2020 MSC: 03B47, 03B52, 06F05, 03B35
Email address: wangsanmin@hotmail.com (SanMin Wang)
Introduction
We introduce an explicit hypersequent calculus H R \mathrm{H}_{\mathrm{R}} H R in the setting of cyclic involutive residuated lattices and prove its strong standard completeness. The standard models are chains whose underlying ordered set is the real unit interval [ 0 , 1 ] [0,1] [ 0 , 1 ] ; neither commutativity nor identification of the monoid unit with the dualizing constant is assumed. The completeness theorem concerns consequence from arbitrary theories in a fixed countable propositional language, with all premises interpreted under the same valuation.
The chain semantics of H R \mathrm{H}_{\mathrm{R}} H R is the class K R \mathcal{K}_{\mathrm{R}} K R specified in § 2 : cyclic involutive residuated chains satisfying unit cyclicity and every ordered finite-history condition. These requirements hold in every densely ordered cyclic involutive residuated chain (Lemma 2.2 ), hence in every standard model.
The main algebraic result is a full standard representation theorem for every countable member of K R \mathcal{K}_{\mathrm{R}} K R (Theorem 7.1 ). It preserves multiplication, both residuals, involution and both designated constants simultaneously. A comparison construction relative to arbitrary theories supplies countable countermodels in K R \mathcal{K}_{\mathrm{R}} K R (Theorem 2.5 ); a single full embedding transports each countermodel and all its premises to a standard chain. This yields strong standard completeness for H R \mathrm{H}_{\mathrm{R}} H R (Theorem 8.2 ). We also obtain effective density elimination for finite proofs without additional assumption leaves, retaining ordinary Cut (Theorem 8.6 ).
Sections 9 and 10 apply these constructions to IUL and HpsUL*. The applications require, respectively, commutative specializations and an involutive extension satisfying all histories, together with preservation of the original endpoints. The resulting bounded standard embeddings yield strong standard completeness for these logics using their ordinary chain-completeness theorems for arbitrary theories in the respective countable languages.
The connection between density elimination and standard completeness is established in the theory of substructural fuzzy logics. Metcalfe and Montagna developed this approach for uninorm logic and several of its extensions [1 ] . Ciabattoni and Metcalfe subsequently gave general sufficient conditions for density elimination by substitution [2 ] . These proof-theoretic criteria require hypotheses on the underlying calculus; they do not establish admissibility for an arbitrary extension by additional structural rules. Our use of ordinary Cut and the entire family of ordered history rules therefore requires a separate argument.
The algebraic counterpart is densification: embedding a chain into a dense chain while preserving its operations and the defining conditions of the class. Metcalfe and Tsinakis show that every dense residuated chain is e e e -cyclic, and relate densification of semilinear varieties to generation by their dense chains as quasivarieties [3 Theorem 1 and Lemma 1] . Baldi and Terui obtain densification through residuated frames for specified integral varieties and commutative varieties [4 ] ; Galatos and Horčík give an explicit one-gap construction for commutative residuated chains [5 Theorems 3.5–3.6] . Preservation of an existing residuated reduct does not by itself make the old designated element dualizing on the extension. This is a separate requirement in the present problem.
The closest proof-theoretic precedents include Wang’s published density-elimination arguments for involutive uninorm logic and for the noncommutative system GpsUL{sup}* [6 , 7 ] . The latter system adds weak commutativity at the monoid unit to pseudo-uninorm logic; its language and rules do not impose a cyclic dualizing element. Its density-elimination and standard-completeness statements are Theorems 4 and 5 of the published article. The former paper concerns commutative involutive logic. Jenei’s later discussion [8 Introduction] records reservations about the verification of its proof. We neither use that proof as an input nor resolve those reservations here.
For commutative involutive chains, Jenei’s group representation [9 , 10 , 11 ] and subsequent densification theorem give a particularly close algebraic comparison. Every countable odd commutative involutive chain embeds into a countable dense chain of the same class [12 Corollary 17] ; the bounded version preserves the endpoints, and yields strong standard completeness of IUL{sup}fp [12 Corollary 19 and Theorem 20] . Here odd means that the monoid unit and the dualizing constant coincide.
Table 1 distinguishes the input classes, constructions and preservation requirements. Our construction retains the full signature for the explicitly axiomatized class K R \mathcal{K}_{R} K R , allowing noncommutative products and d ≠ e d \ne e d = e ; its input conditions are essential to the representation statement.
The proof separates the chain by a closed central support term. The difficult lower branch is reduced to complete group fibres. Canonical completion attains their internal self-support maxima; pullback and a countable elementary substructure retain the exact original group. A lexicographic enlargement then makes that group dense. These preparation steps make it possible to insert points in all remaining gaps while preserving associativity, both actual residuals and all histories. The upper branch requires protected support intervals and a separate reconstruction. Detailed auxiliary construc-
caption: Table 1: Scope of the comparison with [12 Corollaries 17, 19 and Theorem 20] .
:::
tions and the full chain-completeness proof are supplied in Appendices § 12 –§ 15 , with explicit cross-references.
Machine-checked formalization. A companion Lean 4 development provides machine-checked proofs of the main representation and completeness results established here, together with an executable procedure for Den elimination. The source code and reproducibility materials are archived on Zenodo [13 ] . Section E specifies the verified scope, external inputs, and computational guarantees.
The calculus and its chain semantics
A cyclic involutive residuated lattice is a lattice-ordered monoid ( L , ∧ , ∨ , ⋅ , e ) (L,\wedge,\vee,\cdot,e) ( L , ∧ , ∨ , ⋅ , e ) with residuals satisfying
x y ≤ z ⟺ y ≤ x \ z ⟺ x ≤ z / y xy \le z \iff y \le x\backslash z \iff x \le z/y x y ≤ z ⟺ y ≤ x \ z ⟺ x ≤ z / y
and a distinguished element d d d such that x ′ = x \ d = d / x x' = x\backslash d = d/x x ′ = x \ d = d / x and x ′ ′ = x x'' = x x ′′ = x . We use the derived operation x + y = ( x ′ y ′ ) ′ x+y=(x'y')' x + y = ( x ′ y ′ ) ′ ; in particular,
e ′ = d , x \ y = ( y ′ x ) ′ = y + x ′ , y / x = ( x y ′ ) ′ = x ′ + y . (1) e' = d,\qquad x\backslash y=(y'x)'=y+x',\qquad y/x=(xy')'=x'+y. \tag*{(1)} e ′ = d , x \ y = ( y ′ x ) ′ = y + x ′ , y / x = ( x y ′ ) ′ = x ′ + y . (1)
A chain has a total lattice order. A standard chain has underlying ordered set [ 0 , 1 ] [0,1] [ 0 , 1 ] . Neither e e e nor d d d is required to be an endpoint, and multiplication need not be commutative. The class of all standard cyclic involutive residuated chains is denoted by Std.
Formulas are built from countably many variables, their formal duals, e e e , d d d , and ⋅ \cdot ⋅ , + + + , ∧ \wedge ∧ , ∨ \vee ∨ . Involution is extended as a syntactic operation by A ′ ′ = A A''=A A ′′ = A , e ′ = d e'=d e ′ = d , d ′ = e d'=e d ′ = e , and
( A B ) ′ = A ′ + B ′ , ( A + B ) ′ = A ′ B ′ , ( A ∧ B ) ′ = A ′ ∨ B ′ , ( A ∨ B ) ′ = A ′ ∧ B ′ . (AB)'=A'+B',\qquad(A+B)'=A'B',\qquad(A\wedge B)'=A'\vee B',\qquad(A\vee B)'=A'\wedge B'. ( A B ) ′ = A ′ + B ′ , ( A + B ) ′ = A ′ B ′ , ( A ∧ B ) ′ = A ′ ∨ B ′ , ( A ∨ B ) ′ = A ′ ∧ B ′ .
Thus the displayed order is not reversed under involution. Residual formulas abbreviate the terms in (1) . All substitution instances use this convention.
A word Γ \Gamma Γ is a finite ordered list of formulas, possibly empty. A sequent is ⇒ Γ \Rightarrow\Gamma ⇒ Γ , and a hypersequent is a finite multiset of sequents, written with external bars. External permutation is implicit; internal exchange is not. Capital Latin letters G G G , H H H , J J J denote hypersequents, and Greek capitals denote words. A missing external background denotes the empty multiset. All words in the following rules may be empty.
The initial sequents and logical rules are
⇒ A , A ′ I d ⇒ e U n i t , G ∣ ⇒ A , Γ H ∣ ⇒ B , Δ G ∣ H ∣ ⇒ A B , Γ , Δ ( ⋅ ) , G ∣ ⇒ Γ , A , B , Δ G ∣ ⇒ Γ , A + B , Δ ( + ) G ∣ ⇒ Γ , Δ G ∣ ⇒ Γ , d , Δ ( d ) , G ∣ ⇒ Γ , A , Δ G ∣ ⇒ Γ , B , Δ G ∣ ⇒ Γ , A ∧ B , Δ ( ∧ ) , G ∣ ⇒ Γ , A , Δ G ∣ ⇒ Γ , A ∨ B , Δ ( ∨ 1 ) G ∣ ⇒ Γ , B , Δ G ∣ ⇒ Γ , A ∨ B , Δ ( ∨ 2 ) . (2) \begin{aligned}
&\Rightarrow A,A' \quad\mathrm{Id}
&&\Rightarrow e \quad\mathrm{Unit},\\
&\frac{G\mid\Rightarrow A,\Gamma\qquad H\mid\Rightarrow B,\Delta}
{G\mid H\mid\Rightarrow AB,\Gamma,\Delta}\quad(\cdot),\\
&\frac{G\mid\Rightarrow\Gamma,A,B,\Delta}
{G\mid\Rightarrow\Gamma,A+B,\Delta}\quad(+)
&&\frac{G\mid\Rightarrow\Gamma,\Delta}
{G\mid\Rightarrow\Gamma,d,\Delta}\quad(d),\\
&\frac{G\mid\Rightarrow\Gamma,A,\Delta\qquad G\mid\Rightarrow\Gamma,B,\Delta}
{G\mid\Rightarrow\Gamma,A\wedge B,\Delta}\quad(\wedge),\\
&\frac{G\mid\Rightarrow\Gamma,A,\Delta}
{G\mid\Rightarrow\Gamma,A\vee B,\Delta}\quad(\vee_1)
&&\frac{G\mid\Rightarrow\Gamma,B,\Delta}
{G\mid\Rightarrow\Gamma,A\vee B,\Delta}\quad(\vee_2). \tag*{(2)}
\end{aligned} ⇒ A , A ′ Id G ∣ H ∣⇒ A B , Γ , Δ G ∣⇒ A , Γ H ∣⇒ B , Δ ( ⋅ ) , G ∣⇒ Γ , A + B , Δ G ∣⇒ Γ , A , B , Δ ( + ) G ∣⇒ Γ , A ∧ B , Δ G ∣⇒ Γ , A , Δ G ∣⇒ Γ , B , Δ ( ∧ ) , G ∣⇒ Γ , A ∨ B , Δ G ∣⇒ Γ , A , Δ ( ∨ 1 ) ⇒ e Unit , G ∣⇒ Γ , d , Δ G ∣⇒ Γ , Δ ( d ) , G ∣⇒ Γ , A ∨ B , Δ G ∣⇒ Γ , B , Δ ( ∨ 2 ) . (2)
The context of ( ∧ ) (\wedge) ( ∧ ) is shared literally. The structural rules and ordinary Cut are
G ⇒ Γ , Δ G ⇒ Δ , Γ C y c G G ∣ S E W G ∣ S ∣ S G ∣ S E C , \frac{G \Rightarrow\Gamma,\Delta}{G \Rightarrow\Delta,\Gamma}\ \mathrm{Cyc}
\qquad
\frac{G}{G\mid S}\ \mathrm{EW}
\qquad
\frac{G\mid S\mid S}{G\mid S}\ \mathrm{EC}, G ⇒ Δ , Γ G ⇒ Γ , Δ Cyc G ∣ S G EW G ∣ S G ∣ S ∣ S EC ,
G ⇒ Γ 1 , Γ 2 H ⇒ Δ 1 , Δ 2 G ∣ H ⇒ Γ 1 , Δ 2 ∣ ⇒ Δ 1 , Γ 2 C O M , (3) \frac{G\Rightarrow\Gamma_1,\Gamma_2\qquad H\Rightarrow\Delta_1,\Delta_2}{G\mid H\Rightarrow\Gamma_1,\Delta_2\mid\Rightarrow\Delta_1,\Gamma_2}\ \mathrm{COM}, \tag*{(3)} G ∣ H ⇒ Γ 1 , Δ 2 ∣⇒ Δ 1 , Γ 2 G ⇒ Γ 1 , Γ 2 H ⇒ Δ 1 , Δ 2 COM , (3)
G ⇒ Γ , Δ , e G ⇒ Δ , Γ , e W C M G ⇒ Γ , A H ⇒ A ′ , Δ G ∣ H ⇒ Γ , Δ C u t . \frac{G\Rightarrow\Gamma,\Delta,e}{G\Rightarrow\Delta,\Gamma,e}\ \mathrm{WCM}
\qquad
\frac{G\Rightarrow\Gamma,A\qquad H\Rightarrow A',\Delta}{G\mid H\Rightarrow\Gamma,\Delta}\ \mathrm{Cut}. G ⇒ Δ , Γ , e G ⇒ Γ , Δ , e WCM G ∣ H ⇒ Γ , Δ G ⇒ Γ , A H ⇒ A ′ , Δ Cut .
Here S S S is an arbitrary sequent. Cyclically rotating the two active premises and the two outputs of COM gives its equivalent presentation that exchanges a middle block in each premise. The displayed WCM retains its literal final formula e e e ; it is not unrestricted exchange.
For every finite word w w w over { S , T } \{S,T\} { S , T } (with empty word ϵ \epsilon ϵ ) define closed formulas by
( c ϵ , h ϵ ) = ( e , e ) , (c_\epsilon,h_\epsilon)=(e,e), ( c ϵ , h ϵ ) = ( e , e ) ,
( c w S , h w S ) = ( d c w , e + h w ) , ( c w T , h w T ) = ( e + c w , d h w ) . (4) (c_{wS},h_{wS})=(dc_w,e+h_w),\qquad(c_{wT},h_{wT})=(e+c_w,dh_w). \tag*{(4)} ( c w S , h w S ) = ( d c w , e + h w ) , ( c w T , h w T ) = ( e + c w , d h w ) . (4)
The remaining rules, one for every such w w w , are exactly
G ⇒ c w , Σ , Π H ⇒ h w , Ω , Λ J ⇒ Γ , Δ G ∣ H ∣ J ⇒ Π , Σ , Γ , Λ , Ω , Δ R w . (5) \frac{G\Rightarrow c_w,\Sigma,\Pi\qquad H\Rightarrow h_w,\Omega,\Lambda\qquad J\Rightarrow\Gamma,\Delta}{G\mid H\mid J\Rightarrow\Pi,\Sigma,\Gamma,\Lambda,\Omega,\Delta}\ R_w. \tag*{(5)} G ∣ H ∣ J ⇒ Π , Σ , Γ , Λ , Ω , Δ G ⇒ c w , Σ , Π H ⇒ h w , Ω , Λ J ⇒ Γ , Δ R w . (5)
We denote this calculus by H R \mathrm{H_R} H R . Its proofs are finite trees. There is no initial sequent ⇒ d \Rightarrow d ⇒ d , no internal weakening or contraction, and no density rule in H R \mathrm{H_R} H R . Ordinary Cut is retained.
Under a valuation in a chain, put Γ ♯ = A 1 ′ ⋯ A n ′ \Gamma^\sharp=A'_1\cdots A'_n Γ ♯ = A 1 ′ ⋯ A n ′ for Γ = A 1 , … , A n \Gamma=A_1,\ldots,A_n Γ = A 1 , … , A n , with ∅ ♯ = e \varnothing^\sharp=e ∅ ♯ = e . The sequent ⇒ Γ \Rightarrow\Gamma ⇒ Γ means Γ ♯ ≤ d \Gamma^\sharp\le d Γ ♯ ≤ d , and an external bar means disjunction. The empty hypersequent is therefore false. Let K R \mathcal{K}_R K R consist of the cyclic involutive residuated chains satisfying
a b ≤ e ⇒ b a ≤ e (6) ab\le e\ \Rightarrow\ ba\le e \tag*{(6)} ab ≤ e ⇒ ba ≤ e (6)
and, for every w ∈ { S , T } ∗ w\in\{S,T\}^* w ∈ { S , T } ∗ ,
a b ≤ c w , u v ≤ h w ⇒ b a z v u ≤ z ( a , b , u , v , z ∈ L ) . (7) ab\le c_w,\qquad uv\le h_w\ \Rightarrow\ bazvu\le z\qquad(a,b,u,v,z\in L). \tag*{(7)} ab ≤ c w , uv ≤ h w ⇒ ba z v u ≤ z ( a , b , u , v , z ∈ L ) . (7)
The history terms in these conditions are evaluated by (4). In particular, membership in K R \mathcal{K}_R K R requires every finite history, without a fixed bound on its length.
Lemma 2.1 (Rotation on a dense chain). Let L L L be a densely ordered chain. Suppose that f , g : L → L f,g:L\to L f , g : L → L are monotone, g g g has a right adjoint g ∗ g^{*} g ∗ , and f ( g ( t ) ) ≤ t f(g(t))\le t f ( g ( t )) ≤ t for all t t t . Then g ( f ( t ) ) ≤ t g(f(t))\le t g ( f ( t )) ≤ t for all t t t .
Proof. If g ( f ( x ) ) > x g(f(x))>x g ( f ( x )) > x , adjunction implies g ∗ ( x ) < f ( x ) g^{*}(x)<f(x) g ∗ ( x ) < f ( x ) . Choose g ∗ ( x ) < t < f ( x ) g^{*}(x)<t<f(x) g ∗ ( x ) < t < f ( x ) . Then g ( t ) > x g(t)>x g ( t ) > x , whence f ( g ( t ) ) ≥ f ( x ) > t f(g(t))\ge f(x)>t f ( g ( t )) ≥ f ( x ) > t , a contradiction.
Lemma 2.2. Every densely ordered cyclic involutive residuated chain belongs to K R \mathcal{K}_{R} K R .
Proof. If a b ≤ e ab\le e ab ≤ e , apply Lemma 2.1 to f ( t ) = a t f(t)=at f ( t ) = a t and g ( t ) = b t g(t)=bt g ( t ) = b t . Evaluating g ( f ( t ) ) ≤ t g(f(t))\le t g ( f ( t )) ≤ t at e e e gives b a ≤ e ba\le e ba ≤ e . Condition (6), with a , b a,b a , b interchanged, says precisely x \ e = e / x x\backslash e=e/x x \ e = e / x . By (1),
( d x ) ′ = x \ e = e / x = ( x d ) ′ . (dx)'=x\backslash e=e/x=(xd)'. ( d x ) ′ = x \ e = e / x = ( x d ) ′ .
so d d d is central. Consequently e + x = ( d x ′ ) ′ = d \ x e+x=(dx')'=d\backslash x e + x = ( d x ′ ) ′ = d \ x . Induction on w w w now gives
c w z h w ≤ z ( z ∈ L ) . (8) c_w z h_w\le z \qquad(z\in L). \tag*{(8)} c w z h w ≤ z ( z ∈ L ) . (8)
The empty history is immediate. For the two induction steps use centrality of d d d and d ( d \ x ) ≤ x d(d\backslash x)\le x d ( d \ x ) ≤ x to obtain
( d c w ) z ( d \ h w ) = c w z d ( d \ h w ) ≤ c w z h w ≤ z , (dc_w)z(d\backslash h_w)=c_w z d(d\backslash h_w)\le c_w z h_w\le z, ( d c w ) z ( d \ h w ) = c w z d ( d \ h w ) ≤ c w z h w ≤ z ,
( d \ c w ) z ( d h w ) = d ( d \ c w ) z h w ≤ c w z h w ≤ z . (d\backslash c_w)z(dh_w)=d(d\backslash c_w)zh_w\le c_wzh_w\le z. ( d \ c w ) z ( d h w ) = d ( d \ c w ) z h w ≤ c w z h w ≤ z .
No centrality of c w c_w c w or h w h_w h w is assumed. If a b ≤ c w ab\le c_w ab ≤ c w and u v ≤ h w uv\le h_w uv ≤ h w , then a b t u v ≤ c w t h w ≤ t ab tuv\le c_wth_w\le t ab t uv ≤ c w t h w ≤ t . Apply Lemma 2.1 to f ( t ) = a t v f(t)=atv f ( t ) = a t v and g ( t ) = b t u g(t)=btu g ( t ) = b t u ; the latter has right adjoint g ∗ ( s ) = ( b \ s ) / u g^{*}(s)=(b\backslash s)/u g ∗ ( s ) = ( b \ s ) / u . The resulting inequality is b a z v u ≤ z bazvu\le z ba z v u ≤ z , as required.
Theorem 2.3 (Soundness). Every rule of H R \mathrm{H}_{R} H R preserves validity in K R \mathcal{K}_{R} K R . In particular H R ⊢ H \mathrm{H}_{R}\vdash H H R ⊢ H implies S t d ⊨ H \mathrm{Std}\models H Std ⊨ H .
Proof. An active component ⇒ A , Γ \Rightarrow A,\Gamma ⇒ A , Γ is true exactly when Γ # ≤ A \Gamma^{\#}\le A Γ # ≤ A . Thus the two premises of ( ⋅ ) (\cdot) ( ⋅ ) give Γ # Δ # ≤ A B \Gamma^{\#}\Delta^{\#}\le AB Γ # Δ # ≤ A B . The ( + ) (+) ( + ) rule replaces A ′ B ′ A'B' A ′ B ′ by the equal value ( A + B ) ′ (A+B)' ( A + B ) ′ , and the ( d ) (d) ( d ) rule inserts the unit d ′ = e d'=e d ′ = e . Multiplication preserves binary joins in each argument, so ( A ∧ B ) ′ = A ′ ∨ B ′ (A\wedge B)'=A'\vee B' ( A ∧ B ) ′ = A ′ ∨ B ′ verifies ( ∧ ) (\wedge) ( ∧ ) . The join rules follow from monotonicity and order reversal of involution. Id and Unit hold because A A ′ ≤ d AA' \le d A A ′ ≤ d and e ′ = d e' = d e ′ = d . Cyc follows from x y ≤ d xy \le d x y ≤ d if and only if y x ≤ d yx \le d y x ≤ d .
For Cut, write γ = Γ ♯ \gamma= \Gamma^{\sharp} γ = Γ ♯ and δ = Δ ♯ \delta= \Delta^{\sharp} δ = Δ ♯ . The active premises give γ A ′ ≤ d \gamma A' \le d γ A ′ ≤ d , hence γ ≤ A \gamma\le A γ ≤ A , and A δ ≤ d A\delta\le d A δ ≤ d ; therefore γ δ ≤ d \gamma\delta\le d γ δ ≤ d . In WCM the active premise is γ δ d ≤ d \gamma\delta d \le d γ δ d ≤ d , equivalently γ δ ≤ e \gamma\delta\le e γ δ ≤ e , since d / d = e d/d = e d / d = e . Condition (6) gives the conclusion. For COM, put γ i = Γ i ♯ \gamma_i = \Gamma_i^{\sharp} γ i = Γ i ♯ and δ i = Δ i ♯ \delta_i = \Delta_i^{\sharp} δ i = Δ i ♯ . If γ 2 ≤ δ 2 \gamma_2 \le\delta_2 γ 2 ≤ δ 2 , then δ 1 γ 2 ≤ δ 1 δ 2 ≤ d \delta_1\gamma_2 \le\delta_1\delta_2 \le d δ 1 γ 2 ≤ δ 1 δ 2 ≤ d ; if δ 2 ≤ γ 2 \delta_2 \le\gamma_2 δ 2 ≤ γ 2 , then γ 1 δ 2 ≤ γ 1 γ 2 ≤ d \gamma_1\delta_2 \le\gamma_1\gamma_2 \le d γ 1 δ 2 ≤ γ 1 γ 2 ≤ d . One conclusion component is therefore true.
For R w R_w R w , put ( a , b , u , v , r , s ) = ( Σ ♯ , Π ♯ , Ω ♯ , Λ ♯ , Γ ♯ , Δ ♯ ) (a,b,u,v,r,s) = (\Sigma^{\sharp},\Pi^{\sharp},\Omega^{\sharp},\Lambda^{\sharp},\Gamma^{\sharp},\Delta^{\sharp}) ( a , b , u , v , r , s ) = ( Σ ♯ , Π ♯ , Ω ♯ , Λ ♯ , Γ ♯ , Δ ♯ ) . The three active premises give a b ≤ c w ab \le c_w ab ≤ c w , u v ≤ h w uv \le h_w uv ≤ h w , and r s ≤ d rs \le d r s ≤ d . Condition (7) gives v ‾ u s ≤ r s ≤ d \overline{v}us \le rs \le d v u s ≤ r s ≤ d , exactly the ordered conclusion.
If a premise background is true, its inherited copy makes the conclusion true. For shared ( ∧ ) (\wedge) ( ∧ ) this observation uses its one common background. EW and EC preserve external disjunction. This proves soundness in K R K_R K R ; standard soundness follows from Lemma 2.2.
Countable chain completeness
Definition 2.4 (Derivability and satisfaction with assumptions). For any set T T T of formulas, write T ⊢ H R H T \vdash_{\mathrm{HR}} H T ⊢ HR H when the finite hypersequent H H H has a finite H R \mathrm{HR} HR proof allowing ⇒ A \Rightarrow A ⇒ A , for A ∈ T A \in T A ∈ T , as additional leaves. These are specified sequents, not substitution schemata; repeated uses are allowed. No internal structural rule is added. For a valuation v v v in a chain C C C , write v ⊨ T v \models T v ⊨ T if e C ≤ v ( A ) e_C \le v(A) e C ≤ v ( A ) for every A ∈ T A \in T A ∈ T . This agrees with the sequent interpretation, since v ( A ) ′ ≤ d C v(A)' \le d_C v ( A ) ′ ≤ d C if and only if e C ≤ v ( A ) e_C \le v(A) e C ≤ v ( A ) . For a class C C C of chains, write T ⊨ C H T \models_C H T ⊨ C H if every valuation in every member of C C C that satisfies T T T also satisfies H H H . For a formula conclusion we abbreviate T ⊢ H R ⇒ A T \vdash_{\mathrm{HR}} \Rightarrow A T ⊢ HR ⇒ A by T ⊢ H R A T \vdash_{\mathrm{HR}} A T ⊢ HR A . All theories in this paper belong to the fixed countable propositional language; they need not be finite or decidable.
Theorem 2.5 (Countable chain completeness with assumptions). Let T T T be any set of formulas and let H H H be a finite hypersequent. If T ⊬ H R H T \nvdash_{\mathrm{HR}} H T ⊬ HR H , there are a countable chain C ∈ K R C \in K_R C ∈ K R and a valuation v v v in C C C such that v ⊨ T v \models T v ⊨ T and v ⊭ H v \not\models H v ⊨ H .
