Introduction
Free products, also called coproducts, describe how to combine a family of objects in a way that allows morphisms defined on each of the individual factors to determine uniquely a morphism on the resulting object. In many familiar categories, these constructions include disjoint unions of sets and topological spaces, direct sums of vector spaces and modules, or free products of groups. In the category of Banach spaces and contractive linear operators, the corresponding construction coincides with the ℓ 1 \ell_1 ℓ 1 -sum. Although these examples share the same universal property, the structure of a free product depends fundamentally on the class of morphisms under consideration. For vector lattices, when morphisms are the maps preserving the linear and lattice operations, the direct sum with coordinatewise order does not have the required property: its distinct summands are disjoint, whereas the images of independently chosen lattice homomorphisms need not be. A free product must therefore allow lattice operations between elements from different factors without imposing disjointness or any other prescribed relations among them.
The study of these constructions has a substantial history in ordered algebra and functional analysis. The following developments provide background for the questions considered here: In vector lattice theory, Topping [27 ] studied projective and injective objects and proved that there are no nonzero injective objects in the category of vector lattices and linear lattice homomorphisms. This obstruction motivates extension problems relative to restricted classes of embeddings (see § 6 ). Free products of lattice-ordered groups were studied by Holland and Scrimger [10 ] and, in varieties of lattice-ordered groups and in the abelian case, by Martinez [16 , 17 ] . Powell and Tsinakis [24 ] subsequently gave concrete constructions in the abelian case and proved that embeddings of the factors induce embeddings of their free products. This can be seen as the group-theoretic counterpart of Corollary 2.11. The amalgamation results of Pierce [22 , 23 ] and the treatment of push-outs by Hager and Madden [9 ] connect these constructions with extension properties of lattice homomorphisms. Broader accounts of free products and amalgamation appear in the chapters by Powell and Tsinakis [25 ] and by Tsinakis and Powell [28 ] . See also [20 , 26 , 8 ] for further developments.
In the Banach lattice setting, isometric push-outs were constructed in [2 ] and used to obtain amalgamation and relative injectivity results. Tursi [29 ] connected amalgamation with metric Fraïssé theory to construct a separable approximately ultrahomogeneous Banach lattice that is isometrically universal for separable Banach lattices. Also, García-Sánchez, Leung, Taylor and the second author [7 ] adapted push-out constructions in classes of Banach lattices subject to convexity conditions, including p p p -convex lattices and lattices with upper p p p -estimates, together with quantitative results on preservation of embeddings. This connects the categorical constructions with the geometry of lattice norms.
In our previous paper [18 ] , we constructed arbitrary free products of Banach lattices where morphisms are lattice homomorphisms. These can be described as appropriate quotients of free Banach lattices and we could establish several basic results such as their compatibility with isometric embeddings, quotients and projectivity. For nonempty compact Hausdorff spaces K K K , L L L , we also obtained the representation
C ( K ) ∗ B L C ( L ) ≃ C ( K ∗ L ) , C(K) \ast_{\mathcal{BL}} C(L) \simeq C(K \ast L), C ( K ) ∗ BL C ( L ) ≃ C ( K ∗ L ) ,
where K ∗ L K \ast L K ∗ L denotes the topological join, together with an explicit formula for the free-product norm. Through the double suspension theorem, from geometric topology, this representation yields surprising examples of free factors of FBL [ ℓ 1 ] \operatorname{FBL}[\ell_{1}] FBL [ ℓ 1 ] that are not themselves lattice isomorphic to free Banach lattices.
The purpose of the present paper is to understand the vector-lattice structure underlying these constructions and its relation to Banach lattice theory. Three categories enter naturally: vector lattices V L \mathcal{VL} V L , Archimedean vector lattices A V L \mathcal{AVL} A V L , both with lattice homomorphisms, and Banach lattices B L \mathcal{BL} BL with contractive lattice homomorphisms. The relations between these categories raise several questions. When is the ordinary vector-lattice free product of Archimedean factors still Archimedean? Which elements are identified on passing to the Banach lattice free product? Do injective homomorphisms remain injective after taking free products, even when their ranges are not closed? These questions concern both the algebraic relations imposed by the order and the additional structure introduced by a lattice norm and its completion.
We begin with the construction of free products and push-outs in V L \mathcal{VL} V L . Prime ideals reduce the study of embeddings to totally ordered quotients. This gives a direct proof of the classical fact that push-outs preserve injective lattice homomorphisms, and hence that free products preserve embeddings. The arguments use the theory of amalgamation of lattice-ordered groups and vector lattices developed in [22 , 23 , 9 ] . These arguments also provide an algebraic counterpart to the Banach lattice amalgamation methods in [2 ] . The Archimedean category behaves differently: amalgamation can fail, even though its free products still preserve embeddings (Proposition 4.11). Thus stability under free products and stability under general push-outs must be distinguished.
Our first main result characterizes the finitely generated Archimedean vector lattices X X X for which X ∗ V L Y X \ast_{\mathcal{VL}} Y X ∗ V L Y is Archimedean for every finitely generated Archimedean Y Y Y . We prove that this property is equivalent to projectivity of X X X (Proposition 4.6). Equivalently, X X X is finitely presented, embeds into a free vector lattice, or is represented by a polyhedral cone in the Baker–Beynon duality (see [3 , 4 ] ). The latter equivalences are classical, but the free-product condition gives a further characterization of the same class. The proof uses the description of finitely generated Archimedean vector lattices as lattices of piecewise linear functions on closed cones, under which Archimedean free products correspond to products of cones. For every non-polyhedral cone, one can produce a partner for which the ordinary free product is non-Archimedean. For arbitrary Archimedean X X X , the free product with every Archimedean Y Y Y is Archimedean precisely when every finitely generated sublattice of X X X is projective (Corollary 4.7). We also show that this local condition does not imply that X X X itself embeds into a free vector lattice. These results provide a connection between preservation of the Archimedean property, finite presentation and polyhedral geometry.
For Banach lattice factors ( X i ) i ∈ I (X_i)_{i\in I} ( X i ) i ∈ I , we identify precisely the algebraic sublattice generated by their canonical copies in the Banach lattice free product. The canonical maps factor as
\bigast_{i\in I}^{\mathrm{VL}} X_i \longrightarrow\bigast_{i\in I}^{\mathrm{AVL}} X_i \hookrightarrow\bigast_{i\in I}^{\mathrm{BL}} X_i
where the first map is Archimedeanization and the second is injective (Theorem 5.1 ). Moreover, the Banach lattice free product is the completion of the Archimedean free product under the largest lattice norm making the canonical maps contractive. Thus Archimedeanization accounts for all the algebraic identifications required before taking the norm completion. This step is necessary: the vector-lattice free product of two free Banach lattices can fail to be Archimedean (see Example 4.9 ).
The join representation is also useful beyond factors of the form C ( K ) C(K) C ( K ) . We first prove that injective lattice homomorphisms from C ( K ) C(K) C ( K ) and C ( L ) C(L) C ( L ) into arbitrary Banach lattices induce an injective map on their free products (Theorem 3.3 ). We then show that every element of the Banach lattice free product of two Banach lattices comes from the free product of two principal ideals, equipped with their order-unit norms (Proposition 5.2 ). This brings the representation by continuous functions into the study of individual elements of a general free product. It implies that Archimedean free products of order dense sublattices are order dense in the corresponding two-factor Banach lattice free product (Corollary 5.3 ), and that each factor is a regular sublattice of an arbitrary Banach lattice free product (Corollary 5.5 ). We give a countable family for which the algebraic sublattice generated by the factors is not order dense, despite being norm dense. Finally, every uniformly bounded family of injective Banach lattice homomorphisms induces an injective homomorphism between the free products (Proposition 5.6 ). This extends the embedding results of [18 ] and clarifies how injectivity and order approximation behave under completion.
Our final group of results concerns extension problems for vector lattices. Motivated by the connection between amalgamation, relative injectivity and universal disposition in separable Banach lattices developed in [2 ] , we consider the class
C = { X : ∣ Spec p ( X ) ∣ ≤ ℵ 0 } , \mathcal{C} = \{X : |\operatorname{Spec}_{p}(X)| \leq\aleph_{0}\}, C = { X : ∣ Spec p ( X ) ∣ ≤ ℵ 0 } ,
where Spec p ( X ) \operatorname{Spec}_{p}(X) Spec p ( X ) denotes the set of proper prime ideals of a vector lattice X X X . A vector lattice is called C \mathcal{C} C -injective, or countable-prime-injective, if lattice homomorphisms into it extend from sublattices of members of C \mathcal{C} C . Universal disposition for C \mathcal{C} C requires embeddings to admit extensions that are also embeddings.
Writing c = 2 ℵ 0 \mathfrak{c}=2^{\aleph_{0}} c = 2 ℵ 0 , we prove that every member of C \mathcal{C} C has cardinality at most c \mathfrak{c} c , although it may have uncountable dimension (Proposition 6.3 and Example 6.4). This bound permits a transfinite push-out construction which embeds every vector lattice into one of universal disposition for C \mathcal{C} C (Theorem 6.9 ). Such targets are countable-prime-injective, and there exists a nonzero target of universal disposition of cardinality at most 2 ℵ 1 2^{\aleph_{1}} 2 ℵ 1 . The construction takes place in V L \mathcal{VL} V L , since C \mathcal{C} C is closed under sublattices and quotients but not under free products or countable increasing unions.
A different construction gives smaller vector lattices which are C \mathcal{C} C -injective. For every nonprincipal ultrafilter U \mathcal{U} U on N \mathbb{N} N , the ordered ultrapower R N / U \mathbb{R}^{\mathbb{N}}/\mathcal{U} R N / U is countable-prime-injective and has cardinality c \mathfrak{c} c (Theorem 6.14 ). More generally, countable interpolation suffices for a totally ordered real vector space to have this extension property. The proof relates prime quotients and Archimedean classes to the realization of countable order cuts in ultrapowers. Since a totally ordered vector lattice cannot contain coordinatewise ordered R 2 \mathbb{R}^{2} R 2 , these examples also separate relative injectivity from universal disposition. On the other hand, every nonzero C \mathcal{C} C -injective vector lattice has at least c \mathfrak{c} c proper prime ideals and is necessarily non-Archimedean. The obstruction persists even when all vector lattices involved are Archimedean: no nonzero Archimedean vector lattice is injective relative to C ∩ A V L \mathcal{C} \cap\mathcal{AVL} C ∩ A V L (Theorem 6.21 ). This exhibits a limitation of purely order-theoretic extension properties which is relevant to their comparison with extension problems in Banach lattices.
The paper is organized as follows. Section 2 collects the algebraic constructions and the prime-ideal arguments. In Section 3, we develop the C ( K ) C(K) C ( K ) injectivity result. Section 4 studies Archimedeanization and the geometric criterion for free products, and Section 5 establishes the completion and order-theoretic consequences for Banach lattices. Finally, Section 6 deals with universal disposition and relative injectivity with respect to the class of vector lattices with countable prime spectra.
Formalization
The results and numbered definitions in this paper have been formalized in Lean 4.35.0. The development uses Mathlib 4.35.0 and the Banach lattice library BanLat, pinned to commits 56a005855ea1 and c3c5e1bfc4b2, respectively. The formalization is maintained in the FreeProductsVecLat repository; the version linked here is revision ca3135e88ade.
Each numbered statement carries a linked ∀ \forall ∀ symbol and the name of its corresponding Lean declaration. When a statement is split into several formal results, we provide the relevant links together. The links target source lines at the fixed revision above, independently of subsequent changes to the repository. The mathematical sources contain no proof placeholders, and the build includes a transitive check that the registered results use only Lean’s standard axioms.
Statement on AI use
GPT models were used in developing several of the mathematical arguments and constructions in this paper. GPT-5.5 was used to develop the proof, based on prime ideals, that push-outs of vector lattices preserve injective lattice homomorphisms (Proposition 2.10 ). The same model was used for the proof that injective lattice homomorphisms from C ( K ) C(K) C ( K ) spaces induce an injective map on their Banach lattice free products (Theorem 3.3 ), and for the proof that every element of the Banach lattice free product of two Banach lattices belongs to the image of the free product of two principal ideals equipped with their order-unit norms (Proposition 5.2 ).
GPT-5.6 was used to develop the discussion of polyhedral cones and the main characterization result in Section 4 (Proposition 4.6 and Corollary 4.7). In particular, it suggested using cone duality and helped formulate the precise conditions under which vector lattice free products of Archimedean factors remain Archimedean.
GPT-6 Astra was used to develop some of the results and proofs in § 6 , on vector lattices with countable spectra, relative injectivity, and universal disposition. GPT-6 Astra and Sol were also used to assist with the Lean formalization. The authors subsequently reviewed, corrected, and refined the AI-assisted arguments, adapting them to the mathematical framework and exposition of the paper.
Preliminaries on vector lattices
Throughout, all vector spaces and lattices are over the real field. An ideal of a vector lattice is a vector subspace which is solid (x ∈ I x \in I x ∈ I whenever ∣ x ∣ ≤ ∣ y ∣ |x| \le|y| ∣ x ∣ ≤ ∣ y ∣ for some y ∈ I y \in I y ∈ I ). Lattice homomorphisms are linear and preserve finite suprema and infima. We write lat ( A ) \operatorname{lat}(A) lat ( A ) for the vector sublattice generated by a set A A A . For background on vector and Banach lattices, see [14 , 19 ] .
For a vector space E E E , the free vector lattice FVL [ E ] \operatorname{FVL}[E] FVL [ E ] is generated by a linear copy of E E E and has the property that every linear map from E E E into a vector lattice extends uniquely to a lattice homomorphism; see [3 ] .
When E E E is a Banach space, FBL [ E ] \operatorname{FBL}[E] FBL [ E ] denotes the free Banach lattice generated by E E E . It contains a canonical linear isometric copy of E E E , and every bounded linear map from E E E into a Banach lattice extends uniquely to a lattice homomorphism from FBL [ E ] \operatorname{FBL}[E] FBL [ E ] of the same norm. This construction was introduced by Avilés, Rodríguez and Tradacete [1 ] , generalizing the construction over sets due to de Pagter and Wickstead [5 ] . For a systematic study of their structure and their relation to the geometry of the generating Banach space, see Oikhberg, Taylor, Tradacete and Troitsky [21 ] .
For free products of Banach lattices, we follow the conventions of [18 ] .
Given a family of vector lattices ( X i ) i ∈ I (X_i)_{i\in I} ( X i ) i ∈ I , a free product of this family is a vector lattice X X X together with lattice homomorphisms φ i : X i ⟶ X \varphi_i:X_i\longrightarrow X φ i : X i ⟶ X such that, for every family of lattice homomorphisms T i : X i ⟶ Y T_i:X_i\longrightarrow Y T i : X i ⟶ Y , there is a unique lattice homomorphism T : X ⟶ Y T:X\longrightarrow Y T : X ⟶ Y satisfying T i = T φ i T_i=T\varphi_i T i = T φ i for every i ∈ I i\in I i ∈ I . The corresponding definitions in A V L \mathcal{AVL} A V L and B L \mathcal{BL} BL use Archimedean targets and contractive Banach lattice homomorphisms, respectively. By rescaling, a uniformly bounded family of Banach lattice homomorphisms also induces a bounded homomorphism on the Banach lattice free product, of norm at most sup i ∥ T i ∥ \sup_i\|T_i\| sup i ∥ T i ∥ .
Proposition 2.1. ∀ nonempty_freeProduct The free product of a family of vector lattices always exists.
Proof. Let ( X i ) i ∈ I (X_i)_{i\in I} ( X i ) i ∈ I be a family of vector lattices, and let ι i : X i ⟶ Z = ( ⨁ j ∈ I X j ) \iota_i:X_i\longrightarrow Z=\left(\bigoplus_{j\in I}X_j\right) ι i : X i ⟶ Z = ( ⨁ j ∈ I X j ) be the inclusions into their algebraic direct sum. Given lattice homomorphisms T i : X i ⟶ Y T_i:X_i\longrightarrow Y T i : X i ⟶ Y , set S = ⨁ i T i : Z ⟶ Y S=\bigoplus_iT_i:Z\longrightarrow Y S = ⨁ i T i : Z ⟶ Y . The induced lattice homomorphism S ^ : FVL [ Z ] ⟶ Y \widehat{S}:\operatorname{FVL}[Z]\longrightarrow Y S : FVL [ Z ] ⟶ Y satisfies S ^ δ = S \widehat{S}\delta=S S δ = S , where δ : Z ⟶ FVL [ Z ] \delta:Z\longrightarrow\operatorname{FVL}[Z] δ : Z ⟶ FVL [ Z ] is the canonical inclusion.
Consider the ideal J J J of FVL [ Z ] \operatorname{FVL}[Z] FVL [ Z ] generated by elements of the form δ i ∣ x ∣ − ∣ δ i ( x ) ∣ \delta_i|x|-|\delta_i(x)| δ i ∣ x ∣ − ∣ δ i ( x ) ∣ , with i ∈ I i\in I i ∈ I and x ∈ X i x\in X_i x ∈ X i . If q : FVL [ Z ] ⟶ X = FVL [ Z ] / J q:\operatorname{FVL}[Z]\longrightarrow X=\operatorname{FVL}[Z]/J q : FVL [ Z ] ⟶ X = FVL [ Z ] / J denotes the quotient map and φ i = q δ i \varphi_i=q\delta_i φ i = q δ i , then
φ i ∣ x ∣ = q δ i ∣ x ∣ = q ∣ δ i ( x ) ∣ = ∣ φ i ( x ) ∣ . (1) \varphi_i|x|=q\delta_i|x|=q|\delta_i(x)|=|\varphi_i(x)|. \tag*{(1)} φ i ∣ x ∣ = q δ i ∣ x ∣ = q ∣ δ i ( x ) ∣ = ∣ φ i ( x ) ∣. (1)
S ^ \widehat{S} S vanishes on J J J because
S ^ ( δ i ∣ x ∣ − ∣ δ i ( x ) ∣ ) = S i ∣ x ∣ − ∣ S ^ δ i ( x ) ∣ = T i ∣ x ∣ − ∣ T i ( x ) ∣ = 0. \widehat{S}(\delta_i|x| - |\delta_i(x)|) = S_i|x| - |\widehat{S}\delta_i(x)| = T_i|x| - |T_i(x)| = 0. S ( δ i ∣ x ∣ − ∣ δ i ( x ) ∣ ) = S i ∣ x ∣ − ∣ S δ i ( x ) ∣ = T i ∣ x ∣ − ∣ T i ( x ) ∣ = 0.
Hence S ^ \widehat{S} S induces a lattice homomorphism T : X ⟶ Y T:X \longrightarrow Y T : X ⟶ Y with T q = S ^ Tq=\widehat{S} T q = S . The map T T T satisfies
T φ i = T q δ i = S ^ δ i = S ι i = T i . T\varphi_i = Tq\delta_i = \widehat{S}\delta_i = S\iota_i = T_i. T φ i = T q δ i = S δ i = S ι i = T i .
Uniqueness follows because the sublattices φ i ( X i ) \varphi_i(X_i) φ i ( X i ) generate X X X .
As usual, the free product of a family of vector lattices is unique up to lattice isomorphism. We denote by ∗ V L i ∈ I X i \ast_{\mathrm{VL}_{i\in I}}X_i ∗ VL i ∈ I X i the vector lattice free product of the family ( X i ) i ∈ I (X_i)_{i\in I} ( X i ) i ∈ I .
Observation 2.2. ∀ \forall ∀ inclusion_injective The inclusions φ i 0 : Y i 0 ⟶ ∗ V L Y i \varphi_{i_0}:Y_{i_0}\longrightarrow\ast_{\mathrm{VL}}Y_i φ i 0 : Y i 0 ⟶ ∗ VL Y i are injective.
Proof. Choose T i 0 = id Y i 0 T_{i_0}=\operatorname{id}_{Y_{i_0}} T i 0 = id Y i 0 and T i = 0 T_i=0 T i = 0 when i ≠ i 0 i\ne i_0 i = i 0 . Then ( ∗ i T i ) φ i 0 = id Y i 0 (\ast_iT_i)\varphi_{i_0}=\operatorname{id}_{Y_{i_0}} ( ∗ i T i ) φ i 0 = id Y i 0 , so φ i 0 \varphi_{i_0} φ i 0 is injective.
Proposition 2.3. ∀ \forall ∀ isFreeOver_directSum For every family of vector spaces ( X i ) (X_i) ( X i ) ,
FVL [ ⨁ X i ] ≃ ∗ V L FVL [ X i ] . \operatorname{FVL}\left[\bigoplus X_i\right]\simeq\ast_{\mathrm{VL}}\operatorname{FVL}[X_i]. FVL [ ⨁ X i ] ≃ ∗ VL FVL [ X i ] . Proof. The argument in [18 Proposition 5.5] shows that FVL [ ⨁ X i ] \operatorname{FVL}[\bigoplus X_i] FVL [ ⨁ X i ] has the required universal property.
Proposition 2.4. ∀ \forall ∀ exists_quotient_equiv Let J i ⊂ X i J_i\subset X_i J i ⊂ X i be ideals. Then
∗ V L ( X i / J i ) ≃ ( ∗ V L X i ) / J , \ast_{\mathrm{VL}}(X_i/J_i)\simeq(\ast_{\mathrm{VL}}X_i)/J, ∗ VL ( X i / J i ) ≃ ( ∗ VL X i ) / J , where J J J is the ideal generated by the φ X i ( J i ) \varphi_{X_i}(J_i) φ X i ( J i ) .
Proof. Let q i : X i ⟶ X i / J i q_i:X_i\longrightarrow X_i/J_i q i : X i ⟶ X i / J i and q : ∗ V L X i ⟶ Z : = ( ∗ V L X i ) / J q:\ast_{\mathrm{VL}}X_i\longrightarrow Z:=(\ast_{\mathrm{VL}}X_i)/J q : ∗ VL X i ⟶ Z := ( ∗ VL X i ) / J be the quotient maps. The vector lattice ( ∗ V L X i ) / J (\ast_{\mathrm{VL}}X_i)/J ( ∗ VL X i ) / J , together with the homomorphisms ψ i \psi_i ψ i given by ψ i q i = q φ i \psi_iq_i=q\varphi_i ψ i q i = q φ i , is a free product of the family ( X i / J i ) (X_i/J_i) ( X i / J i ) :
Given lattice homomorphisms T i : X i / J i ⟶ Y T_i:X_i/J_i\longrightarrow Y T i : X i / J i ⟶ Y , the induced homomorphism vanishes on each φ i ( J i ) \varphi_i(J_i) φ i ( J i ) , and hence on J J J . The induced homomorphism
T : = ∗ i ( T i q i ) : Z ⟶ Y T:=\ast_i(T_iq_i):Z\longrightarrow Y T := ∗ i ( T i q i ) : Z ⟶ Y satisfies T ψ i = T i T\psi_i=T_i T ψ i = T i and is unique with this property.
Proposition 2.5. ∀ \forall ∀ nonempty_pushout The category of vector lattices with lattice homomorphisms has push-outs. In particular, given lattice homomorphisms
T 1 : X 0 ⟶ X 1 , T 2 : X 0 ⟶ X 2 , T_1:X_0\longrightarrow X_1,\qquad T_2:X_0\longrightarrow X_2, T 1 : X 0 ⟶ X 1 , T 2 : X 0 ⟶ X 2 , there exist a vector lattice P O PO P O and lattice homomorphisms S i S_i S i , i = 1 , 2 i=1,2 i = 1 , 2 , such that the diagram
X 1 → S 1 P O T 1 ↑ ↑ S 2 X 0 → T 2 X 2 \begin{CD}
X_1 @>{S_1}>> PO \\
@A{T_1}AA @AA{S_2}A \\
X_0 @>{T_2}>> X_2
\end{CD} X 1 T 1 ⏐ ↑ X 0 S 1 T 2 P O ⏐ ↑ S 2 X 2
commutes and, for every vector lattice Y Y Y and homomorphisms S i ′ : X i → Y S_i' : X_i \to Y S i ′ : X i → Y making the corresponding square commute, there is a unique lattice homomorphism R : P O → Y R : PO \to Y R : P O → Y such that S i ′ = R S i S_i' = RS_i S i ′ = R S i .
Proof. Let V = X 1 ∗ V L X 2 V = X_1 *_{\mathrm{VL}} X_2 V = X 1 ∗ VL X 2 , with canonical maps φ 1 , φ 2 \varphi_1,\varphi_2 φ 1 , φ 2 , and let J J J be the ideal generated by φ 1 T 1 ( x ) − φ 2 T 2 ( x ) \varphi_1T_1(x) - \varphi_2T_2(x) φ 1 T 1 ( x ) − φ 2 T 2 ( x ) for x ∈ X 0 x \in X_0 x ∈ X 0 . Set P O = V / J PO = V/J P O = V / J and S i = q φ i S_i = q\varphi_i S i = q φ i , where q : V → V / J q : V \to V/J q : V → V / J is the quotient map. These maps satisfy S 1 T 1 = S 2 T 2 S_1T_1 = S_2T_2 S 1 T 1 = S 2 T 2 . For any compatible pair S i ′ : X i → Y S_i' : X_i \to Y S i ′ : X i → Y , the free-product homomorphism V → Y V \to Y V → Y vanishes on J J J , so it factors uniquely through P O PO P O . This proves the universal property. □
Totally ordered vector spaces and prime ideals
Definition 2.6. ∀ IsPrime A prime ideal P P P of a vector lattice X X X is an ideal such that, whenever a , b ∈ X + a,b \in X_+ a , b ∈ X + satisfy a ∧ b = 0 a \wedge b = 0 a ∧ b = 0 , either a ∈ P a \in P a ∈ P or b ∈ P b \in P b ∈ P .
