 ω ( f ) ≥ 0 , with strict inequality exactly when f ∈ m f \in\mathfrak{m} f ∈ m . This proves the remaining assertions.
The following proposition verifies all maximal ideals and all of their localizations.
Proposition 5.2. For the domain in (34) , one has
Max ( D ) = { M } ∪ { P i : i ≥ 1 } , P i = x i D , D M = R , D P i = V i . (36) \operatorname{Max}(D)=\{M\}\cup\{P_i:i\geq1\},\qquad P_i=x_iD,\qquad D_M=R,\qquad D_{P_i}=V_i. \tag*{(36)} Max ( D ) = { M } ∪ { P i : i ≥ 1 } , P i = x i D , D M = R , D P i = V i . (36) In particular, D D D is locally factorial and almost Krull.
Proof. First we show that M M M is maximal. Let ρ : R → L \rho:R\to L ρ : R → L be the residue map obtained by setting all y j y_j y j equal to zero. Its restriction to D D D has kernel M M M , and its image contains k [ x 1 , x 2 , … ] k[x_1,x_2,\ldots] k [ x 1 , x 2 , … ] . Given 0 ≠ g ∈ k [ x 1 , x 2 , … ] 0\neq g\in k[x_1,x_2,\ldots] 0 = g ∈ k [ x 1 , x 2 , … ] , choose ℓ \ell ℓ larger than every index of an x h x_h x h occurring in g g g , and put h = g / ( g 2 + y ℓ ) h=g/(g^2+y_\ell) h = g / ( g 2 + y ℓ ) . The denominator is a unit of R R R , and ρ ( h ) = 1 / g \rho(h)=1/g ρ ( h ) = 1/ g . For each i i i occurring in g g g , one has v i ( y ℓ ) = 1 v_i(y_\ell)=1 v i ( y ℓ ) = 1 ; writing r = v i ( g ) ∈ N 0 r=v_i(g)\in\mathbb{N}_0 r = v i ( g ) ∈ N 0 , we obtain
v i ( h ) = r − min { 2 r , 1 } ≥ 0. v_i(h)=r-\min\{2r,1\}\geq0. v i ( h ) = r − min { 2 r , 1 } ≥ 0.
For all other i i i , v i ( g ) = 0 v_i(g)=0 v i ( g ) = 0 and hence v i ( h ) = 0 v_i(h)=0 v i ( h ) = 0 . Therefore h ∈ D h\in D h ∈ D . It follows that ρ ( D ) = L \rho(D)=L ρ ( D ) = L , so M M M is maximal.
Next, if a ∈ P i a\in P_i a ∈ P i , then a / x i ∈ R a/x_i\in R a / x i ∈ R , its v i v_i v i -value is nonnegative, and its other v j v_j v j -values are unchanged. Thus a / x i ∈ D a/x_i\in D a / x i ∈ D , which proves P i = x i D P_i=x_iD P i = x i D .
We verify that P i P_i P i is maximal without any finiteness assumption on R R R . In the rational function field L L L , the element − x i -x_i − x i is not a square. For j ≠ i j\neq i j = i , the same holds for the residue of − x i -x_i − x i in E j E_j E j : x i x_i x i is one of the independent variables of the purely transcendental field E j E_j E j . Indeed, its x i x_i x i -adic order is one, whereas a square has even order. This argument works in every characteristic.
Let a ∈ D ∖ P i a\in D\setminus P_i a ∈ D ∖ P i and set b = a 2 + x i b=a^2+x_i b = a 2 + x i . Reduction modulo m R \mathfrak{m}_R m R shows that b b b is a unit of R R R . Since v i ( a ) = 0 v_i(a)=0 v i ( a ) = 0 , b b b is also a unit of V i V_i V i . For j ≠ i j\neq i j = i , the residue of b b b in E j E_j E j is nonzero by the preceding nonsquare observation, so b b b is a unit of V j V_j V j . Thus b − 1 ∈ D b^{-1}\in D b − 1 ∈ D . The identity
a b + x i 1 b = 1 \frac{a}{b}+x_i\frac{1}{b}=1 b a + x i b 1 = 1
shows that the image of every a ∉ P i a\notin P_i a ∈ / P i is invertible modulo P i P_i P i . Consequently P i P_i P i is maximal.