Proof outline. Start with the specified leaves T 0 = { ⇒ A : A ∈ T } T_0 = \{\Rightarrow A : A \in T\} T 0 = { ⇒ A : A ∈ T } . The continuation lemma lifts a proof from an additional leaf S S S over any background J J J for which J ∣ S J \mid S J ∣ S is proved. COM on two identities proves ⇒ A ′ , B ∣ ⇒ B ′ , A \Rightarrow A', B \mid\Rightarrow B', A ⇒ A ′ , B ∣⇒ B ′ , A . If adjoining either component made H H H derivable, two continuation substitutions and EC would already prove H H H . Thus one orientation of each formula pair can be adjoined while preserving underivability. Enumerating the pairs yields a total comparison preorder whose formula quotient is a countable cyclic involutive residuated chain. WCM and the actual R w R_w R w instances verify unit cyclicity and every ordered history. Each assumption leaf ⇒ A \Rightarrow A ⇒ A yields ⇒ d \Rightarrow d ⇒ d , A ⇒ e ′ , A A \mathrel{\Rightarrow} e', A A ⇒ e ′ , A , so [ e ] ≤ [ A ] [e] \leq[A] [ e ] ≤ [ A ] in the quotient. A true component of H H H would unfold by Cut and the logical rules to a proof of that component, hence of H H H by EW. The canonical valuation therefore satisfies T T T and falsifies H H H . Appendix A supplies the full continuation, quotient, and truth arguments.
Theorem 2.6 (Countable chain completeness). If a finite hypersequent H H H is not derivable in H R \mathrm{HR} HR , then it is false under a valuation in a countable member of K R \mathcal{K}_{R} K R .
Proof. Take T = ∅ T = \varnothing T = ∅ in Theorem 2.5.
Completion and preservation of the history conditions
An embedding is called full if it preserves the lattice operations, multiplication, both residuals, involution, and the two designated constants. For chains, an order embedding preserving multiplication, involution and the constants is full, by the residual formulas (1).
We use the involutive Dedekind–MacNeille completion theorem [14 Corollary 4.6] ; the published treatment is [15 ] . It supplies a complete cyclic involutive residuated lattice and a full embedding of the original algebra. We write DM ( C ) \operatorname{DM}(C) DM ( C ) for the resulting completion of C C C , equipped with these extended algebraic operations. In the chain case DM ( C ) \operatorname{DM}(C) DM ( C ) is again a chain. The additional history conditions require the following verification.
Theorem 3.1 (Full completion). If C ∈ K R C \in\mathcal{K}_{R} C ∈ K R , then DM ( C ) ∈ K R \operatorname{DM}(C) \in\mathcal{K}_{R} DM ( C ) ∈ K R , and the natural embedding is full.
Proof. Identify C C C with its image in C ^ = DM ( C ) \widehat{C} = \operatorname{DM}(C) C = DM ( C ) . Every x ∈ C ^ x \in\widehat{C} x ∈ C is the join of its old lower approximants:
x = ⋁ { x 0 ∈ C : x 0 ≤ x } . x = \bigvee\{x_0 \in C : x_0 \leq x\}. x = ⋁ { x 0 ∈ C : x 0 ≤ x } .
Multiplication in a complete residuated lattice preserves arbitrary joins separately in each coordinate. If a b ≤ e ab \leq e ab ≤ e , then old a 0 ≤ a a_0 \leq a a 0 ≤ a and b 0 ≤ b b_0 \leq b b 0 ≤ b satisfy a 0 b 0 ≤ e a_0b_0 \leq e a 0 b 0 ≤ e , hence b 0 a 0 ≤ e b_0a_0 \leq e b 0 a 0 ≤ e . Taking the two independent joins gives b a ≤ e ba \leq e ba ≤ e . Fix a finite history w w w , and suppose a b ≤ c w ab \leq c_w ab ≤ c w and u v ≤ h w uv \leq h_w uv ≤ h w in C ^ \widehat{C} C . Fullness preserves the values of the closed guards. For old lower approximants of a , b , u , v , z a,b,u,v,z a , b , u , v , z , the original history gives
b 0 a 0 z 0 v 0 u 0 ≤ z 0 ≤ z . b_0a_0z_0v_0u_0 \leq z_0 \leq z. b 0 a 0 z 0 v 0 u 0 ≤ z 0 ≤ z .
Taking the five independent joins yields b a z v u ≤ z bazvu \leq z ba z v u ≤ z . If an approximating set is empty, its join is the bottom element, and the corresponding product is bottom by complete join preservation. Thus the same argument also covers that case. Each finite history is preserved, with no bound on its length.
Corollary 3.2 (Standard completion). Every nontrivial countable densely ordered member of K R \mathcal{K}_R K R has a full embedding into a member of Std ∩ K R \operatorname{Std} \cap\mathcal{K}_R Std ∩ K R .
Proof. Its MacNeille completion is a complete densely ordered chain with endpoints and a countable order-dense subset. It is therefore order-isomorphic to [ 0 , 1 ] [0,1] [ 0 , 1 ] . Transport the operations along such an isomorphism and apply Theorem 3.1.
Completing an arbitrary chain does not remove its adjacent pairs. Consequently the essential task below is to construct full dense extensions before applying this corollary.
Central support and reconstruction
We write e e e for the monoid identity and d d d for the cyclic dualizer. Thus x ′ = x \ d = d / x x' = x\backslash d = d/x x ′ = x \ d = d / x is an antitone involution, and
x \ y = ( y ′ x ) ′ , y / x = ( x y ′ ) ′ . x\backslash y = (y'x)', \qquad y/x = (xy')'. x \ y = ( y ′ x ) ′ , y / x = ( x y ′ ) ′ .
An embedding is full if it preserves this entire signature, including the two constants. Preservation of order, multiplication, prime and the constants already implies fullness. We use the history constants and the class K R \mathcal{K}_R K R defined in the preceding section. Only unit cyclicity is needed until histories are explicitly invoked in the reconstruction argument.
Lemma 4.1 (Centrality and the history bounds). Unit cyclicity x y ≤ e ⇒ y x ≤ e xy \leq e \Rightarrow yx \leq e x y ≤ e ⇒ y x ≤ e is equivalent to centrality of d d d . Under these equivalent conditions,
e + x = d \ x , c w z h w ≤ z ( w ∈ { S , T } ∗ , z ∈ L ) . e+x=d\backslash x,\qquad c_wzh_w \leq z \quad(w \in\{S,T\}^*,\ z \in L). e + x = d \ x , c w z h w ≤ z ( w ∈ { S , T } ∗ , z ∈ L ) .
Proof. Unit cyclicity in both directions says x \ e = e / x x\backslash e=e/x x \ e = e / x . The residual formulas identify these elements with ( d x ) ′ (dx)' ( d x ) ′ and ( x d ) ′ (xd)' ( x d ) ′ , respectively, proving the equivalence. Centrality gives e + x = ( d x ′ ) ′ = ( x ′ d ) ′ = d \ x e+x=(dx')'=(x'd)'=d\backslash x e + x = ( d x ′ ) ′ = ( x ′ d ) ′ = d \ x . If c w z h w ≤ z c_w z h_w\le z c w z h w ≤ z , then
( d c w ) z ( d \ h w ) = c w z d ( d \ h w ) ≤ c w z h w . (dc_w)z(d\backslash h_w)=c_w z d(d\backslash h_w)\le c_w z h_w. ( d c w ) z ( d \ h w ) = c w z d ( d \ h w ) ≤ c w z h w .
and
( d \ c w ) z ( d h w ) = d ( d \ c w ) z h w ≤ c w z h w . (d\backslash c_w)z(dh_w)=d(d\backslash c_w)zh_w\le c_wzh_w. ( d \ c w ) z ( d h w ) = d ( d \ c w ) z h w ≤ c w z h w .
These are the two induction steps; the empty history is equality. □
Theorem 4.2 (The closed central support). In a unit-cyclic cyclic involutive residuated chain put
t = d 3 \ d 2 , p = d 2 t 2 , q = p d = d 2 t , Q = p t = d 2 t 3 . t=d^3\backslash d^2,\qquad p=d^2t^2,\qquad q=pd=d^2t,\qquad Q=pt=d^2t^3. t = d 3 \ d 2 , p = d 2 t 2 , q = p d = d 2 t , Q = pt = d 2 t 3 . Then p , q , Q p,q,Q p , q , Q are central and
d 3 t = d 2 , p 2 = p , I : = p L = L p = d 2 L , d^3t=d^2,\qquad p^2=p,\qquad I:=pL=Lp=d^2L, d 3 t = d 2 , p 2 = p , I := p L = L p = d 2 L , q Q = Q q = p , Q d = p , q 2 = d 2 . qQ=Qq=p,\qquad Qd=p,\qquad q^2=d^2. q Q = Qq = p , Q d = p , q 2 = d 2 . The set I I I is a monoid with identity p p p . Multiplication by d d d on I I I is an order automorphism, with inverse multiplication by t t t . In particular q , Q q,Q q , Q are mutually inverse central units of I I I .
Proof. Set h = ( d 2 ) ′ = d \ e h=(d^2)'=d\backslash e h = ( d 2 ) ′ = d \ e . The residual formulas and centrality give t = ( d 3 h ) ′ t=(d^3h)' t = ( d 3 h ) ′ . For
b = d 3 ( d 3 h ) = ( d 3 t ) ′ b=d^3(d^3h)=(d^3t)' b = d 3 ( d 3 h ) = ( d 3 t ) ′
we have h ≤ b h\le b h ≤ b and d 3 b = d 3 h d^3b=d^3h d 3 b = d 3 h . Also d h ≤ e dh\le e d h ≤ e . If d b ≤ e db\le e d b ≤ e , then b ≤ h b\le h b ≤ h , and hence b = h b=h b = h . If e ≤ d b e\le db e ≤ d b , then
d 2 ≤ d 3 b = d 3 h = d 2 ( d h ) ≤ d 2 . d^2\le d^3b=d^3h=d^2(dh)\le d^2. d 2 ≤ d 3 b = d 3 h = d 2 ( d h ) ≤ d 2 .
Consequently t = ( d 3 h ) ′ = h t=(d^3h)'=h t = ( d 3 h ) ′ = h and b = ( d 3 t ) ′ = h b=(d^3t)'=h b = ( d 3 t ) ′ = h in this case too. Involution now yields d 3 t = d 2 d^3t=d^2 d 3 t = d 2 .
Multiplying this equality by d t dt d t gives d 4 t 2 = d 2 d^4t^2=d^2 d 4 t 2 = d 2 ; thus p d 2 = d 2 pd^2=d^2 p d 2 = d 2 and p 2 = d 4 t 4 = p p^2=d^4t^4=p p 2 = d 4 t 4 = p . Both inclusions in p L = d 2 L pL=d^2L p L = d 2 L follow from these equations. On this set, d t ( d 2 a ) = d 2 a dt(d^2a)=d^2a d t ( d 2 a ) = d 2 a , so the monotone maps given by d d d and t t t are mutual inverses. In particular multiplication by d 2 d^2 d 2 is injective there. Since d d d is central by Lemma 4.1, so is d 2 d^2 d 2 . For any a ∈ L a\in L a ∈ L we have
p a = d 2 ( t 2 a ) , a p = a ( d 2 t 2 ) = d 2 ( a t 2 ) , pa=d^2(t^2a),\qquad ap=a(d^2t^2)=d^2(at^2), p a = d 2 ( t 2 a ) , a p = a ( d 2 t 2 ) = d 2 ( a t 2 ) ,
so both p a pa p a , a p ap a p belong to d 2 L d^{2}L d 2 L . Moreover, d 2 p = p d 2 = d 2 d^{2}p=pd^{2}=d^{2} d 2 p = p d 2 = d 2 , and centrality of d 2 d^{2} d 2 gives
d 2 ( p a ) = d 2 a = d 2 ( a p ) . d^{2}(pa)=d^{2}a=d^{2}(ap). d 2 ( p a ) = d 2 a = d 2 ( a p ) .
Thus p a = a p pa=ap p a = a p , proving centrality of p p p , and then of q = p d q=pd q = p d . The displayed equations for q , Q , d q,Q,d q , Q , d follow by multiplication. Finally Q a , a Q ∈ I Qa,aQ\in I Q a , a Q ∈ I and q ( Q a ) = p a = a p = q ( a Q ) q(Qa)=pa=ap=q(aQ) q ( Q a ) = p a = a p = q ( a Q ) . Since q q q acts invertibly on I I I , Q a = a Q Qa=aQ Q a = a Q . All cancellation in this proof occurs inside I I I .
The following formulation also applies to the alternative interior support arising in the upper branch.
Proposition 4.3 (Interior projection and exact fibres). Let r ≤ e r\le e r ≤ e be a central idempotent of a cyclic involutive residuated chain L L L . The map
π : L ⟶ J = r L , π ( x ) = r x \pi:L\longrightarrow J=rL,\qquad\pi(x)=rx π : L ⟶ J = r L , π ( x ) = r x is a full surjective homomorphism when J J J is given identity r r r , dualizer q = r d q=rd q = r d , and involution a ∗ = r a ′ a^{*}=ra' a ∗ = r a ′ . In particular,
( r x ) \ J ( r y ) = r ( x \ y ) , ( r y ) / J ( r x ) = r ( y / x ) , ( r x ) ∗ = r x ′ . (rx)\mathbin{\backslash}_{J}(ry)=r(x\backslash y),\qquad(ry)/_{J}(rx)=r(y/x),\qquad(rx)^{*}=rx'. ( r x ) \ J ( r y ) = r ( x \ y ) , ( r y ) / J ( r x ) = r ( y / x ) , ( r x ) ∗ = r x ′ . Every fibre has the exact form
F a : = π − 1 ( a ) = [ a , ( a ∗ ) ′ ] L . (9) F_a:=\pi^{-1}(a)=[a,(a^{*})']_{L}. \tag*{(9)} F a := π − 1 ( a ) = [ a , ( a ∗ ) ′ ] L . (9) If q q q is a central unit of J J J , every nonunit of J J J has a singleton fibre.
Proof. For z ∈ J z\in J z ∈ J we have ( r x ) z = x z (rx)z=xz ( r x ) z = x z . If this is at most r y ry r y , then x z ≤ y xz\le y x z ≤ y , whence z = r z ≤ r ( x \ y ) z=rz\le r(x\backslash y) z = r z ≤ r ( x \ y ) . Conversely, z ≤ r ( x \ y ) z\le r(x\backslash y) z ≤ r ( x \ y ) gives x z ≤ r y xz\le ry x z ≤ r y . This proves the left residual formula; reversing the factor positions proves the right one. Products, minima, maxima and the constants are preserved. The residual formulas give ( r x ) ∗ = r x ′ (rx)^{*}=rx' ( r x ) ∗ = r x ′ , so ∗ \ast ∗ is involutive and cyclic. If r x = a rx=a r x = a , then a ≤ x a\le x a ≤ x and a ∗ = r x ′ ≤ x ′ a^{*}=rx'\le x' a ∗ = r x ′ ≤ x ′ , proving the two bounds in (9). Conversely its two endpoints project to a a a , and monotonicity proves the equality.
Here is the singleton assertion in a form useful below. For any surjective full homomorphism φ : B → A \varphi:B\to A φ : B → A of such chains, a a ′ < d A aa'<d_A a a ′ < d A forces φ − 1 ( a ) \varphi^{-1}(a) φ − 1 ( a ) to be a singleton: if u , w u,w u , w belong to that fibre, then φ ( u w ′ ) = a a ′ < φ ( d B ) \varphi(uw')=aa'<\varphi(d_B) φ ( u w ′ ) = a a ′ < φ ( d B ) , so u w ′ < d B uw'<d_B u w ′ < d B and u ≤ w u\le w u ≤ w ; interchange u , w u,w u , w . A nontrivial fibre over a ∈ J a\in J a ∈ J has a nontrivial complementary fibre. Therefore a a ∗ = a ∗ a = q aa^{*}=a^{*}a=q a a ∗ = a ∗ a = q . If q − 1 q^{-1} q − 1 is the local inverse of q q q , then q − 1 a ∗ q^{-1}a^{*} q − 1 a ∗ is a two-sided inverse of a a a in J J J . This proves the assertion without any assumption concerning one-sided inverses.
Lemma 4.4 (The complete group core). Under the hypotheses of Proposition 4.3 , suppose q q q is a central unit. Write U ( J ) U(J) U ( J ) for the group of units of the monoid J J J with its local identity r r r , and put
G = U ( J ) , K = π − 1 ( G ) . G = U(J), \qquad K = \pi^{-1}(G). G = U ( J ) , K = π − 1 ( G ) . Then K K K is a full subalgebra of L L L ,
r K = G , L = J ∪ K , J ∩ K = G , rK = G, \qquad L = J \cup K, \qquad J \cap K = G, r K = G , L = J ∪ K , J ∩ K = G , and, for a ∈ J a \in J a ∈ J and x ∈ L x \in L x ∈ L ,
a x = a π ( x ) , x a = π ( x ) a . (10) ax = a\pi(x), \qquad xa = \pi(x)a. \tag*{(10)} a x = aπ ( x ) , x a = π ( x ) a . (10) If L ∈ K R L \in\mathcal{K}_{R} L ∈ K R , then K ∈ K R K \in\mathcal{K}_{R} K ∈ K R .
Proof. The group G G G is a full subalgebra of J J J with its local constants: q ∗ = q q − 1 q^{*} = q q^{-1} q ∗ = q q − 1 , g \ J h = g − 1 h g\mathbin{\backslash_{J}}h = g^{-1}h g \ J h = g − 1 h and h / J g = h g − 1 h\mathbin{/_{J}}g = hg^{-1} h / J g = h g − 1 . Minima and maxima select an argument. Thus its inverse image is full and contains e , d e,d e , d . Singleton nonunit fibres give the union and intersection assertions. Since a = r a = a r a = ra = ar a = r a = a r , centrality gives (10). Universal quasi-identities, including all histories, pass to subalgebras.
Definition 4.5 (Complete group core). Under the hypotheses of Lemma 4.4 , the complete group core of L L L at r r r is
K r ( L ) = π − 1 ( U ( r L ) ) , π ( x ) = r x . K_r(L) = \pi^{-1}\bigl(U(rL)\bigr), \qquad\pi(x) = rx. K r ( L ) = π − 1 ( U ( r L ) ) , π ( x ) = r x . Thus every fibre above a unit of the local algebra r L rL r L is included, with its original operations. The word complete refers to this inclusion of entire fibres; it does not mean Dedekind completeness. Lemma 4.4 says that this inverse image is a full subalgebra, rather than a choice of one representative from each fibre.
For the lower branch p ≤ e p \le e p ≤ e , the support and interior results apply with r = p r = p r = p and local dualizer q = p d q = pd q = p d , whose inverse is Q Q Q . Consequently the core at the closed support is exactly K p ( L ) = { x ∈ L : p x ∈ U ( p L ) } K_p(L) = \{x \in L : px \in U(pL)\} K p ( L ) = { x ∈ L : p x ∈ U ( p L )} .
Lemma 4.6 (Central unit dualizers and histories). Every cyclic involutive residuated chain with central invertible dualizer q q q belongs to K R \mathcal{K}_{R} K R . With local identity r r r its history constants are
c w = q k ( w ) , h w = q − k ( w ) , k ( w ) = # S ( w ) − # T ( w ) , c_w = q^{k(w)}, \qquad h_w = q^{-k(w)}, \qquad k(w) = \#S(w) - \#T(w), c w = q k ( w ) , h w = q − k ( w ) , k ( w ) = # S ( w ) − # T ( w ) , where # S ( w ) \#S(w) # S ( w ) and # T ( w ) \#T(w) # T ( w ) count the occurrences of S S S and $T in w w w , and powers use identity r r r .
Proof. The operation x ∘ = q − 1 x ∗ x^\circ= q^{-1}x^* x ∘ = q − 1 x ∗ makes r r r a cyclic dualizer, because
x y ≤ r ⟺ x ( q y ) ≤ q ⟺ y ≤ q − 1 x ∗ . xy \le r \Longleftrightarrow x(qy) \le q \Longleftrightarrow y \le q^{-1}x^*. x y ≤ r ⟺ x ( q y ) ≤ q ⟺ y ≤ q − 1 x ∗ .
For any central unit a a a , ( a x ) ∗ = a − 1 x ∗ (ax)^* = a^{-1}x^* ( a x ) ∗ = a − 1 x ∗ , directly from residuation. Hence ( x ∘ ) ∘ = x (x^\circ)^\circ= x ( x ∘ ) ∘ = x . The unit threshold is cyclic, and multiplication by q − n q^{-n} q − n gives
x y ≤ q n ⟺ y x ≤ q n . xy \le q^n \Longleftrightarrow yx \le q^n. x y ≤ q n ⟺ y x ≤ q n .
On the central units a + b = q − 1 a b a+b=q^{-1}ab a + b = q − 1 ab , so induction gives the history constants displayed above. Rotating both cap premises then gives b a z v u ≤ q k ( w ) z q − k ( w ) = z bazvu \le q^{k(w)}zq^{-k(w)} = z ba z v u ≤ q k ( w ) z q − k ( w ) = z .
Theorem 4.7 (Simultaneous core and support gluing). Let L ∈ K R L \in\mathcal{K}_{R} L ∈ K R have a central idempotent r ≤ e r \le e r ≤ e such that J = r L J=rL J = r L has central unit dualizer q = r d q=rd q = r d . Put G = U ( J ) G=U(J) G = U ( J ) and K = { x ∈ L : r x ∈ G } K=\{x \in L: rx \in G\} K = { x ∈ L : r x ∈ G } . Suppose:
K ↪ K ^ ∈ K R K \hookrightarrow\widehat{K} \in\mathcal{K}_{R} K ↪ K ∈ K R is full, r r r remains central in K ^ \widehat{K} K , and r K ^ = G r\widehat{K}=G r K = G as the old embedded group;
J ↪ T J \hookrightarrow T J ↪ T is full for the local constants r , q r,q r , q , and q q q remains a central unit in T T T .
Then L L L has a full extension B ∈ K R B \in\mathcal{K}_{R} B ∈ K R with underlying set
B = K ^ ∪ ˙ ( T ∖ G ) . B=\widehat{K}\mathbin{\dot\cup}(T\setminus G). B = K ∪ ˙ ( T ∖ G ) . The order replaces every g ∈ G g \in G g ∈ G in T T T by its entire K ^ \widehat{K} K fibre and leaves all other T T T points single. Multiplication by r r r projects B B B onto the floor copy of T T T . Both the given core extension and all its internal inserted points are retained. Countability is preserved.
Proof. Apply Proposition 4.3 in K ^ \widehat{K} K . Its projection π ^ ( k ) = r k \widehat{\pi}(k)=rk π ( k ) = r k is full onto G G G . Define Π : B → T \Pi:B\to T Π : B → T by π ^ \widehat{\pi} π on K ^ \widehat{K} K and the identity outside. Order first by Π \Pi Π , then by the old fibre order. The floor inclusion i : T → B i:T\to B i : T → B sends g ∈ G g\in G g ∈ G to its least fibre point g g g and is the identity elsewhere; hence
i Π ( x ) ≤ x . (11) i\Pi(x)\le x. \tag*{(11)} i Π ( x ) ≤ x . (11)
Keep multiplication on K ^ \widehat{K} K and otherwise put
x y = i ( Π ( x ) Π ( y ) ) . (12) xy=i\bigl(\Pi(x)\Pi(y)\bigr). \tag*{(12)} x y = i ( Π ( x ) Π ( y ) ) . (12)
For g ∈ G g \in G g ∈ G and k ∈ K ^ k \in\widehat{K} k ∈ K , centrality of r r r gives g k = g π ^ ( k ) gk = g\widehat{\pi}(k) g k = g π ( k ) and k g = π ^ ( k ) g kg = \widehat{\pi}(k)g k g = π ( k ) g . Consequently the uniform floor identities are
i ( a ) x = i ( a Π ( x ) ) , x i ( a ) = i ( Π ( x ) a ) . (13) i(a)x = i(a\Pi(x)), \qquad xi(a) = i(\Pi(x)a). \tag*{(13)} i ( a ) x = i ( a Π ( x )) , x i ( a ) = i ( Π ( x ) a ) . (13)
Every product containing an outside factor is therefore the floor of the product of all its projections, independently of bracketing. This proves associativity; products wholly in K ^ \widehat{K} K use its associativity. It also proves that the old e e e is the identity, since Π ( e ) = r \Pi(e) = r Π ( e ) = r .
For a fixed floor factor, monotonicity follows in T T T . A remaining factor k k k belongs to K ^ \widehat{K} K and has unit projection g g g . If the two varying arguments have distinct projections, left or right multiplication by g g g preserves their strict order. If their projections agree, the arguments are equal outside G G G , and inside G G G monotonicity is that of K ^ \widehat{K} K . Thus multiplication is monotone across every boundary as well as within each piece.
Keep prime on K ^ \widehat{K} K , and use the local prime ∗ * ∗ of T T T on T ∖ G T \setminus G T ∖ G . Both G G G and its complement are ∗ * ∗ -invariant: g ∗ = q g − 1 g^* = qg^{-1} g ∗ = q g − 1 belongs to G G G and ∗ * ∗ is an involution. The full projection π ^ \widehat{\pi} π shows that these definitions give an antitone involution on B B B . For a product with an outside factor,
x y ≤ d ⟺ Π ( x ) Π ( y ) ≤ q ⟺ Π ( y ) ≤ Π ( x ) ∗ . xy \le d \Longleftrightarrow\Pi(x)\Pi(y) \le q \Longleftrightarrow\Pi(y) \le\Pi(x)^*. x y ≤ d ⟺ Π ( x ) Π ( y ) ≤ q ⟺ Π ( y ) ≤ Π ( x ) ∗ .
The first equivalence uses that a floor i ( a ) i(a) i ( a ) is at most d d d exactly when a ≤ q a \le q a ≤ q . In the last comparison, if both bases are outside G G G they refer to singleton fibres; if just one is outside G G G equality is impossible. In either case it is exactly y ≤ x ∗ y \le x^* y ≤ x ∗ . For two inputs in K ^ \widehat{K} K this duality already holds. Associativity and cyclic duality now give both residuals by the involutive formulas. They also show that Π \Pi Π is full for the local constants of T T T .
Unit cyclicity holds inside K ^ \widehat{K} K . Otherwise both ordered products are floors, and the assertion follows from unit cyclicity in T T T and r ≤ e r \le e r ≤ e . All closed history constants project to their corresponding constants in T T T . If all five factors in a history instance lie in K ^ \widehat{K} K , use that instance there. Otherwise its conclusion product is a floor, and the projected instance, valid by Lemma 4.6 , gives
b a z v u = i ( Π ( b ) Π ( a ) Π ( z ) Π ( v ) Π ( u ) ) ≤ i Π ( z ) ≤ z . bazvu = i(\Pi(b)\Pi(a)\Pi(z)\Pi(v)\Pi(u)) \le i\Pi(z) \le z. ba z v u = i ( Π ( b ) Π ( a ) Π ( z ) Π ( v ) Π ( u )) ≤ i Π ( z ) ≤ z .
Thus B ∈ K R B \in\mathcal{K}_{R} B ∈ K R for every finite history.
The old chain is K ∪ ( J ∖ G ) K \cup(J \setminus G) K ∪ ( J ∖ G ) . Its mixed products agree with (12) by (10) , and its outside prime agrees with the local prime because those fibres are singletons. Its inclusion is therefore full. Finally r x = i Π ( x ) rx = i\Pi(x) r x = i Π ( x ) by (13) , proving all remaining assertions.
Taking T = J T=J T = J gives fixed-floor core gluing. Taking K ^ = K \widehat{K}=K K = K gives lifting of every full support extension that retains the central unit dualizer. Neither specialization requires that nonunits form an ideal: two outside elements may multiply to an old group element, and the resulting point is its floor, as prescribed by (12).
Theorem 4.8 (Densification at a central unit dualizer). Every countable cyclic involutive residuated chain with central invertible dualizer has a countable full dense extension without endpoints, retaining its dualizer as a central unit. Both chains belong to K R \mathrm{KR} KR .