In this section we allow P = X P = X P = X . The prime spectra in § 6 contain only proper prime ideals.
Proposition 2.7. ∀ isTotal_quotient_iff_isPrime If J J J is an ideal of X X X , then X / J X/J X / J is totally ordered if and only if J J J is prime.
Proof. See [11 Theorem 3.1] . □
Proposition 2.8. ∀ exists_prime_restriction Suppose Y ⊂ X Y \subset X Y ⊂ X is a vector sublattice and P ⊂ X P \subset X P ⊂ X is a prime ideal. Then P ∩ Y P \cap Y P ∩ Y is a prime ideal of Y Y Y . ∀ exists_prime_extension Conversely, if P ⊂ Y P \subset Y P ⊂ Y is a prime ideal of Y Y Y , there is a prime ideal of X X X , denoted by Q Q Q , such that Q ∩ Y = P Q \cap Y = P Q ∩ Y = P .
Proof. Restriction is immediate. For extension, let I I I be the ideal generated by P P P in X X X . Explicitly, I = { x ∈ X : ∣ x ∣ ≤ p for some p ∈ P + } I = \{x \in X : |x| \le p \text{ for some } p \in P_+\} I = { x ∈ X : ∣ x ∣ ≤ p for some p ∈ P + } , so I ∩ Y = P I \cap Y = P I ∩ Y = P . Passing to X / I X/I X / I and Y / P Y/P Y / P reduces the problem to the case where Y Y Y is totally ordered and P = { 0 } P = \{0\} P = { 0 } . By Zorn’s lemma, choose an ideal Q Q Q maximal subject to Q ∩ Y = { 0 } Q \cap Y = \{0\} Q ∩ Y = { 0 } . If Q Q Q were not prime, there would be disjoint a , b ∈ X + ∖ Q a,b \in X_+ \setminus Q a , b ∈ X + ∖ Q . Maximality gives nonzero y a , y b ∈ Y + y_a,y_b \in Y_+ y a , y b ∈ Y + , an element q ∈ Q + q \in Q_+ q ∈ Q + and λ > 0 \lambda> 0 λ > 0 such that
y a ≤ q + λ a , y b ≤ q + λ b . y_a \le q + \lambda a,\qquad y_b \le q + \lambda b. y a ≤ q + λa , y b ≤ q + λb . Since Y Y Y is totally ordered,
0 < y a ∧ y b ≤ ( q + λ a ) ∧ ( q + λ b ) = q , 0 < y_a \wedge y_b \le(q + \lambda a) \wedge(q + \lambda b) = q, 0 < y a ∧ y b ≤ ( q + λa ) ∧ ( q + λb ) = q , contradicting Q ∩ Y = { 0 } Q \cap Y = \{0\} Q ∩ Y = { 0 } . Thus Q Q Q is prime. □
The following amalgamation result for totally ordered vector spaces is a special case of the more general results of Pierce [22 , 23 ] and Hager and Madden [9 ] . We include a direct proof, which is also used in the proof of Proposition 2.10 .
Proposition 2.9. ∀ nonempty_totalAmalgam Let X X X , Y Y Y , Z Z Z be totally ordered vector spaces, and suppose there are injective lattice homomorphisms ι X : Z → X \iota_X : Z \to X ι X : Z → X and ι Y : Z → Y \iota_Y : Z \to Y ι Y : Z → Y , which identify Z Z Z with ordered subspaces of X X X and Y Y Y . Then there exist a totally ordered vector space Z 0 Z_0 Z 0 and injective lattice homomorphisms α : X → Z 0 \alpha: X \to Z_0 α : X → Z 0 and β : Y → Z 0 \beta: Y \to Z_0 β : Y → Z 0 such that
α ι X = β ι Y . \alpha\iota_X = \beta\iota_Y. α ι X = β ι Y .
Proof. Consider the vector-space amalgam
Z 0 = ( X ⊕ Y ) / { ( ι X ( c ) , − ι Y ( c ) ) : c ∈ Z } , Z_0=(X\oplus Y)/\{(\iota_X(c),-\iota_Y(c)):c\in Z\}, Z 0 = ( X ⊕ Y ) / {( ι X ( c ) , − ι Y ( c )) : c ∈ Z } ,
with canonical injective maps α \alpha α and β \beta β , which agree on Z Z Z . The wedge C 0 = α ( X + ) + β ( Y + ) C_0=\alpha(X_+)+\beta(Y_+) C 0 = α ( X + ) + β ( Y + ) is a cone. Indeed, if α ( a ) + β ( b ) = 0 \alpha(a)+\beta(b)=0 α ( a ) + β ( b ) = 0 with a , b ≥ 0 a,b\ge0 a , b ≥ 0 , then a = ι X ( c ) a=\iota_X(c) a = ι X ( c ) and b = − ι Y ( c ) b=-\iota_Y(c) b = − ι Y ( c ) for some c ∈ Z c\in Z c ∈ Z ; hence c ≥ 0 c\ge0 c ≥ 0 and c ≤ 0 c\le0 c ≤ 0 , so a = b = 0 a=b=0 a = b = 0 . Now let w ∈ C 0 ∩ ( − C 0 ) w\in C_0\cap(-C_0) w ∈ C 0 ∩ ( − C 0 ) and write w = α ( a 1 ) + β ( b 1 ) w=\alpha(a_1)+\beta(b_1) w = α ( a 1 ) + β ( b 1 ) and − w = α ( a 2 ) + β ( b 2 ) -w=\alpha(a_2)+\beta(b_2) − w = α ( a 2 ) + β ( b 2 ) , where a 1 , a 2 ∈ X + a_1,a_2\in X_+ a 1 , a 2 ∈ X + and b 1 , b 2 ∈ Y + b_1,b_2\in Y_+ b 1 , b 2 ∈ Y + . Adding gives
α ( a 1 + a 2 ) + β ( b 1 + b 2 ) = 0. \alpha(a_1+a_2)+\beta(b_1+b_2)=0. α ( a 1 + a 2 ) + β ( b 1 + b 2 ) = 0.
The preceding argument yields a 1 + a 2 = 0 a_1+a_2=0 a 1 + a 2 = 0 and b 1 + b 2 = 0 b_1+b_2=0 b 1 + b 2 = 0 . Positivity then forces all four elements to be zero, so w = 0 w=0 w = 0 . Thus C 0 ∩ ( − C 0 ) = { 0 } C_0\cap(-C_0)=\{0\} C 0 ∩ ( − C 0 ) = { 0 } . By Zorn’s lemma, extend C 0 C_0 C 0 to a maximal cone M M M in Z 0 Z_0 Z 0 . It is total: if v ∉ M ∪ ( − M ) v\notin M\cup(-M) v ∈ / M ∪ ( − M ) , then M + R + v M+\mathbb{R}_+v M + R + v is a strictly larger cone, since m + n + ( λ + μ ) v = 0 m+n+(\lambda+\mu)v=0 m + n + ( λ + μ ) v = 0 , with m , n ∈ M m,n\in M m , n ∈ M and λ , μ ≥ 0 \lambda,\mu\ge0 λ , μ ≥ 0 , forces λ = μ = 0 \lambda=\mu=0 λ = μ = 0 and m = n = 0 m=n=0 m = n = 0 . The order defined by M M M therefore makes Z 0 Z_0 Z 0 a totally ordered vector space. The maps α \alpha α and β \beta β preserve positivity and, being injective maps between totally ordered spaces, also reflect it and preserve lattice operations. □
Embeddings and colimits of vector lattices
The following classical fact is discussed by Hager and Madden [9 pp. 183–184] , who also note that their argument for abelian lattice-ordered groups applies to vector lattices. We include a direct proof using prime ideals, following the method of [23 Lemma 2 and the proof of Theorem 1(a)] for abelian lattice-ordered groups. This presentation also makes explicit the parallel with the proof of [2 Theorem 4.4] : prime quotients and totally ordered vector lattices play the role of the Hahn–Banach theorem and the L 1 L_1 L 1 -representations used there.
Proposition 2.10. ∀ \forall ∀ left_injective Let X 0 X_0 X 0 , X 1 X_1 X 1 , and X 2 X_2 X 2 be vector lattices, and let
X 1 → S 1 P O T 1 ↓ ↓ S 2 X 0 → T 2 X 2 \begin{CD}
X_1 @>{S_1}>> PO \\
@V{T_1}VV @VV{S_2}V \\
X_0 @>{T_2}>> X_2
\end{CD} X 1 T 1 ↓ ⏐ X 0 S 1 T 2 P O ↓ ⏐ S 2 X 2 be a push-out square. If T 2 T_2 T 2 is injective, then so is S 1 S_1 S 1 .
Proof. Identify X 0 X_0 X 0 with its image under T 2 T_2 T 2 in X 2 X_2 X 2 . Fix 0 ≠ x ∈ X 1 0\ne x\in X_1 0 = x ∈ X 1 and choose a prime ideal P 1 P_1 P 1 with x ∉ P 1 x\notin P_1 x ∈ / P 1 . The inverse image P 0 = T 1 − 1 ( P 1 ) P_0=T_1^{-1}(P_1) P 0 = T 1 − 1 ( P 1 ) is prime in X 0 X_0 X 0 . By Proposition 2.8 , choose a prime ideal P 2 P_2 P 2 in X 2 X_2 X 2 such that P 2 ∩ X 0 = P 0 P_2\cap X_0=P_0 P 2 ∩ X 0 = P 0 . The induced maps
T ~ 1 : X 0 / P 0 → X 1 / P 1 , T ~ 2 : X 0 / P 0 → X 2 / P 2 \widetilde{T}_1:X_0/P_0\to X_1/P_1,\qquad\widetilde{T}_2:X_0/P_0\to X_2/P_2 T 1 : X 0 / P 0 → X 1 / P 1 , T 2 : X 0 / P 0 → X 2 / P 2
are injective lattice homomorphisms between totally ordered vector spaces. By Proposition 2.9 , they admit an amalgamation into a totally ordered vector space Z Z Z . Writing $q_0,q_1,q_2 for the quotient maps, we obtain the commutative diagram
X 0 → q 0 X 0 / P 0 T 1 ↓ ↓ T ~ 1 X 1 → q 1 X 1 / P 1 T 2 ↓ ↓ T ~ 2 X 2 → q 2 X 2 / P 2 \begin{CD}
X_0 @>q_0>> X_0/P_0 \\
@V T_1 VV @VV \widetilde{T}_1 V \\
X_1 @>q_1>> X_1/P_1 \\
@V T_2 VV @VV \widetilde{T}_2 V \\
X_2 @>q_2>> X_2/P_2
\end{CD} X 0 T 1 ↓ ⏐ X 1 T 2 ↓ ⏐ X 2 q 0 q 1 q 2 X 0 / P 0 ↓ ⏐ T 1 X 1 / P 1 ↓ ⏐ T 2 X 2 / P 2
with R 1 R_1 R 1 and R 2 R_2 R 2 injective. The push-out property gives R : P O → Z R : PO \to Z R : P O → Z with R S 1 = R 1 q 1 RS_1 = R_1q_1 R S 1 = R 1 q 1 . Consequently,
R S 1 ( x ) = R 1 q 1 ( x ) ≠ 0 , RS_1(x) = R_1q_1(x) \ne0, R S 1 ( x ) = R 1 q 1 ( x ) = 0 ,
so S 1 S_1 S 1 is injective.
Corollary 2.11. Let T i : Y i → X i T_i : Y_i \to X_i T i : Y i → X i be injective vector lattice homomorphisms, with i ∈ I i \in I i ∈ I . The free product map
∗ ‾ T i : ∗ V L Y i → ∗ V L X i \overline{*}T_i : *_{\mathrm{VL}} Y_i \to*_{\mathrm{VL}} X_i ∗ T i : ∗ VL Y i → ∗ VL X i is injective.
Proof. Identify each Y i Y_i Y i with its image in X i X_i X i . For two factors, replacing one factor by the larger lattice gives a push-out of its inclusion along its canonical map into the free product. Proposition 2.10 makes the resulting map injective. Replacing the factors successively proves the finite case. For an arbitrary family, every element of a free product involves only finitely many factors. The canonical map from a finite subproduct into the full product has a left inverse obtained by sending all remaining factors to zero, so it is injective. The finite case therefore proves the assertion for arbitrary I I I .
Revisiting free products of C ( K ) C(K) C ( K )
In this section we prove that injective lattice homomorphisms from C ( K 1 ) C(K_1) C ( K 1 ) and C ( K 2 ) C(K_2) C ( K 2 ) into arbitrary Banach lattices induce an injective map on their free products. The proof uses the join representation of free products of C ( K ) C(K) C ( K ) spaces given in [18 ] , together with a simple approximation argument.
Lemma 3.1. Let T : C ( K ) → X T : C(K) \to X T : C ( K ) → X be an injective lattice homomorphism into a Banach lattice X X X . Let k 0 ∈ K k_0 \in K k 0 ∈ K , and let U U U be an open neighborhood of k 0 k_0 k 0 in K K K . Let I U I_U I U be the closed ideal of X X X generated by T { f : f ∣ U = 0 } {T\{f : f|_U = 0\}} T { f : f ∣ U = 0 } , and let q : X → X U = X / I U q : X \to X_U = X/I_U q : X → X U = X / I U be the quotient lattice homomorphism. Then e = q T ( 1 K ) e = qT(1_K) e = q T ( 1 K ) is nonzero. Moreover, if f ∈ C ( K ) f \in C(K) f ∈ C ( K ) and ε > 0 \varepsilon> 0 ε > 0 satisfy ∣ f − f ( k 0 ) 1 K ∣ ≤ ε 1 K |f - f(k_0)1_K| \le\varepsilon1_K ∣ f − f ( k 0 ) 1 K ∣ ≤ ε 1 K on U U U , then
∣ q T ( f ) − f ( k 0 ) e ∣ ≤ ε e . |qT(f) - f(k_0)e| \le\varepsilon e. ∣ q T ( f ) − f ( k 0 ) e ∣ ≤ ε e .
Proof. Choose 0 < g ∈ C ( K ) 0<g\in C(K) 0 < g ∈ C ( K ) such that supp g ⊂ U \operatorname{supp}g\subset U supp g ⊂ U . If f ∈ C ( K ) f\in C(K) f ∈ C ( K ) and f ∣ U = 0 f|_U=0 f ∣ U = 0 , then ∣ f ∣ ∧ g = 0 |f|\wedge g=0 ∣ f ∣ ∧ g = 0 . Since T T T is a lattice homomorphism, ∣ T ( f ) ∣ ∧ T ( g ) = 0 |T(f)|\wedge T(g)=0 ∣ T ( f ) ∣ ∧ T ( g ) = 0 . Hence every generator T ( f ) T(f) T ( f ) of I U I_U I U belongs to the disjoint complement of T ( g ) T(g) T ( g ) , and the disjoint complement of T ( g ) T(g) T ( g ) is a closed ideal. Therefore I U I_U I U is contained in the disjoint complement of T ( g ) T(g) T ( g ) . If q T ( g ) = 0 qT(g)=0 q T ( g ) = 0 , then T ( g ) ∈ I U T(g)\in I_U T ( g ) ∈ I U , so T ( g ) ⊥ T ( g ) T(g)\perp T(g) T ( g ) ⊥ T ( g ) and hence T ( g ) = 0 T(g)=0 T ( g ) = 0 . This contradicts the injectivity of T T T . Thus q T ( g ) ≠ 0 qT(g)\ne0 q T ( g ) = 0 . Since 0 ≤ T ( g ) ≤ ∥ g ∥ ∞ T ( 1 K ) 0\le T(g)\le\lVert g\rVert_\infty T(1_K) 0 ≤ T ( g ) ≤ ∥ g ∥ ∞ T ( 1 K ) , we have that
0 ≠ q T ( g ) ≤ ∥ g ∥ ∞ q T ( 1 K ) = ∥ g ∥ ∞ e , 0\ne qT(g)\le\lVert g\rVert_\infty qT(1_K)=\lVert g\rVert_\infty e, 0 = q T ( g ) ≤ ∥ g ∥ ∞ q T ( 1 K ) = ∥ g ∥ ∞ e ,
so e ≠ 0 e\ne0 e = 0 . □
For convenience, we now recall the representation of free products of C ( K ) C(K) C ( K ) spaces from [18 Theorem 6.2] . First, let us recall the definition of the join. For nonempty compact Hausdorff spaces K 1 K_1 K 1 and K 2 K_2 K 2 , their join is
K 1 ∗ K 2 = ( K 1 × K 2 × [ 0 , 1 ] ) / ∼ , K_1\ast K_2=(K_1\times K_2\times[0,1])/{\sim}, K 1 ∗ K 2 = ( K 1 × K 2 × [ 0 , 1 ]) / ∼ ,
where ( k , l , 0 ) ∼ ( k , l ′ , 0 ) (k,l,0)\sim(k,l',0) ( k , l , 0 ) ∼ ( k , l ′ , 0 ) and ( k , l , 1 ) ∼ ( k ′ , l , 1 ) (k,l,1)\sim(k',l,1) ( k , l , 1 ) ∼ ( k ′ , l , 1 ) for all k , k ′ ∈ K 1 k,k'\in K_1 k , k ′ ∈ K 1 and l , l ′ ∈ K 2 l,l'\in K_2 l , l ′ ∈ K 2 . Write [ k , l , t ] [k,l,t] [ k , l , t ] for the equivalence class of ( k , l , t ) (k,l,t) ( k , l , t ) .
Theorem 3.2. ∀ exists_join_equiv Let K 1 K_1 K 1 and K 2 K_2 K 2 be nonempty compact Hausdorff spaces. There is a lattice isomorphism
C ( K 1 ) ∗ B L C ( K 2 ) ≃ C ( K 1 ∗ K 2 ) C(K_1)\ast_{\mathrm{BL}}C(K_2)\simeq C(K_1\ast K_2) C ( K 1 ) ∗ BL C ( K 2 ) ≃ C ( K 1 ∗ K 2 ) under which the canonical embeddings satisfy
φ 1 ( f ) ( [ k , l , t ] ) = ( 1 − t ) f ( k ) , φ 2 ( g ) ( [ k , l , t ] ) = t g ( l ) . \varphi_1(f)([k,l,t])=(1-t)f(k),\qquad\varphi_2(g)([k,l,t])=tg(l). φ 1 ( f ) ([ k , l , t ]) = ( 1 − t ) f ( k ) , φ 2 ( g ) ([ k , l , t ]) = t g ( l ) . Note that here C ( K 1 ∗ K 2 ) C(K_1\ast K_2) C ( K 1 ∗ K 2 ) is always equipped with its usual supremum norm. The displayed isomorphism is an isomorphism of Banach lattices, in general it is not an isometry for the free product norm.
Theorem 3.3. ∀ induced_injective_of_continuousFunctions Let K 1 , K 2 K_1,K_2 K 1 , K 2 be nonempty compact Hausdorff spaces and let X 1 , X 2 X_1,X_2 X 1 , X 2 be Banach lattices. Suppose there are injective lattice homomorphisms T 1 : C ( K 1 ) → X 1 T_1:C(K_1)\to X_1 T 1 : C ( K 1 ) → X 1 and T 2 : C ( K 2 ) → X 2 T_2:C(K_2)\to X_2 T 2 : C ( K 2 ) → X 2 . The induced lattice homomorphism
T 1 ∗ T 2 : C ( K 1 ) ∗ B L C ( K 2 ) → X 1 ∗ B L X 2 T_1\ast T_2:C(K_1)\ast_{\mathrm{BL}}C(K_2)\to X_1\ast_{\mathrm{BL}}X_2 T 1 ∗ T 2 : C ( K 1 ) ∗ BL C ( K 2 ) → X 1 ∗ BL X 2 is injective.
Proof. Identify the domain with C ( K 1 ∗ K 2 ) C(K_1\ast K_2) C ( K 1 ∗ K 2 ) by the isomorphism from Theorem 3.2 and write T = T 1 ∗ T 2 T=T_1\ast T_2 T = T 1 ∗ T 2 . Fix p 0 = [ k 0 , l 0 , t 0 ] ∈ K 1 ∗ K 2 p_0=[k_0,l_0,t_0]\in K_1\ast K_2 p 0 = [ k 0 , l 0 , t 0 ] ∈ K 1 ∗ K 2 . We will show that T ( h ) = 0 T(h)=0 T ( h ) = 0 implies h ( p 0 ) = 0 h(p_0)=0 h ( p 0 ) = 0 by constructing lattice homomorphisms that approximate evaluation at p 0 p_0 p 0 . Let U U U and V V V be open neighborhoods of k 0 k_0 k 0 and l 0 l_0 l 0 . Applying Lemma 3.1 to T 1 T_1 T 1 and U U U , and to T 2 T_2 T 2 and V V V , gives quotient lattice homomorphisms q 1 : X 1 → X 1 , U q_1:X_1\to X_{1,U} q 1 : X 1 → X 1 , U and q 2 : X 2 → X 2 , V q_2:X_2\to X_{2,V} q 2 : X 2 → X 2 , V . The elements e = q 1 T 1 ( 1 K 1 ) e=q_1T_1(1_{K_1}) e = q 1 T 1 ( 1 K 1 ) and d = q 2 T 2 ( 1 K 2 ) d=q_2T_2(1_{K_2}) d = q 2 T 2 ( 1 K 2 ) are nonzero and positive. Put
Z = X 1 , U ⊗ ^ ∣ π ∣ X 2 , V , Z=X_{1,U}\widehat{\otimes}_{|\pi|}X_{2,V}, Z = X 1 , U ⊗ ∣ π ∣ X 2 , V ,
the Fremlin projective tensor product [6 ] . Its canonical bilinear map is a lattice bimorphism and its norm is a cross norm: ∣ x ⊗ y ∣ = ∣ x ∣ ⊗ ∣ y ∣ |x\otimes y|=|x|\otimes|y| ∣ x ⊗ y ∣ = ∣ x ∣ ⊗ ∣ y ∣ and ∥ x ⊗ y ∥ = ∥ x ∥ ∥ y ∥ \|x\otimes y\|=\|x\|\|y\| ∥ x ⊗ y ∥ = ∥ x ∥∥ y ∥ . Thus w = e ⊗ d > 0 w=e\otimes d>0 w = e ⊗ d > 0 , and the following maps are bounded lattice homomorphisms S 1 : X 1 → Z S_1:X_1\to Z S 1 : X 1 → Z and S 2 : X 2 → Z S_2:X_2\to Z S 2 : X 2 → Z :
S 1 ( x ) = ( 1 − t 0 ) q 1 ( x ) ⊗ d , S 2 ( y ) = t 0 e ⊗ q 2 ( y ) . S_1(x)=(1-t_0)q_1(x)\otimes d,\qquad S_2(y)=t_0e\otimes q_2(y). S 1 ( x ) = ( 1 − t 0 ) q 1 ( x ) ⊗ d , S 2 ( y ) = t 0 e ⊗ q 2 ( y ) .
The universal property gives S = S 1 ∗ B L S 2 : X 1 ∗ B L X 2 → Z S=S_1*_{BL}S_2:X_1*_{BL}X_2\to Z S = S 1 ∗ B L S 2 : X 1 ∗ B L X 2 → Z . Thus Θ = S T : C ( K 1 ∗ K 2 ) → Z \Theta=ST:C(K_1*K_2)\to Z Θ = S T : C ( K 1 ∗ K 2 ) → Z satisfies
Θ ( φ 1 ( f ) ) = ( 1 − t 0 ) q 1 T 1 ( f ) ⊗ d , Θ ( φ 2 ( g ) ) = t 0 e ⊗ q 2 T 2 ( g ) . \Theta(\varphi_1(f))=(1-t_0)q_1T_1(f)\otimes d,\qquad\Theta(\varphi_2(g))=t_0e\otimes q_2T_2(g). Θ ( φ 1 ( f )) = ( 1 − t 0 ) q 1 T 1 ( f ) ⊗ d , Θ ( φ 2 ( g )) = t 0 e ⊗ q 2 T 2 ( g ) .
In particular, since 1 K 1 ∗ K 2 = φ 1 ( 1 K 1 ) + φ 2 ( 1 K 2 ) 1_{K_1*K_2}=\varphi_1(1_{K_1})+\varphi_2(1_{K_2}) 1 K 1 ∗ K 2 = φ 1 ( 1 K 1 ) + φ 2 ( 1 K 2 ) , we have
Θ ( 1 K 1 ∗ K 2 ) = w . \Theta(1_{K_1*K_2})=w. Θ ( 1 K 1 ∗ K 2 ) = w .