Now let 0 ≠ a ∈ D ∖ M 0\neq a\in D\setminus M 0 = a ∈ D ∖ M . The element a a a is a unit of R R R , and Lemma 5.1 gives v i ( a ) = 0 v_i(a)=0 v i ( a ) = 0 for all but finitely many i i i . Set p = ∏ i x i v i ( a ) p=\prod_i x_i^{v_i(a)} p = ∏ i x i v i ( a ) , a finite product. Then a / p a/p a / p is a unit of R R R and has value zero in every V i V_i V i , so a / p ∈ D × a/p\in D^\times a / p ∈ D × . Thus every nonzero element of D ∖ M D\setminus M D ∖ M is a unit times a finite product of the x i x_i x i . A maximal ideal different from M M M contains an element of D ∖ M D\setminus M D ∖ M , hence contains some x i x_i x i , and therefore is P i P_i P i . This proves the assertion about the maximal spectrum.
To prove D M = R D_M=R D M = R , take 0 ≠ a ∈ R 0\neq a\in R 0 = a ∈ R . Only finitely many v i ( a ) v_i(a) v i ( a ) are negative, by Lemma 5.1. Multiplication by a suitable finite product s s s of the x i x_i x i makes all these values nonnegative. Then s a ∈ D sa\in D s a ∈ D and s ∈ D ∖ M s\in D\setminus M s ∈ D ∖ M , so a ∈ D M a\in D_M a ∈ D M . The reverse inclusion holds because every element of D ∖ M D\setminus M D ∖ M is a unit of R R R .
Finally, take 0 ≠ a ∈ V i 0\neq a\in V_i 0 = a ∈ V i and write a = f / g a=f/g a = f / g with nonzero f , g ∈ k [ x 1 , x 2 , … , y 1 , y 2 , … ] f,g\in k[x_1,x_2,\ldots,y_1,y_2,\ldots] f , g ∈ k [ x 1 , x 2 , … , y 1 , y 2 , … ] . Set r = v i ( g ) ≥ 0 r=v_i(g)\geq0 r = v i ( g ) ≥ 0 and s = g / x i r s=g/x_i^r s = g / x i r . Then s ∈ R s\in R s ∈ R , v i ( s ) = 0 v_i(s)=0 v i ( s ) = 0 , and v j ( s ) = v j ( g ) ≥ 0 v_j(s)=v_j(g)\geq0 v j ( s ) = v j ( g ) ≥ 0 for j ≠ i j\neq i j = i . Hence s ∈ D ∖ P i s\in D\setminus P_i s ∈ D ∖ P i . Similarly, s a = f / x i r ∈ R sa=f/x_i^r\in R s a = f / x i r ∈ R has nonnegative value in every V j V_j V j , using v i ( f ) ≥ v i ( g ) v_i(f)\geq v_i(g) v i ( f ) ≥ v i ( g ) at i i i . Thus s a ∈ D sa\in D s a ∈ D and a ∈ D P i a\in D_{P_i} a ∈ D P i . The reverse inclusion is immediate. All the localizations in (36) are UFDs, proving the last assertion. □ \square □
A proper t t t -idempotent ideal
Although all the maximal localizations are Krull, the global t t t -operation has a proper nonzero idempotent ideal. The proof uses the inverse-ideal definition of v v v throughout, so no unrestricted commutation of t t t or v v v with localization is required.
Theorem 5.3. The maximal ideal M M M of (34) satisfies
0 ≠ M = M t = ( M 2 ) t ⊊ D . (37) 0\neq M=M_t=(M^2)_t\subsetneq D. \tag*{(37)} 0 = M = M t = ( M 2 ) t ⊊ D . (37)
Consequently, not every almost Krull domain satisfies the full t t t -cancellation law.
Proof. We first prove M t = M M_t=M M t = M . Let F = ( f 1 , … , f r ) F=(f_1,\ldots,f_r) F = ( f 1 , … , f r ) be a nonzero finitely generated subideal of M M M , omitting zero generators. Each ω ( f h ) > 0 \omega(f_h)>0 ω ( f h ) > 0 , so there is an index i i i for which v i ( f h ) ≥ 1 v_i(f_h)\geq1 v i ( f h ) ≥ 1 for all h h h . Therefore F / x i ⊆ D F/x_i\subseteq D F / x i ⊆ D : division by x i x_i x i preserves membership in R R R and in every V j V_j V j with j ≠ i j\ne i j = i , and the chosen inequalities handle V i V_i V i . Hence x i − 1 ∈ F − 1 x_i^{-1}\in F^{-1} x i − 1 ∈ F − 1 , whereas x i − 1 ∉ D x_i^{-1}\notin D x i − 1 ∈ / D . It follows that 1 ∉ F v 1\notin F_v 1 ∈ / F v . Taking the union over all such F F F gives 1 ∉ M t 1\notin M_t 1 ∈ / M t . Since M M M is maximal and M ⊆ M t ⊆ D M\subseteq M_t\subseteq D M ⊆ M t ⊆ D , we conclude that M t = M M_t=M M t = M .