Proof. Normalize its prime to x ∘ = d − 1 x ′ x^\circ=d^{-1}x' x ∘ = d − 1 x ′ as in Lemma 4.6. The resulting chain is odd. Corollary B.8, proved by the explicit construction in Appendix B, gives a countable dense odd extension without endpoints, in which the old d d d remains central with its inverse. Restore prime by x ′ = d x ∘ x'=dx^\circ x ′ = d x ∘ . The same normalization calculation in reverse proves cyclic duality and involutivity. Multiplication, order and both residuals are unchanged, and the restored prime agrees with the original prime on the old chain. The embedding is full. Lemma 4.6 gives all histories in the source and target.
Example 4.9 (A noncommutative, nonodd central-unit chain). Let H = Z 3 H=\mathbb{Z}^3 H = Z 3 with lexicographic order, in which the first coordinate is compared first, and multiplication
( a , b , c ) ( r , s , t ) = ( a + r , b + s , c + t + a s ) . (a,b,c)(r,s,t)=(a+r,b+s,c+t+as). ( a , b , c ) ( r , s , t ) = ( a + r , b + s , c + t + a s ) .
This is a countable group with
e = ( 0 , 0 , 0 ) , ( a , b , c ) − 1 = ( − a , − b , a b − c ) . e=(0,0,0), \qquad(a,b,c)^{-1}=(-a,-b,ab-c). e = ( 0 , 0 , 0 ) , ( a , b , c ) − 1 = ( − a , − b , ab − c ) .
Associativity follows by comparing the third coordinates of both bracketings: for a third factor ( u , v , w ) (u,v,w) ( u , v , w ) , both are c + t + w + a s + a v + r v c+t+w+as+av+rv c + t + w + a s + a v + r v . The order is invariant under both left and right translations. Indeed, translation preserves the first unequal coordinate; if the first two coordinates coincide, the same quantity is added to the third coordinate of both elements. Thus H H H is a totally ordered residuated group, with
x \ y = x − 1 y , y / x = y x − 1 . x\backslash y=x^{-1}y,\qquad y/x=yx^{-1}. x \ y = x − 1 y , y / x = y x − 1 .
Choose the central element d = ( 0 , 0 , 1 ) d=(0,0,1) d = ( 0 , 0 , 1 ) as dualizer. Then
( a , b , c ) ′ = d ( a , b , c ) − 1 = ( − a , − b , 1 + a b − c ) . (a,b,c)'=d(a,b,c)^{-1}=(-a,-b,1+ab-c). ( a , b , c ) ′ = d ( a , b , c ) − 1 = ( − a , − b , 1 + ab − c ) .
is an antitone involution, and x \ d = x − 1 d = d x − 1 = d / x x\backslash d=x^{-1}d=dx^{-1}=d/x x \ d = x − 1 d = d x − 1 = d / x . Hence this is a cyclic involutive residuated chain, and it is nonodd because d ≠ e d\ne e d = e . Its support terms satisfy
t = d − 1 = ( 0 , 0 , − 1 ) , p = e , q = d , Q = d − 1 . t=d^{-1}=(0,0,-1),\qquad p=e,\qquad q=d,\qquad Q=d^{-1}. t = d − 1 = ( 0 , 0 , − 1 ) , p = e , q = d , Q = d − 1 .
It belongs to K R \mathcal{K}_{R} K R by Lemma 4.6, with c w = d k ( w ) c_w=d^{k(w)} c w = d k ( w ) , h w = d − k ( w ) h_w=d^{-k(w)} h w = d − k ( w ) and k ( w ) = # S ( w ) − # T ( w ) k(w)=\#S(w)-\#T(w) k ( w ) = # S ( w ) − # T ( w ) .
For x = ( 1 , 0 , 0 ) x=(1,0,0) x = ( 1 , 0 , 0 ) and y = ( 0 , 1 , 0 ) y=(0,1,0) y = ( 0 , 1 , 0 ) , direct calculation gives
x y = ( 1 , 1 , 1 ) , y x = ( 1 , 1 , 0 ) , xy=(1,1,1),\qquad yx=(1,1,0), x y = ( 1 , 1 , 1 ) , y x = ( 1 , 1 , 0 ) ,
and
x \ y = ( − 1 , 1 , − 1 ) , y / x = ( − 1 , 1 , 0 ) . x\backslash y=(-1,1,-1),\qquad y/x=(-1,1,0). x \ y = ( − 1 , 1 , − 1 ) , y / x = ( − 1 , 1 , 0 ) .
Thus the left and right multiplication actions, and the two residual operations, differ even in the central-unit case p = e p=e p = e . Both local identities of every group element are still x / x = x \ x = e x/x=x\backslash x=e x / x = x \ x = e .
Corollary 4.10 (Reconstruction from dense group fibres). Let L ∈ K R L\in\mathcal{K}_{R} L ∈ K R be countable, and let r ≤ e r\le e r ≤ e be a central idempotent such that q = r d q=rd q = r d is a central unit of J = r L J=rL J = r L . Put G = U ( J ) G=U(J) G = U ( J ) and K = { x : r x ∈ G } K=\{x:rx\in G\} K = { x : r x ∈ G } . Suppose K K K has a countable full extension K ^ ∈ K R \widehat{K}\in\mathcal{K}_{R} K ∈ K R in which r r r is central, r K ^ = G r\widehat{K}=G r K = G , and every fibre of multiplication by r r r is densely ordered, with singleton fibres allowed. Then L L L has a countable full dense extension without endpoints, and hence a full standard embedding.
Proof. Theorem 4.8 gives a countable dense local extension J ↪ T J\hookrightarrow T J ↪ T without endpoints, retaining q q q as a central unit. Apply Theorem 4.7 to K ^ \widehat{K} K and T T T . For two resulting points with different projections, choose an intermediate projection in T T T and take its floor. Two points with the same projection lie in one of the given dense fibres; a singleton fibre contains no strict pair. There are no endpoints because T T T has none. This proves density and the remaining countability assertion. The full standard completion is supplied by Corollary 3.2.
Corollary 4.11 (Extracting a fixed-floor extension from a standard core). In the setting of Corollary 4.10, suppose the complete core K K K has a full embedding into a standard chain S ∈ K R S\in\mathcal{K}_{R} S ∈ K R in which the old r r r is central. Then L L L has a full standard embedding. In particular this applies to the lower branch r = p ≤ e r=p\le e r = p ≤ e , where centrality of the closed term p p p in S S S is automatic.
Proof. Identify K K K with its image. The old inverse equations make q = r d q=rd q = r d a central unit of r S rS r S . The group G G G is a full local subalgebra of r S rS r S , because its residuals are group division and its prime is g ∗ = q g − 1 g^*=qg^{-1} g ∗ = q g − 1 . By Proposition 4.3 ,
E = { s ∈ S : r s ∈ G } E=\{s\in S:rs\in G\} E = { s ∈ S : r s ∈ G }
is a full subalgebra containing K K K , with r E = G rE=G r E = G . Its fibre over g g g is the entire interval [ g , ( g ∗ ) ] S [g,(g^*)]_S [ g , ( g ∗ ) ] S . Both endpoints belong to the old core. Hence every strict pair in a fibre of E E E has an intermediate point in that fibre, by density of the standard chain. Starting with K K K , choose one such point for every strict same-fibre pair, take the full subalgebra generated by all these choices inside E E E , and repeat countably many times. The union K ‾ \overline{K} K is countable, full, has r K ‾ = G r\overline{K}=G r K = G , and all its fibres are densely ordered. Apply Corollary 4.10 . For r = p r=p r = p , centrality in S S S follows from Theorem 4.2 , since the embedding preserves the defining closed term.
Standard representation of countable lower complete cores
Definition 5.1 (Lower complete core). A member K ∈ K R K\in K_R K ∈ K R is a lower complete core if its closed support p p p satisfies p ≤ e p\le e p ≤ e and p K pK p K is a group with identity p p p . Equivalently, in this case K = K p ( K ) K=K_p(K) K = K p ( K ) . Countability is a separate hypothesis in the representation results.
We prove the representation theorem for a countable lower complete core K K K , and write G = p K G=pK G = p K . Throughout, embeddings preserve the entire original signature. We use the structural identities
p 2 = p ≤ e , p , d central , q = p d ∈ Z ( G ) , d 2 = q 2 (14) p^2=p\le e,\qquad p,d\text{ central},\qquad q=pd\in Z(G),\qquad d^2=q^2 \tag*{(14)} p 2 = p ≤ e , p , d central , q = p d ∈ Z ( G ) , d 2 = q 2 (14)
and the full interior projection π ( x ) = p x \pi(x)=px π ( x ) = p x . Its fibres are convex and the fibre of g ∈ G g\in G g ∈ G has least point g g g . On the group G G G , the identity is p p p , the dualizer is q q q , and g ∗ = q g − 1 g^*=qg^{-1} g ∗ = q g − 1 ; consequently
π ( x y ) = π ( x ) π ( y ) , π ( x − ) = π ( x ) − 1 , \pi(xy)=\pi(x)\pi(y),\qquad\pi(x^{-})=\pi(x)^{-1}, π ( x y ) = π ( x ) π ( y ) , π ( x − ) = π ( x ) − 1 ,
π ( x ′ ) = q π ( x ) − 1 , x − = x \ e = e / x . (15) \pi(x')=q\pi(x)^{-1},\qquad x^{-}=x\backslash e=e/x. \tag*{(15)} π ( x ′ ) = q π ( x ) − 1 , x − = x \ e = e / x . (15)
These facts follow from the structural results 4.1, 4.2, and 4.3. We also use support-base lifting (Theorem 4.7 ) and full completion (Theorem 3.1 ). All the gap insertion and preservation arguments needed here are proved explicitly.
We first construct one simultaneous gap insertion, assuming attained self-supports and internal gaps. The self-support maxima determine actual comparison bounds and hence the greatest cyclic product on the new carrier. This comparison construction alone does not prove associativity. The relative return collapse of Lemma 5.6 excludes the active polar pairs that would be forced by a failed associator in Theorem 5.7. After proving that these properties survive insertion, we supply the preparation that makes them available for an arbitrary complete core and iterate the construction.
The old actions on the split carrier
For an e e e -cyclic cyclic involutive residuated chain A A A , let D ( A ) D(A) D ( A ) be obtained by inserting exactly one point X a , b X_{a,b} X a , b in each adjacent pair a ≺ b a \prec b a ≺ b of A A A . Extend prime by X a , b ′ = X b ′ , a ′ X'_{a,b}=X_{b',a'} X a , b ′ = X b ′ , a ′ . Write r ( X a , b ) = b r(X_{a,b})=b r ( X a , b ) = b and r ( a ) = a r(a)=a r ( a ) = a for a ∈ A a\in A a ∈ A . The construction in this subsection requires neither endpoints nor a support group.
An old labelled context f = ( u , v ) f=(u,v) f = ( u , v ) means f ( x ) = u x v f(x)=uxv f ( x ) = uxv . An admissible return is a context R = ( s , t ) R=(s,t) R = ( s , t ) such that
s u , u s , v t , t v ≤ e . (16) su,\ us,\ vt,\ tv \le e. \tag*{(16)} s u , u s , v t , t v ≤ e . (16)
Then R f ≤ id Rf\le\operatorname{id} R f ≤ id and f R ≤ id fR\le\operatorname{id} f R ≤ id . We call f f f active at ( a , b ) (a,b) ( a , b ) if R f ( b ) = b Rf(b)=b R f ( b ) = b for an admissible return. Equivalently,
u − b v v − = b . (17) u^{-}bv v^{-}=b. \tag*{(17)} u − b v v − = b . (17)
Indeed any admissible s , t s,t s , t satisfy s ≤ u − s\le u^{-} s ≤ u − and t ≤ v − t\le v^{-} t ≤ v − , while ( u − , v − ) (u^{-},v^{-}) ( u − , v − ) is itself admissible.
Lemma 5.2 (The prescribed old actions). If f f f is active at ( a , b ) (a,b) ( a , b ) , then f ( b ) f(b) f ( b ) has an immediate predecessor c c c . The formulas
f ( X a , b ) = { X c , f ( b ) , f is active at ( a , b ) , f ( b ) , f is inactive at ( a , b ) , (18) f(X_{a,b}) =
\begin{cases}
X_{c,f(b)}, & f\text{ is active at }(a,b),\\
f(b), & f\text{ is inactive at }(a,b),
\end{cases} \tag*{(18)} f ( X a , b ) = { X c , f ( b ) , f ( b ) , f is active at ( a , b ) , f is inactive at ( a , b ) , (18) together with the old values define monotone actions on D ( A ) D(A) D ( A ) . They are monotone also in each coefficient, preserve composition of labelled contexts, and satisfy r ( f X ) = f ( r X ) r(fX)=f(rX) r ( f X ) = f ( r X ) . Each action has an actual right adjoint, whose value on an old target is the old right adjoint. In particular, with L u X = u X L_uX=uX L u X = u X and R u X = X u R_uX=Xu R u X = X u ,
L u L v = L u v , R v R u = R u v , L u R v = R v L u , L u ∗ ( Y ) = ( Y ′ u ) ′ , R u ∗ ( Y ) = ( u Y ′ ) ′ . (19) \begin{aligned}
L_uL_v&=L_{uv}, & R_vR_u&=R_{uv}, & L_uR_v&=R_vL_u,\\
L_u^*(Y)&=(Y'u)', & R_u^*(Y)&=(uY')'. \tag*{(19)}
\end{aligned} L u L v L u ∗ ( Y ) = L uv , = ( Y ′ u ) ′ , R v R u R u ∗ ( Y ) = R uv , = ( u Y ′ ) ′ . L u R v = R v L u , (19)
The old unit fixes every point.
The proof is given in Appendix C.1.
For a left action, activity is exactly u − u b = b u^{-}ub=b u − u b = b ; for a right action it is b u u − = b buu^{-}=b b u u − = b . A two-sided admissible return for a one-sided action can always be replaced by a one-sided return: if s u b t = b subt=b s u b t = b and t ≤ e t\leq e t ≤ e , then b ≤ s u b ≤ b b\leq sub\leq b b ≤ s u b ≤ b .
Attainment of coefficient comparisons
For a gap J J J define its positive self-support set by
S J = { u ∈ A : u ≥ e , u X J = X J } . (20) S_J=\{u\in A:u\geq e,\ uX_J=X_J\}. \tag*{(20)} S J = { u ∈ A : u ≥ e , u X J = X J } . (20)
We write (PC) for the condition that each S J S_J S J has a greatest member s J s_J s J . In old operations this condition says that { u ≥ e : u b = b , u − b = b } \{u\geq e:ub=b,\ u^{-}b=b\} { u ≥ e : u b = b , u − b = b } has a maximum for every gap J = ( a , b ) J=(a,b) J = ( a , b ) . Indeed u b = b ub=b u b = b changes the activity equation u − u b = b u^{-}ub=b u − u b = b into u − b = b u^{-}b=b u − b = b .
Lemma 5.3 (Transport and comparison maxima). Assume (PC). Every cut { u ∈ A : u X ≤ Y } \{u\in A:uX\leq Y\} { u ∈ A : u X ≤ Y } , for X , Y ∈ D ( A ) X,Y\in D(A) X , Y ∈ D ( A ) , has a greatest member m ( X , Y ) m(X,Y) m ( X , Y ) . For old x , y x,y x , y and gaps J = ( a , b ) J=(a,b) J = ( a , b ) , K = ( c , h ) K=(c,h) K = ( c , h ) , it is given by
case m ( X , Y ) Y = y ∈ A y / r ( X ) X = x ∈ A , Y = X K c / x X = X J , Y = X K , no coefficient sends X J to X K c / b X = X J , Y = X K , u X J = X K ( u − ) s J = s K / u − (21) \begin{array}{c|c}
\textit{case} & m(X,Y)\\
Y=y\in A & y/r(X)\\
X=x\in A,\ Y=X_K & c/x\\
X=X_J,\ Y=X_K,\ \textit{no coefficient sends }X_J\textit{ to }X_K & c/b\\
X=X_J,\ Y=X_K,\ uX_J=X_K & (u^{-})s_J=s_K/u^{-} \tag*{(21)}
\end{array} case Y = y ∈ A X = x ∈ A , Y = X K X = X J , Y = X K , no coefficient sends X J to X K X = X J , Y = X K , u X J = X K m ( X , Y ) y / r ( X ) c / x c / b ( u − ) s J = s K / u − (21) In the last case the maximum itself sends X J X_J X J to X K X_K X K . Moreover s J 2 = s J s_J^2=s_J s J 2 = s J .
The proof is given in Appendix C.2.
By order reversal, all the following minima now exist:
k ( X , Y ) = min { z ∈ A : z ′ X ≤ Y ′ } = m ( X , Y ′ ) ′ . (22) k(X,Y)=\min\{z\in A:z'X\leq Y'\}=m(X,Y')'. \tag*{(22)} k ( X , Y ) = min { z ∈ A : z ′ X ≤ Y ′ } = m ( X , Y ′ ) ′ . (22)
Conversely, their existence implies (PC) by taking ( X , Y ) = ( X J , X J ′ ) (X,Y)=(X_J,X'_J) ( X , Y ) = ( X J , X J ′ ) . Right-coefficient cuts are attained as well, since X u ≤ Y ⟺ u Y ′ ≤ X ′ Xu\leq Y\Longleftrightarrow uY'\leq X' X u ≤ Y ⟺ u Y ′ ≤ X ′ by the old-action adjunctions.
The greatest cyclic multiplication
Lemma 5.4 (Cyclic selection and associativity with old factors). Under (PC), define on D ( A ) D(A) D ( A )
P ( X , Y ) = min { k ( X , Y ) , r ( X ) Y , X r ( Y ) } . (23) P(X,Y)=\min\{k(X,Y),\,r(X)Y,\,Xr(Y)\}. \tag*{(23)} P ( X , Y ) = min { k ( X , Y ) , r ( X ) Y , X r ( Y )} . (23) This operation is monotone, unital, and retains every prescribed product having an old factor. For z ∈ A z\in A z ∈ A it satisfies
P ( X , Y ) ≤ z ⟺ z ′ X ≤ Y ′ , r ( P ( X , Y ) ) = k ( X , Y ) . (24) P(X,Y)\le z\Longleftrightarrow z'X\le Y',\qquad r(P(X,Y))=k(X,Y). \tag*{(24)} P ( X , Y ) ≤ z ⟺ z ′ X ≤ Y ′ , r ( P ( X , Y )) = k ( X , Y ) . (24) It satisfies all cyclic adjunctions and has actual residuals
X \ Y = P ( Y ′ , X ) ′ , Y / X = P ( X , Y ′ ) . (25) X\backslash Y=P(Y',X)',\qquad Y/X=P(X,Y'). \tag*{(25)} X \ Y = P ( Y ′ , X ) ′ , Y / X = P ( X , Y ′ ) . (25) Furthermore, for u ∈ A u\in A u ∈ A ,
P ( u X , Y ) = u P ( X , Y ) , P ( X , Y u ) = P ( X , Y ) u , P ( X u , Y ) = P ( X , u Y ) , (26) P(uX,Y)=uP(X,Y),\qquad P(X,Yu)=P(X,Y)u,\qquad P(Xu,Y)=P(X,uY), \tag*{(26)} P ( u X , Y ) = u P ( X , Y ) , P ( X , Y u ) = P ( X , Y ) u , P ( X u , Y ) = P ( X , u Y ) , (26) and the two bracketings of every triple have the same old upper anchor:
r ( P ( P ( X , Y ) , Z ) ) = r ( P ( X , P ( Y , Z ) ) ) . (27) r(P(P(X,Y),Z))=r(P(X,P(Y,Z))). \tag*{(27)} r ( P ( P ( X , Y ) , Z )) = r ( P ( X , P ( Y , Z ))) . (27) The proof is given in Appendix § 14 .§ 3 .
Active coefficients and relative return collapse
Lemma 5.5 (Closure of active coefficients). The coefficients having a nonempty active left domain are exactly those having a nonempty active right domain. Their common set Γ ( A ) \Gamma(A) Γ ( A ) is closed under multiplication and under u ↦ u − u\mapsto u^{-} u ↦ u − ; if A A A has a gap, it contains e e e .
Proof. Suppose u X = Y uX=Y u X = Y is active, with X , Y X,Y X , Y new and anchors b , h b,h b , h . The actual right adjoint L u ∗ ( Y ) = ( Y ′ u ) ′ L_u^*(Y)=(Y'u)' L u ∗ ( Y ) = ( Y ′ u ) ′ is at least X X X but strictly below b b b , because u b = h > Y ub=h>Y u b = h > Y . The only point in [ X , b ) [X,b) [ X , b ) is X X X , so Y ′ u = X ′ Y'u=X' Y ′ u = X ′ , an active right action. The opposite argument gives the converse. For the negative return put v = u − v=u^{-} v = u − ; the activity equations give v h = b vh=b v h = b and u v h = h uvh=h uv h = h , so v v v is active on the image gap and returns its midpoint to X X X . This proves closure under u ↦ u − u\mapsto u^{-} u ↦ u − . The unit is active at every gap.
Let u , v ∈ Γ ( A ) u,v\in\Gamma(A) u , v ∈ Γ ( A ) , and put α = u u − ≤ e \alpha=uu^{-}\le e α = u u − ≤ e , β = v − v ≤ e \beta=v^{-}v\le e β = v − v ≤ e . An image upper endpoint for u u u is fixed by α \alpha α , and an active input upper endpoint for v v v is fixed by β \beta β . If α ≤ β \alpha\le\beta α ≤ β , an image upper endpoint h h h for u u u satisfies h = α h ≤ β h ≤ h h=\alpha h\le\beta h\le h h = α h ≤ β h ≤ h , so v v v is active at that image gap and v u vu v u is active by composition. If β ≤ α \beta\le\alpha β ≤ α , take an active input upper endpoint c c c for v v v . Then c = β c ≤ α c ≤ c c=\beta c\le\alpha c\le c c = β c ≤ α c ≤ c , so u u ‾ − c = c u\overline{u}^{-}c=c u u − c = c . The inequalities u u ‾ − ≤ e u\overline{u}^{-}\le e u u − ≤ e and u ‾ − u ≤ e \overline{u}^{-}u\le e u − u ≤ e , the latter by unit cyclicity, make u u u an admissible return for the left action of u ‾ − \overline{u}^{-} u − . Thus u ‾ − \overline{u}^{-} u − is active at this gap by (L3) and the definition of activity. Its image gap has upper endpoint b 0 = u ‾ − c b_0=\overline{u}^{-}c b 0 = u − c , and
u b 0 = c , u ‾ − u b 0 = u ‾ − c = b 0 . ub_0=c,\qquad\overline{u}^{-}ub_0=\overline{u}^{-}c=b_0. u b 0 = c , u − u b 0 = u − c = b 0 .
Consequently u u u is active at the image gap by (L4), and sends it back to the active input gap of v v v . Composition gives an active action of v u vu v u . This argument uses u u u as an admissible return and requires no identity u ‾ − − = u \overline{u}^{--}=u u −− = u . Here u ‾ − = u \ e = e / u \overline{u}^{-}=u\backslash e=e/u u − = u \ e = e / u , whereas prime denotes the given cyclic involution. This proves multiplicative closure.
Say that a lower complete core has internal gaps if
a ≺ b ⟹ p b < b . (28) a\prec b\quad\Longrightarrow\quad pb<b. \tag*{(28)} a ≺ b ⟹ p b < b . (28)
This is equivalent to p a = p b pa=pb p a = p b for every gap. Indeed if p a < p b pa<pb p a < p b , then a < p b ≤ b a<pb\le b a < p b ≤ b and adjacency forces p b = b pb=b p b = b . Conversely, p a = p b = b pa=pb=b p a = p b = b would contradict p a ≤ a < b pa\le a<b p a ≤ a < b . A densely ordered floor group guarantees internal gaps: if p b = b pb=b p b = b were the upper endpoint of a gap, a floor strictly between p a pa p a and b b b would lie strictly between a a a and b b b . The same property holds for a trivial floor group, whose sole point is least in the whole chain.
Lemma 5.6 (Relative return collapse). In a lower complete core with internal gaps, every z ≤ d z\le d z ≤ d with p z = q pz=q p z = q has empty left and right active domains.
Proof. Put t = z − t=z^{-} t = z − . By (L2), p t = q − 1 pt=q^{-1} pt = q − 1 . Centrality of d d d , p p p and d 2 = q 2 d^2=q^2 d 2 = q 2 give
( t z ) 2 ≤ ( t d ) 2 = t 2 d 2 = t 2 q 2 = ( p t ) 2 q 2 = p . (29) (tz)^2\le(td)^2=t^2d^2=t^2q^2=(pt)^2q^2=p. \tag*{(29)} ( t z ) 2 ≤ ( t d ) 2 = t 2 d 2 = t 2 q 2 = ( pt ) 2 q 2 = p . (29) The insertion of p p p is valid because p q 2 = q 2 pq^2=q^2 p q 2 = q 2 ; no interchange of t t t and z z z is used. If z z z were active on the left at a gap with upper endpoint b b b , then t z b = b tzb=b t z b = b , so b = ( t z ) 2 b ≤ p b ≤ b b=(tz)^2b\le pb\le b b = ( t z ) 2 b ≤ p b ≤ b , contrary to (L19). For a right action, use ( z t ) 2 ≤ ( d t ) 2 = p (zt)^2\le(dt)^2=p ( z t ) 2 ≤ ( d t ) 2 = p and b z t = b bzt=b b z t = b .
Theorem 5.7 (Associativity of the internal-gap insertion). If a lower complete core A ∈ K R A\in K_R A ∈ K R has internal gaps and satisfies (PC), the multiplication P P P of (L11) is associative.
Proof. We first record what a strict correction means. If k = k ( X , Y ) < r ( X ) r ( Y ) k=k(X,Y)<r(X)r(Y) k = k ( X , Y ) < r ( X ) r ( Y ) , both X , Y X,Y X , Y are new, since with an old factor their prescribed product has anchor r ( X ) r ( Y ) r(X)r(Y) r ( X ) r ( Y ) . Write their gaps as ( a , b ) , ( c , h ) (a,b),(c,h) ( a , b ) , ( c , h ) . The inequality k ′ X ≤ Y ′ k'X\le Y' k ′ X ≤ Y ′ cannot hold by an inactive action: that would give the old value k ′ b ≤ h ′ k'b\le h' k ′ b ≤ h ′ and hence b h ≤ k bh\le k bh ≤ k , a contradiction. If it is active with upper anchor below c ′ c' c ′ , that anchor is at most h ′ h' h ′ and gives the same contradiction. It therefore has anchor c ′ c' c ′ and equals the unique midpoint Y ′ Y' Y ′ . In particular k ′ k' k ′ is active.
For arbitrary X , Y , Z X,Y,Z X , Y , Z put b = r ( X ) b=r(X) b = r ( X ) , h = r ( Y ) h=r(Y) h = r ( Y ) , t = r ( Z ) t=r(Z) t = r ( Z ) , a = k ( X , Y ) a=k(X,Y) a = k ( X , Y ) , c = k ( Y , Z ) c=k(Y,Z) c = k ( Y , Z ) , and
L = P ( P ( X , Y ) , Z ) , R = P ( X , P ( Y , Z ) ) , K = r ( L ) = r ( R ) . L=P(P(X,Y),Z),\qquad R=P(X,P(Y,Z)),\qquad K=r(L)=r(R). L = P ( P ( X , Y ) , Z ) , R = P ( X , P ( Y , Z )) , K = r ( L ) = r ( R ) .
The equality of anchors follows from Lemma 5.4 . Expanding (L11), using the old-factor equations and preservation of minima by old actions, gives
L = min { K , a Z , b Y t , X ( h t ) } , R = min { K , ( b h ) Z , b Y t , X c } . (30) \begin{aligned}
L&=\min\{K,aZ,bYt,X(ht)\},\\
R&=\min\{K,(bh)Z,bYt,Xc\}. \tag*{(30)}
\end{aligned} L R = min { K , a Z , bY t , X ( h t )} , = min { K , ( bh ) Z , bY t , X c } . (30)
The discarded terms are a t ≥ a Z at\ge aZ a t ≥ a Z and b c ≥ X c bc\ge Xc b c ≥ X c .