Compare this map with Θ 0 : C ( K 1 ∗ K 2 ) → Z \Theta_0:C(K_1*K_2)\to Z Θ 0 : C ( K 1 ∗ K 2 ) → Z , defined by Θ 0 ( h ) = h ( p 0 ) w \Theta_0(h)=h(p_0)w Θ 0 ( h ) = h ( p 0 ) w : on the canonical copies it satisfies
Θ 0 ( φ 1 ( f ) ) = ( 1 − t 0 ) f ( k 0 ) w , Θ 0 ( φ 2 ( g ) ) = t 0 g ( l 0 ) w . \Theta_0(\varphi_1(f))=(1-t_0)f(k_0)w,\qquad\Theta_0(\varphi_2(g))=t_0g(l_0)w. Θ 0 ( φ 1 ( f )) = ( 1 − t 0 ) f ( k 0 ) w , Θ 0 ( φ 2 ( g )) = t 0 g ( l 0 ) w .
Note these constructions depend on U U U and V V V , but for the sake of brevity we suppress this dependence in the notation.
We claim that, for each h ∈ C ( K 1 ∗ K 2 ) h\in C(K_1*K_2) h ∈ C ( K 1 ∗ K 2 ) and η > 0 \eta>0 η > 0 , the neighborhoods U U U and V V V can be chosen so that
∣ Θ ( h ) − Θ 0 ( h ) ∣ = ∣ Θ ( h ) − h ( p 0 ) w ∣ ≤ η w . |\Theta(h)-\Theta_0(h)|=|\Theta(h)-h(p_0)w|\leq\eta w. ∣Θ ( h ) − Θ 0 ( h ) ∣ = ∣Θ ( h ) − h ( p 0 ) w ∣ ≤ η w .
Indeed, first suppose that h h h belongs to the sublattice generated by the canonical copies of C ( K 1 ) C(K_1) C ( K 1 ) and C ( K 2 ) C(K_2) C ( K 2 ) . It is a finite lattice expression in functions f 1 , … , f m ∈ C ( K 1 ) f_1,\ldots,f_m\in C(K_1) f 1 , … , f m ∈ C ( K 1 ) and g 1 , … , g n ∈ C ( K 2 ) g_1,\ldots,g_n\in C(K_2) g 1 , … , g n ∈ C ( K 2 ) . Given δ > 0 \delta>0 δ > 0 , choose U U U and V V V so that
∣ f i − f i ( k 0 ) 1 K 1 ∣ ≤ δ on U , ∣ g j − g j ( l 0 ) 1 K 2 ∣ ≤ δ on V |f_i-f_i(k_0)1_{K_1}|\leq\delta\quad\text{on }U,\qquad|g_j-g_j(l_0)1_{K_2}|\leq\delta\quad\text{on }V ∣ f i − f i ( k 0 ) 1 K 1 ∣ ≤ δ on U , ∣ g j − g j ( l 0 ) 1 K 2 ∣ ≤ δ on V
for all i , j i,j i , j . Lemma 3.1 then gives
∣ q 1 T 1 ( f i ) − f i ( k 0 ) e ∣ ≤ δ e , ∣ q 2 T 2 ( g j ) − g j ( l 0 ) d ∣ ≤ δ d . |q_1T_1(f_i)-f_i(k_0)e|\leq\delta e,\qquad|q_2T_2(g_j)-g_j(l_0)d|\leq\delta d. ∣ q 1 T 1 ( f i ) − f i ( k 0 ) e ∣ ≤ δ e , ∣ q 2 T 2 ( g j ) − g j ( l 0 ) d ∣ ≤ δ d .
Tensoring with d d d or e e e and multiplying by 1 − t 0 1-t_0 1 − t 0 or t 0 t_0 t 0 yields
∣ Θ ( φ 1 ( f i ) ) − Θ 0 ( φ 1 ( f i ) ) ∣ ≤ δ w , ∣ Θ ( φ 2 ( g j ) ) − Θ 0 ( φ 2 ( g j ) ) ∣ ≤ δ w . |\Theta(\varphi_1(f_i))-\Theta_0(\varphi_1(f_i))|\leq\delta w,\qquad|\Theta(\varphi_2(g_j))-\Theta_0(\varphi_2(g_j))|\leq\delta w. ∣Θ ( φ 1 ( f i )) − Θ 0 ( φ 1 ( f i )) ∣ ≤ δ w , ∣Θ ( φ 2 ( g j )) − Θ 0 ( φ 2 ( g j )) ∣ ≤ δ w .
Using the inequalities
∣ ( u ∨ v ) − ( u ′ ∨ v ′ ) ∣ ≤ ∣ u − u ′ ∣ + ∣ v − v ′ ∣ , ∣ ( u ∧ v ) − ( u ′ ∧ v ′ ) ∣ ≤ ∣ u − u ′ ∣ + ∣ v − v ′ ∣ , |(u\vee v)-(u'\vee v')|\leq|u-u'|+|v-v'|,\qquad|(u\wedge v)-(u'\wedge v')|\leq|u-u'|+|v-v'|, ∣ ( u ∨ v ) − ( u ′ ∨ v ′ ) ∣ ≤ ∣ u − u ′ ∣ + ∣ v − v ′ ∣ , ∣ ( u ∧ v ) − ( u ′ ∧ v ′ ) ∣ ≤ ∣ u − u ′ ∣ + ∣ v − v ′ ∣ ,
a standard induction argument gives ∣ Θ ( h ) − Θ 0 ( h ) ∣ ≤ C h δ w |\Theta(h)-\Theta_0(h)|\leq C_h\delta w ∣Θ ( h ) − Θ 0 ( h ) ∣ ≤ C h δ w , where C h C_h C h depends only on the chosen expression, not on δ \delta δ . Taking δ \delta δ sufficiently small proves the claim on the generated sublattice.
For general h h h , choose h ~ ∈ lat ( φ 1 ( C ( K 1 ) ) , φ 2 ( C ( K 2 ) ) ) \widetilde h\in\operatorname{lat}(\varphi_1(C(K_1)),\varphi_2(C(K_2))) h ∈ lat ( φ 1 ( C ( K 1 )) , φ 2 ( C ( K 2 ))) with ∥ h − h ~ ∥ ∞ < η / 3 \|h-\widetilde h\|_\infty<\eta/3 ∥ h − h ∥ ∞ < η /3 . Since both Θ \Theta Θ and Θ 0 \Theta_0 Θ 0 send 1 K 1 ∗ K 2 1_{K_1*K_2} 1 K 1 ∗ K 2 to w w w , each maps h − h ~ h-\widetilde h h − h to an element of absolute value at most ( η / 3 ) w (\eta/3)w ( η /3 ) w . Choose U , V U,V U , V using the preceding case for h ~ \widetilde h h and η / 3 \eta/3 η /3 . The triangle inequality now gives ∣ Θ ( h ) − Θ 0 ( h ) ∣ ≤ η w |\Theta(h)-\Theta_0(h)|\leq\eta w ∣Θ ( h ) − Θ 0 ( h ) ∣ ≤ η w , proving the claim.
Finally, suppose T ( h ) = 0 T(h)=0 T ( h ) = 0 . Then Θ ( h ) = 0 \Theta(h)=0 Θ ( h ) = 0 for every choice of U , V U,V U , V . If h ( p 0 ) ≠ 0 h(p_0)\ne0 h ( p 0 ) = 0 , choose 0 < η < ∣ h ( p 0 ) ∣ 0<\eta<|h(p_0)| 0 < η < ∣ h ( p 0 ) ∣ and apply the claim to obtain
∣ h ( p 0 ) ∣ w = ∣ Θ ( h ) − Θ 0 ( h ) ∣ ≤ η w , |h(p_0)|w=|\Theta(h)-\Theta_0(h)|\leq\eta w, ∣ h ( p 0 ) ∣ w = ∣Θ ( h ) − Θ 0 ( h ) ∣ ≤ η w ,
contradicting w > 0 w > 0 w > 0 . Thus h ( p 0 ) = 0 h(p_0) = 0 h ( p 0 ) = 0 ; since p 0 p_0 p 0 was arbitrary, h = 0 h = 0 h = 0 and T T T is injective. □
Archimedean and non-Archimedean free products
Recall that a vector lattice is Archimedean if 0 ≤ n x ≤ y 0 \le nx \le y 0 ≤ n x ≤ y for every positive integer n n n implies x = 0 x = 0 x = 0 . An ideal is uniformly closed if it contains the limits of its relatively uniformly convergent sequences. Here x n → x x_n \to x x n → x relatively uniformly means that ∣ x n − x ∣ ≤ ε n u \lvert x_n - x\rvert\le\varepsilon_n u ∣ x n − x ∣ ≤ ε n u for some u ≥ 0 u \ge0 u ≥ 0 and some real sequence ε n ↓ 0 \varepsilon_n \downarrow0 ε n ↓ 0 . For an ideal J J J of a vector lattice X X X , the quotient X / J X/J X / J is Archimedean if and only if J J J is uniformly closed; see [14 Theorem 60.2] .
Let a ( X ) \mathfrak{a}(X) a ( X ) be the intersection of all uniformly closed ideals of X X X . This ideal is uniformly closed, so X / a ( X ) X/\mathfrak{a}(X) X / a ( X ) is Archimedean. If Y Y Y is Archimedean and T : X → Y T : X \to Y T : X → Y is a lattice homomorphism, then ker T \ker T ker T is uniformly closed and therefore contains a ( X ) \mathfrak{a}(X) a ( X ) . Thus T T T factors uniquely as
X → T Y q ↓ T ~ ↑ X / a ( X ) \begin{CD}
X @>{T}>> Y\\
@V{q}VV @A{\widetilde{T}}AA\\
X/\mathfrak{a}(X)
\end{CD} X q ↓ ⏐ X / a ( X ) T Y T ⏐ ↑
Consequently, A V L \mathcal{AVL} A V L is a reflective subcategory of V L \mathcal{VL} V L [15 Chapter IV, §3] . We call X / a ( X ) X/\mathfrak{a}(X) X / a ( X ) the Archimedeanization of X X X and denote it by X a X_{\mathrm{a}} X a .
The subcategory of Archimedean vector lattices also has free products and push-outs. These are precisely the Archimedeanizations of the corresponding colimits in V L \mathcal{VL} V L . As the following example shows, colimits in V L \mathcal{VL} V L need not agree with the corresponding colimits in A V L \mathcal{AVL} A V L . This contrasts with the fact that free objects in V L \mathcal{VL} V L are Archimedean.
The following example is essentially due to K. Pierce [22 , 23 ] in the setting of ℓ \ell ℓ -groups.
Example 4.1. ∀ archimedean_amalgam_kills_extra_coordinate The category A V L \mathcal{AVL} A V L of Archimedean vector lattices and lattice homomorphisms does not have the amalgamation property: two injective lattice homomorphisms with a common domain need not admit a commutative square whose remaining maps are both injective. ∀ exists_nonarchimedean_pushout In particular, the canonical maps into their push-out need not be injective. Take, for example, X 0 = { ( x n ) ∈ R N : ( x n ) is eventually constant } X_0 = \{(x_n) \in\mathbb{R}^{\mathbb{N}} : (x_n)\text{ is eventually constant}\} X 0 = {( x n ) ∈ R N : ( x n ) is eventually constant } , X 1 = R N X_1 = \mathbb{R}^{\mathbb{N}} X 1 = R N and X 2 = X 1 ⊕ R X_2 = X_1 \oplus\mathbb{R} X 2 = X 1 ⊕ R . Let j 1 : X 0 → X 1 j_1 : X_0 \to X_1 j 1 : X 0 → X 1 be the inclusion, and define j 2 : X 0 → X 2 j_2 : X_0 \to X_2 j 2 : X 0 → X 2 by j 2 ( ( x n ) ) = ( ( x n ) , lim x n ) j_2((x_n)) = ((x_n), \lim x_n) j 2 (( x n )) = (( x n ) , lim x n ) . For any Archimedean vector lattice Z Z Z and lattice homomorphisms T i : X i → Z T_i : X_i \to Z T i : X i → Z making the diagram
X 0 → j 1 X 1 j 2 ↓ T 1 ↓ X 2 → T 2 Z \begin{CD}
X_0 @>{j_1}>> X_1\\
@V{j_2}VV @V{T_1}VV\\
X_2 @>{T_2}>> Z
\end{CD} X 0 j 2 ↓ ⏐ X 2 j 1 T 2 X 1 T 1 ↓ ⏐ Z commute, we have T 2 ( 0 , 1 ) = 0 T_2(0,1) = 0 T 2 ( 0 , 1 ) = 0 . Indeed, for each n ∈ N n \in\mathbb{N} n ∈ N ,
T 1 ( 1 , 2 , 3 , … ) ≥ T 1 ( 1 , 2 , … , n , n , … ) = T 1 j 1 ( 1 , 2 , … , n , n , … ) = T 2 j 2 ( 1 , 2 , … , n , n , … ) = T 2 ( ( 1 , 2 , … , n , n , … ) , n ) ≥ T 2 ( 0 , n ) = n T 2 ( 0 , 1 ) . \begin{aligned}
T_1(1,2,3,\ldots) &\ge T_1(1,2,\ldots,n,n,\ldots) = T_1j_1(1,2,\ldots,n,n,\ldots) = T_2j_2(1,2,\ldots,n,n,\ldots)\\
&= T_2((1,2,\ldots,n,n,\ldots),n) \ge T_2(0,n) = nT_2(0,1).
\end{aligned} T 1 ( 1 , 2 , 3 , … ) ≥ T 1 ( 1 , 2 , … , n , n , … ) = T 1 j 1 ( 1 , 2 , … , n , n , … ) = T 2 j 2 ( 1 , 2 , … , n , n , … ) = T 2 (( 1 , 2 , … , n , n , … ) , n ) ≥ T 2 ( 0 , n ) = n T 2 ( 0 , 1 ) .
Since Z Z Z is Archimedean, it follows that T 2 ( 0 , 1 ) = 0 T_2(0,1)=0 T 2 ( 0 , 1 ) = 0 . In particular, this holds when Z Z Z is the Archimedean push-out of j 1 j_1 j 1 and j 2 j_2 j 2 .
The preceding example shows that the V L \mathcal{V}\mathcal{L} V L push-out of Archimedean vector lattices need not be Archimedean. We now study the corresponding question for free products.
In what follows, we use Baker’s conventions from [3 ] . A closed cone in R n \mathbb{R}^n R n contains the origin and is invariant under multiplication by nonnegative scalars, though it need not be convex. A closed polyhedral cone is a finite union of closed convex polyhedral cones. An open polyhedral cone is a finite union of nonempty finite intersections of open half-spaces whose boundaries pass through the origin. For a function f f f , its open support is S ( f ) = { x : f ( x ) ≠ 0 } S(f)=\{x:f(x)\ne0\} S ( f ) = { x : f ( x ) = 0 } . In the standard representation of free vector lattices, FVL [ R n ] \operatorname{FVL}[\mathbb{R}^n] FVL [ R n ] consists of continuous positively homogeneous piecewise linear functions on R n \mathbb{R}^n R n , and their open supports are precisely the open polyhedral cones.
For finitely generated Archimedean vector lattices, the A V L \mathcal{A}\mathcal{V}\mathcal{L} A V L free product admits a geometric representation. For a subset A A A of a vector lattice, let I ( A ) I(A) I ( A ) denote the ideal generated by A A A . For a closed cone C ⊆ R n C\subseteq\mathbb{R}^n C ⊆ R n , write
Van ( C ) = { f ∈ FVL [ R n ] : f ∣ C = 0 } . \operatorname{Van}(C)=\{f\in\operatorname{FVL}[\mathbb{R}^n]:f|_C=0\}. Van ( C ) = { f ∈ FVL [ R n ] : f ∣ C = 0 } .
for the ideal of functions vanishing on C C C .
For a closed cone C ⊆ R m C\subseteq\mathbb{R}^m C ⊆ R m , put
L ( C ) = { f ∣ C : f ∈ FVL [ R m ] } . \mathcal{L}(C)=\{f|_C:f\in\operatorname{FVL}[\mathbb{R}^m]\}. L ( C ) = { f ∣ C : f ∈ FVL [ R m ]} .
If C ⊆ R m C\subseteq\mathbb{R}^m C ⊆ R m and D ⊆ R n D\subseteq\mathbb{R}^n D ⊆ R n are closed cones, an ℓ \ell ℓ -map q : C → D q:C\to D q : C → D is a map q = ( q 1 , … , q n ) q=(q_1,\ldots,q_n) q = ( q 1 , … , q n ) whose coordinate functions q i q_i q i belong to L ( C ) \mathcal{L}(C) L ( C ) . In particular, its coordinates extend to continuous positively homogeneous piecewise linear functions on the whole ambient space R m \mathbb{R}^m R m . Let Cone ℓ \operatorname{Cone}^{\ell} Cone ℓ denote the category of closed cones in finite-dimensional Euclidean spaces and ℓ \ell ℓ -maps.
Theorem 4.2 (Baker–Beynon duality). Let A V L f g \mathcal{A}\mathcal{V}\mathcal{L}_{\mathrm{fg}} A V L fg denote the category of finitely generated Archimedean real vector lattices and vector lattice homomorphisms. There is a contravariant equivalence
( Cone ℓ ) o p ≃ A V L f g , C ⟼ L ( C ) , ( q : C → D ) ⟼ ( q ∗ : f ⟼ f ∘ q ) . \left(\operatorname{Cone}^{\ell}\right)^{\mathrm{op}}\simeq\mathcal{A}\mathcal{V}\mathcal{L}_{\mathrm{fg}},\qquad C\longmapsto\mathcal{L}(C),\qquad(q:C\to D)\longmapsto(q^*:f\longmapsto f\circ q). ( Cone ℓ ) op ≃ A V L fg , C ⟼ L ( C ) , ( q : C → D ) ⟼ ( q ∗ : f ⟼ f ∘ q ) .
Proof. The representation X ≅ L ( C ) X\cong\mathcal{L}(C) X ≅ L ( C ) is [3 Theorem 4.3] . The explicit correspondence on morphisms is in [4 Proposition 2.1 and Corollary 1] . □
This defines a notion of duality which gives the following description of free products. Let X 1 , X 2 ∈ A V L f g X_1,X_2\in\mathcal{A}\mathcal{V}\mathcal{L}_{\mathrm{fg}} X 1 , X 2 ∈ A V L fg , and choose closed cones C 1 ⊆ R n C_1\subseteq\mathbb{R}^n C 1 ⊆ R n and C 2 ⊆ R m C_2\subseteq\mathbb{R}^m C 2 ⊆ R m , together with corresponding isomorphisms
X 1 ≅ L ( C 1 ) and X 2 ≅ L ( C 2 ) . X_1\cong\mathcal{L}(C_1)\qquad\text{and}\qquad X_2\cong\mathcal{L}(C_2). X 1 ≅ L ( C 1 ) and X 2 ≅ L ( C 2 ) .
The free product of X 1 X_1 X 1 and X 2 X_2 X 2 in A V L f g \mathcal{A}\mathcal{V}\mathcal{L}_{\mathrm{fg}} A V L fg exists and, up to canonical isomorphism, satisfies
X 1 ∗ A V L f g X 2 ≅ L ( C 1 × C 2 ) . X_1*_{\mathcal{A}\mathcal{V}\mathcal{L}_{\mathrm{fg}}}X_2\cong\mathcal{L}(C_1\times C_2). X 1 ∗ A V L fg X 2 ≅ L ( C 1 × C 2 ) .
Indeed, C 1 × C 2 C_1 \times C_2 C 1 × C 2 , with the coordinate projections, is the categorical product in Cone ℓ \operatorname{Cone}^{\ell} Cone ℓ , and the contravariant equivalence carries products to free products (coproducts).
For two finitely generated Archimedean vector lattices, the A V L f g \mathcal{AVL}_{\mathrm{fg}} A V L fg free product agrees with the A V L \mathcal{AVL} A V L free product because the latter is finitely generated.
Define
J C 1 × C 2 : = I ( j 1 ( Van ( C 1 ) ) ∪ j 2 ( Van ( C 2 ) ) ) ⊆ FVL [ R n + m ] J_{C_1\times C_2}:=I\bigl(j_1(\operatorname{Van}(C_1))\cup j_2(\operatorname{Van}(C_2))\bigr)\subseteq\operatorname{FVL}[\mathbb{R}^{n+m}] J C 1 × C 2 := I ( j 1 ( Van ( C 1 )) ∪ j 2 ( Van ( C 2 )) ) ⊆ FVL [ R n + m ]
where
j 1 ( f ) = f ∘ π 1 , j 2 ( g ) = g ∘ π 2 j_1(f)=f\circ\pi_1,\qquad j_2(g)=g\circ\pi_2 j 1 ( f ) = f ∘ π 1 , j 2 ( g ) = g ∘ π 2
and π 1 : R n + m → R n \pi_1:\mathbb{R}^{n+m}\to\mathbb{R}^n π 1 : R n + m → R n and π 2 : R n + m → R m \pi_2:\mathbb{R}^{n+m}\to\mathbb{R}^m π 2 : R n + m → R m denote the coordinate projections. In particular, J C 1 × C 2 ⊆ Van ( C 1 × C 2 ) J_{C_1\times C_2}\subseteq\operatorname{Van}(C_1\times C_2) J C 1 × C 2 ⊆ Van ( C 1 × C 2 ) .
By Theorem 4.2 and Proposition 2.4 , for finitely generated Archimedean vector lattices
X 1 ≃ L ( C 1 ) and X 2 ≃ L ( C 2 ) , X_1\simeq\mathcal{L}(C_1)\qquad\text{and}\qquad X_2\simeq\mathcal{L}(C_2), X 1 ≃ L ( C 1 ) and X 2 ≃ L ( C 2 ) ,
we have an isomorphism
X 1 ∗ V L X 2 ≃ X 1 ∗ A V L X 2 X_1*_{\mathrm{VL}}X_2\simeq X_1*_{\mathrm{AVL}}X_2 X 1 ∗ VL X 2 ≃ X 1 ∗ AVL X 2
induced by the identity maps when Archimedeanization identifies no distinct vectors, that is, when the inclusion
J C 1 × C 2 ⊆ Van ( C 1 × C 2 ) J_{C_1\times C_2}\subseteq\operatorname{Van}(C_1\times C_2) J C 1 × C 2 ⊆ Van ( C 1 × C 2 )
of ideals in FVL [ R n + m ] \operatorname{FVL}[\mathbb{R}^{n+m}] FVL [ R n + m ] is an equality. We will show that this need not hold in general.
Lemma 4.3. ∀ vanishingIdeal_le_productRelationsIdeal_iff Let C ⊆ R n C\subseteq\mathbb{R}^n C ⊆ R n and D ⊆ R m D\subseteq\mathbb{R}^m D ⊆ R m be closed cones. Then
Van ( C × D ) ⊆ J C × D \operatorname{Van}(C\times D)\subseteq J_{C\times D} Van ( C × D ) ⊆ J C × D if and only if every open polyhedral cone W ⊆ R n + m W\subseteq\mathbb{R}^{n+m} W ⊆ R n + m satisfying W ∩ ( C × D ) = ∅ W\cap(C\times D)=\varnothing W ∩ ( C × D ) = ∅ admits open polyhedral cones U ⊆ R n U\subseteq\mathbb{R}^n U ⊆ R n and V ⊆ R m V\subseteq\mathbb{R}^m V ⊆ R m such that U ∩ C = ∅ U\cap C=\varnothing U ∩ C = ∅ , V ∩ D = ∅ V\cap D=\varnothing V ∩ D = ∅ and
W ⊆ ( U × R m ) ∪ ( R n × V ) . W\subseteq(U\times\mathbb{R}^m)\cup(\mathbb{R}^n\times V). W ⊆ ( U × R m ) ∪ ( R n × V ) .
Proof. Let h ∈ Van ( C × D ) + h\in\operatorname{Van}(C\times D)_+ h ∈ Van ( C × D ) + . We have h ∈ J C × D h\in J_{C\times D} h ∈ J C × D if and only if there are f ∈ Van ( C ) + f\in\operatorname{Van}(C)_+ f ∈ Van ( C ) + , g ∈ Van ( D ) + g\in\operatorname{Van}(D)_+ g ∈ Van ( D ) + such that
h ≤ f π 1 + g π 2 , h\leq f\pi_1+g\pi_2, h ≤ f π 1 + g π 2 ,
or equivalently, by [3 Theorem 3.4] , if and only if there exist f , g f,g f , g as above such that
S ( h ) ⊆ S ( f π 1 ) ∪ S ( g π 2 ) = ( S ( f ) × R m ) ∪ ( R n × S ( g ) ) . S(h)\subseteq S(f\pi_1)\cup S(g\pi_2)=(S(f)\times\mathbb{R}^m)\cup(\mathbb{R}^n\times S(g)). S ( h ) ⊆ S ( f π 1 ) ∪ S ( g π 2 ) = ( S ( f ) × R m ) ∪ ( R n × S ( g )) .