It remains to prove M ⊆ ( M 2 ) t M\subseteq(M^2)_t M ⊆ ( M 2 ) t ; the reverse inclusion follows from M t = M M_t=M M t = M . Fix 0 ≠ a ∈ M 0\ne a\in M 0 = a ∈ M . Choose j j j sufficiently large that 2 1 − j < ω ( a ) 2^{1-j}<\omega(a) 2 1 − j < ω ( a ) . By (35) ,
v i ( y j 2 ) = 2 i + 1 − j ≤ v i ( a ) for all sufficiently large i . v_i(y_j^2)=2^{i+1-j}\leq v_i(a)\qquad\text{for all sufficiently large }i. v i ( y j 2 ) = 2 i + 1 − j ≤ v i ( a ) for all sufficiently large i .
Let T T T be the finite set of indices for which v i ( y j 2 ) > v i ( a ) v_i(y_j^2)>v_i(a) v i ( y j 2 ) > v i ( a ) . For each i ∈ T i\in T i ∈ T , put c i = y j / x i e i j c_i=y_j/x_i^{e_{ij}} c i = y j / x i e ij . This element lies in m R \mathfrak mR m R , has v i ( c i ) = 0 v_i(c_i)=0 v i ( c i ) = 0 , and has positive value in every V h V_h V h with h ≠ i h\ne i h = i . Thus c i ∈ M c_i\in M c i ∈ M .
Consider the finitely generated ideal
G = ( y j 2 , y j + 1 2 , c i 2 ( i ∈ T ) ) D ⊆ M 2 . (38) G=(y_j^2,y_{j+1}^2,c_i^2\ (i\in T))D\subseteq M^2. \tag*{(38)} G = ( y j 2 , y j + 1 2 , c i 2 ( i ∈ T )) D ⊆ M 2 . (38)
For i ∉ T i\notin T i ∈ / T , the generator y j 2 y_j^2 y j 2 gives a ∈ G V i a\in GV_i a ∈ G V i . For i ∈ T i\in T i ∈ T , the generator c i 2 c_i^2 c i 2 is a unit of V i V_i V i , so again a ∈ G V i a\in GV_i a ∈ G V i . Therefore a ∈ G V i a\in GV_i a ∈ G V i for every i i i .
Take u ∈ G − 1 u\in G^{-1} u ∈ G − 1 . Since u y j 2 , u y j + 1 2 ∈ D ⊆ R uy_j^2,uy_{j+1}^2\in D\subseteq R u y j 2 , u y j + 1 2 ∈ D ⊆ R and y j , y j + 1 y_j,y_{j+1} y j , y j + 1 are relatively prime prime elements of the UFD R R R , one has u ∈ R u\in R u ∈ R . Hence a u ∈ R au\in R a u ∈ R . Also u G ⊆ D ⊆ V i uG\subseteq D\subseteq V_i u G ⊆ D ⊆ V i and a ∈ G V i a\in GV_i a ∈ G V i imply a u ∈ V i au\in V_i a u ∈ V i for every i i i . Intersecting yields a u ∈ D au\in D a u ∈ D . We have proved a G − 1 ⊆ D aG^{-1}\subseteq D a G − 1 ⊆ D , so a ∈ G v ⊆ ( M 2 ) t a\in G_v\subseteq(M^2)_t a ∈ G v ⊆ ( M 2 ) t . This proves (37) .
Finally, take A = M A=M A = M , B = M B=M B = M , and C = D C=D C = D . Then
( A B ) t = ( M 2 ) t = M = ( M D ) t = ( A C ) t , B t = M ≠ D = C t . (AB)_t=(M^2)_t=M=(MD)_t=(AC)_t,\qquad B_t=M\ne D=C_t. ( A B ) t = ( M 2 ) t = M = ( M D ) t = ( A C ) t , B t = M = D = C t .
These are nonzero integral ideals, so this is already a failure of the full law on integral ideals. □
The counterexample also separates local cancellation from global cancellation, even within the almost Krull class.
Corollary 5.4. For every maximal ideal N N N of D D D , the domain D N D_N D N satisfies its own full t D N t_{D_N} t D N -cancellation law, but D D D does not satisfy TCL. Moreover, D D D is not a PvMD.
Proof. Each D N D_N D N is Krull. On a Krull domain t = v t=v t = v , and the nonzero fractional divisorial ideals form a group; thus its full t t t -cancellation law holds (see, for example, [11 ] ). The global failure is Theorem 5.3 .