If L ≠ R L\ne R L = R , the one-midpoint carrier and their common anchor force one to be K K K and the other its immediate predecessor. Suppose L < K = R L<K=R L < K = R . The second line implies ( b h ) Z , b Y t , X c ≥ K (bh)Z,bYt,Xc\ge K ( bh ) Z , bY t , X c ≥ K ; since c ≤ h t c\le ht c ≤ h t , also X ( h t ) ≥ K X(ht)\ge K X ( h t ) ≥ K . Thus the first line forces L = a Z < K L=aZ<K L = a Z < K , which is new. Hence a a a is active. Necessarily a < b h a<bh a < bh , so the preceding strict-correction argument makes a ′ a' a ′ active as well. By Lemma 5.5 , a a ′ aa' a a ′ is active. But cyclic duality gives a a ′ ≤ d aa'\le d a a ′ ≤ d , and
π ( a a ′ ) = π ( a ) q π ( a ) − 1 = q . \pi(aa')=\pi(a)q\pi(a)^{-1}=q. π ( a a ′ ) = π ( a ) q π ( a ) − 1 = q .
Lemma 5.6 excludes this. If R < K = L R<K=L R < K = L , apply the same argument in the opposite monoid with the triple reversed; it gives an active right polar pair, and the equality of active coefficient sets gives the identical contradiction. Thus L = R L=R L = R .
Lemma 5.8 (Fullness and all finite histories). The associative insertion in Theorem 5.7 belongs to K R \mathcal{K}_{R} K R , and A ↪ D ( A ) A\hookrightarrow D(A) A ↪ D ( A ) is full.
The proof is given in Appendix § 14 .§ 4 . It treats an arbitrary finite word w w w , rather than reducing the history scheme to the particular short words used in the structural lemmas.
Lemma 5.9 (The insertion invariants). If A A A is a lower complete core with internal gaps and (PC), then D ( A ) D(A) D ( A ) is again such a core, with exactly the same floor group p D ( A ) = p A pD(A)=pA p D ( A ) = p A , the same p p p , q q q , internal gaps, and (PC).
The proof is given in Appendix § 14 .§ 5 .
Example 5.10 (One internal gap and its comparison correction). Let A = { 0 < 1 } A=\{0<1\} A = { 0 < 1 } , with e = 1 e=1 e = 1 , d = 0 d=0 d = 0 , x y = min ( x , y ) xy=\min(x,y) x y = min ( x , y ) , and 0 ′ = 1 0'=1 0 ′ = 1 , 1 ′ = 0 1'=0 1 ′ = 0 . Its residuals are
x \ y = y / x = { 1 , x ≤ y , y , x > y . x\backslash y=y/x=
\begin{cases}
1, & x\le y,\\
y, & x>y.
\end{cases} x \ y = y / x = { 1 , y , x ≤ y , x > y . This is an integral commutative cyclic involutive residuated chain. It belongs to K R \mathcal{K}_{R} K R : unit cyclicity is immediate, and b a z v u ≤ z bazvu\le z ba z v u ≤ z holds for all five elements by integrality and monotonicity, independently of the history premises. The closed support terms are
t = 0 \ 0 = 1 , p = q = Q = 0. t=0\backslash0=1,\qquad p=q=Q=0. t = 0\0 = 1 , p = q = Q = 0. Thus G = p A = { 0 } G=pA=\{0\} G = p A = { 0 } is the trivial floor group and A A A is a lower complete core. Its sole gap J = ( 0 , 1 ) J=(0,1) J = ( 0 , 1 ) is internal, because p 1 = 0 < 1 p1=0<1 p 1 = 0 < 1 .
Write X = X 0 , 1 X=X_{0,1} X = X 0 , 1 , so that D ( A ) = { 0 < X < 1 } D(A)=\{0<X<1\} D ( A ) = { 0 < X < 1 } , X ′ = X X'=X X ′ = X , and r ( X ) = 1 r(X)=1 r ( X ) = 1 . Every old coefficient has u − = u \ e = 1 u^{-}=u\backslash e=1 u − = u \ e = 1 . Consequently the left and right activity equations at J J J reduce to u = 1 u=1 u = 1 . More generally, an old context ( u , v ) (u,v) ( u , v ) is active exactly when u = v = 1 u=v=1 u = v = 1 . The prescribed actions therefore give
0 X = X 0 = 0 , 1 X = X 1 = X . 0X=X0=0,\qquad1X=X1=X. 0 X = X 0 = 0 , 1 X = X 1 = X . Since 1 1 1 is the only old coefficient at least e e e , one has S J = { 1 } S_J=\{1\} S J = { 1 } and s J = 1 s_J=1 s J = 1 ; in particular, (PC) holds. The defining comparison minimum and selected product are
k ( X , X ) = min { z ∈ A : z ′ X ≤ X ′ } = 0 , P ( X , X ) = min { 0 , X , X ′ } = 0. k(X,X)=\min\{z\in A:z'X\le X'\}=0,\qquad P(X,X)=\min\{0,X,X'\}=0. k ( X , X ) = min { z ∈ A : z ′ X ≤ X ′ } = 0 , P ( X , X ) = min { 0 , X , X ′ } = 0. Here 0 0 0 qualifies in the first minimum because 0 ′ X = 1 X = X = X ′ 0'X=1X=X=X' 0 ′ X = 1 X = X = X ′ . The complete multiplication table is therefore
Table 1.
and the residual formulas give
X \ 0 = P ( 1 , X ) ′ = X , X \ X = P ( X , X ) ′ = 1. X\backslash0 = P(1,X)' = X,\qquad X\backslash X = P(X,X)' = 1. X \0 = P ( 1 , X ) ′ = X , X \ X = P ( X , X ) ′ = 1.
The table is associative: 0 0 0 is absorbing, 1 1 1 is the identity, and any product with at least two occurrences of X X X is 0 0 0 . It is commutative and integral, so every finite history holds by the same inequality b a z v u ≤ z bazvu \le z ba z v u ≤ z . The old residuals are unchanged, and the old prime and constants are retained, giving a full embedding A ↪ D ( A ) A \hookrightarrow D(A) A ↪ D ( A ) . This example exhibits the role of the comparison term explicitly: k ( X , X ) = 0 k(X,X)=0 k ( X , X ) = 0 , whereas r ( X ) r ( X ) = 1 r(X)r(X)=1 r ( X ) r ( X ) = 1 .
Preparing an arbitrary countable lower core
We attain (PC) on a changed carrier by canonical extension. The precise completion facts and their source locators are stated in Appendix C.6.
Lemma 5.11 (Canonical completion and attained gap supports). If A ∈ K R A \in\mathcal{K}_{R} A ∈ K R is bounded, then its canonical extension A δ ∈ K R A^{\delta} \in\mathcal{K}_{R} A δ ∈ K R is a full chain extension in which every gap has greatest positive self-support.
The proof is given in Appendix C.6.
Lemma 5.12 (Attained internal supports with the original floor group). Every countable lower complete core K K K has a countable full extension C ∈ K R C \in\mathcal{K}_{R} C ∈ K R with p C = p K = G pC=pK=G pC = p K = G such that every gap whose endpoints have equal degree has greatest positive self-support.
Proof. Apply the full completion theorem to K K K , and then Lemma 5.11, obtaining K ⊆ B ∈ K R K \subseteq B \in\mathcal{K}_{R} K ⊆ B ∈ K R with support maxima at every gap. The old closed term p p p stays idempotent and at most e e e . It stays central in MacNeille completion by join density, and in canonical completion by the closed-product formula followed by join density. Thus the interior-support projection B → p B B \to pB B → pB is full. The old G G G is a full subalgebra of p B pB pB for its local constants p , q p,q p , q . To see the residual claim explicitly, for g , h ∈ G g,h \in G g , h ∈ G and z ∈ p B z \in pB z ∈ pB , multiplication by g g g has order-preserving inverse multiplication by the retained g − 1 g^{-1} g − 1 . Hence g z ≤ h gz \le h g z ≤ h exactly when z ≤ g − 1 h z \le g^{-1}h z ≤ g − 1 h , and the actual residuals are g \ h = g − 1 h g\backslash h=g^{-1}h g \ h = g − 1 h and h / g = h g − 1 h/g=hg^{-1} h / g = h g − 1 . Its local prime is g − 1 q g^{-1}q g − 1 q ; lattice operations select one argument. Therefore
E = { x ∈ B : p x ∈ G } (31) E=\{x \in B:px \in G\} \tag*{(31)} E = { x ∈ B : p x ∈ G } (31) is a full original-signature subalgebra containing K K K , with p E = G pE=G pE = G .
If a ≺ E b a \prec_E b a ≺ E b and p a = p b = g pa = pb = g p a = p b = g , the whole interval [ a , b ] B [a,b]_B [ a , b ] B is in E E E , so it is also a gap of B B B . Its greatest self-support s s s satisfies s b = b sb = b s b = b . Applying the projection and multiplying by the retained inverse of g g g gives p s = p ps = p p s = p , so s ∈ E s \in E s ∈ E . It is therefore the greatest self-support in E E E as well. This internal-support property is first-order in the original finite signature: use
Gap ( a , b ) ≡ a < b & ¬ ∃ z ( a < z < b ) , \operatorname{Gap}(a,b) \equiv a < b \mathbin{\&} \neg\exists z\,(a < z < b), Gap ( a , b ) ≡ a < b & ¬∃ z ( a < z < b ) ,
Fix b ( u ) ≡ e ≤ u & u b = b & ( u \ e ) b = b , \operatorname{Fix}_b(u) \equiv e \leq u \mathbin{\&} ub = b \mathbin{\&} (u\backslash e)b = b, Fix b ( u ) ≡ e ≤ u & u b = b & ( u \ e ) b = b ,
and the sentence asserting that, whenever Gap ( a , b ) \operatorname{Gap}(a,b) Gap ( a , b ) and p a = p b pa = pb p a = p b , the set defined by Fix b \operatorname{Fix}_b Fix b has a greatest member. A countable elementary substructure C ≼ E C \preccurlyeq E C ≼ E containing K K K exists by downward Löwenheim–Skolem. It is a full actual chain, satisfies this sentence and every history, and has p C = G pC = G pC = G , since K ⊆ C ⊆ E K \subseteq C \subseteq E K ⊆ C ⊆ E .
Lemma 5.13 (Dense floor preparation). Every countable lower complete core has a nontrivial countable full extension A ∈ K R A \in\mathcal{K}_R A ∈ K R with a densely ordered floor group, internal gaps, and ( P C ) (PC) ( P C ) .
Proof. Take C C C from Lemma 5.12 and let H = G × → Q H = G \overrightarrow{\times} \mathbb{Q} H = G × Q , with componentwise group product and lexicographic order. Embed g g g as ( g , 0 ) (g,0) ( g , 0 ) . This is a full group embedding for local identity p p p and central dualizer q = ( q , 0 ) q = (q,0) q = ( q , 0 ) . The group H H H is countable, dense, without endpoints and nontrivial. Apply Theorem 4.7 to C C C and H H H to form an extension A A A : the fibre of each old g ∈ G g \in G g ∈ G is unchanged, while every h ∈ H ∖ G h \in H \setminus G h ∈ H ∖ G has the singleton fibre { h } \{h\} { h } . Old products and prime are retained; a product involving a new floor is the floor at the product of the degrees, and its prime is h ↦ q h − 1 h \mapsto qh^{-1} h ↦ q h − 1 . The lifting theorem supplies both actual residuals and every finite history for these precise rules.
Distinct endpoint degrees cannot form a gap, since an intermediate degree in H H H supplies an intermediate floor. Thus every gap lies inside a single unchanged old fibre, and is a gap of C C C . For a positive self-support u u u of such a gap with upper degree g g g , the equation u b = b ub = b u b = b gives π ( u ) g = g \pi(u)g = g π ( u ) g = g , hence π ( u ) = p \pi(u) = p π ( u ) = p . The neutral fibre is unchanged, so every possible self-support is old. The greatest old one exists by the preceding lemma, and its defining equations are preserved. Therefore every gap has ( P C ) (PC) ( P C ) , and all gaps are internal by their equal endpoint degrees.
Theorem 5.14 (Representation of lower complete cores). Every countable lower complete core K ∈ K R K \in\mathcal{K}_R K ∈ K R has a nontrivial countable full dense extension in K R \mathcal{K}_R K R and a full embedding into a standard chain.
Proof. Start with A 0 A_0 A 0 supplied by Lemma 5.13 and put A n + 1 = D ( A n ) A_{n+1}=D(A_n) A n + 1 = D ( A n ) with multiplication (L11). Theorems 5.7 and Lemmas 5.8–5.9 apply at every stage. Each stage is countable, each inclusion is full, and the exact floor group is the same H H H throughout. The union A ∞ = ⋃ n < ω A n A_\infty=\bigcup_{n<\omega}A_n A ∞ = ⋃ n < ω A n is a countable member of K R \mathcal{K}_R K R : each finite tuple, all its operations, and any specified finite-history instance lie in one stage. For x < y x<y x < y in this union, choose a common stage. Either it already has an intermediate point or its next stage inserts one. Thus the union is dense and nontrivial. Corollary 3.2 gives its full standard embedding. Composing the inclusions proves the assertion for K K K . □
Corollary 5.15 (Preservation of old central units). The extensions and standard embedding in Theorem 5.14 can simultaneously preserve the centrality of every old central global unit of K K K .
Proof. Let u u u be any such unit. Fullness retains its old inverse. In a MacNeille completion, multiplication by u u u on either side agrees on old points and preserves joins, so join density makes u u u central everywhere. In a canonical extension, (L25) gives u k = ⋀ { u a : a ∈ A , k ≤ a } = ⋀ { a u : a ∈ A , k ≤ a } = k u uk=\bigwedge\{ua:a\in A,\ k\leq a\}=\bigwedge\{au:a\in A,\ k\leq a\}=ku u k = ⋀ { u a : a ∈ A , k ≤ a } = ⋀ { a u : a ∈ A , k ≤ a } = k u for closed k k k ; join density again gives centrality on all points. The pullback and elementary substructure inherit this identity.
In the floor enlargement, π ( u ) \pi(u) π ( u ) is central in G G G , hence ( π ( u ) , 0 ) (\pi(u),0) ( π ( u ) , 0 ) is central in G × → Q G\mathbin{\overrightarrow{\times}}\mathbb{Q} G × Q . Thus u u u commutes with each new singleton floor by the explicit product rule, and with each old point as before. At a midpoint insertion, u − = u − 1 u^{-}=u^{-1} u − = u − 1 and u − u = u u − = e u^{-}u=uu^{-}=e u − u = u u − = e . Both left and right actions are therefore active at every gap; their equal old upper values u b = b u ub=bu u b = b u determine the same midpoint. Thus u u u remains central at every stage, in the union, and in the final MacNeille completion. These arguments apply to each old central unit without adding any choices or countability requirements. □
The upper-support construction
A full embedding retains the comparison between e e e and the closed support p p p , so the lower construction cannot handle e < p e<p e < p . We summarize the two upper constructions here; their complete proofs are in Appendix D.
Theorem 6.1 (Upper-support representation). Every countable C ∈ K R C\in\mathcal{K}_R C ∈ K R with e ≤ p e\leq p e ≤ p has a countable full dense extension and a full standard embedding. The target retains e ≤ p e\leq p e ≤ p and every original finite history.
Proof. If p = e p=e p = e , then d = q d=q d = q is a central global unit by Theorem 4.2 . Apply Theorem 4.8 and standard completion. This includes a singleton input: then d = e d=e d = e , and the first odd-clone step of Theorem B.7 adds a point on each side of its old point. Thus the dense extension is nontrivial before standard completion. Fullness here concerns the original signature; bounded embeddings preserving both endpoints are treated separately in the application sections.
For p > e p>e p > e put I = p C I=pC I = pC , q = p d q=pd q = p d , f = p ′ f=p' f = p ′ and n = q ′ n=q' n = q ′ . Lemma D.1 gives the local identity p p p and local dualizer f f f , the protected cover n < p n<p n < p in I I I , and the exhaustive alternatives n ∈ U ( I ) n\in U(I) n ∈ U ( I ) and n 2 = n n^2=n n 2 = n .
In the invertible case, put G = U ( I ) G=U(I) G = U ( I ) and K = { x ∈ C : p x ∈ G } K=\{x\in C:px\in G\} K = { x ∈ C : p x ∈ G } , the complete core of Lemma D.2 . Theorem D.15 prepares a countable full extension of K K K with exactly this group and a greatest neutral live coefficient. Theorem D.11 then densifies its protected bands while retaining G G G and their old endpoints. Lemma D.16 restores the remaining support points. Corollary D.17 assembles these steps into a countable full dense extension of the entire input.
In the idempotent case, multiplication by n n n is a full interior projection with central unit local dualizer n d = f nd=f n d = f . In its complete group core, Lemma D.18 unfolds and refolds the active bands, making them dense while retaining their two-point common corner. Lemma D.19 densifies the remaining corner while fixing that common part. Theorem D.20 glues these extensions and reconstructs the entire input, again giving a countable full dense extension.
Standard completion supplies the claimed embeddings in both branches. The dense extensions belong to K R K_R K R , so every finite history holds. Fullness preserves the closed terms e , p e,p e , p and hence their comparison in the standard target.
Full standard representation
The roles of completion and density are distinct. Table 2 records which support group is retained at each stage and when countability and density are obtained. All displayed extensions preserve the original signature and the finite history scheme; this does not assert preservation of arbitrary existing suprema or infima.
Table 2. Preservation and changes of support in the construction. The symbols G G G and H H H refer to the input and prepared groups of the respective branch.
Table 2 (PDF p. 30).
Table 3 gives the main dependency interfaces. In particular, the lower-core insertion proof is independent of the upper construction. The odd-chain results supply support densification for reconstruction and for the relative upper construction, rather than an additional hypothesis on the lower input.
Table 3. Dependencies between the algebraic construction and its logical applications.
Theorem 7.1. Every countable member of K R \mathcal{K}_R K R embeds fully into a standard cyclic involutive residuated chain. In particular, multiplication, both residuals, involution and both designated constants are preserved simultaneously.
Proof. Let L ∈ K R L \in\mathcal{K}_R L ∈ K R be countable and let p p p be its closed central support. If e ≤ p e \leq p e ≤ p , apply Theorem 6.1. Suppose p < e p < e p < e . The full interior projection x ↦ p x x \mapsto px x ↦ p x has image I = p L I = pL I = p L , whose central local dualizer is q = p d q = pd q = p d . Let G = U ( I ) G = U(I) G = U ( I ) and let K = { x ∈ L : p x ∈ G } K = \{x \in L : px \in G\} K = { x ∈ L : p x ∈ G } . Lemma 4.4 makes K K K a countable full subalgebra with p K = G pK = G p K = G . It is therefore a lower complete core, to which Theorem 5.14 applies. Corollary 4.11 reconstructs the entire chain L L L from that standard core embedding and a dense extension of the support. All the embeddings used in this reconstruction are full.
The distinction between the two comparisons with p p p is exhaustive because the order is total. The preparatory properties used in the lower construction, such as attained self-supports and a dense support group, are properties of constructed extensions. They impose no additional hypotheses on L L L .
Strong standard completeness and elimination of density
We now use the full representation theorem proved in this paper: every countable member of K R \mathcal{K}_R K R embeds, in the entire signature, into a standard
chain (Theorem 7.1 ). An embedding preserves the values of all terms under one valuation, including residuals, involution, and all closed history terms.
Corollary 8.1 (Finite strict gaps). Suppose a finite set Σ \Sigma Σ of term comparisons, together with a < b a < b a < b , is realized in a member of K R \mathcal{K}_R K R . If ρ \rho ρ is a fresh variable, then
Σ ∪ { a < ρ , ρ < b } \Sigma\cup\{a < \rho,\ \rho< b\} Σ ∪ { a < ρ , ρ < b } is realized in a standard member of K R \mathcal{K}_R K R . Here comparisons may be equations, weak inequalities, or strict inequalities, and every term is evaluated under the same valuation.
Proof. Take the full subalgebra generated by the finitely many values of variables in the pattern. It is countable, since the signature is finite, and is a nontrivial member of K R \mathcal{K}_R K R : the defining conditions are universal and the strict pair is retained. Apply Theorem 7.1 . The embedding preserves all the old term values and strict comparisons. Choose the value of ρ \rho ρ strictly between the images of a a a and b b b in [ 0 , 1 ] [0,1] [ 0 , 1 ] . Freshness leaves the old pattern unchanged, and the standard target belongs to K R \mathcal{K}_R K R by Lemma 2.2 . □
Theorem 8.2 (Strong standard completeness). For every set T T T of formulas in the fixed countable propositional language and every finite hypersequent H H H ,
T ⊢ H R H ⟺ T ⊨ S t d H . T \vdash_{\mathrm{HR}} H \quad\Longleftrightarrow\quad T \models_{\mathrm{Std}} H. T ⊢ HR H ⟺ T ⊨ Std H . In particular, for every formula A A A ,
T ⊢ H R A ⟺ ∀ S ∈ S t d ∀ v , [ ( ∀ B ∈ T , e S ≤ v ( B ) ) ⟹ e S ≤ v ( A ) ] . T \vdash_{\mathrm{HR}} A \quad\Longleftrightarrow\quad\forall S \in\mathrm{Std}\ \forall v,\ \left[(\forall B \in T,\ e_S \le v(B)) \Longrightarrow e_S \le v(A)\right]. T ⊢ HR A ⟺ ∀ S ∈ Std ∀ v , [ ( ∀ B ∈ T , e S ≤ v ( B )) ⟹ e S ≤ v ( A ) ] . Proof. The rule checks in Theorem 2.3 are pointwise in a valuation. Induction on a proof therefore proves soundness with the specified assumption leaves.
Conversely, suppose T ⊬ H R H T \nvdash_{\mathrm{HR}} H T ⊬ HR H . Theorem 2.5 supplies a countable chain C ∈ K R C \in\mathcal{K}_R C ∈ K R and a valuation v v v satisfying T T T and falsifying H H H . By Theorem 7.1 , choose a full embedding j : C ↪ S j:C \hookrightarrow S j : C ↪ S into a standard chain. For v ′ = j ∘ v v' = j \circ v v ′ = j ∘ v , induction on formulas gives v ′ ( A ) = j ( v ( A ) ) v'(A) = j(v(A)) v ′ ( A ) = j ( v ( A )) . Since j ( e C ) = e S j(e_C) = e_S j ( e C ) = e S and j j j preserves and reflects order, v ′ v' v ′ satisfies every member of T T T . Fullness likewise preserves and reflects the truth of each component of H H H , so v ′ v' v ′ falsifies H H H . Thus T ̸ ⊨ S t d H T \not\models_{\mathrm{Std}} H T ⊨ Std H . All premises are retained by this single embedding; no preservation of infinite suprema or infima is used. □
Theorem 8.3 (Standard theoremhood). For every finite hypersequent H H H ,
H R ⊢ H ⟺ S t d ⊨ H . \mathrm{H}_{\mathrm{R}} \vdash H \Longleftrightarrow\mathrm{Std} \models H. H R ⊢ H ⟺ Std ⊨ H . Proof. Take T = ∅ T = \varnothing T = ∅ in Theorem 8.2. The empty hypersequent remains false and is underivable by soundness.
Remark 8.4 (Scope of strong completeness). The strong result concerns the calculus H R \mathrm{H}_{\mathrm{R}} H R and the class K R \mathcal{K}_{\mathrm{R}} K R ; it does not remove the unit-cyclicity or history conditions from the input representation theorem. All standard cyclic involutive residuated chains belong to K R \mathcal{K}_{\mathrm{R}} K R by Lemma 2.2, which justifies the use of S t d \mathrm{Std} Std here. Standard consequence quantifies over the class of standard algebras, not over one fixed algebra on [ 0 , 1 ] [0,1] [ 0 , 1 ] .
Consider the density rule
G ∣ ⇒ Γ , ρ ∣ ⇒ ρ ′ , Δ G ∣ ⇒ Γ , Δ D e n , ρ ∉ G , Γ , Δ . (32) \frac{G \mid\Rightarrow\Gamma,\rho\mid\Rightarrow\rho',\Delta}{G \mid\Rightarrow\Gamma,\Delta}
\quad\mathrm{Den}, \qquad\rho\notin G,\Gamma,\Delta. \tag*{(32)} G ∣⇒ Γ , Δ G ∣⇒ Γ , ρ ∣⇒ ρ ′ , Δ Den , ρ ∈ / G , Γ , Δ. (32)
The freshness condition includes occurrences of either polarity and occurrences inside formulas.
Lemma 8.5. D e n \mathrm{Den} Den preserves validity over every densely ordered cyclic involutive residuated chain.
Proof. If its conclusion is false under a valuation, every component of G G G is false and γ δ > d \gamma\delta> d γ δ > d , where γ = Γ ♯ \gamma= \Gamma^\sharp γ = Γ ♯ and δ = Δ ♯ \delta= \Delta^\sharp δ = Δ ♯ . By residuation this means δ ′ < γ \delta' < \gamma δ ′ < γ . Choose δ ′ < r < γ \delta' < r < \gamma δ ′ < r < γ and set v ( ρ ) = r v(\rho) = r v ( ρ ) = r . The two new active components would mean, respectively, γ r ′ ≤ d \gamma r' \le d γ r ′ ≤ d , equivalently γ ≤ r \gamma\le r γ ≤ r , and r δ ≤ d r\delta\le d r δ ≤ d , equivalently r ≤ δ ′ r \le\delta' r ≤ δ ′ . Both are false. Freshness retains the falsity of G G G and the old word values, so the premise is false under the extended valuation.
Theorem 8.6 (Unrestricted density elimination). The rule (14) is admissible in H R \mathrm{H}_{\mathrm{R}} H R . Moreover, from any finite H R + D e n \mathrm{H}_{\mathrm{R}}+\mathrm{Den} H R + Den proof one can effectively obtain a finite D e n \mathrm{Den} Den -free H R \mathrm{H}_{\mathrm{R}} H R proof of the same end hypersequent. Ordinary Cut remains available.
Proof. If the Den premise is derivable in H R \mathrm{H}_{\mathrm{R}} H R , it is standard-valid by soundness. Lemma 8.5 makes its conclusion standard-valid, and Theorem 8.3 makes that conclusion derivable in H R \mathrm{H}_{\mathrm{R}} H R . This is admissibility for arbitrary premise proofs, with no restriction on their use of Cut, EC, COM, shared conjunction, or any R w R_w R w .
For effectiveness, the normalized formula syntax and all finite words w ∈ { S , T } ∗ w \in\{S,T\}^{*} w ∈ { S , T } ∗ are computably enumerable. The recursion (4) computes the two guards of each R w R_w R w . Finite proof trees annotated by their rule names, formula data, word splittings, and history indices have a decidable correctness check. Thus all finite H R \mathrm{H}_R H R proofs can be effectively enumerated, including all choices of Cut formulas.
Proceed upward through a supplied finite H R + D e n \mathrm{H}_R + \mathrm{Den} H R + Den proof. Retain an ordinary inference after replacing its premise proofs. At a Den inference, admissibility guarantees an original H R \mathrm{H}_R H R proof of its syntactic conclusion. Enumerate original proofs until one with that conclusion appears, and insert it. Every such search terminates. There are finitely many Den nodes, so this produces the required finite proof effectively. □ \square □
The Den-elimination theorem above concerns proofs without additional assumption leaves. Strong completeness with assumptions is stated separately in Theorem 8.2 . No claim is made here of effective Den elimination relative to an arbitrary theory, elimination of ordinary Cut, or a complexity bound for the proof-enumeration procedure. Relative density elimination would also require the appropriate freshness conditions on assumptions.