Applying the condition to the open polyhedral cone W = S ( h ) W=S(h) W = S ( h ) and [3 Lemma 3.2] , which identifies open polyhedral cones with open supports of elements of the corresponding free vector lattices, provide such f , g f,g f , g . This proves one implication.
For the converse, assume that Van ( C × D ) ⊆ J C × D \operatorname{Van}(C\times D)\subseteq J_{C\times D} Van ( C × D ) ⊆ J C × D , and let W ⊆ R n + m W\subseteq\mathbb{R}^{n+m} W ⊆ R n + m be an open polyhedral cone disjoint from C × D C\times D C × D . By [3 Lemma 3.2] , there is h ∈ FVL [ R n + m ] + h\in\operatorname{FVL}[\mathbb{R}^{n+m}]_+ h ∈ FVL [ R n + m ] + with S ( h ) = W S(h)=W S ( h ) = W . Thus h ∈ Van ( C × D ) ⊆ J C × D h\in\operatorname{Van}(C\times D)\subseteq J_{C\times D} h ∈ Van ( C × D ) ⊆ J C × D . By the definition of the generated ideal, there are f ∈ Van ( C ) + f\in\operatorname{Van}(C)_+ f ∈ Van ( C ) + and g ∈ Van ( D ) + g\in\operatorname{Van}(D)_+ g ∈ Van ( D ) + such that
h ≤ f π 1 + g π 2 . h\leq f\pi_1+g\pi_2. h ≤ f π 1 + g π 2 .
Set U = S ( f ) U=S(f) U = S ( f ) and V = S ( g ) V=S(g) V = S ( g ) . These are open polyhedral cones satisfying U ∩ C = ∅ U\cap C=\varnothing U ∩ C = ∅ and V ∩ D = ∅ V\cap D=\varnothing V ∩ D = ∅ . Since f , g ≥ 0 f,g\ge0 f , g ≥ 0 ,
W = S ( h ) ⊂ S ( f π 1 + g π 2 ) = ( U × R m ) ∪ ( R n × V ) . W=S(h)\subset S(f\pi_1+g\pi_2)=(U\times\mathbb{R}^m)\cup(\mathbb{R}^n\times V). W = S ( h ) ⊂ S ( f π 1 + g π 2 ) = ( U × R m ) ∪ ( R n × V ) .
Hence U , V U,V U , V satisfy the required covering condition.
Lemma 4.4. ∀ \forall ∀ hasPolyhedralCover_of_polyhedral If C ⊆ R n C\subseteq\mathbb{R}^n C ⊆ R n is a closed polyhedral cone and D ⊆ R m D\subseteq\mathbb{R}^m D ⊆ R m is any closed cone, then the covering condition of Lemma 4.3 holds for the pair ( C , D ) (C,D) ( C , D ) .
Proof. Let W ⊆ R n + m W\subseteq\mathbb{R}^{n+m} W ⊆ R n + m be open polyhedral and disjoint from C × D C\times D C × D . Define
U = R n ∖ C and V = π 2 ( W ∩ ( C × R m ) ) . U=\mathbb{R}^n\setminus C\qquad\text{and}\qquad V=\pi_2\bigl(W\cap(C\times\mathbb{R}^m)\bigr). U = R n ∖ C and V = π 2 ( W ∩ ( C × R m ) ) . The cone U U U is open polyhedral. The set V V V is polyhedral by elimination of the x x x variables in finite systems of homogeneous linear inequalities. It is open because V = ⋃ x ∈ C { y : ( x , y ) ∈ W } V=\bigcup_{x\in C}\{y:(x,y)\in W\} V = ⋃ x ∈ C { y : ( x , y ) ∈ W } is a union of open sets. Thus V V V is also an open polyhedral cone. The disjointness of W W W from C × D C\times D C × D gives V ∩ D = ∅ V\cap D=\varnothing V ∩ D = ∅ . Finally, if ( x , y ) ∈ W (x,y)\in W ( x , y ) ∈ W , then either x ∉ C x\notin C x ∈ / C , in which case x ∈ U x\in U x ∈ U , or x ∈ C x\in C x ∈ C , in which case y ∈ V y\in V y ∈ V . Therefore
W ⊆ ( U × R m ) ∪ ( R n × V ) . W\subseteq(U\times\mathbb{R}^m)\cup(\mathbb{R}^n\times V). W ⊆ ( U × R m ) ∪ ( R n × V ) .
For any non-polyhedral cone C C C , we can construct a closed cone D C D_C D C such that the covering condition fails, and hence the corresponding vector lattice free product is not Archimedean.
Lemma 4.5. ∀ \forall ∀ exists_partner_strict_productRelationsIdeal Let C ⊆ R n C\subseteq\mathbb{R}^n C ⊆ R n be a closed nonpolyhedral cone. There is a closed cone D C ⊆ R n + 1 D_C\subseteq\mathbb{R}^{n+1} D C ⊆ R n + 1 such that
J C × D C ⊊ Van ( C × D C ) . J_{C\times D_C}\subsetneq\operatorname{Van}(C\times D_C). J C × D C ⊊ Van ( C × D C ) . Proof. Define
Ω C = { ( y , t ) ∈ R n + 1 : t > 0 , inf x ∈ C ∥ x − y ∥ ∞ < t } and D C = R n + 1 ∖ Ω C . \Omega_C=\left\{(y,t)\in\mathbb{R}^{n+1}:t>0,\ \inf_{x\in C}\lVert x-y\rVert_\infty<t\right\}
\qquad\text{and}\qquad
D_C=\mathbb{R}^{n+1}\setminus\Omega_C. Ω C = { ( y , t ) ∈ R n + 1 : t > 0 , x ∈ C inf ∥ x − y ∥ ∞ < t } and D C = R n + 1 ∖ Ω C . Since C C C is a closed cone, the set Ω C \Omega_C Ω C is an open cone and D C D_C D C is a closed cone. The cone Ω C \Omega_C Ω C is not open polyhedral. Indeed, if it were open polyhedral, its closure would be a closed polyhedral cone: decompose it into finitely many open convex polyhedral cones and take the finite union of their closures. Also, linear sections of polyhedral cones are polyhedral. But
Ω C ‾ ∩ ( R n × { 0 } ) = C × { 0 } \overline{\Omega_C}\cap(\mathbb{R}^n\times\{0\})=C\times\{0\} Ω C ∩ ( R n × { 0 }) = C × { 0 } is not polyhedral, a contradiction. Now let
h ( x , y , t ) = ( t − ∥ x − y ∥ ∞ ) + . h(x,y,t)=(t-\lVert x-y\rVert_\infty)^+. h ( x , y , t ) = ( t − ∥ x − y ∥ ∞ ) + . Its open support is the open convex polyhedral cone
W = { ( x , y , t ) : ∥ x − y ∥ ∞ < t } . W=\{(x,y,t):\lVert x-y\rVert_\infty<t\}. W = {( x , y , t ) : ∥ x − y ∥ ∞ < t } .
By the definition of D C D_C D C , W ∩ ( C × D C ) = ∅ W \cap(C \times D_C) = \varnothing W ∩ ( C × D C ) = ∅ . Suppose the cover condition held for the pair ( C , D C ) (C,D_C) ( C , D C ) . Let U , V U,V U , V be a corresponding pair of covering open polyhedral cones. For each z = ( y , t ) ∈ Ω C z=(y,t) \in\Omega_C z = ( y , t ) ∈ Ω C , choose x ∈ C x \in C x ∈ C satisfying ( x , z ) ∈ W (x,z) \in W ( x , z ) ∈ W . Since U ∩ C = ∅ U \cap C = \varnothing U ∩ C = ∅ , the cover forces z ∈ V z \in V z ∈ V . Hence
Ω C ⊆ V . \Omega_C \subseteq V. Ω C ⊆ V .
On the other hand, V ∩ D C = ∅ V \cap D_C = \varnothing V ∩ D C = ∅ gives V ⊆ Ω C V \subseteq\Omega_C V ⊆ Ω C . Therefore V = Ω C V = \Omega_C V = Ω C , contradicting that Ω C \Omega_C Ω C is not open polyhedral. By Lemma 4.3 ,
J C × D C ⊊ Van ( C × D C ) . J_{C \times D_C} \subsetneq\operatorname{Van}(C \times D_C). J C × D C ⊊ Van ( C × D C ) .
□ \square □
A vector lattice is projective if every lattice homomorphism from it into a quotient vector lattice lifts through the quotient map. All references to projectivity below concern the category V L \mathcal{V}\mathcal{L} V L .
Proposition 4.6. Let X X X be a finitely generated Archimedean vector lattice. The following conditions are equivalent:
For every finitely generated Archimedean vector lattice Y Y Y , the ordinary vector-lattice free product X ∗ V L Y X *_{\mathcal{V}\mathcal{L}} Y X ∗ V L Y is Archimedean;
X X X is isomorphic to a vector sublattice of a free vector lattice;
X X X is projective;
X X X is finitely presented, that is, X ≃ FVL [ R n ] / I X \simeq\operatorname{FVL}[\mathbb{R}^n]/I X ≃ FVL [ R n ] / I for some n n n and some finitely generated ideal I I I ;
In any representation X ≃ L ( C ) X \simeq\mathcal{L}(C) X ≃ L ( C ) , the closed cone C C C is polyhedral.
Proof. The equivalence between (1) and (5) follows from Lemma 4.5 and Lemma 4.4 . Also, the equivalence between (3), (4), and (5) is a consequence of [3 Theorem 5.1 and its proof] . For the implication (3) ⟹ \Longrightarrow ⟹ (2), projectivity splits the canonical quotient map from a finitely generated free vector lattice onto X X X . For (2) ⟹ \Longrightarrow ⟹ (5), finite generation of X X X ensures that the image of an embedding into a free vector lattice lies in a free sublattice on finitely many generators. We may therefore suppose that
T : X ≃ L ( C ) → FVL [ R k ] T : X \simeq\mathcal{L}(C) \to\operatorname{FVL}[\mathbb{R}^k] T : X ≃ L ( C ) → FVL [ R k ]
is injective. By duality, T T T is induced by an ℓ \ell ℓ -map φ : R k → C \varphi: \mathbb{R}^k \to C φ : R k → C . Its image is a closed polyhedral cone: on each of finitely many closed convex polyhedral cones covering R k \mathbb{R}^k R k , the map is linear. If φ ( R k ) ≠ C \varphi(\mathbb{R}^k) \ne C φ ( R k ) = C , we could choose f ∈ FVL [ R n ] f \in\operatorname{FVL}[\mathbb{R}^n] f ∈ FVL [ R n ] with S ( f ) = R n ∖ φ ( R k ) S(f) = \mathbb{R}^n \setminus\varphi(\mathbb{R}^k) S ( f ) = R n ∖ φ ( R k ) . Then f ∣ C ≠ 0 f|_C \ne0 f ∣ C = 0 , whereas
T ( f ∣ C ) = f φ = 0 , T(f|_C) = f\varphi= 0, T ( f ∣ C ) = f φ = 0 ,
a contradiction. Hence C = φ ( R k ) C = \varphi(\mathbb{R}^k) C = φ ( R k ) is polyhedral. □ \square □
Corollary 4.7. Let X X X be an Archimedean vector lattice. Then X ∗ V L Y X *_{\mathcal{V}\mathcal{L}} Y X ∗ V L Y is Archimedean for every Archimedean Y Y Y if and only if every finitely generated sublattice of X X X satisfies one of the equivalent conditions from Proposition 4.6 .
Proof. Suppose first that X ∗ V L Y X \ast_{\mathrm{VL}} Y X ∗ VL Y is Archimedean for every Archimedean Y Y Y . If A ⊆ X A \subseteq X A ⊆ X is finitely generated and B B B is finitely generated and Archimedean, then Corollary 2.11 embeds A ∗ V L B A \ast_{\mathrm{VL}} B A ∗ VL B into X ∗ V L B X \ast_{\mathrm{VL}} B X ∗ VL B . Hence the former is Archimedean, and Proposition 4.6 applies to A A A .
Conversely, suppose every finitely generated sublattice of X X X is projective, and let Y Y Y be Archimedean. The free product X ∗ V L Y X \ast_{\mathrm{VL}} Y X ∗ VL Y is the directed union of the embedded products A ∗ V L B A \ast_{\mathrm{VL}} B A ∗ VL B , where A ⊆ X A \subseteq X A ⊆ X and B ⊆ Y B \subseteq Y B ⊆ Y are finitely generated. Each such product is Archimedean by Proposition 4.6 . If 0 ≤ n z ≤ w 0 \leq nz \leq w 0 ≤ n z ≤ w for all n n n , a single such product contains z z z , w w w and reflects these inequalities. Thus z = 0 z = 0 z = 0 , proving that the full free product is Archimedean. □
Example 4.8. Not every Archimedean vector lattice such that every finitely generated sublattice is projective embeds in a free vector lattice. Let
X = { f : N → R : f ( N ) is finite } , X = \{f : \mathbb{N} \to\mathbb{R} : f(\mathbb{N})\text{ is finite}\}, X = { f : N → R : f ( N ) is finite } , with pointwise vector-lattice operations. Then:
∀ \forall ∀ finiteRangeSequences_locally_projective Every finitely generated vector sublattice of X X X embeds into a finitely generated free vector lattice;
∀ \forall ∀ finiteRangeSequences_not_free_embeddable X X X does not embed into FVL [ E ] \operatorname{FVL}[E] FVL [ E ] for any vector space E E E .
Indeed, let f 1 , … , f d ∈ X f_1,\ldots,f_d \in X f 1 , … , f d ∈ X . The map
n ↦ ( f 1 ( n ) , … , f d ( n ) ) n \mapsto(f_1(n),\ldots,f_d(n)) n ↦ ( f 1 ( n ) , … , f d ( n )) has finite range. Its fibers give a finite partition N = P 1 ⊔ ⋯ ⊔ P r \mathbb{N} = P_1 \sqcup\cdots\sqcup P_r N = P 1 ⊔ ⋯ ⊔ P r , and the vector sublattice generated by f 1 , … , f d f_1,\ldots,f_d f 1 , … , f d is contained in the lattice of functions constant on every P i P_i P i . The latter lattice is isomorphic to R r \mathbb{R}^r R r , which is projective and therefore embeds into a finitely generated free vector lattice. The same embedding restricts to the sublattice generated by f 1 , … , f d f_1,\ldots,f_d f 1 , … , f d , proving the first assertion.
Now recall that a component of an element e ≥ 0 e \geq0 e ≥ 0 is an element w w w satisfying
0 ≤ w ≤ e and w ∧ ( e − w ) = 0. 0 \leq w \leq e \quad\text{and}\quad w \wedge(e-w) = 0. 0 ≤ w ≤ e and w ∧ ( e − w ) = 0. For any set S S S , every positive element of FVL [ R S ] \operatorname{FVL}[\mathbb{R}^{S}] FVL [ R S ] has only finitely many components. Indeed, let e ∈ FVL [ R S ] + e \in\operatorname{FVL}[\mathbb{R}^{S}]_{+} e ∈ FVL [ R S ] + . It depends on a finite set T ⊆ S T \subseteq S T ⊆ S , so it may be regarded as an element of FVL [ R T ] \operatorname{FVL}[\mathbb{R}^{T}] FVL [ R T ] . In the standard representation of FVL \operatorname{FVL} FVL as piecewise-linear functions, the open polyhedral cone
P = { x ∈ R T : e ( x ) > 0 } P = \{x \in\mathbb{R}^{T} : e(x) > 0\} P = { x ∈ R T : e ( x ) > 0 } has only finitely many connected components. If w w w is a component of e e e , choose a finite set U ⊇ T U \supseteq T U ⊇ T such that w ∈ FVL [ R U ] w \in\operatorname{FVL}[\mathbb{R}^{U}] w ∈ FVL [ R U ] . On P × R U ∖ T P \times\mathbb{R}^{U\setminus T} P × R U ∖ T , the continuous function w / e w/e w / e takes values only in { 0 , 1 } \{0,1\} { 0 , 1 } . It is therefore constant on each connected component. Outside this set, w = 0 w = 0 w = 0 because 0 ≤ w ≤ e 0 \leq w \leq e 0 ≤ w ≤ e . Thus a component is determined by a subset of the finitely many connected components of P P P , and only finitely many components are possible.
On the other hand, the constant function 1 ∈ X 1 \in X 1 ∈ X has the distinct components
1 M ( n ) = { 1 , n ∈ M , 0 , n ∉ M , ( M ⊆ N ) . 1_M(n) =
\begin{cases}
1, & n \in M,\\
0, & n \notin M,
\end{cases}
\qquad(M \subseteq\mathbb{N}). 1 M ( n ) = { 1 , 0 , n ∈ M , n ∈ / M , ( M ⊆ N ) .
An embedding preserves components and their distinctness. Hence no embedding from X X X into any FVL [ R S ] \operatorname{FVL}[\mathbb{R}^{S}] FVL [ R S ] can exist.
Example 4.9. ∀ freeBanach_pair_not_archimedean The V L \mathcal{V}\mathcal{L} V L free product of free Banach lattices need not be Archimedean. ∀ freeBanach_comparison_not_injective Thus, for Banach spaces E E E and F F F , the map I \mathcal{I} I in the diagram
FVL [ E ⊕ F ] ≃ FVL [ E ] ∗ V L FVL [ F ] ⟶ FBL [ E ] ∗ B L FBL [ F ] ≃ FBL [ E ⊕ 1 F ] ≃ FVL [ E ] ∗ V L FVL [ F ] ↓ ↗ I FBL [ E ] ∗ V L FBL [ F ] \begin{aligned}
\operatorname{FVL}[E\oplus F] &\simeq\operatorname{FVL}[E] *_{\mathcal{V}\mathcal{L}} \operatorname{FVL}[F] \longrightarrow\operatorname{FBL}[E] *_{\mathcal{B}\mathcal{L}} \operatorname{FBL}[F] \simeq\operatorname{FBL}[E\oplus_{1} F] \\
&\phantom{\simeq\operatorname{FVL}[E] *_{\mathcal{V}\mathcal{L}} \operatorname{FVL}[F]}\downarrow\hspace{7em}\nearrow\mathcal{I}\\
\operatorname{FBL}[E] *_{\mathcal{V}\mathcal{L}} \operatorname{FBL}[F] &
\end{aligned} FVL [ E ⊕ F ] FBL [ E ] ∗ V L FBL [ F ] ≃ FVL [ E ] ∗ V L FVL [ F ] ⟶ FBL [ E ] ∗ BL FBL [ F ] ≃ FBL [ E ⊕ 1 F ] ≃ FVL [ E ] ∗ V L FVL [ F ] ↓ ↗ I is not injective in general.
Indeed, take the closed cone
C = { ( x , y , z ) ∈ R 3 : z = x 2 + y 2 } . C=\{(x,y,z)\in\mathbb{R}^{3}:z=\sqrt{x^{2}+y^{2}}\}. C = {( x , y , z ) ∈ R 3 : z = x 2 + y 2 } .
It is not polyhedral: its intersection with the plane x 3 = 1 x_{3}=1 x 3 = 1 is a circle, and a convex subset of that circle contains at most one point. As in the proof of Lemma 4.5 , consider the cones
Ω C and D C = R 4 ∖ Ω C . \Omega_{C}\qquad\text{and}\qquad D_{C}=\mathbb{R}^{4}\setminus\Omega_{C}. Ω C and D C = R 4 ∖ Ω C .
It follows that
J C × D C ⊊ Van ( C × D C ) . J_{C\times D_{C}}\subsetneq\operatorname{Van}(C\times D_{C}). J C × D C ⊊ Van ( C × D C ) .
To embed the two factors into free Banach lattices, choose continuous surjections
q C : S 1 → C ∩ S 2 , and q D C : S 3 → D C ∩ S 3 . q_{C}:S^{1}\to C\cap S^{2},\qquad\text{and}\qquad q_{D_{C}}:S^{3}\to D_{C}\cap S^{3}. q C : S 1 → C ∩ S 2 , and q D C : S 3 → D C ∩ S 3 .
For the first map, take q C ( s , t ) = 2 − 1 / 2 ( s , t , 1 ) q_{C}(s,t)=2^{-1/2}(s,t,1) q C ( s , t ) = 2 − 1/2 ( s , t , 1 ) . Observe that D C ∩ S 3 D_{C}\cap S^{3} D C ∩ S 3 is homeomorphic to the closed Euclidean ball B 3 B^{3} B 3 : writing a ( u ) = dist ∞ ( u , C ) a(u)=\operatorname{dist}_{\infty}(u,C) a ( u ) = dist ∞ ( u , C ) for u ∈ S 2 u\in S^{2} u ∈ S 2 , a homeomorphism is
Φ ( ρ u ) = ( u , a ( u ) + 1 − ρ − 1 ) 1 + ( a ( u ) + 1 − ρ − 1 ) 2 ( 0 < ρ ≤ 1 ) , Φ ( 0 ) = ( 0 , − 1 ) . \Phi(\rho u)=\frac{(u,a(u)+1-\rho^{-1})}{\sqrt{1+(a(u)+1-\rho^{-1})^{2}}}\quad(0<\rho\leq1),\qquad\Phi(0)=(0,-1). Φ ( ρ u ) = 1 + ( a ( u ) + 1 − ρ − 1 ) 2 ( u , a ( u ) + 1 − ρ − 1 ) ( 0 < ρ ≤ 1 ) , Φ ( 0 ) = ( 0 , − 1 ) .
Every point of D C ∩ S 3 D_{C}\cap S^{3} D C ∩ S 3 other than ( 0 , − 1 ) (0,-1) ( 0 , − 1 ) has the unique form ( u , s ) / 1 + s 2 (u,s)/\sqrt{1+s^{2}} ( u , s ) / 1 + s 2 with s ≤ a ( u ) s\leq a(u) s ≤ a ( u ) , corresponding to ρ = ( a ( u ) + 1 − s ) − 1 \rho=(a(u)+1-s)^{-1} ρ = ( a ( u ) + 1 − s ) − 1 . Boundedness of a a a gives continuity at ρ = 0 \rho=0 ρ = 0 , so the continuous bijection is a homeomorphism. The projection P : S 3 → B 3 P:S^{3}\to B^{3} P : S 3 → B 3 onto the first three coordinates is surjective; hence q D C = Φ P : S 3 → D C ∩ S 3 q_{D_{C}}=\Phi P:S^{3}\to D_{C}\cap S^{3} q D C = Φ P : S 3 → D C ∩ S 3 is a continuous surjection.
Pullback along q C q_{C} q C and q D C q_{D_{C}} q D C gives embeddings
L ( C ) ↪ C ( S 1 ) ≃ FBL [ ℓ 2 2 ] , L ( D C ) ↪ C ( S 3 ) ≃ FBL [ ℓ 2 4 ] . \mathcal{L}(C)\hookrightarrow C(S^{1})\simeq\operatorname{FBL}[\ell_{2}^{2}],\qquad\mathcal{L}(D_{C})\hookrightarrow C(S^{3})\simeq\operatorname{FBL}[\ell_{2}^{4}]. L ( C ) ↪ C ( S 1 ) ≃ FBL [ ℓ 2 2 ] , L ( D C ) ↪ C ( S 3 ) ≃ FBL [ ℓ 2 4 ] .
By Corollary 2.11, these induce an embedding
L ( C ) ∗ V L L ( D C ) ↪ FBL [ ℓ 2 2 ] ∗ V L FBL [ ℓ 2 4 ] . \mathcal{L}(C)*_{\mathcal{V}\mathcal{L}}\mathcal{L}(D_{C})\hookrightarrow\operatorname{FBL}[\ell_{2}^{2}]*_{\mathcal{V}\mathcal{L}}\operatorname{FBL}[\ell_{2}^{4}]. L ( C ) ∗ V L L ( D C ) ↪ FBL [ ℓ 2 2 ] ∗ V L FBL [ ℓ 2 4 ] .
Therefore the latter cannot be Archimedean.
Proposition 4.10. ∀ \forall ∀ exists_join_generated_equiv Let X 1 X_1 X 1 , X 2 X_2 X 2 be nonzero Archimedean vector lattices with specified strong units u 1 , u 2 u_1,u_2 u 1 , u 2 . Identify their order-unit completions with C ( K 1 ) C(K_1) C ( K 1 ) and C ( K 2 ) C(K_2) C ( K 2 ) , respectively, so that u i u_i u i corresponds to 1 K i 1_{K_i} 1 K i . Then
X 1 ∗ A V L X 2 X_1 *_{\mathrm{AVL}} X_2 X 1 ∗ AVL X 2 is lattice isomorphic to the sublattice
lat ( φ 1 ( X 1 ) , φ 2 ( X 2 ) ) ⊂ C ( K 1 ∗ K 2 ) . \operatorname{lat}(\varphi_1(X_1),\varphi_2(X_2)) \subset C(K_1 * K_2). lat ( φ 1 ( X 1 ) , φ 2 ( X 2 )) ⊂ C ( K 1 ∗ K 2 ) .