For a direct verification of the failure of the PvMD property, set F = ( y 1 , y 2 ) D F=(y_1,y_2)D F = ( y 1 , y 2 ) D . If u ∈ F − 1 u\in F^{-1} u ∈ F − 1 , then u y 1 , u y 2 ∈ R uy_1,uy_2\in R u y 1 , u y 2 ∈ R , so coprimality in the UFD R R R implies u ∈ R u\in R u ∈ R . It follows that F F − 1 ⊆ m R ∩ D = M FF^{-1}\subseteq\mathfrak mR\cap D=M F F − 1 ⊆ m R ∩ D = M , and hence ( F F − 1 ) t ⊆ M t = M ≠ D (FF^{-1})_t\subseteq M_t=M\ne D ( F F − 1 ) t ⊆ M t = M = D . Thus the finitely generated ideal F F F is not t t t -invertible. □
The same ideal makes the finite-type descent obstruction explicit.
Corollary 5.5. For the canonical locally divisorial star operation
I κ = ⋂ N ∈ Max ( D ) ( I D N ) v N , I_\kappa=\bigcap_{N\in\operatorname{Max}(D)}(ID_N)_{v_N}, I κ = N ∈ Max ( D ) ⋂ ( I D N ) v N , one has κ f = t \kappa_f=t κ f = t , while κ \kappa κ is cancellative and t t t is not. In fact, M κ = D M_\kappa=D M κ = D but M t = M ≠ D M_t=M\ne D M t = M = D .
Proof. By Proposition 5.2 , M D M = m R MD_M=mR M D M = m R , whose divisorial closure is R R R , since it contains the relatively prime elements y 1 y_1 y 1 and y 2 y_2 y 2 . For each i i i , the element y 1 / x 1 e i y_1/x_1^{e_i} y 1 / x 1 e i belongs to M M M and is a unit of V i V_i V i , so M V i = V i MV_i=V_i M V i = V i . Hence M κ = D M^\kappa=D M κ = D .
The assertions κ f = t \kappa_f=t κ f = t and cancellativity of κ \kappa κ follow from Proposition 4.1. The failure for t t t is Theorem 5.3 . □
Remark 5.6. The example does not satisfy the eventual-constancy hypothesis used in the finite-patching argument for the original Heinzer–Ohm domain. Already v i ( y 1 ) = 2 i − 1 v_i(y_1)=2^{i-1} v i ( y 1 ) = 2 i − 1 is not eventually constant. After normalization, the limiting valuation has ω ( y j ) = 2 − j \omega(y_j)=2^{-j} ω ( y j ) = 2 − j , so its positive values on M M M have no least element. Choosing generators with successively smaller positive weights is exactly what makes the witnesses in (38) possible. Thus the failure is not explained merely by non-PvMD behavior or by the strict inequality t < κ t<\kappa t < κ : it is witnessed by the explicit identity ( M 2 ) t = M ≠ D (M^2)_t=M\ne D ( M 2 ) t = M = D .
Descent criteria and the final placement
The positive and negative examples distinguish an operation gap from a cancellation defect. This section expresses that defect in the class monoid and in residual ideals, computes it in the weighted construction, and combines the examples into the final placement theorem.
Global and fiberwise descent criteria
The obstruction to descent can be expressed first in the monoid of all fractional t t t -ideals and then inside individual κ \kappa κ -fibers. Let I t ( D ) \mathcal{I}_t(D) I t ( D ) be the commutative monoid of fractional t t t -ideals under I ⊙ J = ( I J ) t I\mathbin{\odot}J=(IJ)_t I ⊙ J = ( I J ) t , and let P ( D ) \mathcal{P}(D) P ( D ) be its subgroup of principal fractional ideals. We call S t ( D ) = I t ( D ) / P ( D ) \mathcal{S}_t(D)=\mathcal{I}_t(D)/\mathcal{P}(D) S t ( D ) = I t ( D ) / P ( D ) the t t t -class monoid; thus I I I and J J J determine the same class exactly when I = x J I=xJ I = x J for some 0 ≠ x ∈ K 0\ne x\in K 0 = x ∈ K . Unlike the t t t -class group, this monoid contains all fractional t t t -ideals, not only the t t t -invertible ones. Its unit group is the t t t -class group, so it need not be reduced in the monoid-theoretic sense of having only the identity as a unit; compare [22 ] .
The following lemma permits cancellation to be checked after removing scalar multiples.
Lemma 6.1. If D D D is completely integrally closed, then I t ( D ) \mathcal{I}_t(D) I t ( D ) is cancellative if and only if S t ( D ) \mathcal{S}_t(D) S t ( D ) is cancellative.