Transfer to involutive uninorm logic
We use IUL for involutive uninorm logic and I U L f p \mathrm{IUL}^{fp} IUL f p for its extension by e ↔ d e \leftrightarrow d e ↔ d . An IUL-chain is a bounded commutative involutive residuated chain in the signature
( ∧ , ∨ , ⋅ , → , e , d , ⊥ , ⊤ ) , x ′ = x → d , x ′ ′ = x . (\wedge,\vee,\cdot,\to,e,d,\bot,\top), \qquad x' = x \to d, \qquad x'' = x. ( ∧ , ∨ , ⋅ , → , e , d , ⊥ , ⊤ ) , x ′ = x → d , x ′′ = x .
The designated condition for a formula is e ≤ v ( φ ) e \le v(\varphi) e ≤ v ( φ ) ; neither e e e nor d d d is required to be a lattice endpoint. In particular, no equation e = d e = d e = d is imposed. Standard chains have ordered carrier [ 0 , 1 ] [0,1] [ 0 , 1 ] , with ⊥ = 0 \bot= 0 ⊥ = 0 and ⊤ = 1 \top= 1 ⊤ = 1 .
Jenei’s finite strong standard completeness theorem concerns I U L f p \mathrm{IUL}^{fp} IUL f p [8 Theorem 2] ; his later densification theorem yields strong standard completeness for the same fixed-point extension [12 Theorem 20] . The distinction is essential for embeddings. If e ≠ d e \ne d e = d in an input chain, an embedding preserving both constants cannot have an odd target: the target equation e = d e = d e = d would identify their two images.
We apply the full representation theorem to general IUL. The representation constructions already supply comparison maxima, associativity, cyclic duality, actual residuals, and density. For this application we additionally prove that they preserve commutativity, and then recover the two lattice-bound constants by restriction to the old endpoint interval. These steps give the corollary below. The proof-theoretic argument of [6 ] is not used.
The history conditions on commutative inputs
Lemma 9.1 (Automatic histories). Every commutative involutive residuated chain belongs to K R \mathcal{K}_{R} K R . In particular, the reduct of every IUL-chain obtained by forgetting ⊥ , ⊤ \bot, \top ⊥ , ⊤ belongs to K R \mathcal{K}_{R} K R . No density or oddness assumption is needed.
Proof. Unit cyclicity is immediate from commutativity. The involutive residual formula gives e + x = ( d x ′ ) ′ = d → x e+x=(dx')'=d\to x e + x = ( d x ′ ) ′ = d → x . Thus the closed history terms of (4) satisfy
( c ϵ , h ϵ ) = ( e , e ) , (c_{\epsilon},h_{\epsilon})=(e,e), ( c ϵ , h ϵ ) = ( e , e ) , ( c w S , h w S ) = ( d c w , d → h w ) , ( c w T , h w T ) = ( d → c w , d h w ) . (c_{wS},h_{wS})=(dc_{w},d\to h_{w}),\qquad(c_{wT},h_{wT})=(d\to c_{w},dh_{w}). ( c w S , h w S ) = ( d c w , d → h w ) , ( c w T , h w T ) = ( d → c w , d h w ) . We prove c w h w ≤ e c_{w}h_{w}\le e c w h w ≤ e by induction on w w w . The empty history gives equality. For the two induction steps, commutativity and residuation give
c w S h w S = c w ( d ( d → h w ) ) ≤ c w h w , c_{wS}h_{wS}=c_{w}\bigl(d(d\to h_{w})\bigr)\le c_{w}h_{w}, c w S h w S = c w ( d ( d → h w ) ) ≤ c w h w , c w T h w T = ( d ( d → c w ) ) h w ≤ c w h w . c_{wT}h_{wT}=\bigl(d(d\to c_{w})\bigr)h_{w}\le c_{w}h_{w}. c w T h w T = ( d ( d → c w ) ) h w ≤ c w h w . If a b ≤ c w ab\le c_{w} ab ≤ c w and u v ≤ h w uv\le h_{w} uv ≤ h w , then
b a z v u = ( a b ) ( u v ) z ≤ c w h w z ≤ z . bazvu=(ab)(uv)z\le c_{w}h_{w}z\le z. ba z v u = ( ab ) ( uv ) z ≤ c w h w z ≤ z . This is precisely (7) , for every finite history and every choice of the five elements.
Commutativity of the extension constructions
Fullness of an embedding preserves products of old elements, but does not by itself impose commutativity on new elements. We therefore check the actual constructions, including products of two new points.
Lemma 9.2 (Symmetry of the lower insertion). Suppose the old chain A A A in Lemmas 5.2–5.4 is commutative and satisfies the hypotheses ensuring the existence of the minima in (L10). Then the multiplication P P P of (L11) is commutative.
Proof. For an old coefficient u u u and a gap with upper endpoint b b b , left activity means u − u b = b u^{-}ub=b u − u b = b , while right activity means b u u − = b buu^{-}=b b u u − = b . These conditions coincide, and the old upper values u b ub u b and b u bu b u coincide. The prescribed actions in Lemma 5.2 therefore satisfy
u X = X u ( u ∈ A , X ∈ D ( A ) ) . uX = Xu \qquad(u \in A,\ X \in D(A)). u X = X u ( u ∈ A , X ∈ D ( A )) .
For an old z z z , their right-adjoint formulas yield
z ′ X ≤ Y ′ ⟺ X ≤ ( Y z ′ ) ′ ⟺ Y z ′ ≤ X ′ ⟺ z ′ Y ≤ X ′ . z'X \le Y' \Longleftrightarrow X \le(Yz')'
\Longleftrightarrow Yz' \le X'
\Longleftrightarrow z'Y \le X'. z ′ X ≤ Y ′ ⟺ X ≤ ( Y z ′ ) ′ ⟺ Y z ′ ≤ X ′ ⟺ z ′ Y ≤ X ′ .
Consequently, the two sets whose minima define k ( X , Y ) k(X,Y) k ( X , Y ) and k ( Y , X ) k(Y,X) k ( Y , X ) are equal. The assumed attainment gives k ( X , Y ) = k ( Y , X ) k(X,Y)=k(Y,X) k ( X , Y ) = k ( Y , X ) . Since r ( X ) , r ( Y ) r(X),r(Y) r ( X ) , r ( Y ) are old, the other two entries in the minimum are exchanged when X , Y X,Y X , Y are exchanged. Hence
P ( X , Y ) = min { k ( X , Y ) , r ( X ) Y , X r ( Y ) } = min { k ( Y , X ) , r ( Y ) X , Y r ( X ) } = P ( Y , X ) . P(X,Y) = \min\{k(X,Y),r(X)Y,Xr(Y)\}
= \min\{k(Y,X),r(Y)X,Yr(X)\} = P(Y,X). P ( X , Y ) = min { k ( X , Y ) , r ( X ) Y , X r ( Y )} = min { k ( Y , X ) , r ( Y ) X , Y r ( X )} = P ( Y , X ) .
This argument proves symmetry; it does not infer attainment or associativity from symmetry.
Proposition 9.3 (Commutative specialization of the constructions). For a commutative input, the completion, preparation, insertion, and reconstruction procedures used in Theorem 7.1 can be performed with commutative intermediate algebras and a commutative standard target.
Proof. We verify preservation in the constituent constructions.
Completions and restrictions. In a MacNeille completion, separate preservation of joins and join density extend the old equation x y = y x xy=yx x y = y x to all pairs of completed elements. In the canonical extension, the closed-product formula (L25) is symmetric in its two arguments; expansion as joins of closed elements extends this symmetry to the whole algebra. In the proper-filter completion of Lemma D.13, old commutativity gives
F ∗ G = ↑ { a b : a ∈ F , b ∈ G } = G ∗ F , F * G = \mathord{\uparrow}\{ab : a \in F,b \in G\} = G * F, F ∗ G = ↑ { ab : a ∈ F , b ∈ G } = G ∗ F ,
so the induced product of sets of filters is symmetric as well. Subalgebras, the indicated inverse images, elementary substructures, and increasing unions inherit commutativity.
Odd extensions and support groups. For a commutative odd input, the local identities in Appendix B satisfy ℓ ( x ) = ρ ( x ) = u \ell(x) = \rho(x) = u ℓ ( x ) = ρ ( x ) = u . Every arrow is therefore an endomorphism at its local identity. For z > u z > u z > u , the absorption identities give
α x ( z ) = x z x ′ = z x x ′ = z u ′ = z , \alpha_x(z) = xzx' = zxx' = zu' = z, α x ( z ) = x z x ′ = z x x ′ = z u ′ = z ,
and α x ( u ) = u \alpha_x(u) = u α x ( u ) = u . Thus every tail map is the identity. The groups of matching isomorphisms and the quotient groups are Abelian; old matching products are commutative. The new matching product (36) uses their commutative product and the symmetric operation min ( i , j ) \min(i,j) min ( i , j ) on marks. In (42), both arguments are lifted to the same maximum of their two local identities; interchanging the arguments therefore leaves the product unchanged. The full clone, its relative restriction in Corollary B.9, and their iterations are commutative. Normalizing or restoring a central unit dualizer changes prime but not multiplication. The degree groups obtained from commutative inputs are Abelian, as are the lexicographic enlargements by Q \mathbb{Q} Q or Z \mathbb{Z} Z .
Lower cores and interior gluing. The preceding completion and restriction arguments apply to Lemma 5.12. The dense floor preparation uses an Abelian enlarged group. At each subsequent insertion, Lemma 9.2 applies; its attainment hypotheses are supplied by the cited preparation and invariant lemmas. In Theorem 4.7, a product involving an outside factor has the form i ( Π ( x ) Π ( y ) ) i(\Pi(x)\Pi(y)) i ( Π ( x ) Π ( y )) , which is symmetric when the support algebra is commutative. Products in the core already commute. This covers the lower representation and its whole-chain reconstruction, including the central-unit support extension.
Upper coefficients and middle intervals. Use the notation of Lemma D.7. Commutativity of U U U gives p u = q u p_u = q_u p u = q u and
L ( u , v ) ⟺ R ( v , u ) . L(u,v) \Longleftrightarrow R(v,u). L ( u , v ) ⟺ R ( v , u ) .
Its raw product
( u , r ) ( v , s ) = ( u v , min ( { 1 } ∪ { r : R ( u , v ) } ∪ { s : L ( u , v ) } ) ) (u,r)(v,s) = \left(uv,\min\left(\{1\}\cup\{r:R(u,v)\}\cup\{s:L(u,v)\}\right)\right) ( u , r ) ( v , s ) = ( uv , min ( { 1 } ∪ { r : R ( u , v )} ∪ { s : L ( u , v )} ) )
is consequently symmetric. The closed-cut product I ⊗ K = j ( ↓ I K ) I\otimes K = j(\downarrow IK) I ⊗ K = j ( ↓ I K ) remains symmetric, as does its restriction to the selected degree fibres. For the middle action extensions, equal old left and right actions have equal activity tests in Lemma D.5. They transport a point to the same labelled point or collapse it to the same old anchor. The completion in Lemma D.6 and the joins in (47) therefore preserve λ A = ρ A \lambda_A = \rho_A λ A = ρ A . Cyclic adjunction gives, for every old coefficient u u u ,
u X ≤ Y ′ ⟺ Y u ≤ X ′ ⟺ u Y ≤ X ′ . uX \le Y' \Longleftrightarrow Yu \le X' \Longleftrightarrow uY \le X'. u X ≤ Y ′ ⟺ Y u ≤ X ′ ⟺ u Y ≤ X ′ .
If X , Y X,Y X , Y have degrees g , k g,k g , k , the boundary degree b = ( g k ) − 1 δ b=(gk)^{-1}\delta b = ( g k ) − 1 δ is unchanged by interchanging them, because the degree group is Abelian. Their boundary comparison sets are equal, including the case in which they are empty. Whenever the comparison element is defined, it follows that
κ b ( X , Y ′ ) = κ b ( Y , X ′ ) . \kappa_b(X,Y')=\kappa_b(Y,X'). κ b ( X , Y ′ ) = κ b ( Y , X ′ ) .
In (49), the old anchor product is symmetric, the comparison branch has just been shown symmetric, and the remaining branch is the support top of degree g k = k g gk=kg g k = k g . All three cases are therefore symmetric. Coefficient–middle products commute because λ A = ρ A \lambda_A=\rho_A λ A = ρ A ; coefficient–reverse products commute because the two residuals of the commutative coefficient algebra agree. The other prescribed products are support tops of the same product degree. This verifies commutativity in Theorem D.11.
Upper reconstruction. Adjoining external bounds uses symmetric absorbing-product rules. The neutral-maximum preparation uses the proper-filter completion and subalgebras already treated above. Protected top gluing in Lemma D.16 evaluates a mixed product as Π ( x ) Π ( y ) \Pi(x)\Pi(y) Π ( x ) Π ( y ) in a commutative support algebra. In the idempotent upper branch, the unfolded product (51) is symmetric over the Abelian group H × Z H\times\mathbb{Z} H × Z . Its upper-invertible extension has just been treated; restriction to the retained positive bands preserves symmetry, and the separate floor rules (54) are symmetric over H H H . The deletion in Lemma D.19 is a full subalgebra restriction. Finally, in (60), the two mixed products are ( p a ) x (pa)x ( p a ) x and x ( p a ) x(pa) x ( p a ) inside the commutative corner extension, so they agree. This covers both upper branches and their whole-chain gluing.
At every use, the cited results supply the existence, fullness, associativity, residuals, and density required for that construction; the preceding calculations supply commutativity. The final MacNeille completion preserves it by the first paragraph.
Preserving the lattice bounds
Lemma 9.4 (Restriction to the old endpoint interval). Let A A A be a nontrivial bounded commutative involutive residuated chain. Suppose i : A ↪ S i:A\hookrightarrow S i : A ↪ S preserves
order, multiplication, prime, e e e , and d d d , where S S S is a commutative standard chain; the embedding is not assumed to preserve the bounds of S S S . Then A A A has an embedding into a standard IUL-chain preserving the entire bounded signature.
Proof. Put a = i ( ⊥ A ) a=i(\bot_A) a = i ( ⊥ A ) and b = i ( ⊤ A ) b=i(\top_A) b = i ( ⊤ A ) . The bottom of a bounded residuated lattice is absorbing, and ⊤ A e = ⊤ A \top_A e=\top_A ⊤ A e = ⊤ A with e ≤ ⊤ A e\le\top_A e ≤ ⊤ A implies ⊤ A 2 = ⊤ A \top_A^2=\top_A ⊤ A 2 = ⊤ A . Thus
a 2 = a , b 2 = b , a ′ = b , b ′ = a . a^2=a,\qquad b^2=b,\qquad a'=b,\qquad b'=a. a 2 = a , b 2 = b , a ′ = b , b ′ = a .
The closed real interval S 0 = [ a , b ] S_0=[a,b] S 0 = [ a , b ] contains e S = i ( e A ) e_S=i(e_A) e S = i ( e A ) and d S = i ( d A ) d_S=i(d_A) d S = i ( d A ) . For x , y ∈ S 0 x,y\in S_0 x , y ∈ S 0 , monotonicity gives
a = a 2 ≤ x y ≤ b 2 = b . a=a^2\le xy\le b^2=b. a = a 2 ≤ x y ≤ b 2 = b .
The involution exchanges the endpoints and therefore preserves S 0 S_0 S 0 . Since x → y = ( x y ′ ) ′ x\mathbin{\to}y=(xy')' x → y = ( x y ′ ) ′ , the interval is closed under the ambient residual itself, as well as under the inherited lattice operations. It is a full subalgebra with bottom a a a and top b b b ; no truncation or alteration of the residual is involved. Nontriviality and injectivity give a < b a<b a < b . Transport the operations of S 0 S_0 S 0 along t ↦ ( t − a ) / ( b − a ) t\mapsto(t-a)/(b-a) t ↦ ( t − a ) / ( b − a ) . The resulting algebra has carrier [ 0 , 1 ] [0,1] [ 0 , 1 ] , and the composite embedding preserves all operations and all four constants e , d , ⊥ , ⊤ e,d,\bot,\top e , d , ⊥ , ⊤ . □
Standard embeddings and strong standard completeness
Corollary 9.5 (Bounded standard embeddings for IUL). Every nontrivial countable IUL-chain embeds into a standard IUL-chain preserving its entire bounded signature.
Proof. Let A A A be a nontrivial countable IUL-chain. Lemma 9.1 places its reduct in K R \mathcal{K}_R K R . Theorem 7.1, specialized as in Proposition 9.3, supplies a full embedding of that reduct into a commutative standard chain. Lemma 9.4 then restores preservation of the bounded signature. □
For any class of algebras in the signature of this section, T ⊨ φ T\models\varphi T ⊨ φ denotes consequence under the designated condition e ≤ v ( ψ ) e\le v(\psi) e ≤ v ( ψ ) , with all premises interpreted by the same valuation. The same convention will be used for pseudo-uninorm logic in the next section.
Corollary 9.6 (Strong standard completeness for IUL). In a countable propositional language, for every set T T T of formulas and every formula φ \varphi φ ,
T ⊢ I U L φ ⟺ T ⊨ S t d ( I U L ) φ . T \vdash_{\mathrm{IUL}} \varphi\quad\Longleftrightarrow\quad T \models_{\mathrm{Std(IUL)}} \varphi. T ⊢ IUL φ ⟺ T ⊨ Std ( IUL ) φ . No fixed-point axiom e ↔ d e \leftrightarrow d e ↔ d is imposed.
Proof. Use ordinary chain completeness for arbitrary theories [16 Theorem 1, with Definitions 2 and 4–6] . Its constant f f f is our d d d , and commutativity identifies its two residuals. If T ⊬ I U L φ T \nvdash_{\mathrm{IUL}} \varphi T ⊬ IUL φ , there are an IUL-chain A A A and a valuation v v v with e A ≤ v ( γ ) e_A \leq v(\gamma) e A ≤ v ( γ ) for every γ ∈ T \gamma\in T γ ∈ T , whereas v ( φ ) < e A v(\varphi) < e_A v ( φ ) < e A . Let A 0 A_0 A 0 be the subalgebra generated by the values of all propositional variables and all signature constants, including ⊥ \bot ⊥ and ⊤ \top ⊤ . It is countable because the language is countable and the signature is finitary. It remains an IUL-chain, retains all these inequalities, and is nontrivial since v ( φ ) < e A v(\varphi) < e_A v ( φ ) < e A .
Corollary 9.5 supplies a full bounded standard embedding j : A 0 ↪ S j : A_0 \hookrightarrow S j : A 0 ↪ S . For v S = j ∘ v v_S = j \circ v v S = j ∘ v , with v v v now valued in A 0 A_0 A 0 , induction on formulas gives v S ( ψ ) = j ( v ( ψ ) ) v_S(\psi) = j(v(\psi)) v S ( ψ ) = j ( v ( ψ )) for every formula ψ \psi ψ . Thus e S ≤ v S ( γ ) e_S \leq v_S(\gamma) e S ≤ v S ( γ ) for every γ ∈ T \gamma\in T γ ∈ T , while v S ( φ ) < e S v_S(\varphi) < e_S v S ( φ ) < e S . This is a standard countermodel. The reverse implication follows from soundness. Only the ordinary theory-chain-completeness result is used from the cited source, not a standard-completeness result. □
Corollary 9.7 (Standard embeddings and standard completeness for IUL). Every nontrivial countable IUL-chain embeds into a standard IUL-chain preserving its entire bounded signature. Consequently, using the ordinary chain completeness of IUL,
I U L ⊢ φ ⟺ φ is valid in every standard IUL-chain . \mathrm{IUL} \vdash\varphi\quad\Longleftrightarrow\quad\varphi\text{ is valid in every standard IUL-chain}. IUL ⊢ φ ⟺ φ is valid in every standard IUL-chain . Proof. The embedding assertion is Corollary 9.5; the theoremhood equivalence is the case T = ∅ T = \varnothing T = ∅ of Corollary 9.6. □
Transfer to pseudo-uninorm logic
We use the convention
a b ≤ c ⟺ b ≤ a \ c ⟺ a ≤ c / b . ab \le c \Longleftrightarrow b \le a\backslash c \Longleftrightarrow a \le c/b. ab ≤ c ⟺ b ≤ a \ c ⟺ a ≤ c / b .
A p s U L ∗ \mathrm{psUL}^{*} psUL ∗ -chain is a bounded residuated chain
A = ( A , ∧ , ∨ , ⋅ , \ , / , e , ⊥ A , ⊤ A ) A = (A,\wedge,\vee,\cdot,\backslash,/ ,e,\bot_{A},\top_{A}) A = ( A , ∧ , ∨ , ⋅ , \ , / , e , ⊥ A , ⊤ A )
satisfying unit cyclicity
a b ≤ e ⟺ b a ≤ e . (33) ab \le e \Longleftrightarrow ba \le e. \tag*{(33)} ab ≤ e ⟺ ba ≤ e . (33)
In Wang’s notation [7 ] , a → b = a \ b a \to b = a\backslash b a → b = a \ b , a ⇝ b = b / a a \rightsquigarrow b = b/a a ⇝ b = b / a , and t = e t=e t = e . The added weak commutativity axiom is equivalent to e / a ≤ a \ e e/a \le a\backslash e e / a ≤ a \ e for every a a a . Indeed, this inequality sends b a ≤ e ba \le e ba ≤ e to a b ≤ e ab \le e ab ≤ e ; interchanging a , b a,b a , b gives the converse. Thus it is equivalent to (15). No involution is assumed in A A A .
The empty-theory target is the theoremhood statement of [7 Theorem 5] ; Corollary 10.7 below strengthens it to arbitrary theories:
H p s U L ∗ ⊢ φ ⟺ v ( φ ) ≥ e in every standard p s U L ∗ -chain . (34) \mathrm{HpsUL}^{*} \vdash\varphi\Longleftrightarrow v(\varphi) \ge e \text{ in every standard }\mathrm{psUL}^{*}\text{-chain}. \tag*{(34)} HpsUL ∗ ⊢ φ ⟺ v ( φ ) ≥ e in every standard psUL ∗ -chain . (34)
Here “standard” means that the lattice reduct is [ 0 , 1 ] [0,1] [ 0 , 1 ] , with its usual bounds. This convention does not require the monoid unit to be strictly below 1 1 1 . The star is essential: the target is H p s U L ∗ \mathrm{HpsUL}^{*} HpsUL ∗ , not the unextended H p s U L \mathrm{HpsUL} HpsUL .
The representation input is Theorem 7.1 , with K R K_{R} K R defined by (6) and all ordered history conditions (7) . The transfer below supplies the embedding and endpoint arguments needed to apply that theorem to non-involutive inputs. It uses ordinary chain completeness, but neither density elimination nor the standard completeness result of [7 ] .
The upper-copy involutive extension
The following extension is the construction of Galatos–Raftery [17 Section 9, Theorem 9.1] . We give the operation-level verification needed for the present application.
Take a disjoint order-dual copy A ˉ = { a ˉ : a ∈ A } \bar{A}=\{\bar{a}:a\in A\} A ˉ = { a ˉ : a ∈ A } and two new endpoints. Order
R ( A ) = { 0 R } ∪ ˙ A ∪ ˙ A ˉ ∪ ˙ { 1 R } R(A)=\{0_{R}\}\mathbin{\dot\cup}A\mathbin{\dot\cup}\bar{A}\mathbin{\dot\cup}\{1_{R}\} R ( A ) = { 0 R } ∪ ˙ A ∪ ˙ A ˉ ∪ ˙ { 1 R }
by 0 R < a < b ˉ < 1 R 0_R<a<\bar{b}<1_R 0 R < a < b ˉ < 1 R for all a , b ∈ A a,b\in A a , b ∈ A , and a ˉ ≤ b ˉ \bar{a}\le\bar{b} a ˉ ≤ b ˉ if and only if b ≤ a b\le a b ≤ a . Define an order-reversing involution by
a ′ = a ˉ , ( a ˉ ) ′ = a , 0 R ′ = 1 R , 1 R ′ = 0 R , d = e ˉ . a'=\bar{a},\qquad(\bar{a})'=a,\qquad0_R'=1_R,\qquad1_R'=0_R,\qquad d=\bar{e}. a ′ = a ˉ , ( a ˉ ) ′ = a , 0 R ′ = 1 R , 1 R ′ = 0 R , d = e ˉ .
The product of internal elements is
a b = ( a b ) A , a b ˉ = b ˉ / a , a ˉ b = b ˉ \ a , a ˉ b ˉ = 1 R . (35) ab=(ab)_A,\qquad a\bar{b}=\bar{b}/a,\qquad\bar{a}b=\bar{b}\backslash a,\qquad\bar{a}\bar{b}=1_R. \tag*{(35)} ab = ( ab ) A , a b ˉ = b ˉ / a , a ˉ b = b ˉ \ a , a ˉ b ˉ = 1 R . (35)
Set 0 R x = x 0 R = 0 R 0_Rx=x0_R=0_R 0 R x = x 0 R = 0 R for all x x x , and 1 R x = x 1 R = 1 R 1_Rx=x1_R=1_R 1 R x = x 1 R = 1 R whenever x ≠ 0 R x\ne0_R x = 0 R . The zero rule takes precedence, so 0 R 1 R = 1 R 0 R = 0 R 0_R1_R=1_R0_R=0_R 0 R 1 R = 1 R 0 R = 0 R .
Lemma 10.1. For every residuated chain A A A , R ( A ) \mathrm{R}(A) R ( A ) is a cyclic involutive residuated chain. The inclusion of A A A preserves its lattice operations, multiplication, monoid unit, and both residuals. Furthermore,
x y = 0 R ⟺ x = 0 R or y = 0 R . (36) xy=0_R\quad\Longleftrightarrow\quad x=0_R\text{ or }y=0_R. \tag*{(36)} x y = 0 R ⟺ x = 0 R or y = 0 R . (36)
Proof. The identity e e e acts as identity on both copies, since a / e = a = e \ a a/e=a=e\backslash a a / e = a = e \ a , and on the endpoints by definition. Equation (36) follows directly from the product definitions.
We check associativity exhaustively by the types of the three factors. For three factors in A A A , use old associativity. When exactly one factor is in A ˉ \bar{A} A ˉ , the three positions reduce to
c / ( a b ) = ( c / b ) / a , c \ ( b / a ) = ( c \ b ) / a , ( b c ) \ a = c \ ( b \ a ) . \begin{aligned}
c/(ab)&=(c/b)/a,\\
c\backslash(b/a)&=(c\backslash b)/a,\\
(bc)\backslash a&=c\backslash(b\backslash a).
\end{aligned} c / ( ab ) c \ ( b / a ) ( b c ) \ a = ( c / b ) / a , = ( c \ b ) / a , = c \ ( b \ a ) .
For example, the middle identity holds because an element t t t is below either side exactly when c t a ≤ b cta\le b c t a ≤ b . The other two follow in the same way from t a b ≤ c tab\le c t ab ≤ c and b c t ≤ a bct\le a b c t ≤ a , respectively. When at least two factors are in A ˉ \bar{A} A ˉ , both bracketings give 1 R 1_R 1 R . A factor 0 R 0_R 0 R makes both bracketings 0 R 0_R 0 R . If 1 R 1_R 1 R occurs without 0 R 0_R 0 R , both bracketings give 1 R 1_R 1 R , using (36) . This covers all triples.