Proof. The proof is analogous to [18 Theorem 6.2] . □
Although A V L \mathcal{AVL} A V L lacks the amalgamation property, its free products still preserve embeddings. The join representation gives a proof for finitely generated sublattices, and the general case follows by assembling these finite pieces.
Proposition 4.11. ∀ \forall ∀ induced_injective Let Y i ⊂ X i Y_i \subset X_i Y i ⊂ X i be Archimedean vector lattices, for i = 1 , 2 i=1,2 i = 1 , 2 . Then the induced lattice homomorphism
Y 1 ∗ A V L Y 2 ⟶ X 1 ∗ A V L X 2 Y_1 *_{\mathrm{AVL}} Y_2 \longrightarrow X_1 *_{\mathrm{AVL}} X_2 Y 1 ∗ AVL Y 2 ⟶ X 1 ∗ AVL X 2 is injective.
Proof. First suppose that all four lattices are finitely generated. Choose strong units in Y i Y_i Y i and X i X_i X i . The inclusions are bounded for the associated order-unit norms and extend to lattice homomorphisms J i : C ( K i ) → C ( L i ) J_i:C(K_i)\to C(L_i) J i : C ( K i ) → C ( L i ) between their completions. These extensions are injective. Indeed, Y i Y_i Y i is order dense in C ( K i ) C(K_i) C ( K i ) , since it is uniformly dense and contains the unit. A nonzero kernel of J i J_i J i would therefore contain a nonzero positive element of Y i Y_i Y i , contradicting injectivity of the original inclusion. Theorem 3.3 makes the induced map on the Banach lattice free products injective. Restricting it to the sublattices in Proposition 4.10 proves the assertion in this case.
For arbitrary X 1 , X 2 X_1,X_2 X 1 , X 2 , consider the directed system A 1 ∗ A V L A 2 A_1 *_{\mathrm{AVL}} A_2 A 1 ∗ AVL A 2 , where A i A_i A i ranges over the finitely generated sublattices of X i X_i X i . Whenever A i ⊆ B i A_i \subseteq B_i A i ⊆ B i for i = 1 , 2 i=1,2 i = 1 , 2 , the transition map A 1 ∗ A V L A 2 → B 1 ∗ A V L B 2 A_1 *_{\mathrm{AVL}} A_2 \to B_1 *_{\mathrm{AVL}} B_2 A 1 ∗ AVL A 2 → B 1 ∗ AVL B 2 is the lattice homomorphism induced by the inclusions of the factors. These maps are compatible by uniqueness in the universal property. The finite case shows that all transition maps are embeddings. Identify these lattices along the transition maps and take their directed union V V V . This union is Archimedean: if 0 ≤ n z ≤ w 0 \le nz \le w 0 ≤ n z ≤ w for every n n n , a single member of the directed system contains z , w z,w z , w and these inequalities hold there so z = 0 z=0 z = 0 .
The canonical maps X i → V X_i \to V X i → V are compatible with the maps at each stage. For an Archimedean target Z Z Z and lattice homomorphisms T i : X i → Z T_i:X_i\to Z T i : X i → Z , the induced maps on the directed system agree on overlaps, by uniqueness in their universal properties. They therefore give a unique lattice homomorphism V → Z V\to Z V → Z extending both T i T_i T i . Thus V V V is X 1 ∗ A V L X 2 X_1 *_{\mathrm{AVL}} X_2 X 1 ∗ AVL X 2 .
Apply the same description to Y 1 , Y 2 Y_1,Y_2 Y 1 , Y 2 . Each finite generated sublattice of their free product embeds into a finite generated sublattice for X 1 , X 2 X_1,X_2 X 1 , X 2 , and hence into V V V . Since every element belongs to a finite generated sublattice, the induced map on the full free products is injective. □
Consequences for the free product of Banach lattices
Maximal norm and completion
We now identify the algebraic sublattice generated by the factors inside the Banach lattice free product. It is their Archimedean free product, and the Banach lattice free product is recovered by completing it under the maximal compatible lattice norm.
Theorem 5.1. ∀ \forall ∀ archimedean_completion Given a family ( X i ) i ∈ I (X_i)_{i\in I} ( X i ) i ∈ I of Banach lattices, the lattice homomorphism
I : = ∗ id X i ‾ : ∗ A V L X i → ∗ B L X i \mathcal{I}:=\overline{\ast\operatorname{id}_{X_i}}:\ast_{\mathrm{AVL}} X_i\to\ast_{\mathrm{BL}}X_i I := ∗ id X i : ∗ AVL X i → ∗ BL X i is injective. Let M \mathcal{M} M be the set of lattice seminorms p p p on ∗ A V L X i \ast_{\mathrm{AVL}}X_i ∗ AVL X i such that
p ( x i ) ≤ ∥ x i ∥ X i p(x_i)\leq\lVert x_i\rVert_{X_i} p ( x i ) ≤ ∥ x i ∥ X i for all x i ∈ X i x_i\in X_i x i ∈ X i and i ∈ I i\in I i ∈ I . If we endow ∗ A V L X i \ast_{\mathrm{AVL}}X_i ∗ AVL X i with the lattice norm
∥ z ∥ = sup p ∈ M p ( z ) , \lVert z\rVert=\sup_{p\in\mathcal{M}}p(z), ∥ z ∥ = p ∈ M sup p ( z ) , then I \mathcal{I} I is a lattice isometric embedding with dense range.
Proof. We first prove injectivity when I = { 1 , 2 } I=\{1,2\} I = { 1 , 2 } , writing the factors as X , Y X,Y X , Y . Let 0 ≠ z ∈ X ∗ A V L Y 0\ne z\in X\ast_{\mathrm{AVL}}Y 0 = z ∈ X ∗ AVL Y . Write
z = Φ ( φ X ( x 1 ) , … , φ X ( x n ) , φ Y ( y 1 ) , … , φ Y ( y m ) ) z=\Phi\bigl(\varphi_X(x_1),\ldots,\varphi_X(x_n),\varphi_Y(y_1),\ldots,\varphi_Y(y_m)\bigr) z = Φ ( φ X ( x 1 ) , … , φ X ( x n ) , φ Y ( y 1 ) , … , φ Y ( y m ) )
and put
u = ∣ x 1 ∣ ∨ ⋯ ∨ ∣ x n ∣ , v = ∣ y 1 ∣ ∨ ⋯ ∨ ∣ y m ∣ . u=|x_1|\vee\cdots\vee|x_n|,\qquad v=|y_1|\vee\cdots\vee|y_m|. u = ∣ x 1 ∣ ∨ ⋯ ∨ ∣ x n ∣ , v = ∣ y 1 ∣ ∨ ⋯ ∨ ∣ y m ∣.
Let X u X_u X u and Y v Y_v Y v be the principal ideals generated by u u u and v v v , respectively, which become Banach lattices when equipped with the corresponding order unit norms. We have a commutative diagram
X u ∗ A V L Y v → X u ∗ B L Y v ↓ ↓ X ∗ A V L Y → I X ∗ B L Y \begin{CD}
X_u\ast_{\mathrm{AVL}}Y_v @>>> X_u\ast_{\mathrm{BL}}Y_v\\
@VVV @VVV\\
X\ast_{\mathrm{AVL}}Y @>{\mathcal{I}}>> X\ast_{\mathrm{BL}}Y
\end{CD} X u ∗ AVL Y v ↓ ⏐ X ∗ AVL Y I X u ∗ BL Y v ↓ ⏐ X ∗ BL Y
The element
0 ≠ z ~ = Φ ( φ X u ( x 1 ) , … , φ X u ( x n ) , φ Y v ( y 1 ) , … , φ Y v ( y m ) ) 0\ne\widetilde z=\Phi\bigl(\varphi_{X_u}(x_1),\ldots,\varphi_{X_u}(x_n),\varphi_{Y_v}(y_1),\ldots,\varphi_{Y_v}(y_m)\bigr) 0 = z = Φ ( φ X u ( x 1 ) , … , φ X u ( x n ) , φ Y v ( y 1 ) , … , φ Y v ( y m ) )
in X u ∗ A V L Y v X_u\ast_{\mathrm{AVL}}Y_v X u ∗ AVL Y v maps to z z z under the left vertical arrow. The top horizontal arrow is injective by Proposition 4.10 , and the right vertical arrow is injective by Theorem 3.3 , using the order-unit representations of X u X_u X u and Y v Y_v Y v . Commutativity gives I ( z ) ≠ 0 \mathcal{I}(z)\ne0 I ( z ) = 0 .
The finite case follows by induction. Embed the Archimedean free product of the first factors into their Banach lattice free product using the induction hypothesis, apply Proposition 4.11 to adjoin the last factor, and then use the two-factor case. For arbitrary I I I , every element involves finitely many factors. The canonical maps from a finite subproduct into the full product have left inverses obtained by setting the other factors to zero, in both categories. Hence the finite case gives injectivity for arbitrary families.
The norm inherited from ∗ B L X i \ast_{\mathrm{BL}}X_i ∗ BL X i belongs to M \mathcal{M} M . Every z ∈ ∗ A V L X i z\in\ast_{\mathrm{AVL}}X_i z ∈ ∗ AVL X i can be dominated by a finite sum of elements from the free factors, so the supremum is finite and defines a lattice norm on ∗ A V L X i \ast_{AVL} X_i ∗ A V L X i . Let Z 0 Z_0 Z 0 be its norm completion. For every family of contractive Banach lattice homomorphisms T i : X i → Z T_i:X_i\to Z T i : X i → Z , the induced vector lattice homomorphism
∗ T i : ∗ A V L X i → Z \ast T_i:\ast_{AVL}X_i\to Z ∗ T i : ∗ A V L X i → Z
defines an element of M \mathcal{M} M via p ( z ) = ∥ ( ∗ T i ) ( z ) ∥ Z p(z)=\lVert(\ast T_i)(z)\rVert_Z p ( z ) = ∥( ∗ T i ) ( z ) ∥ Z , so ∗ T i \ast T_i ∗ T i is contractive and extends to a unique contractive lattice homomorphism from Z 0 Z_0 Z 0 . This is the universal property of free products of Banach lattices, and [18 Remark 3.2] completes the proof. □
Order density and regularity
A sublattice A A A of X X X is order dense if every 0 < x ∈ X 0<x\in X 0 < x ∈ X dominates some 0 < a ∈ A 0<a\in A 0 < a ∈ A . It is regular if every net decreasing to zero in A A A also decreases to zero in X X X . Every order dense sublattice is regular.
The completion approach in the previous subsection also admits a local description: every element comes from a free product of principal ideals in the two factors. Since these ideals have strong units, this brings the topological join approach into play for elements of the completed free product as well.
Proposition 5.2. ∀ exists_mem_range_principal_freeProduct Let X , Y X,Y X , Y be Banach lattices. For every z ∈ X ∗ B L Y z\in X\ast_{BL}Y z ∈ X ∗ B L Y , there are elements u ∈ X + u\in X_+ u ∈ X + , v ∈ Y + v\in Y_+ v ∈ Y + such that z z z belongs to the range of the canonical map
X u ∗ B L Y v → X ∗ B L Y . X_u\ast_{BL}Y_v\to X\ast_{BL}Y. X u ∗ B L Y v → X ∗ B L Y . Here the principal ideals X u X_u X u and Y v Y_v Y v carry their order-unit norms.
Proof. We first establish a formula for the free-product norm on X ∗ A V L Y X\ast_{AVL}Y X ∗ A V L Y . For w ∈ X ∗ A V L Y w\in X\ast_{AVL}Y w ∈ X ∗ A V L Y , set
p ( w ) = inf { ∥ a ∥ X + ∥ b ∥ Y : a ∈ X + , b ∈ Y + , ∣ w ∣ ≤ φ X ( a ) + φ Y ( b ) } . p(w)=\inf\{\lVert a\rVert_X+\lVert b\rVert_Y:a\in X_+,\ b\in Y_+,\ |w|\leq\varphi_X(a)+\varphi_Y(b)\}. p ( w ) = inf {∥ a ∥ X + ∥ b ∥ Y : a ∈ X + , b ∈ Y + , ∣ w ∣ ≤ φ X ( a ) + φ Y ( b )} . Every element of the algebraic sublattice generated by the factors is dominated in modulus by a sum of positive elements from the factors, so p ( w ) p(w) p ( w ) is finite. Adding admissible majorants gives subadditivity, and rescaling them gives absolute homogeneity. Moreover, p ( w ) ≤ p ( w ′ ) p(w)\leq p(w') p ( w ) ≤ p ( w ′ ) whenever ∣ w ∣ ≤ ∣ w ′ ∣ |w|\leq|w'| ∣ w ∣ ≤ ∣ w ′ ∣ . Thus p p p is a lattice seminorm, and
p ( φ X ( a ) ) ≤ ∥ a ∥ X , p ( φ Y ( b ) ) ≤ ∥ b ∥ Y ( a ∈ X , b ∈ Y ) . p(\varphi_X(a))\leq\lVert a\rVert_X,\qquad p(\varphi_Y(b))\leq\lVert b\rVert_Y\qquad(a\in X,\ b\in Y). p ( φ X ( a )) ≤ ∥ a ∥ X , p ( φ Y ( b )) ≤ ∥ b ∥ Y ( a ∈ X , b ∈ Y ) . Theorem 5.1 therefore gives p ( w ) ≤ ∥ w ∥ ∗ p(w)\leq\lVert w\rVert_* p ( w ) ≤ ∥ w ∥ ∗ . Conversely, every pair a , b a,b a , b admissible in the definition of p ( w ) p(w) p ( w ) satisfies
∥ w ∥ ∗ ≤ ∥ φ X ( a ) + φ Y ( b ) ∥ ∗ ≤ ∥ a ∥ X + ∥ b ∥ Y . \lVert w\rVert_*\leq\lVert\varphi_X(a)+\varphi_Y(b)\rVert_*\leq\lVert a\rVert_X+\lVert b\rVert_Y. ∥ w ∥ ∗ ≤ ∥ φ X ( a ) + φ Y ( b ) ∥ ∗ ≤ ∥ a ∥ X + ∥ b ∥ Y . Taking the infimum yields ∥ w ∥ ∗ ≤ p ( w ) \lVert w\rVert_*\leq p(w) ∥ w ∥ ∗ ≤ p ( w ) , so p ( w ) = ∥ w ∥ ∗ . p(w)=\lVert w\rVert_*. p ( w ) = ∥ w ∥ ∗ .
Now fix z ∈ X ∗ B L Y z\in X\ast_{BL}Y z ∈ X ∗ B L Y . By Theorem 5.1 , choose z n ∈ X ∗ A V L Y z_n\in X\ast_{AVL}Y z n ∈ X ∗ A V L Y with
∥ z − z n ∥ ∗ < 4 − n ( n ≥ 1 ) . \lVert z-z_n\rVert_*<4^{-n}\qquad(n\geq1). ∥ z − z n ∥ ∗ < 4 − n ( n ≥ 1 ) . Set z 0 = 0 z_0=0 z 0 = 0 and
d n = z n − z n − 1 . d_n=z_n-z_{n-1}. d n = z n − z n − 1 . Then
∑ n = 1 ∞ 2 n ∥ d n ∥ ∗ < ∞ , \sum_{n=1}^{\infty}2^n\lVert d_n\rVert_*<\infty, n = 1 ∑ ∞ 2 n ∥ d n ∥ ∗ < ∞ ,
because for n ≥ 2 n \ge2 n ≥ 2 ,
∥ d n ∥ ∗ ≤ ∥ z − z n ∥ ∗ + ∥ z − z n − 1 ∥ ∗ < 4 − n + 4 − ( n − 1 ) . \lVert d_n\rVert_* \le\lVert z-z_n\rVert_*+\lVert z-z_{n-1}\rVert_*<4^{-n}+4^{-(n-1)}. ∥ d n ∥ ∗ ≤ ∥ z − z n ∥ ∗ + ∥ z − z n − 1 ∥ ∗ < 4 − n + 4 − ( n − 1 ) .
Also
z = ∑ n = 1 ∞ d n z=\sum_{n=1}^{\infty}d_n z = n = 1 ∑ ∞ d n
in X ∗ B L Y X*_{\mathrm{BL}}Y X ∗ BL Y .
Since p ( d n ) = ∥ d n ∥ ∗ p(d_n)=\lVert d_n\rVert_* p ( d n ) = ∥ d n ∥ ∗ , for each n n n we can choose elements u n ∈ X + u_n\in X_+ u n ∈ X + , v n ∈ Y + v_n\in Y_+ v n ∈ Y + such that
∣ d n ∣ ≤ φ X ( u n ) + φ Y ( v n ) |d_n|\le\varphi_X(u_n)+\varphi_Y(v_n) ∣ d n ∣ ≤ φ X ( u n ) + φ Y ( v n )
and
∥ u n ∥ X + ∥ v n ∥ Y ≤ ∥ d n ∥ ∗ + 4 − n . \lVert u_n\rVert_X+\lVert v_n\rVert_Y\le\lVert d_n\rVert_*+4^{-n}. ∥ u n ∥ X + ∥ v n ∥ Y ≤ ∥ d n ∥ ∗ + 4 − n .
Then
∑ n = 1 ∞ 2 n ( ∥ u n ∥ X + ∥ v n ∥ Y ) < ∞ . \sum_{n=1}^{\infty}2^n(\lVert u_n\rVert_X+\lVert v_n\rVert_Y)<\infty. n = 1 ∑ ∞ 2 n (∥ u n ∥ X + ∥ v n ∥ Y ) < ∞.
List the elements of X X X and Y Y Y occurring in chosen lattice-linear expressions for the z n z_n z n as sequences ( x n ) (x_n) ( x n ) and ( y n ) (y_n) ( y n ) , respectively. Define
x = ∑ n = 1 ∞ ∣ x n ∣ 2 n ( 1 + ∥ x n ∥ X ) , y = ∑ n = 1 ∞ ∣ y n ∣ 2 n ( 1 + ∥ y n ∥ Y ) . x=\sum_{n=1}^{\infty}\frac{|x_n|}{2^n(1+\lVert x_n\rVert_X)},\qquad
y=\sum_{n=1}^{\infty}\frac{|y_n|}{2^n(1+\lVert y_n\rVert_Y)}. x = n = 1 ∑ ∞ 2 n ( 1 + ∥ x n ∥ X ) ∣ x n ∣ , y = n = 1 ∑ ∞ 2 n ( 1 + ∥ y n ∥ Y ) ∣ y n ∣ .
Both series converge in norm, since the norm of each summand is at most 2 − n 2^{-n} 2 − n . Moreover,
∣ x n ∣ ≤ 2 n ( 1 + ∥ x n ∥ X ) x , ∣ y n ∣ ≤ 2 n ( 1 + ∥ y n ∥ Y ) y . |x_n|\le2^n(1+\lVert x_n\rVert_X)x,\qquad
|y_n|\le2^n(1+\lVert y_n\rVert_Y)y. ∣ x n ∣ ≤ 2 n ( 1 + ∥ x n ∥ X ) x , ∣ y n ∣ ≤ 2 n ( 1 + ∥ y n ∥ Y ) y .
Let
u : = x ∨ ∑ n = 1 ∞ 2 n u n ∈ X + , v : = y ∨ ∑ n = 1 ∞ 2 n v n . u:=x\vee\sum_{n=1}^{\infty}2^n u_n\in X_+,\qquad
v:=y\vee\sum_{n=1}^{\infty}2^n v_n. u := x ∨ n = 1 ∑ ∞ 2 n u n ∈ X + , v := y ∨ n = 1 ∑ ∞ 2 n v n .
In particular, u n ≤ 2 − n u u_n\le2^{-n}u u n ≤ 2 − n u , v n ≤ 2 − n v v_n\le2^{-n}v v n ≤ 2 − n v , and every x n x_n x n belongs to X u X_u X u , while every y n y_n y n belongs to Y v Y_v Y v .
The principal ideals X u X_u X u and Y v Y_v Y v , with their order-unit norms ∥ a ∥ u = inf { λ ≥ 0 : ∣ a ∣ ≤ λ u } \lVert a\rVert_u=\inf\{\lambda\ge0:|a|\le\lambda u\} ∥ a ∥ u = inf { λ ≥ 0 : ∣ a ∣ ≤ λ u } and ∥ b ∥ v = inf { λ ≥ 0 : ∣ b ∣ ≤ λ v } \lVert b\rVert_v=\inf\{\lambda\ge0:|b|\le\lambda v\} ∥ b ∥ v = inf { λ ≥ 0 : ∣ b ∣ ≤ λ v } , are Banach lattices. Their inclusions into X X X and Y Y Y are bounded, since ∥ a ∥ X ≤ ∥ u ∥ X ∥ a ∥ u \lVert a\rVert_X\le\lVert u\rVert_X\lVert a\rVert_u ∥ a ∥ X ≤ ∥ u ∥ X ∥ a ∥ u and ∥ b ∥ Y ≤ ∥ v ∥ Y ∥ b ∥ v \lVert b\rVert_Y\le\lVert v\rVert_Y\lVert b\rVert_v ∥ b ∥ Y ≤ ∥ v ∥ Y ∥ b ∥ v . Consider the induced bounded lattice homomorphism
j : X u ∗ B L Y v ⟶ X ∗ B L Y . j:X_u*_{\mathrm{BL}}Y_v\longrightarrow X*_{\mathrm{BL}}Y. j : X u ∗ BL Y v ⟶ X ∗ BL Y .
The order-unit representations of X u X_u X u and Y v Y_v Y v as C ( K ) C(K) C ( K ) spaces allow us to apply Theorem 3.3 , so j j j is injective.
Every d n d_n d n has a preimage
d ~ n ∈ X u ∗ A V L Y v ⊂ X u ∗ B L Y v , \widetilde d_n\in X_u*_{\mathrm{AVL}}Y_v\subset X_u*_{\mathrm{BL}}Y_v, d n ∈ X u ∗ AVL Y v ⊂ X u ∗ BL Y v ,
obtained by taking the difference of the chosen expressions for z n z_n z n and z n − 1 z_{n-1} z n − 1 in these principal ideals. Since an injective lattice homomorphism reflects order, the bound on d n d_n d n gives
∣ d ~ n ∣ ≤ φ X u ( u n ) + φ Y v ( v n ) , |\widetilde d_n|\le\varphi_{X_u}(u_n)+\varphi_{Y_v}(v_n), ∣ d n ∣ ≤ φ X u ( u n ) + φ Y v ( v n ) ,
so the free-product norm satisfies
∥ d ~ n ∥ X u ∗ B L Y v ≤ ∥ u n ∥ u + ∥ v n ∥ v ≤ 2 − n + 1 . \lVert\widetilde{d}_n\rVert_{X_u *_{BL} Y_v} \le\lVert u_n\rVert_u+\lVert v_n\rVert_v\le2^{-n+1}. ∥ d n ∥ X u ∗ B L Y v ≤ ∥ u n ∥ u + ∥ v n ∥ v ≤ 2 − n + 1 .
Thus ∑ n d ~ n \sum_n \widetilde{d}_n ∑ n d n converges in X u ∗ B L Y v X_u*_{BL}Y_v X u ∗ B L Y v . Define
z ~ = ∑ n = 1 ∞ d ~ n . \widetilde{z}=\sum_{n=1}^{\infty}\widetilde{d}_n. z = n = 1 ∑ ∞ d n .
By continuity of j j j ,
j z ~ = ∑ n = 1 ∞ d n = z . j\widetilde{z}=\sum_{n=1}^{\infty}d_n=z. j z = n = 1 ∑ ∞ d n = z .
Corollary 5.3. ∀ archimedean_induced_orderDense Let X ′ , Y ′ X',Y' X ′ , Y ′ be order dense sublattices of the Banach lattices X X X and Y Y Y . Then
X ′ ∗ A V L Y ′ ⊂ X ∗ B L Y X'*_{AVL}Y'\subset X*_{BL}Y X ′ ∗ A V L Y ′ ⊂ X ∗ B L Y is an order dense sublattice. ∀ archimedean_induced_regular In particular, it is regular. The same conclusion holds for any finite family of factors and order dense sublattices.
Proof. Let 0 < z ∈ X ∗ B L Y 0<z\in X*_{BL}Y 0 < z ∈ X ∗ B L Y . By Proposition 5.2, choose u u u and v v v such that z z z belongs to the image of j : X u ∗ B L Y v → X ∗ B L Y j:X_u*_{BL}Y_v\to X*_{BL}Y j : X u ∗ B L Y v → X ∗ B L Y . Let z ~ \widetilde{z} z be the preimage of z z z ; then z ~ > 0 \widetilde{z}>0 z > 0 . By [18 Proposition 6.7] , we can choose
w ∈ ( X ′ ∩ X u ) ∗ A V L ( Y ′ ∩ Y v ) ⊂ X u ∗ A V L Y v w\in(X'\cap X_u)*_{AVL}(Y'\cap Y_v)\subset X_u*_{AVL}Y_v w ∈ ( X ′ ∩ X u ) ∗ A V L ( Y ′ ∩ Y v ) ⊂ X u ∗ A V L Y v
such that 0 < w ≤ z ~ 0<w\le\widetilde{z} 0 < w ≤ z . Injectivity of j j j then implies
0 < j w ≤ z , j w ∈ X ′ ∗ A V L Y ′ . 0<jw\le z,\qquad jw\in X'*_{AVL}Y'. 0 < j w ≤ z , j w ∈ X ′ ∗ A V L Y ′ .