Proof. Suppose first that I t ( D ) \mathcal{I}_t(D) I t ( D ) is cancellative and that [ A ] [ B ] = [ A ] [ C ] [A][B]=[A][C] [ A ] [ B ] = [ A ] [ C ] in S t ( D ) \mathcal{S}_t(D) S t ( D ) . Then ( A B ) t = x ( A C ) t = ( A ( x C ) ) t (AB)_t=x(AC)_t=(A(xC))_t ( A B ) t = x ( A C ) t = ( A ( x C ) ) t for some 0 ≠ x ∈ K 0\ne x\in K 0 = x ∈ K , so cancellation in I t ( D ) \mathcal{I}_t(D) I t ( D ) gives B = x C B=xC B = x C and hence [ B ] = [ C ] [B]=[C] [ B ] = [ C ] . Conversely, suppose S t ( D ) \mathcal{S}_t(D) S t ( D ) is cancellative and ( A B ) t = ( A C ) t (AB)_t=(AC)_t ( A B ) t = ( A C ) t for fractional t t t -ideals A A A , B B B , C C C . Cancellation in the quotient gives B = x C B=xC B = x C for some 0 ≠ x ∈ K 0\ne x\in K 0 = x ∈ K . Set J = ( A C ) t J=(AC)_t J = ( A C ) t . The product equality gives x J = J xJ=J x J = J . Choose 0 ≠ j ∈ J 0\ne j\in J 0 = j ∈ J and 0 ≠ d ∈ D 0\ne d\in D 0 = d ∈ D with d J ⊆ D dJ\subseteq D dJ ⊆ D . Since x n J = J = x − n J x^nJ=J=x^{-n}J x n J = J = x − n J for every n ≥ 0 n\ge0 n ≥ 0 , both d j x n d jx^n d j x n and d j x − n djx^{-n} d j x − n belong to D D D . Thus x x x and x − 1 x^{-1} x − 1 are almost integral over D D D . Complete integral closure gives x , x − 1 ∈ D × x,x^{-1}\in D^\times x , x − 1 ∈ D × and B = C B=C B = C . □
The class monoid gives a formal reformulation of the descent question; this is a cancellation criterion, not an intrinsic classification of the underlying domains.
Theorem 6.2. Let D D D be almost Krull, with canonical operation κ \kappa κ . The following conditions are equivalent.
(i) Cancellation for κ \kappa κ descends to κ f = t \kappa_f=t κ f = t .
(ii) D D D satisfies TCL.
(iii) The t t t -class monoid S t ( D ) \mathcal{S}_t(D) S t ( D ) is cancellative.
(iv) D D D has full t t t -linkage.
These conditions do not imply κ = t \kappa=t κ = t , even when κ = v \kappa=v κ = v and every height-one prime is divisorial.
Proof. Since κ f = t \kappa_f=t κ f = t and κ \kappa κ is always cancellative, (i) and (ii) are the same assertion. By Proposition 4.1(i) and Lemma 6.1 , (ii) and (iii) are equivalent. The equivalence of (ii) and (iv) is Theorem 2.3 . Finally, Corollary 4.4 shows that the conditions may hold with t < κ t<\kappa t < κ , and Corollary 4.8 shows that this can occur even with κ = v \kappa=v κ = v .
For fractional t t t -ideals A A A and B B B , define the residual enlargement
ρ A ( B ) = ( ( ( A B ) t : A ) ) t . (39) \rho_A(B)=\left(((AB)_t:A)\right)_t. \tag*{(39)} ρ A ( B ) = ( (( A B ) t : A ) ) t . (39)
The next theorem locates failure of cancellation within a single κ \kappa κ -fiber.
Theorem 6.3. Let κ \kappa κ be a cancellative star operation on D D D with κ f = t \kappa_f=t κ f = t . For fractional t t t -ideals A A A and B B B , the following statements hold.
(i) B ⊆ ρ A ( B ) B\subseteq\rho_A(B) B ⊆ ρ A ( B ) and ( A ρ A ( B ) ) t = ( A B ) t (A\rho_A(B))_t=(AB)_t ( A ρ A ( B ) ) t = ( A B ) t .
(ii) The map ρ A \rho_A ρ A is idempotent.
(iii) ( ρ A ( B ) ) κ = B κ (\rho_A(B))_\kappa=B_\kappa ( ρ A ( B ) ) κ = B κ .
(iv) The ideal A A A is t t t -cancellative if and only if ρ A ( B ) = B \rho_A(B)=B ρ A ( B ) = B for every fractional t t t -ideal B B B .
Consequently, cancellation descends from κ \kappa κ to t t t precisely when all residual enlargements inside common κ \kappa κ -fibers are trivial.