We next check the cyclic comparison law
x y ≤ z ′ ⟺ y z ≤ x ′ . (37) xy\le z'\quad\Longleftrightarrow\quad yz\le x'. \tag*{(37)} x y ≤ z ′ ⟺ y z ≤ x ′ . (37)
If x , y , z x,y,z x , y , z include 0 R 0_R 0 R , both comparisons hold. If they include 1 R 1_R 1 R but not 0 R 0_R 0 R , both comparisons fail, by (36) . For three factors in A A A , both comparisons hold, since A < A ˉ A < \bar{A} A < A ˉ . For at least two factors in A ˉ \bar{A} A ˉ , both comparisons fail. The remaining case, two old elements and one upper-copy element, is exactly the chain of equivalences
a b ≤ c ⟺ c ˉ / b ≤ a ˉ ⟺ a ˉ \ c ≤ b ˉ . ab \leq c \quad\Longleftrightarrow\quad\bar{c}/b \leq\bar{a} \quad\Longleftrightarrow\quad\bar{a}\backslash c \leq\bar{b}. ab ≤ c ⟺ c ˉ / b ≤ a ˉ ⟺ a ˉ \ c ≤ b ˉ .
These are the three cyclic placements of a , b , c ˉ a,b,\bar{c} a , b , c ˉ .
Define
x \ R y = ( y ′ x ) ′ , y / R x = ( x y ′ ) ′ . (38) x\backslash_R y = (y'x)', \qquad y/Rx = (xy')'. \tag*{(38)} x \ R y = ( y ′ x ) ′ , y / R x = ( x y ′ ) ′ . (38)
Applying (19) to ( x , t , y ′ ) (x,t,y') ( x , t , y ′ ) , twice if necessary, gives
x t ≤ y ⟺ t ≤ ( y ′ x ) ′ ⟺ x ≤ ( t y ′ ) ′ . xt \leq y \quad\Longleftrightarrow\quad t \leq(y'x)' \quad\Longleftrightarrow\quad x \leq(ty')'. x t ≤ y ⟺ t ≤ ( y ′ x ) ′ ⟺ x ≤ ( t y ′ ) ′ .
Thus these are actual left and right residuals, and in particular the product is monotone in each coordinate. Moreover, x \ R d = ( e x ) ′ = x ′ = d / R x x\backslash_R d = (ex)' = x' = d/Rx x \ R d = ( e x ) ′ = x ′ = d / R x , so d d d is cyclic dualizing. For old a , b ∈ A a,b \in A a , b ∈ A , the definitions give
a \ R b = ( b ˉ a ) ′ = a \ A b , b / R a = ( a b ˉ ) ′ = b / A a . a\backslash_R b = (\bar{b}a)' = a\backslash_A b,\qquad b/Ra = (a\bar{b})' = b/_A a. a \ R b = ( b ˉ a ) ′ = a \ A b , b / R a = ( a b ˉ ) ′ = b / A a .
The total order restricts to the old order, so the lattice operations are also preserved. The old bounds need not be the new endpoints.
Why every ordered history is satisfied
Lemma 10.2. If A A A satisfies unit cyclicity, then R ( A ) ∈ K R \mathrm{R}(A) \in K_R R ( A ) ∈ K R .
Proof. For x , y ≠ 0 R x,y \ne0_R x , y = 0 R , the comparison x y ≤ e xy \leq e x y ≤ e can hold only when both x , y ∈ A x,y \in A x , y ∈ A : every other product lies in A ˉ \bar{A} A ˉ or equals 1 R 1_R 1 R . Hence unit cyclicity in R ( A ) \mathrm{R}(A) R ( A ) follows from unit cyclicity in A A A ; the zero cases are symmetric.
Since d = e ˉ d = \bar{e} d = e ˉ , we have
d 2 = 1 R , e + e = ( d 2 ) ′ = 0 R , d 0 R = 0 R , e + 0 R = ( d 1 R ) ′ = 0 R . d^2 = 1_R,\qquad e+e = (d^2)' = 0_R,\qquad d0_R = 0_R,\qquad e+0_R = (d1_R)' = 0_R. d 2 = 1 R , e + e = ( d 2 ) ′ = 0 R , d 0 R = 0 R , e + 0 R = ( d 1 R ) ′ = 0 R . Consequently
( c S , h S ) = ( d , 0 R ) , ( c T , h T ) = ( 0 R , d ) . (c_S,h_S) = (d,0_R),\qquad(c_T,h_T) = (0_R,d). ( c S , h S ) = ( d , 0 R ) , ( c T , h T ) = ( 0 R , d ) . The recursion (4) preserves a zero coordinate in either position. Induction on the word length therefore yields
w ≠ ϵ ⟹ c w = 0 R or h w = 0 R . (39) w \ne\epsilon\quad\Longrightarrow\quad c_w = 0_R\ \text{or}\ h_w = 0_R. \tag*{(39)} w = ϵ ⟹ c w = 0 R or h w = 0 R . (39)
If a b ≤ c w ab \leq c_w ab ≤ c w and u v ≤ h w uv \leq h_w uv ≤ h w for such a word, one of a b ab ab , u v uv uv is 0 R 0_R 0 R . By (18), at least one of a a a , b b b , u u u , v v v equals 0 R 0_R 0 R , so b a z v u = 0 R ≤ z bazvu = 0_R \leq z ba z v u = 0 R ≤ z . For w = ϵ w = \epsilon w = ϵ , unit cyclicity gives
a b ≤ e , u v ≤ e ⟹ b a ≤ e , v u ≤ e ⟹ b a z v u ≤ e z e = z . ab \leq e,\quad uv \leq e \quad\Longrightarrow\quad ba \leq e,\quad vu \leq e \quad\Longrightarrow\quad bazvu \leq eze = z. ab ≤ e , uv ≤ e ⟹ ba ≤ e , v u ≤ e ⟹ ba z v u ≤ ez e = z .
Thus all histories, including the empty one, hold.
This is the additional check that permits application of Theorem 7.1. The original chain need not carry an involution, a dualizer, or any extra history axioms. The reflection is countable whenever A A A is.
Recovering the original bounds
Lemma 10.3. Let A A A be a nontrivial bounded residuated chain and let i : A ⟶ S i : A \longrightarrow S i : A ⟶ S be an order embedding into a residuated chain on [ 0 , 1 ] [0,1] [ 0 , 1 ] preserving multiplication, both residuals, and e e e , but not necessarily the bounds. Then i i i factors through a full bounded embedding into a residuated chain on [ 0 , 1 ] [0,1] [ 0 , 1 ] , obtained by restricting S S S to the interval between the images of the two old bounds and rescaling.
Proof. Put α = i ( ⊥ A ) \alpha= i(\bot_A) α = i ( ⊥ A ) and β = i ( ⊤ A ) \beta= i(\top_A) β = i ( ⊤ A ) . In every bounded residuated lattice the bottom is absorbing. Since e ≤ ⊤ A e \leq\top_A e ≤ ⊤ A , the top is idempotent. Preservation of old operations gives
α 2 = α , β 2 = β , α β = β α = α , α \ S β = β = β / S α . (40) \alpha^2 = \alpha,\quad\beta^2 = \beta,\quad\alpha\beta= \beta\alpha= \alpha,\quad\alpha\backslash_S\beta= \beta= \beta/_S\alpha. \tag*{(40)} α 2 = α , β 2 = β , α β = β α = α , α \ S β = β = β / S α . (40)
The real interval J = [ α , β ] J = [\alpha,\beta] J = [ α , β ] contains i ( e ) i(e) i ( e ) . For x , y ∈ J x,y \in J x , y ∈ J ,
α = α 2 ≤ x y ≤ β 2 = β . \alpha= \alpha^2 \leq xy \leq\beta^2 = \beta. α = α 2 ≤ x y ≤ β 2 = β .
Moreover, both ambient residuals stay in J J J :
α ≤ x \ S y ≤ α \ S β = β , α ≤ y / S x ≤ β / S α = β . (41) \alpha\leq x\backslash_S y \leq\alpha\backslash_S\beta= \beta,\quad\alpha\leq y/_S x \leq\beta/_S\alpha= \beta. \tag*{(41)} α ≤ x \ S y ≤ α \ S β = β , α ≤ y / S x ≤ β / S α = β . (41)
The upper bounds use the order properties of residuals. The lower bounds follow respectively from x α ≤ β α = α ≤ y x\alpha\leq\beta\alpha= \alpha\leq y xα ≤ β α = α ≤ y and α x ≤ α β = α ≤ y \alpha x \leq\alpha\beta= \alpha\leq y α x ≤ α β = α ≤ y . Thus restriction retains the actual residuals; no truncation is used. The lattice operations restrict automatically, and the new bounds are α , β \alpha,\beta α , β . Nontriviality and injectivity imply α < β \alpha< \beta α < β . Transport all operations along
θ : J ⟶ [ 0 , 1 ] , θ ( t ) = t − α β − α . \theta: J \longrightarrow[0,1],\quad\theta(t) = \frac{t-\alpha}{\beta-\alpha}. θ : J ⟶ [ 0 , 1 ] , θ ( t ) = β − α t − α .
Then θ ∘ i \theta\circ i θ ∘ i preserves the entire bounded residuated signature. □
Lemma 10.4. Every densely ordered residuated chain is unit-cyclic.
Proof. Suppose a b ≤ e ab \le e ab ≤ e and b a > e ba > e ba > e . Residuation gives b \ e < a b\backslash e < a b \ e < a . Choose b \ e < r < a b\backslash e < r < a b \ e < r < a by density. Then b r > e br > e b r > e , and hence
a ≤ a ( b r ) = ( a b ) r ≤ r < a , a \le a(br) = (ab)r \le r < a, a ≤ a ( b r ) = ( ab ) r ≤ r < a , a contradiction. Interchange a , b a,b a , b for the converse.
Corollary 10.5 (Full standard embedding for pseudo-uninorm chains). Every nontrivial countable bounded unit-cyclic residuated chain embeds into a standard psUL∗ {}^{*} ∗ -chain, preserving
∧ , ∨ , ⋅ , \ , / , e , ⊥ , ⊤ . \wedge, \vee, \cdot, \backslash, /, e, \bot, \top. ∧ , ∨ , ⋅ , \ , / , e , ⊥ , ⊤. Proof. By Lemmas 10.1 and 10.2, A A A embeds as a residuated reduct into the countable member R ( A ) R(A) R ( A ) of K R \mathcal{K}_{R} K R . Apply Theorem 7.1 to obtain a full embedding R ( A ) → S R(A) \to S R ( A ) → S , where S S S has underlying order [ 0 , 1 ] [0,1] [ 0 , 1 ] . The composite preserves the old residuated signature. Lemma 10.3 makes it preserve the two old bounds as well. The resulting standard residuated chain is unit-cyclic by Lemma 10.4 , and is therefore a standard psUL∗ {}^{*} ∗ -chain.
Corollary 10.6 (Countable densification for pseudo-uninorm chains). Every nontrivial countable psUL∗ {}^{*} ∗ -chain has a full bounded embedding into a countable densely ordered psUL∗ {}^{*} ∗ -chain.
Proof. In the standard target of Corollary 10.5, take the subalgebra generated by the image of A A A together with Q ∩ ( 0 , 1 ) \mathbb{Q} \cap(0,1) Q ∩ ( 0 , 1 ) . Both endpoints already belong to the image, since the embedding preserves the bounds. The signature is finite, so this subalgebra is countable. The included rational points make its order dense. All defining operations, bounds, and unit cyclicity are retained.
Strong standard completeness
For the theory-level statement, the consequence relation is that of the calculus in [16 ] , Definitions 2 and 6, Theorem 1. For the displayed signature it is restricted to theories and conclusions not containing the additional constant f f f of that presentation. The result also covers its full language with f f f : interpret f f f as an arbitrary distinguished element, include its value among the generators below, and interpret it in the target as its image under the embedding. It is not identified with the dualizer of the auxiliary involutive extension.
Corollary 10.7 (Strong standard completeness for H p s U L ∗ \mathrm{HpsUL}^{*} HpsUL ∗ ). In a countable propositional language, for every theory T T T and formula φ \varphi φ in the chosen presentation just specified,
T ⊢ H p s U L ∗ φ ⟺ T ⊨ S t d ( p s U L ∗ ) φ . T \vdash_{\mathrm{HpsUL}^{*}} \varphi\quad\Longleftrightarrow\quad T \models_{\mathrm{Std}(\mathrm{psUL}^{*})} \varphi. T ⊢ HpsUL ∗ φ ⟺ T ⊨ Std ( psUL ∗ ) φ . In the presentation with f f f , the standard models in this formula include all interpretations of that additional constant.
Proof. Ordinary chain completeness for arbitrary theories [16 Theorem 1, with Definitions 2 and 4–6] gives, from T ⊬ H p s U L ∗ φ T \nvdash_{\mathrm{HpsUL}^{*}} \varphi T ⊬ HpsUL ∗ φ , a p s U L ∗ \mathrm{psUL}^{*} psUL ∗ -chain A A A and a valuation v v v satisfying e A ≤ v ( γ ) e_A \leq v(\gamma) e A ≤ v ( γ ) for all γ ∈ T \gamma\in T γ ∈ T and v ( φ ) < e A v(\varphi) < e_A v ( φ ) < e A . Take the subalgebra A 0 A_0 A 0 generated by all variable values and all constants of the chosen signature, including both bounds and, when present, f A f_A f A . It is countable, nontrivial, remains unit-cyclic, and retains all these premise and conclusion values.
By Corollary 10.5, its residuated reduct has a standard embedding j : A 0 ↪ S j \colon A_0 \hookrightarrow S j : A 0 ↪ S preserving both residuals, multiplication, lattice operations, the unit, and the two bounds. If f f f is present, set f S = j ( f A ) f_S = j(f_A) f S = j ( f A ) . Induction on formulas then gives v S ( ψ ) = j ( v ( ψ ) ) v_S(\psi) = j(v(\psi)) v S ( ψ ) = j ( v ( ψ )) for every formula ψ \psi ψ , where v S = j ∘ v v_S = j \circ v v S = j ∘ v . In particular, all members of T T T remain designated and v S ( φ ) < e S v_S(\varphi) < e_S v S ( φ ) < e S . This refutes standard consequence. The converse is soundness. No density-elimination or standard-completeness theorem is used as an input.
Corollary 10.8 (Standard completeness for H p s U L ∗ \mathrm{HpsUL}^{*} HpsUL ∗ ). The standard theoremhood equivalence (34) holds.
Proof. Take T = ∅ T = \varnothing T = ∅ in Corollary 10.7, in the displayed signature without the additional constant f f f .
Remark 10.9 (Scope). The reflection construction is due to [17 ] ; the history and endpoint checks above make the representation theorem applicable to it. Corollary 10.6 supplies the lower densification link in Wang’s Figure 4, while Corollary 10.5 gives a direct standard embedding. The logical conclusions use ordinary theory-chain completeness, not Wang’s density-elimination or standard-completeness result. The monoid unit may equal 1, as it must for an integral input whose bounds are preserved; this is allowed by the standard-algebra convention in [7 Theorem 5] .
Technical appendices
The following appendices contain the detailed auxiliary proofs used in the main text. Numbered references point to the main sections; lettered references point to the appendices. All constructions use the original signature and the class K R \mathcal{K}_{R} K R defined in Section 2. No result from an unpublished working manuscript is used as a theorem.
The countable comparison construction
For a set T \mathcal{T} T of specified single-component sequents, write T ⊢ H \mathcal{T} \vdash H T ⊢ H if H H H has a finite H R \mathrm{H}_{R} H R proof allowing members of T \mathcal{T} T as additional leaves. An additional leaf is its specified sequent, not a substitution schema. Repeated uses of it are allowed. The original axioms retain all their instances.
Lemma A.1 (External continuation). For a specified sequent S S S and finite hypersequents J J J , K K K ,
T ∪ { S } ⊢ K , T ⊢ J ∣ S ⟹ T ⊢ J ∣ K . \mathcal{T} \cup\{S\} \vdash K,\qquad\mathcal{T} \vdash J \mid S \quad\Longrightarrow\quad\mathcal{T} \vdash J \mid K. T ∪ { S } ⊢ K , T ⊢ J ∣ S ⟹ T ⊢ J ∣ K . Proof. Induct on the first supplied proof, maintaining the entire multiset J J J in the constructed external background. Replace each extra leaf S S S by the second supplied proof. At every other leaf add J J J by EW. Unary logical rules, Cyc, WCM, and EW retain J J J . At EC contract the two occurrences selected by the old inference, leaving the extra background intact.
At a product, Cut, or COM inference, the lifted premises supply two copies of J J J ; replay that inference and use EC on literally identical components until the multiplicities of J J J are restored. The same construction contracts the three copies supplied at R w R_w R w . For shared conjunction, the lifted premises have the identical background J ∣ G J \mid G J ∣ G and the identical ordered contexts; apply the displayed ( ∧ ) (\wedge) ( ∧ ) rule with that background. No unequal components are contracted or aligned. These are all the rules. The resulting proof is finite, since both supplied proofs are finite.
Lemma A.2 (Comparison orientation). If T ⊬ H \mathcal{T} \nvdash H T ⊬ H , then for every pair of formulas A A A , B B B at least one of
T ∪ { ⇒ A ′ , B } ⊬ H , T ∪ { ⇒ B ′ , A } ⊬ H \mathcal{T} \cup\{\Rightarrow A', B\} \nvdash H,\qquad\mathcal{T} \cup\{\Rightarrow B', A\} \nvdash H T ∪ { ⇒ A ′ , B } ⊬ H , T ∪ { ⇒ B ′ , A } ⊬ H holds.
Proof. Set S ⇒ A ′ , B S \mathrel{\Rightarrow} A', B S ⇒ A ′ , B and T ⇒ B ′ , A T \mathrel{\Rightarrow} B', A T ⇒ B ′ , A . COM applied to ⇒ A ′ , A \mathrel{\Rightarrow} A', A ⇒ A ′ , A and ⇒ B ′ , B \mathrel{\Rightarrow} B', B ⇒ B ′ , B proves S ∣ T S \mid T S ∣ T . If both enlarged sets proved H H H , first lift the continuation from S S S over the full background T T T , obtaining T ⊢ T ∣ H \mathcal{T} \vdash T \mid H T ⊢ T ∣ H . Lift the continuation from T T T over the full background H H H to obtain T ⊢ H ∣ H \mathcal{T} \vdash H \mid H T ⊢ H ∣ H . EC then gives T ⊢ H \mathcal{T} \vdash H T ⊢ H , a contradiction. This uses both alternatives through their continuations; it never projects one component from a proved hypersequent.
Lemma A.3 (The comparison algebra). For any set T T T of specified single-component sequents, define
A ⪯ B ⟺ T ⊢ ⇒ A ′ , B . (42) A \preceq B \quad\Longleftrightarrow\quad\mathcal{T} \vdash\Rightarrow A', B. \tag*{(42)} A ⪯ B ⟺ T ⊢⇒ A ′ , B . (42) This is a preorder whose symmetric part is a congruence for all formula operations. Its quotient is a cyclic involutive residuated lattice with unit [ e ] [e] [ e ] and dualizer [ d ] [d] [ d ] . If every pair of formulas is comparable, the quotient is a chain.
Proof. Id proves reflexivity and Cut proves transitivity. Cyc gives
A ⪯ B ⟺ B ′ ⪯ A ′ . (43) A \preceq B \quad\Longleftrightarrow\quad B' \preceq A'. \tag*{(43)} A ⪯ B ⟺ B ′ ⪯ A ′ . (43) We give the remaining calculations to specify the ordered quotient without invoking a completeness theorem for another calculus. The ( + ) (+) ( + ) rule has a derivable inverse in every ordered context:
T ⊢ ⇒ Γ , A + B , Δ ⟺ T ⊢ ⇒ Γ , A , B , Δ . (44) \mathcal{T} \vdash\Rightarrow\Gamma, A + B, \Delta\quad\Longleftrightarrow\quad\mathcal{T} \vdash\Rightarrow\Gamma, A, B, \Delta. \tag*{(44)} T ⊢⇒ Γ , A + B , Δ ⟺ T ⊢⇒ Γ , A , B , Δ. (44) Indeed, product applied to ⇒ A ′ , A \Rightarrow A', A ⇒ A ′ , A and ⇒ B ′ , B \Rightarrow B', B ⇒ B ′ , B proves ⇒ ( A + B ) ′ , A , B \Rightarrow(A + B)', A, B ⇒ ( A + B ) ′ , A , B . Rotate the supplied left component to ⇒ Δ , Γ , A + B \Rightarrow\Delta, \Gamma, A + B ⇒ Δ , Γ , A + B , cut against that proof, and rotate back. The reverse implication is ( + ) (+) ( + ) . No internal exchange occurs in this calculation.
If A ⪯ B A \preceq B A ⪯ B and C ⪯ D C \preceq D C ⪯ D , cyclically display their proofs as ⇒ B , A ′ \Rightarrow B, A' ⇒ B , A ′ and ⇒ D , C ′ \Rightarrow D, C' ⇒ D , C ′ . Product gives ⇒ B D , A ′ , C ′ \Rightarrow BD, A', C' ⇒ B D , A ′ , C ′ ; fold A ′ , C ′ A', C' A ′ , C ′ and rotate to obtain A C ⪯ B D AC \preceq BD A C ⪯ B D . The two product bracketings yield actual proofs
⇒ ( A B ) C , A ′ , B ′ , C ′ , ⇒ A ( B C ) , A ′ , B ′ , C ′ \Rightarrow(AB)C, A', B', C', \qquad\Rightarrow A(BC), A', B', C' ⇒ ( A B ) C , A ′ , B ′ , C ′ , ⇒ A ( B C ) , A ′ , B ′ , C ′ from three identities. Folding the negative word using the opposite bracketing proves both comparisons between ( A B ) C (AB)C ( A B ) C and A ( B C ) A(BC) A ( B C ) . For the unit laws, product with ⇒ e \Rightarrow e ⇒ e proves A ⪯ A e A \preceq Ae A ⪯ A e and A ⪯ e A A \preceq eA A ⪯ e A . Conversely insert d d d in ⇒ A ′ , A \Rightarrow A', A ⇒ A ′ , A , and fold A ′ , d A', d A ′ , d or d , A ′ d, A' d , A ′ ; this proves A e ⪯ A Ae \preceq A A e ⪯ A and e A ⪯ A eA \preceq A e A ⪯ A .
Apply a join injection to the negative formula of an identity to obtain A ∧ B ⪯ A A \wedge B \preceq A A ∧ B ⪯ A and A ∧ B ⪯ B A \wedge B \preceq B A ∧ B ⪯ B . If C ⪯ A C \preceq A C ⪯ A and C ⪯ B C \preceq B C ⪯ B , shared conjunction with context C ′ C' C ′ gives C ⪯ A ∧ B C \preceq A \wedge B C ⪯ A ∧ B . Join injections give A ⪯ A ∨ B A \preceq A \vee B A ⪯ A ∨ B and B ⪯ A ∨ B B \preceq A \vee B B ⪯ A ∨ B ; its least-upper-bound property follows by order reversal from the meet property. These properties also give monotonicity of both lattice operations. The operations therefore respect symmetric equivalence, and the quotient has the claimed lattice-ordered monoid structure.
By the definitions of the residual formulas and (26), C ⪯ A \ B C \preceq A\backslash B C ⪯ A \ B means that ⇒ C ′ , B , A ′ \Rightarrow C', B, A' ⇒ C ′ , B , A ′ has a proof. Rotating and folding gives ⇒ A ′ + C ′ , B \Rightarrow A' + C', B ⇒ A ′ + C ′ , B , namely A C ⪯ B AC \preceq B A C ⪯ B . Every step is reversible. Similarly, C ⪯ B / A C \preceq B/A C ⪯ B / A unfolds to ⇒ C ′ , A ′ , B \Rightarrow C', A', B ⇒ C ′ , A ′ , B and folds to C A ⪯ B CA \preceq B C A ⪯ B . Thus both residuation laws hold. Involution and its order reversal follow from A ′ ′ = A A'' = A A ′′ = A and (25); finally
A \ d = ( e A ) ′ ≡ A ′ , d / A = ( A e ) ′ ≡ A ′ . A\backslash d = (eA)' \equiv A', \qquad d/A = (Ae)' \equiv A'. A \ d = ( e A ) ′ ≡ A ′ , d / A = ( A e ) ′ ≡ A ′ .
This verifies the common cyclic dualizer and completes the congruence assertion. Totality passes directly to the quotient order. □
Proof of Theorem 2.5. Enumerate all pairs of formulas as ( A n , B n ) (A_n, B_n) ( A n , B n ) , n ≥ 0 n \ge0 n ≥ 0 . Starting with T 0 = { ⇒ A : A ∈ T } \mathcal{T}_0 = \{\Rightarrow A : A \in T\} T 0 = { ⇒ A : A ∈ T } , use Lemma A.2 to adjoin one of ⇒ A n ′ , B n \Rightarrow A'_n, B_n ⇒ A n ′ , B n and ⇒ B n ′ , A n \Rightarrow B'_n, A_n ⇒ B n ′ , A n while keeping H H H underivable; call the result T n + 1 \mathcal{T}_{n+1} T n + 1 . Put T = ⋃ n T n \mathcal{T} = \bigcup_n \mathcal{T}_n T = ⋃ n T n . A proof is finite, so T ⊬ H \mathcal{T} \nvdash H T ⊬ H : any contrary proof would use extra leaves from a single T n \mathcal{T}_n T n . This argument does not require T 0 \mathcal{T}_0 T 0 to be finite. Every pair is now comparable. Lemma A.3 gives a countable cyclic involutive residuated chain C C C of formula classes.
We verify all additional conditions on this quotient. If [ A ] [ B ] ≤ [ e ] [A][B] \le[e] [ A ] [ B ] ≤ [ e ] , the comparison proof unfolds to ⇒ A ′ , B ′ , e \Rightarrow A', B', e ⇒ A ′ , B ′ , e . WCM gives ⇒ B ′ , A ′ , e \Rightarrow B', A', e ⇒ B ′ , A ′ , e , which folds to [ B ] [ A ] ≤ [ e ] [B][A] \le[e] [ B ] [ A ] ≤ [ e ] . Fix any finite history w w w . If [ A ] [ B ] ≤ [ c w ] [A][B] \le[c_w] [ A ] [ B ] ≤ [ c w ] and [ U ] [ V ] ≤ [ h w ] [U][V] \le[h_w] [ U ] [ V ] ≤ [ h w ] , unfold and rotate their comparison proofs to
T ⊢ ⇒ c w , A ′ , B ′ , T ⊢ ⇒ h w , U ′ , V ′ . \mathcal{T} \vdash\Rightarrow c_w, A', B', \qquad\mathcal{T} \vdash\Rightarrow h_w, U', V'. T ⊢⇒ c w , A ′ , B ′ , T ⊢⇒ h w , U ′ , V ′ .
For an arbitrary formula Z Z Z , use the identity ⇒ Z ′ , Z \Rightarrow Z', Z ⇒ Z ′ , Z as the third premise of R w R_w R w , with
( Σ , Π , Ω , Λ , Γ , Δ ) = ( A ′ , B ′ , U ′ , V ′ , Z ′ , Z ) . (\Sigma, \Pi, \Omega, \Lambda, \Gamma, \Delta) = (A', B', U', V', Z', Z). ( Σ , Π , Ω , Λ , Γ , Δ ) = ( A ′ , B ′ , U ′ , V ′ , Z ′ , Z ) .
The conclusion and its first-five-formula folding give
T ⊢ ⇒ B ′ , A ′ , Z ′ , V ′ , U ′ , Z , [ B ] [ A ] [ Z ] [ V ] [ U ] ≤ [ Z ] . \mathcal{T} \vdash\Rightarrow B', A', Z', V', U', Z, \qquad[B][A][Z][V][U] \le[Z]. T ⊢⇒ B ′ , A ′ , Z ′ , V ′ , U ′ , Z , [ B ] [ A ] [ Z ] [ V ] [ U ] ≤ [ Z ] .
Every element has a formula representative and every closed history formula has its quotient term value. This proves (7) for all elements and every finite w w w , so C ∈ K B C \in\mathcal{K}_{B} C ∈ K B .