For finitely many factors, argue by induction: the Archimedean free product of the first factors is order dense in their Banach lattice free product, so the two-factor assertion applies when the last factor is adjoined. Associativity identifies the resulting sublattice with the full Archimedean free product. □
Example 5.4. ∀ countable_freeProduct_not_orderDense Order density need not hold for infinitely many free factors. ∀ countable_freeProduct_closed_ideal_disjoint In fact, even the weaker property that closed principal ideals contain nonzero elements from the A V L AVL A V L free product can fail. Take X n = F B L [ R ] X_n=FBL[\mathbb{R}] X n = F B L [ R ] for every n ∈ N n\in\mathbb{N} n ∈ N . Then
∗ B L X n ≃ F B L [ ℓ 1 ] *_{BL}X_n\simeq FBL[\ell_1] ∗ B L X n ≃ F B L [ ℓ 1 ] as in [18 Proposition 5.5] . Let
x = ∑ n = 1 ∞ e n 2 n ∈ ℓ 1 and h = ∣ δ x ∣ ∈ F B L [ ℓ 1 ] . x=\sum_{n=1}^{\infty}\frac{e_n}{2^n}\in\ell_1\quad\text{and}\quad h=|\delta_x|\in FBL[\ell_1]. x = n = 1 ∑ ∞ 2 n e n ∈ ℓ 1 and h = ∣ δ x ∣ ∈ F B L [ ℓ 1 ] . The closed ideal generated by h h h , denoted by I h ‾ \overline{I_h} I h , is contained in the ideal of functions vanishing on
Z ( x ) = { y ∗ ∈ ℓ ∞ : y ∗ ( x ) = 0 } . Z(x)=\{y^*\in\ell_\infty:y^*(x)=0\}. Z ( x ) = { y ∗ ∈ ℓ ∞ : y ∗ ( x ) = 0 } .
Now let
f ∈ I h ‾ ∩ V L ∗ X n . f \in\overline{I_h} \cap{}^{*}_{\mathrm{VL}} X_n. f ∈ I h ∩ VL ∗ X n .
There is a finite set S ⊂ N S \subset\mathbb{N} S ⊂ N such that f f f depends only on the coordinates ( y ∗ ( e n ) ) n ∈ S (y^{*}(e_n))_{n\in S} ( y ∗ ( e n ) ) n ∈ S . We show that f = 0 f=0 f = 0 . Fix arbitrary scalar values ( α n ) n ∈ S (\alpha_n)_{n\in S} ( α n ) n ∈ S . Choose m ∉ S m\notin S m ∈ / S and define y ∗ ∈ ℓ ∞ y^{*}\in\ell_\infty y ∗ ∈ ℓ ∞ by
y ∗ ( e n ) = α n ( n ∈ S ) , y ∗ ( e m ) = − 2 m ∑ n ∈ S 2 − n α n , y^{*}(e_n)=\alpha_n \quad(n\in S), \qquad y^{*}(e_m)=-2^m\sum_{n\in S}2^{-n}\alpha_n, y ∗ ( e n ) = α n ( n ∈ S ) , y ∗ ( e m ) = − 2 m n ∈ S ∑ 2 − n α n ,
and y ∗ ( e k ) = 0 y^{*}(e_k)=0 y ∗ ( e k ) = 0 for all other k k k . Then y ∗ y^{*} y ∗ is bounded and
y ∗ ( x ) = ∑ n ∈ S 2 − n α n + 2 − m y ∗ ( e m ) = 0 , y^{*}(x)=\sum_{n\in S}2^{-n}\alpha_n+2^{-m}y^{*}(e_m)=0, y ∗ ( x ) = n ∈ S ∑ 2 − n α n + 2 − m y ∗ ( e m ) = 0 ,
so y ∗ ∈ Z ( x ) y^{*}\in Z(x) y ∗ ∈ Z ( x ) . Since f ∈ I h ‾ f\in\overline{I_h} f ∈ I h , it follows that f ( y ∗ ) = 0 f(y^{*})=0 f ( y ∗ ) = 0 . But f f f depends only on the coordinates in S S S , and the values ( α n ) n ∈ S (\alpha_n)_{n\in S} ( α n ) n ∈ S were arbitrary. Therefore f f f vanishes for every possible value of its finite coordinate variables, so f = 0 f=0 f = 0 .
Corollary 5.5. Let ( X i ) i ∈ I (X_i)_{i\in I} ( X i ) i ∈ I be a family of Banach lattices and i 0 ∈ I i_0\in I i 0 ∈ I . Then
X i 0 ⊂ B L ∗ X i X_{i_0}\subset{}^{*}_{\mathrm{BL}}X_i X i 0 ⊂ BL ∗ X i is a regular sublattice.
Proof. Since
B L ∗ X i ≅ X i 0 B L ∗ B L ∗ X i , {}^{*}_{\mathrm{BL}}X_i \cong X_{i_0}{}^{*}_{\mathrm{BL}}{}^{*}_{\mathrm{BL}}X_i, BL ∗ X i ≅ X i 0 BL ∗ BL ∗ X i , the claim follows from Corollary 5.3 and [18 Proposition 6.10] .
Preservation of injectivity
Proposition 5.6. Let ( T i : X i → Y i ) i ∈ I (T_i:X_i\to Y_i)_{i\in I} ( T i : X i → Y i ) i ∈ I be a uniformly bounded family of injective lattice homomorphisms between Banach lattices. Then
∗ T i : B L ∗ X i → B L ∗ Y i {}^{*}T_i:{}^{*}_{\mathrm{BL}}X_i\to{}^{*}_{\mathrm{BL}}Y_i ∗ T i : BL ∗ X i → BL ∗ Y i is injective.
Proof. For F ⊆ I F\subseteq I F ⊆ I , write
X F = B L ∗ X i , Y F = B L ∗ Y i , X_F={}^{*}_{\mathrm{BL}}X_i,\qquad Y_F={}^{*}_{\mathrm{BL}}Y_i, X F = BL ∗ X i , Y F = BL ∗ Y i , and let T F : X F → Y F T_F:X_F\to Y_F T F : X F → Y F be the homomorphism induced by ( T i ) i ∈ F (T_i)_{i\in F} ( T i ) i ∈ F . The uniform bound on the T i T_i T i ensures that these maps are bounded.
First, suppose that I = { 1 , 2 } I=\{1,2\} I = { 1 , 2 } . By Theorem 5.1 and Proposition 4.11, the restriction of T I T_I T I to X 1 A V L ∗ X 2 X_1{}^{*}_{\mathrm{AVL}}X_2 X 1 AVL ∗ X 2 is injective. This sublattice is order dense in X I X_I X I by Corollary 5.3. If 0 ≠ z ∈ ker T I 0\ne z\in\ker T_I 0 = z ∈ ker T I , choose w ∈ X 1 A V L ∗ X 2 w\in X_1{}^{*}_{\mathrm{AVL}}X_2 w ∈ X 1 AVL ∗ X 2 with 0 < w ≤ ∣ z ∣ 0<w\leq|z| 0 < w ≤ ∣ z ∣ . Then
0 ≤ T I ( w ) ≤ T I ( ∣ z ∣ ) = ∣ T I ( z ) ∣ = 0 , 0\leq T_I(w)\leq T_I(|z|)=|T_I(z)|=0, 0 ≤ T I ( w ) ≤ T I ( ∣ z ∣ ) = ∣ T I ( z ) ∣ = 0 , contradicting injectivity on that sublattice. The general finite case follows by induction using associativity of Banach lattice free products.
Now let I I I be infinite. For every finite F ⊆ I F \subseteq I F ⊆ I , the universal property gives a canonical contractive map j F X : X F → X I j_F^X:X_F \to X_I j F X : X F → X I and a contractive lattice homomorphism P F X : X I → X F P_F^X:X_I \to X_F P F X : X I → X F induced by the canonical maps on the factors indexed by F F F and the zero maps on the remaining factors. Since P F X j F X = id X F P_F^Xj_F^X=\operatorname{id}_{X_F} P F X j F X = id X F , the map j F X j_F^X j F X is an isometric lattice embedding, and we identify X F X_F X F with its image in X I X_I X I . Define P F Y : Y I → Y F P_F^Y:Y_I \to Y_F P F Y : Y I → Y F in the same way.
We claim that if z ∈ X I z \in X_I z ∈ X I satisfies P F X z = 0 P_F^Xz=0 P F X z = 0 for every finite F ⊆ I F \subseteq I F ⊆ I , then z = 0 z=0 z = 0 . By Theorem 5.1 , choose z n ∈ ∗ i ∈ I A V L X i z_n \in\ast_{i\in I}^{\mathrm{AVL}}X_i z n ∈ ∗ i ∈ I AVL X i with
∥ z − z n ∥ ≤ 1 / n \lVert z-z_n\rVert\leq1/n ∥ z − z n ∥ ≤ 1/ n
and let F n F_n F n be a finite set of indices containing all factors occurring in a lattice-linear expression for z n z_n z n . Then z n ∈ X F n z_n \in X_{F_n} z n ∈ X F n and P F n X z n = z n P_{F_n}^Xz_n=z_n P F n X z n = z n . Our assumption gives P F n X z = 0 P_{F_n}^Xz=0 P F n X z = 0 , so
∥ z n ∥ = ∥ P F n X ( z n − z ) ∥ ≤ ∥ z n − z ∥ ≤ 1 / n . \lVert z_n\rVert=\lVert P_{F_n}^X(z_n-z)\rVert\leq\lVert z_n-z\rVert\leq1/n. ∥ z n ∥ = ∥ P F n X ( z n − z )∥ ≤ ∥ z n − z ∥ ≤ 1/ n .
Consequently, ∥ z ∥ ≤ ∥ z − z n ∥ + ∥ z n ∥ ≤ 2 / n \lVert z\rVert\leq\lVert z-z_n\rVert+\lVert z_n\rVert\leq2/n ∥ z ∥ ≤ ∥ z − z n ∥ + ∥ z n ∥ ≤ 2/ n for every n n n , which proves the claim.
For every finite F ⊆ I F\subseteq I F ⊆ I , the diagram
X I → T I Y I P F X ↓ ↓ P F Y X F → T F Y F \begin{CD}
X_I @>T_I>> Y_I \\
@V P_F^X VV @VV P_F^Y V \\
X_F @>T_F>> Y_F
\end{CD} X I P F X ↓ ⏐ X F T I T F Y I ↓ ⏐ P F Y Y F
commutes, since both compositions agree on every free factor. If z ∈ ker T I z\in\ker T_I z ∈ ker T I , it follows that
T F ( P F X z ) = P F Y ( T I z ) = 0. T_F(P_F^Xz)=P_F^Y(T_Iz)=0. T F ( P F X z ) = P F Y ( T I z ) = 0.
The finite case shows that T F T_F T F is injective, so P F X z = 0 P_F^Xz=0 P F X z = 0 for every finite F F F . The claim now gives z = 0 z=0 z = 0 , proving that T I T_I T I is injective. □
Vector lattices with countable spectra
In this section, we apply the amalgamation results to extension problems for vector lattices with countably many prime ideals. These problems are analogous to those concerning injectivity and universal disposition for separable Banach lattices studied in [2 ] , but here size is measured by the cardinality of the prime spectrum. All vector lattices are real, and no Archimedean assumption is imposed unless explicitly stated. For background on prime ideals in vector lattices, see [14 Chapter 5] or [11 ] .
Throughout this section, the spectrum of a vector lattice X X X consists of its proper prime ideals:
Spec p ( X ) = { P ⊊ X : P is an ideal and X / P is totally ordered } . \operatorname{Spec}_{p}(X)=\{P\subsetneq X:P\text{ is an ideal and }X/P\text{ is totally ordered}\}. Spec p ( X ) = { P ⊊ X : P is an ideal and X / P is totally ordered } .
Let C \mathcal{C} C denote the class of vector lattices with countable spectrum:
C = { X : ∣ Spec p ( X ) ∣ ≤ ℵ 0 } . \mathcal{C}=\{X:|\operatorname{Spec}_{p}(X)|\leq\aleph_{0}\}. C = { X : ∣ Spec p ( X ) ∣ ≤ ℵ 0 } .
Thus countable spectrum includes finite spectra, and Spec p ( 0 ) = ∅ \operatorname{Spec}_{p}(0)=\varnothing Spec p ( 0 ) = ∅ .
The size of a vector lattice with countable spectrum
For a subset M M M of a vector lattice X X X , write I X ( M ) I_X(M) I X ( M ) for the ideal generated by M M M .
Lemma 6.1. Let Y ⊂ X Y \subset X Y ⊂ X be a vector sublattice.
(1) ∀ mem_generated_ideal_iff For every ideal J J J of Y Y Y ,
I X ( J ) ∩ Y = J . I_X(J) \cap Y = J. I X ( J ) ∩ Y = J . (2) ∀ spectrum_cardinal_sublattice
∣ Spec p ( Y ) ∣ ≤ ∣ Spec p ( X ) ∣ . |\operatorname{Spec}_p(Y)| \le|\operatorname{Spec}_p(X)|. ∣ Spec p ( Y ) ∣ ≤ ∣ Spec p ( X ) ∣. (3) ∀ spectrum_cardinal_le_of_surjective If K K K is an ideal of X X X , then
∣ Spec p ( X / K ) ∣ ≤ ∣ Spec p ( X ) ∣ . |\operatorname{Spec}_p(X/K)| \le|\operatorname{Spec}_p(X)|. ∣ Spec p ( X / K ) ∣ ≤ ∣ Spec p ( X ) ∣.
In particular, C \mathcal{C} C is closed under sublattices, quotients and homomorphic images.
Proof. Since J J J is an ideal of Y Y Y ,
I X ( J ) = { x ∈ X : ∣ x ∣ ≤ j for some j ∈ J + } . I_X(J) = \{x \in X : |x| \le j \text{ for some } j \in J_+\}. I X ( J ) = { x ∈ X : ∣ x ∣ ≤ j for some j ∈ J + } .
If x x x also belongs to Y Y Y , solidity in Y Y Y gives x ∈ J x \in J x ∈ J . This proves the first assertion and yields the canonical embedding Y / J → X / I X ( J ) Y/J \to X/I_X(J) Y / J → X / I X ( J ) . By Proposition 2.8 , every proper prime ideal of Y Y Y extends to a prime ideal of X X X . Such an extension is proper, and distinct prime ideals of Y Y Y require distinct extensions. Notice that restriction in the other direction can give the whole sublattice: for P ∈ Spec p ( X ) P \in\operatorname{Spec}_p(X) P ∈ Spec p ( X ) , either P ∩ Y = Y P \cap Y = Y P ∩ Y = Y or P ∩ Y ∈ Spec p ( Y ) P \cap Y \in\operatorname{Spec}_p(Y) P ∩ Y ∈ Spec p ( Y ) . Proper prime ideals of X / K X/K X / K correspond bijectively to proper prime ideals of X X X containing K K K . Finally, homomorphic images are quotients by kernels. □
We also recall prime ideal separation. If 0 ≠ x ∈ X 0 \ne x \in X 0 = x ∈ X , extend the zero prime ideal of the one-dimensional sublattice R ∣ x ∣ \mathbb{R}|x| R ∣ x ∣ to a prime ideal of X X X which does not contain x x x . Thus, the natural homomorphism
ι : X ⟶ ∏ P ∈ Spec p ( X ) X / P \iota: X \longrightarrow\prod_{P \in\operatorname{Spec}_p(X)} X/P ι : X ⟶ P ∈ Spec p ( X ) ∏ X / P
x ⟼ ( x + P ) P ∈ Spec p ( X ) (2) x \longmapsto(x+P)_{P \in\operatorname{Spec}_p(X)} \tag*{(2)} x ⟼ ( x + P ) P ∈ Spec p ( X ) (2)
is injective, so X X X can be regarded as a sublattice of a product of totally ordered vector lattices.
There is no distinction between the category of totally ordered vector spaces and that of totally ordered vector lattices. For a totally ordered real vector space L L L , denote its set of Archimedean classes by
Γ L = ( L ∖ { 0 } ) / ≃ , \Gamma_L = (L \setminus\{0\})/{\simeq}, Γ L = ( L ∖ { 0 }) / ≃ ,
where
x ≃ y ⟺ ∣ x ∣ ≤ C ∣ y ∣ and ∣ y ∣ ≤ D ∣ x ∣ for some C , D > 0. x \simeq y \quad\Longleftrightarrow\quad|x| \le C|y| \text{ and } |y| \le D|x| \quad\text{for some } C,D>0. x ≃ y ⟺ ∣ x ∣ ≤ C ∣ y ∣ and ∣ y ∣ ≤ D ∣ x ∣ for some C , D > 0.
Write v ( x ) v(x) v ( x ) for the class of x x x , and order the classes by
v ( x ) ≥ v ( y ) ⟺ ∣ x ∣ ≤ C ∣ y ∣ for some C > 0 , v(x) \ge v(y) \quad\Longleftrightarrow\quad|x| \le C|y| \text{ for some } C>0, v ( x ) ≥ v ( y ) ⟺ ∣ x ∣ ≤ C ∣ y ∣ for some C > 0 ,
and set v ( 0 ) = ∞ v(0)=\infty v ( 0 ) = ∞ . For nonzero x , y x,y x , y ,
v ( x ) > v ( y ) ⟺ ∣ x ∣ < ϵ ∣ y ∣ for every ϵ > 0. v(x)>v(y)\Longleftrightarrow|x|<\epsilon|y|\text{ for every }\epsilon>0. v ( x ) > v ( y ) ⟺ ∣ x ∣ < ϵ ∣ y ∣ for every ϵ > 0.
We have v ( r x ) = v ( x ) v(rx)=v(x) v ( r x ) = v ( x ) for r ≠ 0 r\ne0 r = 0 and v ( x + y ) ≥ min { v ( x ) , v ( y ) } v(x+y)\ge\min\{v(x),v(y)\} v ( x + y ) ≥ min { v ( x ) , v ( y )} . If the two values are distinct, say v ( x ) < v ( y ) v(x)<v(y) v ( x ) < v ( y ) , then v ( x + y ) = v ( x ) v(x+y)=v(x) v ( x + y ) = v ( x ) , and x + y > 0 x+y>0 x + y > 0 exactly when x > 0 x>0 x > 0 .
Lemma 6.2. For every nonzero totally ordered real vector space L L L ,
∀ \forall ∀ archimedeanClasses_cardinal_le_spectrum ∣ Γ L ∣ ≤ ∣ Spec p ( L ) ∣ |\Gamma_L|\le|\operatorname{Spec}_p(L)| ∣ Γ L ∣ ≤ ∣ Spec p ( L ) ∣ ;
∀ \forall ∀ cardinal_le_power_archimedeanClasses ∣ L ∣ ≤ c ∣ Γ L ∣ |L|\le\mathfrak{c}^{|\Gamma_L|} ∣ L ∣ ≤ c ∣ Γ L ∣ .
Proof. Choose e γ > 0 e_\gamma>0 e γ > 0 in each class γ ∈ Γ L \gamma\in\Gamma_L γ ∈ Γ L , and let
D γ = { x : ∣ x ∣ ≤ r e γ for some r > 0 } , N γ = { x : n ∣ x ∣ ≤ e γ for every n ≥ 1 } . D_\gamma=\{x:|x|\le re_\gamma\text{ for some }r>0\},\qquad N_\gamma=\{x:n|x|\le e_\gamma\text{ for every }n\ge1\}. D γ = { x : ∣ x ∣ ≤ r e γ for some r > 0 } , N γ = { x : n ∣ x ∣ ≤ e γ for every n ≥ 1 } .
These are ideals, and N γ N_\gamma N γ is proper and prime in L L L . If γ > η \gamma>\eta γ > η , then e γ ∈ N η ∖ N γ e_\gamma\in N_\eta\setminus N_\gamma e γ ∈ N η ∖ N γ , proving the first inequality.
For x ∈ D γ x\in D_\gamma x ∈ D γ define
q γ ( x ) = sup { r ∈ R : r e γ ≤ x } . q_\gamma(x)=\sup\{r\in\mathbb{R}:re_\gamma\le x\}. q γ ( x ) = sup { r ∈ R : r e γ ≤ x } .
This is a well-defined real number, and, for every ϵ > 0 \epsilon>0 ϵ > 0 ,
( q γ ( x ) − ϵ ) e γ ≤ x ≤ ( q γ ( x ) + ϵ ) e γ . (q_\gamma(x)-\epsilon)e_\gamma\le x\le(q_\gamma(x)+\epsilon)e_\gamma. ( q γ ( x ) − ϵ ) e γ ≤ x ≤ ( q γ ( x ) + ϵ ) e γ .
By definition, if a , b ∈ R a,b\in\mathbb{R} a , b ∈ R and z ∈ D γ z\in D_\gamma z ∈ D γ satisfy a e γ ≤ z ≤ b e γ ae_\gamma\le z\le be_\gamma a e γ ≤ z ≤ b e γ , then a ≤ q γ ( z ) ≤ b a\le q_\gamma(z)\le b a ≤ q γ ( z ) ≤ b . For x , y ∈ D γ x,y\in D_\gamma x , y ∈ D γ , adding the preceding bounds gives
( q γ ( x ) + q γ ( y ) − 2 ϵ ) e γ ≤ x + y ≤ ( q γ ( x ) + q γ ( y ) + 2 ϵ ) e γ . (q_\gamma(x)+q_\gamma(y)-2\epsilon)e_\gamma\le x+y\le(q_\gamma(x)+q_\gamma(y)+2\epsilon)e_\gamma. ( q γ ( x ) + q γ ( y ) − 2 ϵ ) e γ ≤ x + y ≤ ( q γ ( x ) + q γ ( y ) + 2 ϵ ) e γ .
Consequently,
∣ q γ ( x + y ) − q γ ( x ) − q γ ( y ) ∣ ≤ 2 ϵ . |q_\gamma(x+y)-q_\gamma(x)-q_\gamma(y)|\le2\epsilon. ∣ q γ ( x + y ) − q γ ( x ) − q γ ( y ) ∣ ≤ 2 ϵ .
Since this holds for every ϵ > 0 \epsilon>0 ϵ > 0 , we obtain q γ ( x + y ) = q γ ( x ) + q γ ( y ) q_\gamma(x+y)=q_\gamma(x)+q_\gamma(y) q γ ( x + y ) = q γ ( x ) + q γ ( y ) . Similarly, multiplying the bounds for x x x by r ∈ R r\in\mathbb{R} r ∈ R , reversing the inequalities when r < 0 r<0 r < 0 , yields
( r q γ ( x ) − ∣ r ∣ ϵ ) e γ ≤ r x ≤ ( r q γ ( x ) + ∣ r ∣ ϵ ) e γ . (rq_\gamma(x)-|r|\epsilon)e_\gamma\le rx\le(rq_\gamma(x)+|r|\epsilon)e_\gamma. ( r q γ ( x ) − ∣ r ∣ ϵ ) e γ ≤ r x ≤ ( r q γ ( x ) + ∣ r ∣ ϵ ) e γ .
Thus ∣ q γ ( r x ) − r q γ ( x ) ∣ ≤ ∣ r ∣ ϵ |q_\gamma(rx)-rq_\gamma(x)|\le|r|\epsilon ∣ q γ ( r x ) − r q γ ( x ) ∣ ≤ ∣ r ∣ ϵ for every ϵ > 0 \epsilon>0 ϵ > 0 , which implies q γ ( r x ) = r q γ ( x ) q_\gamma(rx)=rq_\gamma(x) q γ ( r x ) = r q γ ( x ) . This proves that q γ q_\gamma q γ is linear.
If q γ ( x ) = 0 q_\gamma(x)=0 q γ ( x ) = 0 , the bounds for x x x give ∣ x ∣ ≤ ϵ e γ |x|\le\epsilon e_\gamma ∣ x ∣ ≤ ϵ e γ for every ϵ > 0 \epsilon>0 ϵ > 0 . Taking ϵ = 1 / n \epsilon=1/n ϵ = 1/ n shows that n ∣ x ∣ ≤ e γ n|x|\le e_\gamma n ∣ x ∣ ≤ e γ for every n ≥ 1 n\ge1 n ≥ 1 , so x ∈ N γ x\in N_\gamma x ∈ N γ . Conversely, if x ∈ N γ x\in N_\gamma x ∈ N γ , then
− 1 n e γ ≤ x ≤ 1 n e γ ( n ≥ 1 ) , -\frac{1}{n}e_\gamma\le x\le\frac{1}{n}e_\gamma\qquad(n\ge1), − n 1 e γ ≤ x ≤ n 1 e γ ( n ≥ 1 ) ,
and hence ∣ q γ ( x ) ∣ ≤ 1 / n |q_\gamma(x)|\le1/n ∣ q γ ( x ) ∣ ≤ 1/ n for every n ≥ 1 n\ge1 n ≥ 1 . Therefore q γ ( x ) = 0 q_\gamma(x)=0 q γ ( x ) = 0 , and ker q γ = N γ \ker q_\gamma=N_\gamma ker q γ = N γ .