Proof. Put E = ( A B ) t E=(AB)_t E = ( A B ) t and H = ( E : A ) H=(E:A) H = ( E : A ) . Since A B ⊆ E AB\subseteq E A B ⊆ E , one has B ⊆ H ⊆ H t = ρ A ( B ) B\subseteq H\subseteq H_t=\rho_A(B) B ⊆ H ⊆ H t = ρ A ( B ) . Also A H ⊆ E AH\subseteq E A H ⊆ E , while B ⊆ H B\subseteq H B ⊆ H gives E ⊆ ( A H ) t E\subseteq(AH)_t E ⊆ ( A H ) t . Hence ( A ρ A ( B ) ) t = ( A H ) t = E (A\rho_A(B))_t=(AH)_t=E ( A ρ A ( B ) ) t = ( A H ) t = E , proving (i).
Applying the same formula again gives ρ A ( ρ A ( B ) ) = ( ( ( A ρ A ( B ) ) t : A ) ) t = ( ( E : A ) ) t = ρ A ( B ) \rho_A(\rho_A(B))=(((A\rho_A(B))_t:A))_t=((E:A))_t=\rho_A(B) ρ A ( ρ A ( B )) = ((( A ρ A ( B ) ) t : A ) ) t = (( E : A ) ) t = ρ A ( B ) .
Since t = κ f ≤ κ t=\kappa_f\leq\kappa t = κ f ≤ κ , the inclusions A B ⊆ A H ⊆ E ⊆ ( A B ) κ AB\subseteq AH\subseteq E\subseteq(AB)_\kappa A B ⊆ A H ⊆ E ⊆ ( A B ) κ imply ( A H ) κ = ( A B ) κ (AH)_\kappa=(AB)_\kappa ( A H ) κ = ( A B ) κ . Cancelling A A A for κ \kappa κ yields H κ = B κ H_\kappa=B_\kappa H κ = B κ . From H ⊆ H t ⊆ H κ H\subseteq H_t\subseteq H_\kappa H ⊆ H t ⊆ H κ , one obtains ( ρ A ( B ) ) κ = B κ (\rho_A(B))_\kappa=B_\kappa ( ρ A ( B ) ) κ = B κ , proving (iii).
If A A A is t t t -cancellative, (i) gives ρ A ( B ) = B \rho_A(B)=B ρ A ( B ) = B . Conversely, assume this equality for every B B B and let ( A B ) t = ( A C ) t (AB)_t=(AC)_t ( A B ) t = ( A C ) t . Then C ⊆ ( ( A B ) t : A ) ⊆ ρ A ( B ) = B C\subseteq((AB)_t:A)\subseteq\rho_A(B)=B C ⊆ (( A B ) t : A ) ⊆ ρ A ( B ) = B . Interchanging B B B and C C C gives the reverse inclusion, proving (iv).
The preceding fiber criterion and the well-behaved-prime description locate any possible failure away from height one.
Corollary 6.4. Let D D D be almost Krull. For fractional t t t -ideals A A A , B B B and every height-one prime P P P ,
ρ A ( B ) D P = B D P . (40) \rho_A(B)D_P=BD_P. \tag*{(40)} ρ A ( B ) D P = B D P . (40) If D D D does not satisfy TCL, then there exist fractional t t t -ideals A A A , B B B and a maximal t t t -ideal M M M of height greater than one such that B D M ⊊ ρ A ( B ) D M BD_M\subsetneq\rho_A(B)D_M B D M ⊊ ρ A ( B ) D M . Such an M M M is necessarily not well-behaved.
Proof. By Theorem 6.3 (iii), ( ρ A ( B ) ) κ = B κ (\rho_A(B))_\kappa=B_\kappa ( ρ A ( B ) ) κ = B κ . Localizing this equality at a height-one prime and applying Proposition 4.2(ii) gives (40).
If D D D is not in TCL, Theorem 6.3 (iv) gives B ⊊ ρ A ( B ) B\subsetneq\rho_A(B) B ⊊ ρ A ( B ) for some A A A , B B B . Both ideals are t t t -ideals and hence w w w -ideals, so each is the intersection of its localizations at the maximal t t t -ideals. Their localizations must therefore differ at some M ∈ t -Max ( D ) M\in t\text{-Max}(D) M ∈ t -Max ( D ) . Equation (40) excludes height one, and Proposition 4.2(i) shows that M M M is not well-behaved.
The Arnold–Matsuda ideal in Proposition 4.7 shows that a higher-height non-well-behaved maximal t t t -ideal with M t ⊆ M κ M_t\subseteq M_\kappa M t ⊆ M κ need not cause a residual defect. The weighted construction realizes such a defect explicitly.
Proposition 6.5. For the domain and maximal ideal M M M of Theorem 5.3 , one has
ρ M ( M ) = D ⊋ M , M D P = D P for every height-one prime P . (41) \rho_M(M)=D\supsetneq M,\qquad MD_P=D_P\quad\text{for every height-one prime }P. \tag*{(41)} ρ M ( M ) = D ⊋ M , M D P = D P for every height-one prime P . (41) Moreover, M M M has infinite height and is not well-behaved, whereas M D P i = D P i MD_{P_i}=D_{P_i} M D P i = D P i for all i i i . Thus the residual enlargement is visible at the maximal localization D M D_M D M and at no other maximal localization.