Take the canonical valuation v ( P ) = [ P ] v(P) = [P] v ( P ) = [ P ] . Induction on formulas gives v ( A ) = [ A ] v(A) = [A] v ( A ) = [ A ] . If A ∈ T A \in T A ∈ T , the assumption leaf ⇒ A \Rightarrow A ⇒ A and the dualizer rule give T ⊢ ⇒ d , A T \vdash\Rightarrow d,A T ⊢⇒ d , A . Since d = e ′ d = e' d = e ′ , the quotient order gives e C = [ e ] ≤ [ A ] = v ( A ) e_C = [e] \leq[A] = v(A) e C = [ e ] ≤ [ A ] = v ( A ) . Hence v ⊨ T v \models T v ⊨ T .
Suppose that a component ⇒ A 1 , … , A k \Rightarrow A_1,\ldots,A_k ⇒ A 1 , … , A k of H H H were true. Put W = A 1 ′ ⋯ A k ′ W = A'_1\cdots A'_k W = A 1 ′ ⋯ A k ′ with any fixed bracketing, or W = e W = e W = e when k = 0 k = 0 k = 0 . Truth says [ W ] ≤ [ d ] [W] \leq[d] [ W ] ≤ [ d ] , so T ⊢ ⇒ W ′ , d T \vdash\Rightarrow W',d T ⊢⇒ W ′ , d . Cut the last d d d against ⇒ e \Rightarrow e ⇒ e to obtain T ⊢ ⇒ W ′ T \vdash\Rightarrow W' T ⊢⇒ W ′ . For k > 0 k > 0 k > 0 , repeated unfolding in (26) gives the original ordered component. For k = 0 k = 0 k = 0 , W ′ = d W' = d W ′ = d , and a second Cut against ⇒ e \Rightarrow e ⇒ e gives the empty sequent. In either case EW adds the remaining components of H H H , contrary to T ⊬ H T \nvdash H T ⊬ H . Every component is therefore false. For the empty hypersequent falsity is immediate.
The orientation construction is an existence argument, not a decision procedure for choosing the underivability-preserving branch. The proof uses all external backgrounds and allows arbitrary ordinary Cut, external contraction, shared conjunction, and finite histories.
A two-ended construction for odd chains
Throughout this section A A A is an odd cyclic involutive residuated chain: its dualizer is its identity e e e . We construct a full extension in which every old point has a new point on each side. The use of two local identities is essential in the noncommutative case.
The construction treats an element as an arrow between its right and left local identities. We first show how to lift arrows to common identities and compare them there. We then split the local layers while retaining those transports. This produces three ordered copies of the old chain, with the middle copy preserving all original operations.
Lemma B.1 (Local identities, matching products and tails). Put
ℓ ( x ) = x / x = ( x x ′ ) ′ , ρ ( x ) = x \ x = ( x ′ x ) ′ , E = { u ≥ e : u 2 = u } , \ell(x) = x/x = (xx')', \qquad\rho(x) = x\backslash x = (x'x)', \qquad E = \{u \geq e : u^2 = u\}, ℓ ( x ) = x / x = ( x x ′ ) ′ , ρ ( x ) = x \ x = ( x ′ x ) ′ , E = { u ≥ e : u 2 = u } , and H v , u = { x : ℓ ( x ) = v , ρ ( x ) = u } H_{v,u} = \{x : \ell(x) = v,\ \rho(x) = u\} H v , u = { x : ℓ ( x ) = v , ρ ( x ) = u } . Then ℓ ( x ) , ρ ( x ) ∈ E \ell(x),\rho(x) \in E ℓ ( x ) , ρ ( x ) ∈ E , and
ℓ ( x ) x = x = x ρ ( x ) , x x ′ = ℓ ( x ) ′ , x ′ x = ρ ( x ) ′ . (45) \ell(x)x = x = x\rho(x), \qquad xx' = \ell(x)', \qquad x'x = \rho(x)'. \tag*{(45)} ℓ ( x ) x = x = x ρ ( x ) , x x ′ = ℓ ( x ) ′ , x ′ x = ρ ( x ) ′ . (45) ℓ ( x ′ ) = ρ ( x ) , ρ ( x ′ ) = ℓ ( x ) , x ρ ( x ) ′ = ℓ ( x ) ′ x . (46) \ell(x') = \rho(x), \qquad\rho(x') = \ell(x), \qquad x\rho(x)' = \ell(x)'x. \tag*{(46)} ℓ ( x ′ ) = ρ ( x ) , ρ ( x ′ ) = ℓ ( x ) , x ρ ( x ) ′ = ℓ ( x ) ′ x . (46)
For u < v u < v u < v in E E E ,
u v = v u = v , u u ′ = u ′ u = u ′ , v u ′ = u ′ v = v . (47) uv = vu = v,\qquad uu' = u'u = u',\qquad vu' = u'v = v. \tag*{(47)} uv = v u = v , u u ′ = u ′ u = u ′ , v u ′ = u ′ v = v . (47)
Products at matching identities satisfy H v , u H u , w ⊆ H v , w H_{v,u}H_{u,w} \subseteq H_{v,w} H v , u H u , w ⊆ H v , w . For x ∈ H v , u x \in H_{v,u} x ∈ H v , u the maps
α x : [ u , ∞ ) ∩ E ⟶ [ v , ∞ ) ∩ E , α x ( u ) = v , α x ( z ) = x z x ′ ( z > u ) \alpha_x : [u,\infty) \cap E \longrightarrow[v,\infty) \cap E,\qquad\alpha_x(u) = v,\qquad\alpha_x(z) = xzx' \ (z > u) α x : [ u , ∞ ) ∩ E ⟶ [ v , ∞ ) ∩ E , α x ( u ) = v , α x ( z ) = x z x ′ ( z > u )
and α x ′ \alpha_{x'} α x ′ are inverse order isomorphisms. Furthermore
α x ( z ) x = x z , ℓ ( x z ) = α x ( z ) , ρ ( x z ) = z . (48) \alpha_x(z)x = xz,\qquad\ell(xz) = \alpha_x(z),\qquad\rho(xz) = z. \tag*{(48)} α x ( z ) x = x z , ℓ ( x z ) = α x ( z ) , ρ ( x z ) = z . (48)
For z > u z > u z > u , the element x z xz x z has a matching inverse z x ′ zx' z x ′ : their products are α x ( z ) \alpha_x(z) α x ( z ) and z z z .
Proof. Residual evaluation and e ≤ ℓ ( x ) , ρ ( x ) e \leq\ell(x),\rho(x) e ≤ ℓ ( x ) , ρ ( x ) give the local identity equations. Since x ρ ( x ) 2 = x x\rho(x)^2 = x x ρ ( x ) 2 = x , residuation gives ρ ( x ) 2 ≤ ρ ( x ) \rho(x)^2 \leq\rho(x) ρ ( x ) 2 ≤ ρ ( x ) ; the reverse inequality follows from ρ ( x ) ≥ e \rho(x) \geq e ρ ( x ) ≥ e . The left version is identical. The remaining equations follow from the involutive formulas and associativity. For a ≥ e a \geq e a ≥ e , the identities imply ℓ ( a ) , ρ ( a ) ≤ a \ell(a),\rho(a) \leq a ℓ ( a ) , ρ ( a ) ≤ a . Thus a positive idempotent u u u has ℓ ( u ) = ρ ( u ) = u \ell(u) = \rho(u) = u ℓ ( u ) = ρ ( u ) = u and u u ′ = u ′ u = u ′ uu' = u'u = u' u u ′ = u ′ u = u ′ . Two positive idempotents multiply to their maximum by monotonicity. For u < v u < v u < v , put b = v u ′ b = vu' b = v u ′ . Cyclic duality gives b > e b > e b > e , while b ≤ v b \leq v b ≤ v and v b = b vb = b v b = b . Hence v ≤ ℓ ( b ) ≤ b ≤ v v \leq\ell(b) \leq b \leq v v ≤ ℓ ( b ) ≤ b ≤ v , so v u ′ = v vu' = v v u ′ = v . The proof for u ′ v u'v u ′ v uses right identities.
For matching x ∈ H v , u x \in H_{v,u} x ∈ H v , u and y ∈ H u , w y \in H_{u,w} y ∈ H u , w , put a = ℓ ( x y ) ≥ v a = \ell(xy) \geq v a = ℓ ( x y ) ≥ v . Multiplying a x y = x y axy = xy a x y = x y by y ′ y' y ′ gives a x u ′ = x u ′ axu' = xu' a x u ′ = x u ′ , hence a v ′ x = v ′ x av'x = v'x a v ′ x = v ′ x . If a > v a > v a > v , absorption gives a v ′ = a av' = a a v ′ = a and consequently a x = v ′ x ≤ x ax = v'x \leq x a x = v ′ x ≤ x . Thus a x = x ax = x a x = x and a ≤ ℓ ( x ) = v a \leq\ell(x) = v a ≤ ℓ ( x ) = v , a contradiction. Hence ℓ ( x y ) = v \ell(xy) = v ℓ ( x y ) = v . The symmetric calculation gives ρ ( x y ) = w \rho(xy) = w ρ ( x y ) = w .
For z > u z > u z > u , cyclic duality gives a = x z x ′ > e a = xzx' > e a = x z x ′ > e , because x z x ′ ≤ e xzx' \leq e x z x ′ ≤ e would be equivalent to x z ≤ x xz \leq x x z ≤ x , or z ≤ u z \leq u z ≤ u . Absorption yields a 2 = x z u ′ z x ′ = a a^2 = xzu'zx' = a a 2 = x z u ′ z x ′ = a and a x = x z u ′ = x z ax = xzu' = xz a x = x z u ′ = x z . Since x z > x xz > x x z > x , this forces a > v a > v a > v . Also
( x z ) ( z x ′ ) = a , ( z x ′ ) ( x z ) = z . (xz)(zx') = a,\qquad(zx')(xz) = z. ( x z ) ( z x ′ ) = a , ( z x ′ ) ( x z ) = z .
If b = ℓ ( x z ) b = \ell(xz) b = ℓ ( x z ) , then b ≥ a b \geq a b ≥ a and b ( x z ) = x z b(xz) = xz b ( x z ) = x z ; multiplying by z x ′ zx' z x ′ gives b a = a ba = a ba = a . Positive idempotents multiply to their maximum, so b = a b = a b = a . Similarly ρ ( x z ) = z \rho(xz) = z ρ ( x z ) = z . Finally x ′ ( x z x ′ ) x = u ′ z u ′ = z x'(xzx')x = u'zu' = z x ′ ( x z x ′ ) x = u ′ z u ′ = z . Interchanging x , x ′ x,x' x , x ′ proves the inverse assertion. Monotonicity and the endpoint values now give the stated tail order isomorphisms.
We regard x ∈ H v , u x \in H_{v,u} x ∈ H v , u as an arrow v ← u v \leftarrow u v ← u . An arrow is an isomorphism when it has an inverse for matching multiplication with positive identities v , u v,u v , u ; this inverse is denoted x † x^{\dagger} x † . It need not equal x ′ x' x ′ .
Lemma B.2 (The two kinds of local layer). For u ∈ E u \in E u ∈ E write f u = u ′ f_u=u' f u = u ′ . The endofibre H u , u H_{u,u} H u , u is closed under product, prime and both residuals, with identity u u u , dualizer f u f_u f u , and x x ′ = x ′ x = f u xx'=x'x=f_u x x ′ = x ′ x = f u for each of its elements. The element f u f_u f u is central in this endofibre. If u > e u>e u > e , then f u < u f_u<u f u < u is a cover within it. If f u 2 < f u f_u^2<f_u f u 2 < f u , the endofibre is a group . If f u 2 = f u f_u^2=f_u f u 2 = f u , then
{ x ∈ H u , u : f u x = f u } = { f u , u } . (49) \{x\in H_{u,u}:f_ux=f_u\}=\{f_u,u\}. \tag*{(49)} { x ∈ H u , u : f u x = f u } = { f u , u } . (49) For u = e u=e u = e the endofibre is the global group of units.
Proof. Matching products and prime preserve the endofibre, so the residual formulas do too. Its local duality follows from x \ f u = ( u x ) ′ = x † x\backslash f_u=(ux)'=x^\dagger x \ f u = ( ux ) ′ = x † . Centrality is f u x = ( x x ′ ) x = x ( x ′ x ) = x f u f_ux=(xx')x=x(x'x)=xf_u f u x = ( x x ′ ) x = x ( x ′ x ) = x f u . If f u < z < u f_u<z<u f u < z < u in the endofibre, then z ≥ e z\ge e z ≥ e would imply u = ℓ ( z ) ≤ z u=\ell(z)\le z u = ℓ ( z ) ≤ z . If z ≤ e z\le e z ≤ e , then z ′ ≥ e z'\ge e z ′ ≥ e and u = ℓ ( z ′ ) ≤ z ′ u=\ell(z')\le z' u = ℓ ( z ′ ) ≤ z ′ , giving z ≤ f u z\le f_u z ≤ f u . Both alternatives are impossible.
Suppose f u 2 < f u f_u^2<f_u f u 2 < f u . The element h = ( f u 2 ) ′ h=(f_u^2)' h = ( f u 2 ) ′ belongs to the endofibre and satisfies h > u h>u h > u and f u h = h f u ≤ u f_uh=hf_u\le u f u h = h f u ≤ u . If f u h ≤ f u f_uh\le f_u f u h ≤ f u , then h ≤ f u \ f u = ( u f u ) ′ = u h\le f_u\backslash f_u=(uf_u)'=u h ≤ f u \ f u = ( u f u ) ′ = u , a contradiction. The cover just proved gives f u h = h f u = u f_uh=hf_u=u f u h = h f u = u . Thus f u f_u f u is a central unit; f u − 1 x ′ f_u^{-1}x' f u − 1 x ′ is a two-sided inverse of each x x x . If instead f u 2 = f u f_u^2=f_u f u 2 = f u , an element with f u x = f u f_ux=f_u f u x = f u satisfies x ≤ f u \ f u = u x\le f_u\backslash f_u=u x ≤ f u \ f u = u . It also satisfies x ≥ f u x\ge f_u x ≥ f u , because f u ≤ e f_u\le e f u ≤ e and f u x ≤ x f_ux\le x f u x ≤ x . The cover proves (31). At u = e u=e u = e , the equations x x ′ = x ′ x = e xx'=x'x=e x x ′ = x ′ x = e identify exactly the global units.
Partition E E E into { e } \{e\} { e } and
E I = { u > e : f u 2 = f u } , E J = { u > e : f u 2 < f u } . E_I=\{u>e:f_u^2=f_u\},\qquad E_J=\{u>e:f_u^2<f_u\}. E I = { u > e : f u 2 = f u } , E J = { u > e : f u 2 < f u } .
There are no other cases, since f u ≤ e f_u\le e f u ≤ e .
Lemma B.3 (Types and quotient arrows). A nonempty H v , u H_{v,u} H v , u has either u , v ∈ E J u,v\in E_J u , v ∈ E J or u , v ∈ { e } ∪ E I u,v\in\{e\}\cup E_I u , v ∈ { e } ∪ E I . Every arrow of the first kind is an isomorphism . No isomorphism connects e e e to another object. On the second kind put
σ ( x ) = f v x = x f u . \sigma(x)=f_vx=xf_u. σ ( x ) = f v x = x f u . The images, with matching multiplication, form a groupoid P P P , whose identity at u u u is f u f_u f u and whose inverse is σ ( x ) − 1 = σ ( x ′ ) \sigma(x)^{-1}=\sigma(x') σ ( x ) − 1 = σ ( x ′ ) . The image of the isomorphisms
is a subgroupoid. Every fibre of σ \sigma σ has one or two points. It has two points exactly over an isomorphism image with endpoints in E I E_I E I ; these are the lower point σ ( x ) \sigma(x) σ ( x ) and the unique upper isomorphism x x x . If an endpoint is e e e , the fibre is single.
Proof. If f u 2 = f u f_u^2=f_u f u 2 = f u , the intertwining equation gives f v 2 x = f v x f_v^2x=f_vx f v 2 x = f v x and hence f v 3 = f v 2 f_v^3=f_v^2 f v 3 = f v 2 after multiplying by x ′ x' x ′ . If v ∈ E J v\in E_J v ∈ E J , Lemma B.2 allows cancellation and yields f v = v f_v=v f v = v , a contradiction. Interchanging the endpoints proves type agreement. When both endpoints belong to E J E_J E J , the arrow f u − 1 x ′ = x ′ f v − 1 f_u^{-1}x'=x'f_v^{-1} f u − 1 x ′ = x ′ f v − 1 is a matching inverse of x x x . If g x = e gx=e g x = e and x g = v xg=v xg = v , unit cyclicity gives v ≤ e v\le e v ≤ e , so v = e v=e v = e .
On the remaining objects all f u f_u f u are idempotent. Matching composition gives
σ ( x ) σ ( y ) = σ ( x y ) , σ ( x ) σ ( x ′ ) = f v , σ ( x ′ ) σ ( x ) = f u . \sigma(x)\sigma(y)=\sigma(xy),\qquad\sigma(x)\sigma(x')=f_v,\qquad\sigma(x')\sigma(x)=f_u. σ ( x ) σ ( y ) = σ ( x y ) , σ ( x ) σ ( x ′ ) = f v , σ ( x ′ ) σ ( x ) = f u .
Also σ 2 = σ \sigma^2=\sigma σ 2 = σ . These equations prove the groupoid assertion. Isomorphisms compose and invert, proving the subgroupoid assertion. Suppose x < y x<y x < y have equal image in H v , u H_{v,u} H v , u , with u , v ∈ E I u,v\in E_I u , v ∈ E I . The products y x ′ yx' y x ′ and x ′ y x'y x ′ y project to f v , f u f_v,f_u f v , f u . Moreover y x ′ > f v yx'>f_v y x ′ > f v , for otherwise y ≤ f v / x ′ = ( x ′ v ′ ) = x y\le f_v/x'=(x'v')=x y ≤ f v / x ′ = ( x ′ v ′ ) = x ; likewise x ′ y > f u x'y>f_u x ′ y > f u . Equation (31) therefore gives y x ′ = v yx'=v y x ′ = v and x ′ y = u x'y=u x ′ y = u . Thus y y y is an isomorphism and x = x x ′ y = f v y = σ ( y ) x=xx'y=f_vy=\sigma(y) x = x x ′ y = f v y = σ ( y ) . Conversely an isomorphism y y y satisfies σ ( y ) < y \sigma(y)<y σ ( y ) < y : equality, multiplied by y † y^\dagger y † , would give f v = v f_v=v f v = v . The preceding argument proves uniqueness of the upper point. If an endpoint is e e e , the identity f e = e f_e=e f e = e and intertwining make σ \sigma σ the identity on that hom set. □
Write C u = { x : ρ ( x ) = u } C_u=\{x:\rho(x)=u\} C u = { x : ρ ( x ) = u } for an entire right column; its arrows can have different left endpoints. All column orders below are the inherited chain orders, not independently chosen orders on hom sets.
Lemma B.4 (Column order and lifts). On a column of type { e } ∪ E I \{e\}\cup E_I { e } ∪ E I , the map σ \sigma σ is monotone, is strictly increasing on isomorphisms, and has adjacent split pairs in the entire column. For z > ρ ( x ) z>\rho(x) z > ρ ( x ) ,
σ ( x ) z = x z (50) \sigma(x)z=xz \tag*{(50)} σ ( x ) z = x z (50) when σ ( x ) \sigma(x) σ ( x ) is defined. If ρ ( x ) < ρ ( y ) \rho(x)<\rho(y) ρ ( x ) < ρ ( y ) , then
x < y ⟺ x ρ ( y ) ≤ y . (51) x<y\quad\Longleftrightarrow\quad x\rho(y)\le y. \tag*{(51)} x < y ⟺ x ρ ( y ) ≤ y . (51) Matching right multiplication by an isomorphism is an order isomorphism between the entire corresponding columns. The same holds for matching
multiplication in the quotient groupoid P P P . For every u > e u > e u > e there is no point of C u C_u C u strictly between f u f_u f u and u u u . For an isomorphism a ∈ C u a \in C_u a ∈ C u , the point f u f_u f u is its immediate predecessor in C u C_u C u .
Proof. On C u C_u C u , the projection is the single monotone multiplication x ↦ x f u x \mapsto xf_u x ↦ x f u . Equal projected values are equal elements of the old chain and therefore have the same left endpoint; the preceding fibre classification applies. It proves strictness on isomorphisms and adjacency of every split pair in the whole column. Equation (50) is absorption f u z = z f_u z = z f u z = z . For (51) , put v = ρ ( y ) > ρ ( x ) v = \rho(y) > \rho(x) v = ρ ( y ) > ρ ( x ) . We have x < x v x < xv x < xv , because equality would imply v ≤ ρ ( x ) v \le\rho(x) v ≤ ρ ( x ) . Thus x v ≤ y xv \le y xv ≤ y implies x < y x < y x < y . Conversely x < y x < y x < y gives x v ≤ y v = y xv \le yv = y xv ≤ y v = y .
Matching multiplication by a a a and by a † a^\dagger a † are monotone mutual inverses on the whole columns, by the local identity equations. For a quotient arrow b b b , the same reasoning uses b b − 1 = f ℓ ( b ) bb^{-1} = f_{\ell(b)} b b − 1 = f ℓ ( b ) and b − 1 b = f ρ ( b ) b^{-1}b = f_{\rho(b)} b − 1 b = f ρ ( b ) ; every quotient arrow is fixed by its corresponding f f f . The proof of the cover f u < u f_u < u f u < u in Lemma B.2 uses only the right identity: if f u < z < u f_u < z < u f u < z < u , either e ≤ z e \le z e ≤ z gives u = ρ ( z ) ≤ z u = \rho(z) \le z u = ρ ( z ) ≤ z , or z ≤ e z \le e z ≤ e gives u = ℓ ( z ′ ) ≤ z ′ u = \ell(z') \le z' u = ℓ ( z ′ ) ≤ z ′ and z ≤ f u z \le f_u z ≤ f u . Thus it proves the cover in C u C_u C u . Translate this cover by the appropriate matching isomorphism to obtain a f u < a af_u < a a f u < a .
For later calculations put R z x = x z R_zx = xz R z x = x z when z ≥ ρ ( x ) z \ge\rho(x) z ≥ ρ ( x ) . Its target is α x ( z ) \alpha_x(z) α x ( z ) by Lemma B.1. Its left-lift counterpart is L z x = z x = R α x − 1 ( z ) x L_zx = zx = R_{\alpha_x^{-1}(z)}x L z x = z x = R α x − 1 ( z ) x . For arbitrary x , y x,y x , y , putting k = max ( ρ ( x ) , ℓ ( y ) ) k = \max(\rho(x), \ell(y)) k = max ( ρ ( x ) , ℓ ( y )) gives
x y = ( R k x ) ( R α y − 1 ( k ) y ) , ℓ ( x y ) = α x ( k ) , ρ ( x y ) = α y − 1 ( k ) . (52) xy = (R_kx)(R_{\alpha_y^{-1}(k)}y), \qquad\ell(xy) = \alpha_x(k), \qquad\rho(xy) = \alpha_y^{-1}(k). \tag*{(52)} x y = ( R k x ) ( R α y − 1 ( k ) y ) , ℓ ( x y ) = α x ( k ) , ρ ( x y ) = α y − 1 ( k ) . (52)
Indeed x y = ( x k ) ( k y ) xy = (xk)(ky) x y = ( x k ) ( k y ) because k k k is one of the matching old identities, and matching composition then gives the endpoints. On its domain tail,
α x y = α x α y , R z ( x y ) = ( R α y ( z ) x ) ( R z y ) (53) \alpha_{xy} = \alpha_x\alpha_y, \qquad R_z(xy) = (R_{\alpha_y(z)}x)(R_zy) \tag*{(53)} α x y = α x α y , R z ( x y ) = ( R α y ( z ) x ) ( R z y ) (53)
whenever ρ ( x ) = ℓ ( y ) \rho(x) = \ell(y) ρ ( x ) = ℓ ( y ) and z ≥ ρ ( y ) z \ge\rho(y) z ≥ ρ ( y ) . At the initial object these are the local identity equations. Above it, both factors are strict lifts, and their targets are computed by Lemma B.1; associativity of the old product proves the formulas. Also R w R z x = R w x R_wR_zx = R_wx R w R z x = R w x for w ≥ z w \ge z w ≥ z , because positive idempotents multiply to their maximum.
The new objects and arrows
Form a new ordered object set E B E^B E B by inserting u − u^- u − immediately below each u > e u > e u > e , and u + u^+ u + immediately above each u ∈ { e } ∪ E I u \in\{e\} \cup E_I u ∈ { e } ∪ E I . Different old bases retain their order. Every new object has type I I I . Retain all old arrows and add:
for every old isomorphism a : v ← u a:v\leftarrow u a : v ← u with u , v > e u,v>e u , v > e , two arrows a 0 − , a 1 − : v − ← u − a_0^-,a_1^-:v^-\leftarrow u^- a 0 − , a 1 − : v − ← u − ;
for every quotient arrow a : v ← u a:v\leftarrow u a : v ← u of P P P , two arrows a 0 + , a 1 + : v + ← u + a_0^+,a_1^+:v^+\leftarrow u^+ a 0 + , a 1 + : v + ← u + .
Here 0 < 1 0<1 0 < 1 gives the order of the copies of one arrow. Each new column is ordered by its entire underlying old or quotient column, then by this binary mark. At matching new objects define
a i ε b j ε = ( a b ) min ( i , j ) ε , ( a i ε ) ′ = ( a − 1 ) 1 − i ε ( ε ∈ { − , + } ) . (54) a_i^\varepsilon b_j^\varepsilon=(ab)_{\min(i,j)}^\varepsilon,\qquad\left(a_i^\varepsilon\right)'=(a^{-1})_{1-i}^\varepsilon\quad(\varepsilon\in\{-,+\}). \tag*{(54)} a i ε b j ε = ( ab ) m i n ( i , j ) ε , ( a i ε ) ′ = ( a − 1 ) 1 − i ε ( ε ∈ { − , + }) . (54)
The inverse is the matching groupoid inverse: for minus arrows, a − 1 = a † a^{-1}=a^\dagger a − 1 = a † . The upper identity copy is the new object’s identity. Old matched products and prime remain unchanged.
Represent an old arrow by itself, a minus arrow by its old isomorphism, and a plus arrow by its old quotient representative; write x ˉ \bar{x} x ˉ for this representative, forgetting its mark. The lift to the arrow’s own source is itself. For a strict lift to an object s s s with old base z z z define
R s B x = { x z ‾ , s = z , ( x z ‾ ) 1 − , s = z − , ( σ ( x z ‾ ) ) 1 + , s = z + . (55) R_s^B x=
\begin{cases}
\overline{xz},&s=z,\\
(\overline{xz})_1^-,&s=z^-,\\
(\sigma(\overline{xz}))_1^+,&s=z^+.
\end{cases} \tag*{(55)} R s B x = ⎩ ⎨ ⎧ x z , ( x z ) 1 − , ( σ ( x z ) ) 1 + , s = z , s = z − , s = z + . (55)
If x x x is old and s = ρ ( x ) + s=\rho(x)^+ s = ρ ( x ) + , the last expression is ( σ ( x ) ) 1 + (\sigma(x))_1^+ ( σ ( x ) ) 1 + . Every other strict same-base lift begins at a minus object, whose representative is already an isomorphism. At a higher old base, x z ‾ \overline{xz} x z is a strict old lift and hence an isomorphism. Thus every expression is defined in its indicated groupoid. Its target defines a tail map α x B \alpha_x^B α x B . At a higher base z z z this is α x ˉ B ( z ) \alpha_{\bar{x}}^B(z) α x ˉ B ( z ) with the same available mark − , 0 , + -,0,+ − , 0 , + . At the source it is the target of x x x ; same-base transitions retain their marks and change only the base.