Let f γ : L → R f_\gamma:L\to\mathbb{R} f γ : L → R be a linear functional extending q γ q_\gamma q γ . For x ≠ 0 x\ne0 x = 0 , taking γ = v ( x ) \gamma=v(x) γ = v ( x ) gives x ∈ D γ ∖ N γ x\in D_\gamma\setminus N_\gamma x ∈ D γ ∖ N γ , so f γ ( x ) ≠ 0 f_\gamma(x)\ne0 f γ ( x ) = 0 . Consequently x ↦ ( f γ ( x ) ) γ ∈ Γ L x\mapsto(f_\gamma(x))_{\gamma\in\Gamma_L} x ↦ ( f γ ( x ) ) γ ∈ Γ L is a linear injection into R Γ L \mathbb{R}^{\Gamma_L} R Γ L , proving the second inequality. □ \square □
Proposition 6.3. ∀ \forall ∀ cardinal_le_continuum Every vector lattice in C \mathcal{C} C has cardinality at most c \mathfrak{c} c . ∀ \forall ∀ exists_small_sublattice_representatives There is a set of at most 2 c 2^{\mathfrak{c}} 2 c representatives of the isomorphism types in C \mathcal{C} C , and of the embedding types A ↪ B A\hookrightarrow B A ↪ B with B ∈ C B\in\mathcal{C} B ∈ C .
Proof. If X ∈ C X \in C X ∈ C , each prime quotient X / P X/P X / P has countable spectrum by Lemma 6.1 , and therefore has cardinality at most c ℵ 0 = c \mathfrak{c}^{\aleph_0} = \mathfrak{c} c ℵ 0 = c by Lemma 6.2 . The embedding of X X X into a countable product of these quotients given in (1) yields ∣ X ∣ ≤ c |X| \le\mathfrak{c} ∣ X ∣ ≤ c .
Structures of cardinality at most c \mathfrak{c} c can be coded on subsets of a fixed set of size c \mathfrak{c} c . The vector-lattice language has c \mathfrak{c} c operation symbols, all of finite arity, so there are at most 2 c 2^{\mathfrak{c}} 2 c such structures. Specifying a sublattice or an embedding does not increase this bound. □
The cardinality bound in Proposition 6.3 is sharp: a vector lattice with countable spectrum can have Hamel dimension c \mathfrak{c} c .
Example 6.4. Countable spectrum does not imply countable generation. Consider L = R N L = \mathbb{R}^{\mathbb{N}} L = R N with the lexicographic order. Its proper ideals are 0 0 0 and
L n = { x ∈ R N : x 1 = ⋯ = x n = 0 } ( n ≥ 1 ) . L_n = \{x \in\mathbb{R}^{\mathbb{N}} : x_1 = \cdots= x_n = 0\} \qquad(n \ge1). L n = { x ∈ R N : x 1 = ⋯ = x n = 0 } ( n ≥ 1 ) .
Indeed, let J J J be a nonzero proper ideal, and let i i i be the least index that occurs as the first nonzero coordinate of an element of J J J . A vector in J J J with first nonzero coordinate i i i dominates, up to a positive scalar, every vector whose first i − 1 i-1 i − 1 coordinates vanish. Hence J = L i − 1 J = L_{i-1} J = L i − 1 , with L 0 = L L_0 = L L 0 = L ; properness forces i ≥ 2 i \ge2 i ≥ 2 . ∀ lex_sequences_countable_spectrum Thus L ∈ C L \in C L ∈ C . However, the vectors ( 1 , t , t 2 , … ) (1,t,t^2,\ldots) ( 1 , t , t 2 , … ) , t > 0 t > 0 t > 0 , form a linearly independent family, so dim R L = c \dim_{\mathbb{R}} L = \mathfrak{c} dim R L = c . Every linear subspace of a totally ordered space is a sublattice. ∀ lex_sequences_not_countably_generated Hence L L L needs c \mathfrak{c} c lattice generators.
Universal disposition and the push-out construction
Definition 6.5. ∀ IsCountablePrimeInjective A vector lattice Z Z Z is injective relative to C C C , or countable-prime-injective , if every lattice homomorphism T : A → Z T : A \to Z T : A → Z , where A A A is a sublattice of B ∈ C B \in C B ∈ C , extends to a lattice homomorphism T ~ : B → Z \widetilde{T} : B \to Z T : B → Z .
Definition 6.6. ∀ HasCountablePrimeUniversalDisposition A vector lattice U U U is of universal disposition for C C C if, for every embedding i : A → B i : A \to B i : A → B with B ∈ C B \in C B ∈ C and every embedding j : A → U j : A \to U j : A → U , there is an embedding k : B → U k : B \to U k : B → U such that k i = j ki = j k i = j .
The bound on the size of the prime spectrum is imposed on the ambient lattice B B B ; its sublattice A A A then also belongs to C C C . Neither Z Z Z nor U U U is required to have countable spectrum. Taking A = 0 A = 0 A = 0 in Definition 6.6 shows that U U U contains every member of C C C as a sublattice.
Theorem 6.7. ∀ isCountablePrimeInjective_of_universalDisposition Every vector lattice U U U of universal disposition for C C C is countable-prime-injective. ∀ extension_with_generated_kernel More precisely, for A ⊂ B ∈ C A \subset B \in C A ⊂ B ∈ C and a lattice homomorphism T : A → U T : A \to U T : A → U , there is an extension T ~ : B → U \widetilde{T} : B \to U T : B → U satisfying
ker T ~ = I B ( ker T ) . \ker\widetilde{T} = I_B(\ker T). ker T = I B ( ker T ) . Proof. Put K = ker T K = \ker T K = ker T and J = I B ( K ) J = I_B(K) J = I B ( K ) . Lemma 6.1 gives an embedding A / K → B / J A/K \to B/J A / K → B / J , and B / J ∈ C B/J \in C B / J ∈ C . The map induced by T T T on A / K A/K A / K is an embedding into U U U , so universal disposition extends it to an embedding H : B / J → U H : B/J \to U H : B / J → U . Composing H H H with the quotient map B → B / J B \to B/J B → B / J gives the required extension, with kernel exactly J J J . □
Example 6.8. ∀ \forall ∀ real_pair_uncountable_spectrum The class C \mathcal{C} C is not closed under free products, even with two one-dimensional factors. Let x , y x,y x , y be the canonical positive generators of R ∗ V L R \mathbb{R} \ast_{\mathrm{VL}} \mathbb{R} R ∗ VL R . For each t > 0 t>0 t > 0 , the homomorphism ϕ t \phi_t ϕ t into R \mathbb{R} R determined by ϕ t ( x ) = 1 \phi_t(x)=1 ϕ t ( x ) = 1 and ϕ t ( y ) = t \phi_t(y)=t ϕ t ( y ) = t is onto and its kernel is a proper prime ideal. These kernels are distinct, since y − t x y-tx y − t x belongs to ker ϕ t \ker\phi_t ker ϕ t but not to ker ϕ s \ker\phi_s ker ϕ s when s ≠ t s\ne t s = t . Thus this free product has at least c \mathfrak{c} c proper prime ideals.
∀ \forall ∀ cylinder_union_uncountable_spectrum Similarly, the class C \mathcal{C} C is not closed under countable increasing unions. On K = { 0 , 1 } N K=\{0,1\}^{\mathbb{N}} K = { 0 , 1 } N , let X n X_n X n be the space of functions depending on the first n n n coordinates. Then X n ≅ R n X_n\cong\mathbb{R}^n X n ≅ R n has finite spectrum, whereas distinct points of K K K give distinct evaluation kernels on ⋃ n X n \bigcup_n X_n ⋃ n X n . This union has at least c \mathfrak{c} c proper prime ideals.
The quotient argument in Theorem 6.7 avoids any assumption that a push-out remains in C \mathcal{C} C . For the existence result, we use the bound on underlying cardinality instead.
Theorem 6.9. ∀ \forall ∀ exists_countablePrimeUniversalDisposition Every vector lattice X X X embeds into a vector lattice U U U of universal disposition for C \mathcal{C} C . With
λ = ( max { ∣ X ∣ , 2 c } ) c , \lambda=\left(\max\{|X|,2^{\mathfrak{c}}\}\right)^{\mathfrak{c}}, λ = ( max { ∣ X ∣ , 2 c } ) c , one can arrange that ∣ U ∣ ≤ λ |U|\le\lambda ∣ U ∣ ≤ λ . ∀ \forall ∀ exists_small_universalDisposition In fact, U U U can be chosen of universal disposition for all vector lattices of underlying cardinality at most c \mathfrak{c} c .
Proof. We have λ c = λ \lambda^{\mathfrak{c}}=\lambda λ c = λ and c + ≤ 2 c ≤ λ \mathfrak{c}^{+}\le2^{\mathfrak{c}}\le\lambda c + ≤ 2 c ≤ λ . By the coding argument in Proposition 6.3 , choose at most 2 c 2^{\mathfrak{c}} 2 c representatives of embeddings i : A ↪ B i:A\hookrightarrow B i : A ↪ B with ∣ B ∣ ≤ c |B|\le\mathfrak{c} ∣ B ∣ ≤ c .
Construct an increasing continuous chain ( U α ) α < c + (U_\alpha)_{\alpha<\mathfrak{c}^{+}} ( U α ) α < c + , beginning with U 0 = X U_0=X U 0 = X , with all stages of cardinality at most λ \lambda λ . At each successor stage, list all pairs
( i : A ↪ B , j : A ↪ U α ) , (i:A\hookrightarrow B,\;j:A\hookrightarrow U_\alpha), ( i : A ↪ B , j : A ↪ U α ) ,
where i i i is one of the chosen representatives. There are at most 2 c λ c = λ 2^{\mathfrak{c}}\lambda^{\mathfrak{c}}=\lambda 2 c λ c = λ such pairs. Process them successively, forming the push-out of i i i and the composite of j j j with the inclusion of U α U_\alpha U α into the current lattice. Proposition 2.10 , applied in both directions, makes both canonical maps embeddings. At internal limit stages take unions.
A push-out is generated by the images of its two factors, so for factors V V V and B B B its cardinality is at most max { c , ∣ V ∣ , ∣ B ∣ } \max\{\mathfrak{c},|V|,|B|\} max { c , ∣ V ∣ , ∣ B ∣ } . Therefore the whole successor stage U α + 1 U_{\alpha+1} U α + 1 has cardinality at most λ \lambda λ . At outer limits take U α = ⋃ β < α U β U_\alpha=\bigcup_{\beta<\alpha}U_\beta U α = ⋃ β < α U β , preserving the same bound.
Let U = ⋃ α < c + U α U=\bigcup_{\alpha<\mathfrak{c}^{+}}U_\alpha U = ⋃ α < c + U α . For an embedding i : A → B i:A\to B i : A → B with ∣ B ∣ ≤ c |B|\le\mathfrak{c} ∣ B ∣ ≤ c and an embedding j : A → U j:A\to U j : A → U , regularity of c + \mathfrak{c}^{+} c + places j ( A ) j(A) j ( A ) inside some U α U_\alpha U α . Up to the chosen representative, this diagram was processed at the next stage, giving the required embedding B → U B\to U B → U . Finally, Proposition 6.3 includes every member of C \mathcal{C} C among the considered objects. □ \square □
Corollary 6.10. ∀ \forall ∀ exists_nontrivial_countablePrimeUniversalDisposition There is a nonzero vector lattice U U U of universal disposition for C \mathcal{C} C such that
∣ U ∣ ≤ 2 c , ∣ Spec p ( U ) ∣ ≤ 2 2 c . |U|\le2^{\mathfrak{c}},\qquad|\operatorname{Spec}_{p}(U)|\le2^{2^{\mathfrak{c}}}. ∣ U ∣ ≤ 2 c , ∣ Spec p ( U ) ∣ ≤ 2 2 c . ∀ \forall ∀ exists_countablePrimeInjective Every vector lattice embeds into a countable-prime-injective vector lattice.
Proof. For the first assertion take X = R X = \mathbb{R} X = R in Theorem 6.9 and use ∣ Spec p ( U ) ∣ ≤ 2 ∣ U ∣ |\operatorname{Spec}_{p}(\mathcal{U})| \leq2^{|\mathcal{U}|} ∣ Spec p ( U ) ∣ ≤ 2 ∣ U ∣ . The second follows from that result and Theorem 6.7 . □
The construction uses transfinite induction in c + \mathfrak{c}^{+} c + to capture every domain of cardinality at most c \mathfrak{c} c . Example 6.4 shows why countable generation cannot be used to replace this argument by an inductive construction in ω 1 \omega_{1} ω 1 .
Ordered ultrapowers and relative injectivity
For a nonprincipal ultrafilter U \mathcal{U} U on N \mathbb{N} N , consider
X U = R N / U . X_{\mathcal{U}} = \mathbb{R}^{\mathbb{N}}/\mathcal{U}. X U = R N / U .
Two sequences represent the same element when they agree on a set in U \mathcal{U} U , and
[ ( x n ) ] ≤ [ ( y n ) ] ⟺ { n : x n ≤ y n } ∈ U . [(x_n)] \leq[(y_n)] \quad\Longleftrightarrow\quad\{n : x_n \leq y_n\} \in\mathcal{U}. [( x n )] ≤ [( y n )] ⟺ { n : x n ≤ y n } ∈ U .
With coordinatewise operations and real scalars identified with constant sequences, this is a totally ordered real vector space of cardinality c \mathfrak{c} c . We will prove that it is C C C -injective. In particular, the target can be much smaller than the bound supplied by the earlier transfinite push-out construction.
For subsets L , R L, R L , R of a totally ordered space, write L < R L < R L < R when ℓ < r \ell< r ℓ < r for every ℓ ∈ L \ell\in L ℓ ∈ L and r ∈ R r \in R r ∈ R . We say that a vector lattice has the countable interpolation property if, whenever L , R L, R L , R are countable and L < R L < R L < R , there is z z z with L < z < R L < z < R L < z < R .
The following lemma is a consequence of the countable saturation of ultrapowers over nonprincipal ultrafilters on N \mathbb{N} N [12 Theorem 5.6] , applied to the underlying order in the language { < } \{<\} { < } . We include a direct proof for completeness.
Lemma 6.11. ∀ orderedUltrapower_hasCountableInterpolation X U X_{\mathcal{U}} X U has the countable interpolation property .
Proof. Let L , R ⊆ X U L, R \subseteq X_{\mathcal{U}} L , R ⊆ X U be countable sets with L < R L < R L < R . Suppose first that both sets are nonempty. Write L = { ℓ j : j ≥ 1 } L = \{\ell_j : j \geq1\} L = { ℓ j : j ≥ 1 } and R = { r j : j ≥ 1 } R = \{r_j : j \geq1\} R = { r j : j ≥ 1 } , repeating elements if either set is finite, and choose representatives
ℓ j = [ ( ℓ j , n ) n ] , r j = [ ( r j , n ) n ] . \ell_j = [(\ell_{j,n})_n], \qquad r_j = [(r_{j,n})_n]. ℓ j = [( ℓ j , n ) n ] , r j = [( r j , n ) n ] .
Since ℓ j < r k \ell_j < r_k ℓ j < r k for every j , k j,k j , k , the definition of the ultrapower order gives
{ n : ℓ j , n < r k , n } ∈ U . \{n : \ell_{j,n} < r_{k,n}\} \in\mathcal{U}. { n : ℓ j , n < r k , n } ∈ U .
For m , n ≥ 1 m,n \geq1 m , n ≥ 1 , put
a m ( n ) = max 1 ≤ j ≤ m ℓ j , n , b m ( n ) = min 1 ≤ j ≤ m r j , n , a_m(n) = \max_{1 \leq j \leq m} \ell_{j,n}, \qquad b_m(n) = \min_{1 \leq j \leq m} r_{j,n}, a m ( n ) = 1 ≤ j ≤ m max ℓ j , n , b m ( n ) = 1 ≤ j ≤ m min r j , n ,
and define
E m = { n ≥ m : a m ( n ) < b m ( n ) } = { n : n ≥ m } ∩ ⋂ 1 ≤ j , k ≤ m { n : ℓ j , n < r k , n } . E_m = \{n \geq m : a_m(n) < b_m(n)\} = \{n : n \geq m\} \cap\bigcap_{1 \leq j,k \leq m} \{n : \ell_{j,n} < r_{k,n}\}. E m = { n ≥ m : a m ( n ) < b m ( n )} = { n : n ≥ m } ∩ 1 ≤ j , k ≤ m ⋂ { n : ℓ j , n < r k , n } .
Each E m E_m E m belongs to U \mathcal{U} U , since U \mathcal{U} U is closed under finite intersections and contains every cofinite subset of N \mathbb{N} N . Moreover, E m + 1 ⊆ E m E_{m+1} \subseteq E_m E m + 1 ⊆ E m .
For every n ≥ 1 n \geq1 n ≥ 1 , let
k ( n ) = max ( { 0 } ∪ { m ∈ { 1 , … , n } : n ∈ E m } ) , k(n) = \max\bigl(\{0\} \cup\{m \in\{1,\ldots,n\} : n \in E_m\}\bigr), k ( n ) = max ( { 0 } ∪ { m ∈ { 1 , … , n } : n ∈ E m } ) ,
and set
z n = { a k ( n ) ( n ) + b k ( n ) ( n ) 2 , k ( n ) > 0 , 0 , k ( n ) = 0. z_n =
\begin{cases}
\frac{a_{k(n)}(n)+b_{k(n)}(n)}{2}, & k(n)>0,\\
0, & k(n)=0.
\end{cases} z n = { 2 a k ( n ) ( n ) + b k ( n ) ( n ) , 0 , k ( n ) > 0 , k ( n ) = 0.
If k ( n ) > 0 k(n)>0 k ( n ) > 0 , then n ∈ E k ( n ) n\in E_{k(n)} n ∈ E k ( n ) , so the midpoint defining z n z_n z n lies strictly between a k ( n ) ( n ) a_{k(n)}(n) a k ( n ) ( n ) and b k ( n ) ( n ) b_{k(n)}(n) b k ( n ) ( n ) . Fix j ≥ 1 j\geq1 j ≥ 1 . For every n ∈ E j n\in E_j n ∈ E j , we have k ( n ) ≥ j k(n)\geq j k ( n ) ≥ j , and hence
ℓ j , n ≤ a k ( n ) ( n ) < z n < b k ( n ) ( n ) ≤ r j , n . \ell_{j,n}\leq a_{k(n)}(n)<z_n<b_{k(n)}(n)\leq r_{j,n}. ℓ j , n ≤ a k ( n ) ( n ) < z n < b k ( n ) ( n ) ≤ r j , n .
Thus E j ⊆ { n : ℓ j , n < z n < r j , n } E_j\subseteq\{n:\ell_{j,n}<z_n<r_{j,n}\} E j ⊆ { n : ℓ j , n < z n < r j , n } , and the latter set belongs to U \mathcal{U} U . Consequently z = [ ( z n ) ] z=[(z_n)] z = [( z n )] satisfies ℓ j < z < r j \ell_j<z<r_j ℓ j < z < r j for every j j j , proving L < z < R L<z<R L < z < R .
If L ≠ ∅ L\neq\varnothing L = ∅ and R = ∅ R=\varnothing R = ∅ , enumerate L L L and choose representatives as above, and set
z n = 1 + max 1 ≤ j ≤ n ℓ j , n . z_n=1+\max_{1\leq j\leq n}\ell_{j,n}. z n = 1 + 1 ≤ j ≤ n max ℓ j , n .
For every fixed j j j , the inequality ℓ j , n < z n \ell_{j,n}<z_n ℓ j , n < z n holds for all n ≥ j n\geq j n ≥ j , so ℓ j < [ ( z n ) ] \ell_j<[(z_n)] ℓ j < [( z n )] . Similarly, if L = ∅ L=\varnothing L = ∅ and R ≠ ∅ R\neq\varnothing R = ∅ , use
z n = min 1 ≤ j ≤ n r j , n − 1 z_n=\min_{1\leq j\leq n}r_{j,n}-1 z n = 1 ≤ j ≤ n min r j , n − 1
to obtain [ ( z n ) ] < r j [(z_n)]<r_j [( z n )] < r j for every j j j .
We need the following property of cofinality of cuts in a totally ordered vector space with countable set of Archimedean classes. This follows from the cofinality formulas for cuts in divisible ordered abelian groups given in [13 Theorems 4.5 and 4.8] , but we include a direct proof for completeness.
Lemma 6.12. ∀ \forall ∀ countable_cut_points Let V V V be a totally ordered real vector space such that Γ V \Gamma_V Γ V is at most countable. Let D D D be a vector subspace of V V V and b ∈ V ∖ D b\in V\setminus D b ∈ V ∖ D . Then
L b = { d ∈ D : d < b } , R b = { d ∈ D : d > b } L_b=\{d\in D:d<b\},\qquad R_b=\{d\in D:d>b\} L b = { d ∈ D : d < b } , R b = { d ∈ D : d > b } have, respectively, a countable cofinal subset and a countable coinitial subset.
Proof. We use the terminology for Archimedean classes introduced in § 6.1 . The nonempty set
Δ = { v ( b − d ) : d ∈ D } ⊆ Γ V \Delta=\{v(b-d):d\in D\}\subseteq\Gamma_V Δ = { v ( b − d ) : d ∈ D } ⊆ Γ V
is at most countable.
Suppose first that Δ \Delta Δ has no largest element. Choose d n ∈ D d_n\in D d n ∈ D such that
δ n = v ( b − d n ) \delta_n=v(b-d_n) δ n = v ( b − d n )
is strictly increasing and cofinal in Δ \Delta Δ . Set
h n = ∣ d n + 1 − d n ∣ , ℓ n = d n − 2 h n , r n = d n + 2 h n . h_n=\lvert d_{n+1}-d_n\rvert,\qquad\ell_n=d_n-2h_n,\qquad r_n=d_n+2h_n. h n = ∣ d n + 1 − d n ∣ , ℓ n = d n − 2 h n , r n = d n + 2 h n .
Since δ n + 1 > δ n \delta_{n+1}>\delta_n δ n + 1 > δ n , the valuation identities give
v ( d n + 1 − d n ) = δ n , v ( b − d n + 1 ) > δ n . v(d_{n+1}-d_n)=\delta_n,\qquad v(b-d_{n+1})>\delta_n. v ( d n + 1 − d n ) = δ n , v ( b − d n + 1 ) > δ n .
Consequently,
ℓ n < b < r n , v ( b − ℓ n ) = v ( r n − b ) = δ n . \ell_n<b<r_n,\qquad v(b-\ell_n)=v(r_n-b)=\delta_n. ℓ n < b < r n , v ( b − ℓ n ) = v ( r n − b ) = δ n .
Indeed, write b − d n = ( d n + 1 − d n ) + ( b − d n + 1 ) b-d_n=(d_{n+1}-d_n)+(b-d_{n+1}) b − d n = ( d n + 1 − d n ) + ( b − d n + 1 ) and observe that the second summand is infinitesimal relative to h n h_n h n . Both 2 h n + ( b − d n ) 2h_n+(b-d_n) 2 h n + ( b − d n ) and 2 h n − ( b − d n ) 2h_n-(b-d_n) 2 h n − ( b − d n ) are therefore positive and have the same valuation as h n h_n h n .
Given d ∈ L b d\in L_b d ∈ L b , choose n n n with δ n > v ( b − d ) \delta_n>v(b-d) δ n > v ( b − d ) . Then 0 < b − ℓ n < b − d 0<b-\ell_n<b-d 0 < b − ℓ n < b − d , so d < ℓ n d<\ell_n d < ℓ n . Similarly, given d ∈ R b d\in R_b d ∈ R b , some n n n satisfies r n < d r_n<d r n < d . Hence { ℓ n : n ≥ 1 } \{\ell_n:n\ge1\} { ℓ n : n ≥ 1 } and { r n : n ≥ 1 } \{r_n:n\ge1\} { r n : n ≥ 1 } have the required properties.