Proof. Since D D D is almost Krull, it is completely integrally closed. This gives ( M : M ) = D (M:M)=D ( M : M ) = D : if u M ⊆ M uM\subseteq M u M ⊆ M , choose 0 ≠ m ∈ M 0\ne m\in M 0 = m ∈ M and obtain m u n ∈ D mu^n\in D m u n ∈ D for all n ≥ 0 n\ge0 n ≥ 0 . Now Theorem 5.3 gives ρ M ( M ) = ( ( M : M ) ) t = D \rho_M(M)=((M:M))_t=D ρ M ( M ) = (( M : M ) ) t = D . By Corollary 5.5, M κ = D M_\kappa=D M κ = D ; Proposition 4.2(ii) then implies M D P = D P MD_P=D_P M D P = D P at every height-one prime. In particular, 1 1 1 belongs to all these localizations but does not belong to F v F_v F v for any finitely generated F ⊆ M F\subseteq M F ⊆ M , since M t = M M_t=M M t = M .
The localization D M = R D_M=R D M = R contains strictly increasing chains of prime ideals of arbitrary finite length, ( 0 ) ⊊ ( y 1 ) R ⊊ ( y 1 , y 2 ) R ⊊ ⋯ ⊊ m R (0)\subsetneq(y_1)R\subsetneq(y_1,y_2)R\subsetneq\cdots\subsetneq\mathfrak mR ( 0 ) ⊊ ( y 1 ) R ⊊ ( y 1 , y 2 ) R ⊊ ⋯ ⊊ m R . Hence ht M = ∞ \operatorname{ht}M=\infty ht M = ∞ . Its extension M D M = m R MD_M=\mathfrak mR M D M = m R has t R t_R t R -closure R R R , because R R R is a UFD and m R \mathfrak mR m R contains the relatively prime elements y 1 , y 2 y_1,y_2 y 1 , y 2 . Thus M M M is not well-behaved. The description of the maximal spectrum in Proposition 5.2 , together with M V i = V i MV_i=V_i M V i = V i , proves the last assertion. □ \square □
A known prime-power criterion further describes the difference between strict descent and equality of operations. We say that D D D satisfies the strong t t t -prime-power condition if every proper nonzero t t t -ideal I I I with prime radical P P P has the form I = ( P n ) t I=(P^n)_t I = ( P n ) t for some n ≥ 1 n\ge1 n ≥ 1 . Olberding and Reinhart prove for a finitary ideal system that the corresponding almost Dedekind property is equivalent to strong prime-power together with cancellation of every nontrivial ideal [27 Proposition 3.6] . Specializing their theorem to the t t t -system gives the following statement.
Proposition 6.6. Let D D D be a nonfield domain. Then D D D is t t t -almost Dedekind if and only if D D D satisfies TCL and the strong t t t -prime-power condition. Consequently, for a nonfield almost Krull domain,
κ = t ⟺ D is a PvMD ⟺ { D satisfies TCL , D satisfies strong t -prime-power . (42) \kappa=t\Longleftrightarrow D\text{ is a PvMD}\Longleftrightarrow
\begin{cases}
D\text{ satisfies TCL},\\
D\text{ satisfies strong }t\text{-prime-power}.
\end{cases} \tag*{(42)} κ = t ⟺ D is a PvMD ⟺ { D satisfies TCL , D satisfies strong t -prime-power . (42)
Proof. Apply [27 Proposition 3.6] to the multiplicative monoid D ∖ { 0 } D\setminus\{0\} D ∖ { 0 } with its finitary t t t -system. Its t t t -ideals are the nonzero parts of the ring-theoretic t t t -ideals, its prime t t t -ideals are the nonzero parts of prime t t t -ideals of D D D , and its local discrete valuation monoids correspond to the DVRs D P D_P D P . Integral and fractional t t t -cancellation are equivalent by a common scaling. The cited proposition therefore gives the first assertion, and the second follows from Proposition 4.1(iv). □ \square □
In particular, both the Heinzer–Ohm and Arnold–Matsuda domains must fail the strong t t t -prime-power condition. This additional condition distinguishes equality t = κ t=\kappa t = κ from mere preservation of cancellation; it should not be confused with the failure of cancellation exhibited in Theorem 5.3 .
Answers to the Kabbaj–Mimouni questions
The examples now settle both directions of comparison between almost Krullness and full t t t -linkage. The following theorem collects the answers to Question 1.1 and determines the missing t t t -linkage node in [20 Figure 1] .
Theorem 6.7. The following assertions hold.
(i) Full t t t -linkage is exactly TCL, and t t t -AD ⊆ \subseteq ⊆ TCL ⊆ \subseteq ⊆ CIC ⊆ \subseteq ⊆ v v v -domain.
(ii) The classes TCL and PvMD are incomparable, and TCL ∩ \cap ∩ PvMD = t =t = t -AD.
(iii) Full t t t -cancellation is strictly stronger than the a.b. property of t t t .
(iv) The classes TCL and AK are incomparable.
(v) Within the almost Krull class, AK ∩ \cap ∩ PvMD ⊊ \subsetneq ⊊ AK ∩ \cap ∩ TCL ⊊ \subsetneq ⊊ AK. The middle class contains a non-PvMD with κ = v \kappa= v κ = v .
(vi) For every almost Krull domain, κ \kappa κ is cancellative and κ f = t \kappa_f = t κ f = t . When t < κ t < \kappa t < κ , cancellation may either descend to t t t or fail to descend. Full cancellation for the local t D N tD_N t D N -operations at all maximal ideals does not imply TCL.
Proof. The equivalence in (i) is Theorem 2.3 , and the containments follow from Proposition 2.6 . The first containment is strict by Theorem 3.6 ; the second is strict by Theorem 5.3 , since an almost Krull domain is completely integrally closed.
For (ii), the Heinzer–Ohm domain belongs to TCL ∖ \setminus ∖ PvMD. A nondiscrete rank-one valuation domain V V V gives the reverse separation. Indeed, V V V is a PvMD, t = d t = d t = d , and its nonzero maximal ideal N N N satisfies N 2 = N ≠ V N^2 = N \ne V N 2 = N = V . To see idempotence, for a positive value γ \gamma γ choose 0 < δ < γ 0 < \delta< \gamma 0 < δ < γ in the nondiscrete value group and factor an element of value γ \gamma γ into elements of values δ \delta δ and γ − δ \gamma- \delta γ − δ . Thus V V V fails full cancellation. The intersection assertion is Proposition 2.6 (iv).
By Proposition 2.6 (i), TCL implies that t t t is a.b. Conversely, the valuation domain just used has an a.b. t = d t = d t = d operation, because every finitely generated fractional ideal is principal, but it does not satisfy TCL. This proves (iii).
For (iv), Theorem 3.3 gives TCL ∖ \setminus ∖ AK ≠ ∅ \ne\varnothing = ∅ , whereas Theorem 5.3 gives AK ∖ \setminus ∖ TCL ≠ ∅ \ne\varnothing = ∅ . The first strict inclusion in (v) is Corollary 4.4 ; the second is Theorem 5.3 . The additional assertion is Corollary 4.8. Finally, Proposition 4.1, together with Corollary 4.4 and Corollaries 5.4 and 5.5, proves (vi). □
Returning to Question 1.1 , full t t t -linkage lies strictly between t t t -almost Dedekindness and complete integral closure. It is incomparable with almost Krullness and with the PvMD property, while its intersection with the latter is the t t t -almost Dedekind class already identified in [20 Corollary 3.4] . Thus the missing node is not another name for one of those neighboring domain classes: its natural interpretation is cancellativity of the entire fractional t t t -ideal monoid.
The distinction can be read directly from the residual ideal. In the Heinzer–Ohm and Arnold–Matsuda examples, finite valuation patching forces ρ A ( B ) = B \rho_A(B)=B ρ A ( B ) = B for every pair of fractional t t t -ideals. In the weighted construction, ρ M ( M ) = D ⊋ M \rho_M(M)=D\supsetneq M ρ M ( M ) = D ⊋ M , although the enlargement is invisible at every height-one prime. Thus the almost Krull counterexample does not contradict the positive patching theorems: its limiting positive weights have no minimum, and their eventual-constancy hypothesis fails. The positive and negative examples therefore explain why local Krullness alone cannot determine the missing node.
Competing interests: The author declares that there are no competing interests.
Availability of data and material: No datasets were generated or analysed during the current study.
Declaration of generative AI and AI-assisted technologies in the manuscript preparation process. During the preparation of this work the author used ChatGPT, an AI assistant developed by OpenAI, in order to explore different mathematical contents about the topics of this paper. After using this tool/service, the author reviewed and edited the content as needed and take full responsibility for the content of the published article.
Funding. This research was supported by the Basic Science Research Program through the National Research Foundation of Korea (NRF), funded by the Ministry of Education (grant no. RS-2026-25589739).
Department of Computer Engineering, Hoseo University, Asan 31499, Republic of Korea
Email address: hkkim@hoseo.edu