Suppose now that Δ \Delta Δ has largest element δ = v ( b − d 0 ) \delta=v(b-d_0) δ = v ( b − d 0 ) , where d 0 ∈ D d_0\in D d 0 ∈ D . We claim that
δ ∉ v ( D ∖ { 0 } ) . (3) \delta\notin v(D\setminus\{0\}). \tag*{(3)} δ ∈ / v ( D ∖ { 0 }) . (3)
If this failed, choose u ∈ D u\in D u ∈ D , u > 0 u>0 u > 0 , with v ( u ) = δ v(u)=\delta v ( u ) = δ , and put w = b − d 0 w=b-d_0 w = b − d 0 . Since ∣ w ∣ ≍ u |w|\asymp u ∣ w ∣ ≍ u , the number
a = sup { s ∈ R : s u ≤ w } a=\sup\{s\in\mathbb{R}:su\le w\} a = sup { s ∈ R : s u ≤ w }
is a well-defined real number. For every η > 0 \eta>0 η > 0 ,
( a − η ) u < w < ( a + η ) u , (a-\eta)u<w<(a+\eta)u, ( a − η ) u < w < ( a + η ) u ,
and thus ∣ w − a u ∣ < η u |w-au|<\eta u ∣ w − a u ∣ < η u . The vector w − a u w-au w − a u is nonzero, since b ∉ D b\notin D b ∈ / D , and therefore
v ( b − ( d 0 + a u ) ) > δ , v\bigl(b-(d_0+au)\bigr)>\delta, v ( b − ( d 0 + a u ) ) > δ ,
contrary to the choice of δ \delta δ . This proves (3) .
Replacing b b b by − b -b − b if necessary, it suffices to consider w > 0 w>0 w > 0 . For each γ ∈ v ( D ∖ { 0 } ) \gamma\in v(D\setminus\{0\}) γ ∈ v ( D ∖ { 0 }) , choose u γ ∈ D u_\gamma\in D u γ ∈ D with u γ > 0 u_\gamma>0 u γ > 0 and v ( u γ ) = γ v(u_\gamma)=\gamma v ( u γ ) = γ . Define
L 0 = { d 0 } ∪ { d 0 + n u γ : γ > δ , n ≥ 1 } , L_0=\{d_0\}\cup\{d_0+nu_\gamma:\gamma>\delta,\ n\ge1\}, L 0 = { d 0 } ∪ { d 0 + n u γ : γ > δ , n ≥ 1 } ,
R 0 = { d 0 + n − 1 u γ : γ < δ , n ≥ 1 } . R_0=\{d_0+n^{-1}u_\gamma:\gamma<\delta,\ n\ge1\}. R 0 = { d 0 + n − 1 u γ : γ < δ , n ≥ 1 } .
These sets are at most countable and satisfy L 0 ⊆ L b L_0\subseteq L_b L 0 ⊆ L b and R 0 ⊆ R b R_0\subseteq R_b R 0 ⊆ R b .
To verify cofinality, let d ∈ L b d\in L_b d ∈ L b . If d ≤ d 0 d\le d_0 d ≤ d 0 , use d 0 ∈ L 0 d_0\in L_0 d 0 ∈ L 0 . Otherwise 0 < d − d 0 < w 0<d-d_0<w 0 < d − d 0 < w , so (3) implies v ( d − d 0 ) > δ v(d-d_0)>\delta v ( d − d 0 ) > δ . A sufficiently large integer multiple of the chosen representative of this class dominates d − d 0 d-d_0 d − d 0 . Thus d ≤ ℓ d\le\ell d ≤ ℓ for some ℓ ∈ L 0 \ell\in L_0 ℓ ∈ L 0 . Similarly, if d ∈ R b d\in R_b d ∈ R b , then d − d 0 > w > 0 d-d_0>w>0 d − d 0 > w > 0 and v ( d − d 0 ) < δ v(d-d_0)<\delta v ( d − d 0 ) < δ . A sufficiently small positive multiple n − 1 u v ( d − d 0 ) n^{-1}u_{v(d-d_0)} n − 1 u v ( d − d 0 ) is less than d − d 0 d-d_0 d − d 0 , proving that R 0 R_0 R 0 is coinitial in R b R_b R b . □
Lemma 6.13. ∀ ordered_lattice_embedding_extension Let Y Y Y be a totally ordered real vector space with the countable interpolation property. If V V V is a totally ordered real vector space with at most countably many Archimedean classes, then every ordered linear embedding j : D → Y j:D\to Y j : D → Y , defined on a vector subspace D ⊆ V D\subseteq V D ⊆ V , extends to an ordered linear embedding V → Y V\to Y V → Y .
Proof. By Zorn’s lemma, choose a maximal extension S : X → Y S:X\to Y S : X → Y of j j j among ordered linear embeddings defined on vector subspaces of V V V . Suppose that b ∈ V ∖ X b\in V\setminus X b ∈ V ∖ X . By Lemma 6.12 , there are at most countable sets L 0 ⊆ { e ∈ X : e < b } L_0\subseteq\{e\in X:e<b\} L 0 ⊆ { e ∈ X : e < b } and R 0 ⊆ { e ∈ X : e > b } R_0\subseteq\{e\in X:e>b\} R 0 ⊆ { e ∈ X : e > b } that are cofinal and coinitial in these sets, respectively. Countable interpolation yields z ∈ Y z\in Y z ∈ Y with
S ( L 0 ) < z < S ( R 0 ) . S(L_0)<z<S(R_0). S ( L 0 ) < z < S ( R 0 ) .
By cofinality and coinitiality,
S ( e ) < z ( e ∈ X , e < b ) , z < S ( e ) ( e ∈ X , b < e ) . S(e) < z \quad(e \in X,\ e < b), \qquad z < S(e) \quad(e \in X,\ b < e). S ( e ) < z ( e ∈ X , e < b ) , z < S ( e ) ( e ∈ X , b < e ) .
Define
S ′ ( e + t b ) = S ( e ) + t z ( e ∈ X , t ∈ R ) . S'(e+tb)=S(e)+tz \quad(e \in X,\ t \in\mathbb{R}). S ′ ( e + t b ) = S ( e ) + t z ( e ∈ X , t ∈ R ) .
Since b ∉ X b \notin X b ∈ / X , this defines a linear extension of S S S . If t ≠ 0 t \ne0 t = 0 , the sign of e + t b e+tb e + t b is determined by the comparison of b b b with − e / t -e/t − e / t . The displayed inequalities show that S ′ ( e + t b ) S'(e+tb) S ′ ( e + t b ) has exactly the same sign. For t = 0 t=0 t = 0 , this follows from the corresponding property of S S S . Hence S ′ S' S ′ is an ordered linear embedding, contradicting maximality. Therefore X = V X=V X = V .
Theorem 6.14. ∀ orderedUltrapower_isCountablePrimeInjective For every nonprincipal ultrafilter U U U on N \mathbb{N} N , the totally ordered vector lattice X U X_U X U is C C C -injective. ∀ isCountablePrimeInjective_of_countableInterpolation More generally, every totally ordered real vector space with the countable interpolation property has this property.
Proof. Let X X X have the countable interpolation property, let A ⊂ B ∈ C A \subset B \in C A ⊂ B ∈ C , and let T : A → X T:A \to X T : A → X be a nonzero lattice homomorphism. Then P = ker T P=\ker T P = ker T is a proper prime of A A A , because A / P A/P A / P is lattice isomorphic to the totally ordered space T ( A ) T(A) T ( A ) . By Proposition 2.8 , choose a proper prime Q Q Q of B B B with Q ∩ A = P Q \cap A=P Q ∩ A = P .
Let q : B → V = B / Q q:B \to V=B/Q q : B → V = B / Q be the quotient map. By Lemmas 6.1 and 6.2 , the totally ordered space V V V has at most countably many Archimedean classes. The map
j 0 : q ( A ) ⟶ X , j 0 ( q ( a ) ) = T ( a ) , j_0:q(A)\longrightarrow X,\qquad j_0(q(a))=T(a), j 0 : q ( A ) ⟶ X , j 0 ( q ( a )) = T ( a ) ,
is a well-defined ordered linear embedding. By Lemma 6.13 , it extends to an ordered linear embedding j : V → X j:V \to X j : V → X . Such a map between totally ordered spaces is a lattice homomorphism, so T ~ = j q \widetilde{T}=jq T = j q extends T T T as required. The case of X U X_U X U follows from Lemma 6.11 .
Corollary 6.15. ∀ hyperreal_injective_not_universalDisposition C C C -injectivity does not imply universal disposition for C C C .
Proof. The totally ordered lattice X U X_U X U cannot contain a lattice copy of coordinatewise ordered R 2 \mathbb{R}^2 R 2 , since it has no pair of disjoint nonzero positive elements. But R 2 \mathbb{R}^2 R 2 has exactly two proper primes, and universal disposition applied to 0 ⊂ R 2 0 \subset\mathbb{R}^2 0 ⊂ R 2 requires such a copy.
Observation 6.16. ∀ exists_extension_of_countable_prime_quotient_classes The proof of Theorem 6.14 only requires each prime quotient B / Q B/Q B / Q to have countably many Archimedean classes. It imposes no countability assumption on the Hamel dimension of A A A or B B B . The extension property is preserved by products and by retractions that are lattice homomorphisms. ∀ isCountablePrimeInjective_pi For products, extend each coordinate separately; ∀ isCountablePrimeInjective_of_retract for a retract, compose an extension with the retraction.
Properties of the C C C -injective vector lattices
Proposition 6.17. Let Z ≠ 0 Z \ne0 Z = 0 be C C C -injective. Then:
(1) ∀ exists_forall_smul_le_of_isCountablePrimeInjective For every z > 0 z > 0 z > 0 there is y ∈ Z y \in Z y ∈ Z such that n z ≤ y nz \le y n z ≤ y for all n ≥ 1 n \ge1 n ≥ 1 .
(2) ∀ not_isVLArchimedean_of_isCountablePrimeInjective Z Z Z is non-Archimedean .
(3) ∀ not_strongOrderUnit_of_isCountablePrimeInjective Z Z Z has no strong order unit .
(4) ∀ exists_disjoint_positive_sum_of_universalDisposition If U U U is of universal disposition for C C C , then every u > 0 u > 0 u > 0 is the sum of two nonzero disjoint positive elements .
(5) ∀ exists_positive_forall_nsmul_le_of_universalDisposition If U U U is of universal disposition for C C C , then for every u > 0 u > 0 u > 0 there is x > 0 x > 0 x > 0 with n x ≤ u nx \le u n x ≤ u for all n ≥ 1 n \ge1 n ≥ 1 .
Proof. The lexicographic plane R l e x 2 \mathbb{R}^{2}_{\mathrm{lex}} R lex 2 , with its first coordinate dominant, has two proper primes. Extend the homomorphism ( 0 , t ) ↦ t z (0,t) \mapsto tz ( 0 , t ) ↦ t z to this plane and let y y y be the image of ( 1 , 0 ) (1,0) ( 1 , 0 ) . Then y ≥ n z y \ge nz y ≥ n z for all n n n . This forbids the Archimedean property. If z z z were a strong order unit, y ≤ r z y \le rz y ≤ r z for some real r r r , also a contradiction.
For the first assertion about U U U , extend t ↦ t u t \mapsto tu t ↦ t u along the diagonal embedding t ↦ ( t , t ) t \mapsto(t,t) t ↦ ( t , t ) into coordinatewise R 2 \mathbb{R}^{2} R 2 . The images of its coordinate vectors are the required disjoint summands. For the second, extend t ↦ t u t \mapsto tu t ↦ t u along t ↦ ( t , 0 ) t \mapsto(t,0) t ↦ ( t , 0 ) into R l e x 2 \mathbb{R}^{2}_{\mathrm{lex}} R lex 2 . The image of ( 0 , 1 ) (0,1) ( 0 , 1 ) is the required nonzero element. □
Proposition 6.18. ∀ exists_injective_hom_of_isCountablePrimeInjective Every nonzero C C C -injective vector lattice Z Z Z contains a lattice copy of every totally ordered member of C C C . ∀ continuum_le_cardinal_of_nontrivial In particular, ∣ Z ∣ ≥ c |Z| \ge\mathfrak{c} ∣ Z ∣ ≥ c . ∀ orderedUltrapower_cardinal This lower bound is attained by X U X_U X U .
Proof. Let C ∈ C C \in\mathcal{C} C ∈ C be totally ordered and form W = C ⊕ l e x R W = C \oplus_{\mathrm{lex}} \mathbb{R} W = C ⊕ lex R , with the C C C coordinate dominant. Its proper ideals are 0 0 0 and J ⊕ R J \oplus\mathbb{R} J ⊕ R , where J J J ranges over the proper ideals of C C C . Thus W ∈ C W \in\mathcal{C} W ∈ C , and every nonzero ideal of W W W contains e = ( 0 , 1 ) e = (0,1) e = ( 0 , 1 ) . For z > 0 z > 0 z > 0 in Z Z Z , extend t e ↦ t z te \mapsto tz t e ↦ t z to W W W . The extension has zero kernel, because a nonzero kernel would contain e e e . Restriction to C × { 0 } C \times\{0\} C × { 0 } gives the desired embedding. Taking the lexicographic product from Example 6.4 yields ∣ Z ∣ ≥ c |Z| \ge\mathfrak{c} ∣ Z ∣ ≥ c ; the ultrapowers X U X_U X U have cardinality exactly c \mathfrak{c} c . □
For an ordinal α > 0 \alpha> 0 α > 0 , let H α = R ( α ) H_{\alpha} = \mathbb{R}^{(\alpha)} H α = R ( α ) , ordered by the sign of the coefficient at the largest support index. Its proper ideals are precisely the initial-coordinate spaces H β H_{\beta} H β , β < α \beta< \alpha β < α : a vector with largest support index γ \gamma γ generates the ideal H γ + 1 H_{\gamma+1} H γ + 1 , and all ideals are unions of such ideals. In particular, ∣ Spec p ( H α ) ∣ = ∣ α ∣ |\operatorname{Spec}_{p}(H_{\alpha})| = |\alpha| ∣ Spec p ( H α ) ∣ = ∣ α ∣ , and every nonzero ideal of H α H_{\alpha} H α contains e 0 e_{0} e 0 .
Theorem 6.19. ∀ exists_omega_one_colex_embedding Every nonzero C C C -injective vector lattice Z Z Z contains a lattice copy of H ω 1 H_{\omega_{1}} H ω 1 . ∀ continuum_le_spectrum_of_isCountablePrimeInjective Its prime spectrum satisfies
∣ Spec p ( Z ) ∣ ≥ c . |\operatorname{Spec}_{p}(Z)| \ge\mathfrak{c}. ∣ Spec p ( Z ) ∣ ≥ c . ∀ countable_spectrum_not_injective In particular, no nonzero member of C C C is C C C -injective.
Proof. Fix z > 0 z > 0 z > 0 in Z Z Z and embed H 1 H_{1} H 1 by e 0 ↦ z e_{0} \mapsto z e 0 ↦ z . Construct compatible embeddings T α : H α → Z T_{\alpha}: H_{\alpha} \to Z T α : H α → Z for 1 ≤ α < ω 1 1 \le\alpha< \omega_{1} 1 ≤ α < ω 1 . At a successor ordinal H α + 1 H_{\alpha+1} H α + 1 has countable spectrum, so relative injectivity extends T α T_\alpha T α to a homomorphism on H α + 1 H_{\alpha+1} H α + 1 . This extension is injective: any nonzero kernel would contain e 0 e_0 e 0 . At limit ordinals take unions. The final union embeds H ω 1 H_{\omega_1} H ω 1 into Z Z Z .
For the stronger spectrum bound, let F n F_n F n consist of 0, 1 and the first n n n terms of an enumeration of Q ∩ ( 0 , 1 ) \mathbb{Q} \cap(0,1) Q ∩ ( 0 , 1 ) . Give H ( F n ) = R F n H(F_n)=\mathbb{R}^{F_n} H ( F n ) = R F n the order determined by its largest nonzero coordinate. Successively extending e 0 ↦ z e_0 \mapsto z e 0 ↦ z gives compatible embeddings H ( F n ) → Z H(F_n) \to Z H ( F n ) → Z : these test lattices have finite spectra, and a nonzero kernel would again contain e 0 e_0 e 0 . Their union embeds H ( Q ∩ [ 0 , 1 ] ) = R Q ∩ [ 0 , 1 ] H(\mathbb{Q}\cap[0,1])=\mathbb{R}^{\mathbb{Q}\cap[0,1]} H ( Q ∩ [ 0 , 1 ]) = R Q ∩ [ 0 , 1 ] into Z Z Z . For each t ∈ ( 0 , 1 ) t \in(0,1) t ∈ ( 0 , 1 ) , the span of the e q e_q e q with q < t q<t q < t is a proper prime ideal of this chain. These ideals are distinct, so Lemma 6.1 gives ∣ Spec p ( Z ) ∣ ≥ c |\operatorname{Spec}_p(Z)| \ge\mathfrak{c} ∣ Spec p ( Z ) ∣ ≥ c .
An obstruction using Archimedean vector lattices
Proposition 6.17 uses a non-Archimedean vector lattice. The obstruction persists when all vector lattices involved are required to be Archimedean and to have countable spectrum.
Let 1 = ( 1 , 1 , … ) \mathbf{1}=(1,1,\ldots) 1 = ( 1 , 1 , … ) , b = ( 1 , 2 , 3 , … ) b=(1,2,3,\ldots) b = ( 1 , 2 , 3 , … ) , and let c 00 c_{00} c 00 denote the finitely supported real sequences. Give
A = c 00 + R 1 , B = c 00 + R 1 + R b A=c_{00}+\mathbb{R}\mathbf{1},\qquad B=c_{00}+\mathbb{R}\mathbf{1}+\mathbb{R}b A = c 00 + R 1 , B = c 00 + R 1 + R b
their coordinatewise orders. These are the eventually constant and eventually affine sequences, respectively. Since an affine sequence has constant sign eventually, both spaces are vector sublattices of R N \mathbb{R}^{\mathbb{N}} R N and hence are Archimedean.
Lemma 6.20. Writing P n A = { a ∈ A : a n = 0 } P_n^A=\{a\in A:a_n=0\} P n A = { a ∈ A : a n = 0 } and P n B = { x ∈ B : x n = 0 } P_n^B=\{x\in B:x_n=0\} P n B = { x ∈ B : x n = 0 } , we have:
(1) ∀ \forall ∀ eventuallyConstant_spectrum
Spec p ( A ) = { P n A : n ≥ 1 } ∪ { c 00 } . \operatorname{Spec}_p(A)=\{P_n^A:n\ge1\}\cup\{c_{00}\}. Spec p ( A ) = { P n A : n ≥ 1 } ∪ { c 00 } . (2) ∀ \forall ∀ eventuallyAffine_spectrum
Spec p ( B ) = { P n B : n ≥ 1 } ∪ { c 00 , A } . \operatorname{Spec}_p(B)=\{P_n^B:n\ge1\}\cup\{c_{00},A\}. Spec p ( B ) = { P n B : n ≥ 1 } ∪ { c 00 , A } . (3) ∀ \forall ∀ sequence_lattices_countable_spectrum A , B ∈ C A,B\in\mathcal{C} A , B ∈ C .
(4) ∀ \forall ∀ affine_sublattice_eq_top B B B is generated as a vector lattice by 1 \mathbf{1} 1 and b b b .
Proof. In either space, if a prime ideal P P P does not contain e n e_n e n , disjointness with e n e_n e n forces every vector whose n n n th coordinate is zero to belong to P P P . Conversely, if x ∈ P x\in P x ∈ P , the inequality ∣ x n ∣ e n ≤ ∣ x ∣ |x_n|e_n\le|x| ∣ x n ∣ e n ≤ ∣ x ∣ forces x n = 0 x_n=0 x n = 0 . Thus P P P is the corresponding coordinate kernel. All other primes contain c 00 c_{00} c 00 .
Now A / c 00 ≅ R A/c_{00}\cong\mathbb{R} A / c 00 ≅ R , while B / c 00 ≅ R l e x 2 B/c_{00}\cong\mathbb{R}^2_{\mathrm{lex}} B / c 00 ≅ R lex 2 under [ a 1 + s b ] ↦ ( s , a ) [a\mathbf{1}+sb]\mapsto(s,a) [ a 1 + s b ] ↦ ( s , a ) . The latter has exactly the two proper ideals 0 0 0 and { 0 } × R \{0\}\times\mathbb{R} { 0 } × R , whose inverse images are c 00 c_{00} c 00 and A A A . This gives the stated descriptions of the spectra. Finally,
e n = ( 1 − ∣ b − n 1 ∣ ) + ( n ≥ 1 ) , e_n=(\mathbf{1}-|b-n\mathbf{1}|)^+\qquad(n\ge1), e n = ( 1 − ∣ b − n 1 ∣ ) + ( n ≥ 1 ) ,
so 1 , b \mathbf{1},b 1 , b generate B B B .
Theorem 6.21. ∀ \forall ∀ archimedean_nonextendible_hom No nonzero Archimedean vector lattice is injective relative to C ∩ A V L \mathcal{C}\cap\mathsf{AVL} C ∩ AVL . ∀ \forall ∀ archimedean_not_universalDisposition There is no Archimedean vector lattice of universal disposition for this class either.
Proof. If Z ≠ 0 Z \ne0 Z = 0 is Archimedean, choose u > 0 u > 0 u > 0 in Z Z Z and define the lattice homomorphism
T : A → Z , T ( a ) = ( lim n → ∞ a n ) u . T:A\to Z,\qquad T(a)=\left(\lim_{n\to\infty}a_n\right)u. T : A → Z , T ( a ) = ( n → ∞ lim a n ) u .
For every integer m ≥ 1 m \ge1 m ≥ 1 , the vector a ( m ) = b ∧ m 1 a^{(m)}=b\wedge m1 a ( m ) = b ∧ m 1 lies in A A A and has eventual value m m m . Any extension T ~ : B → Z \widetilde{T}:B\to Z T : B → Z would satisfy
0 ≤ m u = T ( a ( m ) ) ≤ T ~ ( b ) ( m ≥ 1 ) , 0\le mu=T(a^{(m)})\le\widetilde{T}(b)\qquad(m\ge1), 0 ≤ m u = T ( a ( m ) ) ≤ T ( b ) ( m ≥ 1 ) ,
contrary to the Archimedean property.
For universal disposition, put D = A ⊕ R D=A\oplus\mathbb{R} D = A ⊕ R with coordinatewise order, and consider the embeddings
i : A → B , i ( a ) = a , j : A → D , j ( a ) = ( a , lim a ) . i:A\to B,\qquad i(a)=a,\qquad j:A\to D,\qquad j(a)=(a,\lim a). i : A → B , i ( a ) = a , j : A → D , j ( a ) = ( a , lim a ) .
The lattice D D D is Archimedean and has countable spectrum: the proper primes of a finite direct product come from a prime in one factor. If homomorphisms f : B → Z f:B\to Z f : B → Z and g : D → Z g:D\to Z g : D → Z into an Archimedean lattice satisfy f i = g j fi=gj f i = g j , then
f ( b ) ≥ f ( a ( m ) ) = g ( a ( m ) , m ) ≥ m g ( 0 , 1 ) ≥ 0 ( m ≥ 1 ) . f(b)\ge f(a^{(m)})=g(a^{(m)},m)\ge mg(0,1)\ge0\qquad(m\ge1). f ( b ) ≥ f ( a ( m ) ) = g ( a ( m ) , m ) ≥ m g ( 0 , 1 ) ≥ 0 ( m ≥ 1 ) .
Hence g ( 0 , 1 ) = 0 g(0,1)=0 g ( 0 , 1 ) = 0 , and g g g cannot be injective. If an Archimedean U U U had the stated universal disposition, first embed D D D into U U U using 0 ⊂ D 0\subset D 0 ⊂ D and then extend the resulting embedding of A A A along i : A → B i:A\to B i : A → B . This would give such an amalgam with g g g injective, a contradiction. □ \square □
∀ \forall ∀ exists_countable_spectrum_nonarchimedean_pushout In the ordinary V L \mathcal{V}\mathcal{L} V L push-out of i i i and j j j , Proposition 2.10 makes the image e e e of ( 0 , 1 ) ∈ D (0,1)\in D ( 0 , 1 ) ∈ D nonzero. If y y y is the image of b ∈ B b\in B b ∈ B , the same calculation gives 0 < m e ≤ y 0<me\le y 0 < m e ≤ y for every m ≥ 1 m\ge1 m ≥ 1 . Thus even a diagram of Archimedean lattices with countable spectra can force a non-Archimedean push-out; its Archimedeanization sends e e e to zero.
The existence results above therefore belong to V L \mathcal{V}\mathcal{L} V L . They provide no uniqueness theorem for universal disposition relative to C \mathcal{C} C : a back-and-forth construction would require control of the intermediate sublattices, and Example 6.8 shows why countable spectrum alone does not supply the usual closure assumptions.
Acknowledgements
Research partially supported by grants PID2024-162214NB-I00 and CEX2023-001347-S funded by MCIN/AEI/10.13039/501100011033. Martínez-Fernández was also supported by grant PREX2023-000080 funded by MCIN/AEI/10.13039/501100011033.
References
Instituto de Ciencias Matemáticas, Consejo Superior de Investigaciones Científicas, C/ Nicolás Cabrera 13–15, 28049 Madrid, Spain
Email address: gonzalo.martinez@icmat.es
Instituto de Ciencias Matemáticas, Consejo Superior de Investigaciones Científicas, C/ Nicolás Cabrera 13–15, 28049 Madrid, Spain
Email address: pedro.tradacete@icmat.es