Lemma B.5 (Coherent transports). Each α x B \alpha_x^B α x B is an order isomorphism of the upper tails of its source and target. The maps R s B R_s^B R s B are monotone on columns and satisfy
R t B R s B x = R t B x ( t ≥ s ≥ ρ ( x ) ) , (56) R_t^B R_s^B x=R_t^B x\qquad(t\ge s\ge\rho(x)), \tag*{(56)} R t B R s B x = R t B x ( t ≥ s ≥ ρ ( x )) , (56) α R s B x B = α x B ∣ [ s , ∞ ) , (57) \alpha_{R_s^B x}^B=\left.\alpha_x^B\right|_{[s,\infty)}, \tag*{(57)} α R s B x B = α x B [ s , ∞ ) , (57) R s B ( x y ) = ( R α y B ( s ) B x ) ( R s B y ) , α x y B = α x B α y B (58) R_s^B(xy)=\left(R_{\alpha_y^B(s)}^B x\right)\left(R_s^B y\right),\qquad\alpha_{xy}^B=\alpha_x^B\alpha_y^B \tag*{(58)} R s B ( x y ) = ( R α y B ( s ) B x ) ( R s B y ) , α x y B = α x B α y B (58)
for matching x , y x,y x , y and s ≥ ρ ( y ) s \ge\rho(y) s ≥ ρ ( y ) . Every strict lift is an isomorphism, and
( R s B x ) † = R α x ′ B ( s ) B ( x ′ ) ( s > ρ ( x ) ) . (59) \left(R_s^B x\right)^\dagger= R_{\alpha_{x'}^B(s)}^B(x') \qquad(s > \rho(x)). \tag*{(59)} ( R s B x ) † = R α x ′ B ( s ) B ( x ′ ) ( s > ρ ( x )) . (59)
Proof. At a higher base, old tail isomorphisms preserve the available types: the corresponding old strict lift is an isomorphism and therefore connects two equal types. At an initial old object, the same-base tail is 0 < + 0 < + 0 < + for type { e } ∪ E I \{e\} \cup E_I { e } ∪ E I , and only 0 0 0 for type E J E_J E J . At an initial minus object the remaining same-base tail is − < 0 < + - < 0 < + − < 0 < + in type I I I and − < 0 - < 0 − < 0 in type J J J . A plus object has no higher same-base object. These lists coincide at source and target, so the tail maps are order isomorphisms.
Higher-base lifts are old monotone multiplications, optionally followed by σ \sigma σ and insertion as an upper copy. Same-base lifts are inclusion of the isomorphism column, the monotone quotient map, or their composite. Forgetting a binary mark is monotone. This proves column monotonicity.
Coherence at different bases follows from old coherence and σ ( x ) z = x z \sigma(x)z = xz σ ( x ) z = x z for a strict higher-base lift. At one base the only possible composite is inclusion followed by σ \sigma σ along − < 0 < + - < 0 < + − < 0 < + , which is exactly the prescribed − < + - < + − < + map. This proves (38); taking targets proves (39). In (40), equality at the source is matched composition. Otherwise both right-hand factors undergo strict lifts. At higher bases the equality is (35); at the same base it follows because inclusion, σ \sigma σ , and mark-forgetting each preserve matching multiplication. Both output marks are 1, whose product is again 1. Taking targets proves the transport identity. Finally (41) at a higher base is the old strict-lift inverse formula. For minus-to-old it is the definition of † \dagger † ; for old-to-plus it is σ ( x ′ ) = σ ( x ) − 1 \sigma(x') = \sigma(x)^{-1} σ ( x ′ ) = σ ( x ) − 1 ; minus-to-plus is their composite. These exhaust the strict same-base cases.
For arbitrary arrows define, with k = max ( ρ ( x ) , ℓ ( y ) ) k = \max(\rho(x),\ell(y)) k = max ( ρ ( x ) , ℓ ( y )) ,
x ⊙ y = ( R k B x ) ( R ( α y B ) − 1 ( k ) B y ) . (60) x \mathbin{\odot} y = \left(R_k^B x\right)\left(R_{(\alpha_y^B)^{-1}(k)}^B y\right). \tag*{(60)} x ⊙ y = ( R k B x ) ( R ( α y B ) − 1 ( k ) B y ) . (60)
The right side is a matching product. Order equal right columns as specified above, and different columns by
x < y ⟺ { R ρ ( y ) B x ≤ y , ρ ( x ) < ρ ( y ) , x < R ρ ( x ) B y , ρ ( x ) > ρ ( y ) . (61) x < y \quad\Longleftrightarrow\quad
\begin{cases}
R_{\rho(y)}^B x \le y, & \rho(x) < \rho(y),\\
x < R_{\rho(x)}^B y, & \rho(x) > \rho(y).
\end{cases} \tag*{(61)} x < y ⟺ { R ρ ( y ) B x ≤ y , x < R ρ ( x ) B y , ρ ( x ) < ρ ( y ) , ρ ( x ) > ρ ( y ) . (61)
Lemma B.6 (Associativity and the precise new order). The product (42) is associative and has identity e e e . The relation (43) is a total order extending that of A A A . As an ordered set with involution, the extension is
B ( A ) ≅ A × → { − 1 , 0 , 1 } , ( x , i ) ′ = ( x ′ , − i ) , (62) B(A) \cong A \overrightarrow{\times} \{-1,0,1\}, \qquad(x,i)'=(x',-i), \tag*{(62)} B ( A ) ≅ A × { − 1 , 0 , 1 } , ( x , i ) ′ = ( x ′ , − i ) , (62) where ( x , 0 ) (x,0) ( x , 0 ) is the old point.
Proof. The old identity lifts to each new upper identity, proving the unit laws. For associativity write the arrows as x : a ← b x:a\leftarrow b x : a ← b , y : c ← d y:c\leftarrow d y : c ← d , z : f ← g z:f\leftarrow g z : f ← g and put
β 0 = ( α y B ) − 1 ( max ( b , c ) ) , β = max ( β 0 , f ) , γ = ( α z B ) − 1 ( β ) . \beta_0=(\alpha_y^B)^{-1}(\max(b,c)), \qquad\beta=\max(\beta_0,f), \qquad\gamma=(\alpha_z^B)^{-1}(\beta). β 0 = ( α y B ) − 1 ( max ( b , c )) , β = max ( β 0 , f ) , γ = ( α z B ) − 1 ( β ) . The coherence identities reduce either bracketing to the matching product of
R α y B ( β ) B x , R β B y , R γ B z . R^B_{\alpha_y^B(\beta)}x,\qquad R^B_\beta y,\qquad R^B_\gamma z. R α y B ( β ) B x , R β B y , R γ B z . For the second bracketing use that a tail order isomorphism preserves maxima. Matching products are associative, including their binary marks, so the total product is associative.
Trichotomy of (43) follows from column trichotomy. A strict inequality between lifts to a common source forces the corresponding original strict inequality: the opposite weak inequality would lift monotonically in the opposite direction. For transitivity of x < y < z x<y<z x < y < z , lift to the largest of their sources. If that largest source is the first or last, the needed weak or strict comparison follows directly by composing the two column comparisons. If only the middle source is largest, the comparisons there are R x ≤ y < R z Rx\le y<Rz R x ≤ y < R z , so R x < R z Rx<Rz R x < R z . If the first and last sources agree, this forces their strict comparison. If the first is larger, it forces x < R ρ ( x ) z x<R_{\rho(x)}z x < R ρ ( x ) z ; if the last is larger, it forces R ρ ( z ) x ≤ z R_{\rho(z)}x\le z R ρ ( z ) x ≤ z . Otherwise monotonicity would contradict the strict comparison at the middle source. These are exactly the two clauses of (43). Thus the relation is transitive. Its restriction to old points is (33).
We give the positions of all new points, which also proves antitonicity rather than assuming it. For an old isomorphism a a a with source u > e u>e u > e , the point a 1 a_1 a 1 is immediately below a a a relative to the old order, and a 0 a_0 a 0 is immediately above a f u af_u a f u . For a quotient representative a a a , the point a 0 a_0 a 0 is immediately below the bottom of its old σ \sigma σ fibre, whereas a 1 a_1 a 1 is immediately above its top. Here “relative to the old order” means that no old point lies in between. To verify these statements against an old point y y y , separate the cases ρ ( y ) < u \rho(y)<u ρ ( y ) < u , ρ ( y ) = u \rho(y)=u ρ ( y ) = u and ρ ( y ) > u \rho(y)>u ρ ( y ) > u . In the first case (43) tests its lifted old coordinate against a
lower or upper copy; the lower-copy test is strict, the upper-copy test weak. In the equal case use the whole-column adjacent pairs of Lemma B.4 . In the last case both members of an old split pair have the same strict lift, and so do a a a and a f u af_u a f u by absorption. Substitution in (61) gives the stated comparisons in each case.
Consequently every old point x x x has one designated new lower point b ( x ) b(x) b ( x ) and upper point a ( x ) a(x) a ( x ) , as follows:
Table 1 (PDF p. 58).
For the first row, f u − 1 z = z f_u^{-1}z=z f u − 1 z = z at every z > u z>u z > u : multiply f u z = z f_uz=z f u z = z by the local inverse and use u z = z uz=z u z = z . Thus the previous strict-lift comparison applies to its upper point too. Every new point occurs exactly once in this table.
It remains to check a ( x ) < b ( y ) a(x)<b(y) a ( x ) < b ( y ) for old x < y x<y x < y . An intermediate old point already proves it, so assume x < y x<y x < y is a cover. At equal sources, either quotient coordinates differ, or the points form a split pair, or they form a type-J J J predecessor pair. The underlying column order proves the first case, and 0 < 1 0<1 0 < 1 proves the latter two. If ρ ( x ) < ρ ( y ) \rho(x)<\rho(y) ρ ( x ) < ρ ( y ) , the cover forces x ρ ( y ) = y x\rho(y)=y x ρ ( y ) = y , since otherwise this old strict lift lies between them. Thus y y y is an isomorphism, and the lift of a ( x ) a(x) a ( x ) to ρ ( y ) − \rho(y)^{-} ρ ( y ) − is y 1 − = b ( y ) y_1^{-}=b(y) y 1 − = b ( y ) . The first clause of (61) gives the strict comparison. If ρ ( x ) > ρ ( y ) \rho(x)>\rho(y) ρ ( x ) > ρ ( y ) , put h = y ρ ( x ) h=y\rho(x) h = y ρ ( x ) . The immediate predecessor in the column gives x ≤ h f ρ ( x ) < y x\le h f_{\rho(x)}<y x ≤ h f ρ ( x ) < y , so the cover forces x = h f ρ ( x ) x=h f_{\rho(x)} x = h f ρ ( x ) . Now a ( x ) = h 0 − a(x)=h_0^{-} a ( x ) = h 0 − , while the lift of b ( y ) b(y) b ( y ) to ρ ( x ) − \rho(x)^{-} ρ ( x ) − is h 1 − h_1^{-} h 1 − . The second clause gives the result. This proves the lexicographic description in (62) .
Prime exchanges b ( x ) b(x) b ( x ) with a ( x ′ ) a(x') a ( x ′ ) . For type J J J use x ′ = x † f ℓ ( x ) x'=x^\dagger f_{\ell(x)} x ′ = x † f ℓ ( x ) . For a split pair it exchanges the lower point with the upper point over the inverse coordinate: the upper point h h h has h ′ = h † f ℓ ( h ) h'=h^\dagger f_{\ell(h)} h ′ = h † f ℓ ( h ) , and ( σ ( h ) ) ′ = h † (\sigma(h))'=h^\dagger ( σ ( h ) ) ′ = h † , by the residual formulas for a matching isomorphism. On singleton quotient points it is groupoid inversion. Together with (54) , these formulas prove ( x , i ) ′ = ( x ′ , − i ) (x,i)'=(x',-i) ( x , i ) ′ = ( x ′ , − i ) , hence antitonicity.
Theorem B.7 (The odd extension). With the operations above, B ( A ) B(A) B ( A ) is an odd cyclic involutive residuated chain. The inclusion of A A A is full, and every old point has a new point on each side; every strict old pair receives an intermediate point. Every old central unit remains central.
Proof. We prove cyclic duality. In a column with source s > e s > e s > e , comparison with e e e is strict comparison with its column identity s s s , because R s B e = s R^{B}_{s}e=s R s B e = s . In column e e e it is the usual comparison with e e e . For two arrows a , b a,b a , b with the same source,
a ≤ b ⟺ a ⊙ b ′ ≤ e . (63) a \leq b \quad\Longleftrightarrow\quad a \mathbin{\odot} b' \leq e. \tag*{(63)} a ≤ b ⟺ a ⊙ b ′ ≤ e . (63)
In an old column this is old duality and old matching product. In a new column, matching right translation by the inverse of the underlying coordinate of b b b is a strict order isomorphism onto its target column. At unequal coordinates it turns the comparison into strict comparison with the identity coordinate. At equal coordinates the product mark is min ( i , 1 − j ) \min(i,1-j) min ( i , 1 − j ) , which is 0 0 0 exactly when i ≤ j i \leq j i ≤ j . This proves (45).
If ρ ( x ) < ρ ( y ) \rho(x) < \rho(y) ρ ( x ) < ρ ( y ) , then x ⊙ y ′ = ( R ρ ( y ) B x ) y ′ x \mathbin{\odot} y' = (R^{B}_{\rho(y)}x)y' x ⊙ y ′ = ( R ρ ( y ) B x ) y ′ , so (43) and (45) give x < y ⟺ x ⊙ y ′ ≤ e x < y \Longleftrightarrow x \mathbin{\odot} y' \leq e x < y ⟺ x ⊙ y ′ ≤ e . If ρ ( x ) > ρ ( y ) \rho(x) > \rho(y) ρ ( x ) > ρ ( y ) , set b = R ρ ( x ) B y b = R^{B}_{\rho(x)}y b = R ρ ( x ) B y . It is a strict-lift isomorphism with target s = α B ( ρ ( x ) ) > ℓ ( y ) ≥ e s = \alpha_{B}(\rho(x)) > \ell(y) \geq e s = α B ( ρ ( x )) > ℓ ( y ) ≥ e . Equation (41) gives x ⊙ y ′ = x b † x \mathbin{\odot} y' = xb^{\dagger} x ⊙ y ′ = x b † . Right multiplication by b † b^{\dagger} b † is a strict order isomorphism on the whole relevant column: for an old column this is Lemma B.4; for a new one the underlying groupoid translation is an order isomorphism and the upper inverse copy preserves each binary mark. Hence
x < b ⟺ x b † < s ⟺ x ⊙ y ′ ≤ e . x < b \quad\Longleftrightarrow\quad xb^{\dagger} < s \quad\Longleftrightarrow\quad x \mathbin{\odot} y' \leq e. x < b ⟺ x b † < s ⟺ x ⊙ y ′ ≤ e .
Together with the equal-source case this proves x ≤ y ⟺ x ⊙ y ′ ≤ e x \leq y \Longleftrightarrow x \mathbin{\odot} y' \leq e x ≤ y ⟺ x ⊙ y ′ ≤ e universally. Antitonicity and involutivity now give
x ⊙ y ≤ e ⟺ y ≤ x ′ ⟺ y ⊙ x ≤ e . x \mathbin{\odot} y \leq e \quad\Longleftrightarrow\quad y \leq x' \quad\Longleftrightarrow\quad y \mathbin{\odot} x \leq e. x ⊙ y ≤ e ⟺ y ≤ x ′ ⟺ y ⊙ x ≤ e .
Associativity yields both residual adjunctions; for example
x ⊙ y ≤ z ⟺ x ⊙ y ⊙ z ′ ≤ e ⟺ y ≤ ( z ′ ⊙ x ) ′ . x \mathbin{\odot} y \leq z \quad\Longleftrightarrow\quad x \mathbin{\odot} y \mathbin{\odot} z' \leq e \quad\Longleftrightarrow\quad y \leq(z' \mathbin{\odot} x)'. x ⊙ y ≤ z ⟺ x ⊙ y ⊙ z ′ ≤ e ⟺ y ≤ ( z ′ ⊙ x ) ′ .
These adjunctions also imply monotonicity of multiplication. Thus this is an odd cyclic involutive residuated chain and the old inclusion is full. The assigned local endpoints are the actual residual endpoints: matching multiplication gives x x ′ = ℓ ( x ) ′ xx'=\ell(x)' x x ′ = ℓ ( x ) ′ and x ′ x = ρ ( x ) x'x=\rho(x) x ′ x = ρ ( x ) , including the lower identity copy at a new object, and the residual formulas give x / x = ℓ ( x ) x/x=\ell(x) x / x = ℓ ( x ) and x \ x = ρ ( x ) x\backslash x=\rho(x) x \ x = ρ ( x ) . Its type is I I I , because that lower identity is idempotent. The order description proves the insertion assertions.
Finally, every old global unit c c c has both endpoints e e e . Conjugation by c c c preserves old endpoints, types, quotient projections and matching inverses. Extending it to the new arrows by
a i ε ↦ ( c a c − 1 ) i ε a_i^\varepsilon\mapsto(cac^{-1})_i^\varepsilon a i ε ↦ ( c a c − 1 ) i ε
preserves all the defining formulas. This extended map equals actual conjugation by c c c in B ( A ) B(A) B ( A ) : multiplication by c c c on either side lifts it strictly to the relevant new object, using an upper mark, which leaves i i i unchanged under the minimum rule. In the minus family its lifted representatives are c v cv c v and u c − 1 uc^{-1} u c − 1 , giving c a ca c a and c a c − 1 cac^{-1} c a c − 1 ; in the plus family they are c f v cfv c f v and f u c − 1 fuc^{-1} f u c − 1 , and the quotient identity equations give the same products. On old arrows this is old conjugation. If c c c was central, its old conjugation and therefore its extended conjugation are the identity. Thus it stays central. □
Corollary B.8 (Odd densification). Every countable odd cyclic involutive residuated chain has a countable full dense extension without endpoints. All its old central units remain central with their old inverses.
Proof. Iterate Theorem B.7 countably many times. Each step adds two points per old point, so is countable. A pair or point in the union occurs at a finite stage and receives the required intermediate or outer points at the next. Equations, operations and residuals are preserved at every stage. Centrality of each original central unit holds on every stage and hence on the union. □
Corollary B.9 (Relative odd densification). Suppose A A A is countable and directly finite, meaning x y = e ⇒ y x = e xy=e \Rightarrow yx=e x y = e ⇒ y x = e . There is a countable full odd extension A ↪ D A \hookrightarrow D A ↪ D with exactly the same unit group G G G and the following properties: D D D is directly finite; every strict pair with a nonunit endpoint has an intermediate point; every old cover between two units remains a cover; and all old central units remain central. If G G G is nontrivial, D D D has no endpoints.
Proof. In a directly finite odd chain there are no arrows between e e e and another object. Indeed ρ ( x ) = e \rho(x)=e ρ ( x ) = e gives x ′ x = e x'x=e x ′ x = e , and direct finiteness gives x x ′ = e xx'=e x x ′ = e , so ℓ ( x ) = e \ell(x)=e ℓ ( x ) = e ; the converse is symmetric. Delete the newly introduced object e + e^+ e + and all its arrows from B ( A ) B(A) B ( A ) . There are no arrows between this deleted component and any other component, because the corresponding old quotient arrows would have an endpoint e e e and another endpoint different from e e e .
The retained set is product- and prime-closed. More explicitly, all its nonunits have both endpoints strictly above e e e , while old units have endpoints e. Formula (60) shows that a product involving a nonunit still has both endpoints above e e e ; a product of old units is its old product. Thus no retained product falls in the deleted component. Prime exchanges the two endpoints, so is retained as well. The resulting full subalgebra has exactly the old global units. A product equal to e e e must have two old unit factors, which proves direct finiteness. The deleted new points are precisely the two neighbours assigned to each old unit in (44). Every nonunit retains its two neighbours, and a cover between old units receives no new point. Centrality follows by restriction from the full clone. Iterate this restricted construction. Every strict pair with a nonunit endpoint is separated at the next stage, and old unit covers persist at every stage. A nonunit point gets points on both sides. If G G G is nontrivial, every unit has a smaller and a larger old unit, by multiplication by a positive unit or its inverse. Hence the union has no endpoints. □ \square □
Corollary B.10 (Relative densification with a central unit dualizer). Let J J J be a countable directly finite cyclic involutive residuated chain with central invertible dualizer q q q . It has a countable full extension T T T with the same group of units, retaining every old central unit and every old cover between units, in which every strict pair with a nonunit endpoint has an intermediate point. If J J J has a central unit n < r n < r n < r that immediately precedes its identity r r r , the only covers of T T T are g n < g gn < g g n < g for g ∈ U ( J ) g \in U(J) g ∈ U ( J ) , and T T T has no endpoints.
Proof. Normalize to the odd prime q − 1 x ∗ q^{-1}x^* q − 1 x ∗ as in Lemma 4.6, apply Corollary B.9, and restore prime by x ∗ = q x ∘ x^* = qx^\circ x ∗ = q x ∘ . The old q q q remains central with its inverse, so these operations are well defined and preserve the full old signature. Direct finiteness and units concern the unchanged ordered monoid. If n < r n < r n < r is the indicated cover, multiplication by a unit gives a cover g n < g gn < g g n < g in J J J . Conversely every cover between units must be of this form, because translating its upper endpoint to r r r gives the unique predecessor n n n . Every remaining strict pair is separated either by a nonunit already between its endpoints or by the relative construction. The nontrivial unit group supplies the no-endpoints assertion. □ \square □
Technical proofs for the lower-core construction
Old actions and their actual adjoints
Proof of Lemma 5.2. Let R f ( b ) = b Rf(b) = b R f ( b ) = b for an admissible return. If z < f ( b ) z < f(b) z < f ( b ) , then f R ( z ) ≤ z < f ( b ) fR(z) \le z < f(b) f R ( z ) ≤ z < f ( b ) , whence R ( z ) < b R(z) < b R ( z ) < b and R ( z ) ≤ a R(z) \le a R ( z ) ≤ a . Thus z ≤ R ∗ ( a ) z \le R^*(a) z ≤ R ∗ ( a ) . On the other hand R ∗ ( a ) < f ( b ) R^{*}(a)<f(b) R ∗ ( a ) < f ( b ) , since R f ( b ) = b > a Rf(b)=b>a R f ( b ) = b > a . Therefore c = R ∗ ( a ) c=R^{*}(a) c = R ∗ ( a ) is the immediate predecessor of f ( b ) f(b) f ( b ) , independently of the return chosen. Also f ( a ) ≤ c f(a)\le c f ( a ) ≤ c because R f ( a ) ≤ a Rf(a)\le a R f ( a ) ≤ a . These inequalities give monotonicity across the inserted gap. For distinct upper anchors b < t b<t b < t , the only possible failure would be an active later gap with f ( b ) = f ( t ) f(b)=f(t) f ( b ) = f ( t ) . Its return would give R f ( b ) = t > b Rf(b)=t>b R f ( b ) = t > b , contrary to R f ≤ id Rf\le\operatorname{id} R f ≤ id . This proves monotonicity on the whole carrier.
For coefficient monotonicity, suppose u ≤ u ~ u\le\tilde{u} u ≤ u ~ and v ≤ v ~ v\le\tilde{v} v ≤ v ~ . The old upper values are ordered. A failure at a new input could therefore occur only if they were equal, the larger context were active, and the smaller inactive. But a return ( s , t ) (s,t) ( s , t ) for the larger context is admissible for the smaller, and
b = s u ~ v ~ t = s u v t ≤ b b=s\tilde{u}\tilde{v}t=suvt\le b b = s u ~ v ~ t = s uv t ≤ b
when their upper values agree. Thus the smaller context is active as well, excluding that failure.
For composition write f = ( u , v ) f=(u,v) f = ( u , v ) and h = ( w , z ) h=(w,z) h = ( w , z ) , so h f = ( w u , v z ) hf=(wu,vz) h f = ( w u , v z ) . If ( s , t ) (s,t) ( s , t ) returns h f hf h f at b b b , then ( s w , z t ) (sw,zt) ( s w , z t ) returns f f f at b b b , and ( u s , t v ) (us,tv) ( u s , t v ) returns h h h at f ( b ) f(b) f ( b ) . Their contractive inequalities are cyclic rotations at e e e of s w u ≤ e swu\le e s w u ≤ e and v z t ≤ e vzt\le e v z t ≤ e . Their return equations follow from s w u b v z t = b swubvzt=b s w u b v z t = b , and from this equality multiplied on the left by u u u and on the right by v v v . Conversely, if ( s , t ) (s,t) ( s , t ) returns f f f and ( S , T ) (S,T) ( S , T ) returns h h h at its image, then ( s S , T t ) (sS,Tt) ( s S , T t ) returns h f hf h f : for example s S w u ≤ s u ≤ e sSwu\le su\le e s S w u ≤ s u ≤ e and v z T t ≤ v t ≤ e vzTt\le vt\le e v z T t ≤ v t ≤ e , and the other two inequalities follow by e e e -cyclicity. The endpoint equations compose. Thus h f hf h f is active exactly when both successive actions are active. In that case both evaluations give the unique new point below h f ( b ) hf(b) h f ( b ) ; in every other case both give old h f ( b ) hf(b) h f ( b ) . This proves composition. The identity label is active and fixes every inserted point.
At an old target y y y , a new input is eligible for f X ≤ y fX\le y f X ≤ y exactly when its upper anchor b b b satisfies f ( b ) ≤ y f(b)\le y f ( b ) ≤ y . The greatest eligible input is consequently old f ∗ ( y ) f^{*}(y) f ∗ ( y ) . At a new target Y = X c , h Y=X_{c,h} Y = X c , h put a = f ∗ ( c ) a=f^{*}(c) a = f ∗ ( c ) . An eligible new input above a a a must map actively to Y Y Y . Its lower endpoint is eligible for old c c c , and its upper endpoint is not, so that lower endpoint must be a a a . There is at most one such gap. Its midpoint, if eligible, is the maximum; otherwise a a a is the maximum. These are actual right adjoints on every target.
It remains to identify them with prime. On the old chain the dual context f † = ′ f ∗ ′ f^{\dagger}={}'f^{*}{'} f † = ′ f ∗ ′ is ( v , u ) (v,u) ( v , u ) . If f f f actively transports ( a , b ) (a,b) ( a , b ) to ( c , h ) (c,h) ( c , h ) with return R R R , then f ∗ ( c ) = a f^{*}(c)=a f ∗ ( c ) = a and R ∗ ( a ) = c R^{*}(a)=c R ∗ ( a ) = c . Indeed f ( a ) ≤ c < f ( b ) f(a)\le c<f(b) f ( a ) ≤ c < f ( b ) and adjunction determine the first equality, while the second was proved above. Hence f † f^{\dagger} f † sends c ′ c' c ′ to a ′ a' a ′ , and the dual return R † = ( t , s ) R^{\dagger}=(t,s) R † = ( t , s ) sends a ′ a' a ′ to c ′ c' c ′ ; both composites are contractive. It actively transports ( h ′ , c ′ ) (h',c') ( h ′ , c ′ ) to ( b ′ , a ′ ) (b',a') ( b ′ , a ′ ) . Dualizing again gives the converse. Thus the adjoint of f f f at a new target Y Y Y is new exactly when f † f^{\dagger} f † is active at Y ′ Y' Y ′ , and in that case the two outputs are paired by prime. In the remaining case it is old f ∗ ( c ) f^*(c) f ∗ ( c ) , whose prime is the inactive value f † ( c ′ ) f^{\dagger}(c') f † ( c ′ ) . Therefore f^*='f^{\dagger}' on the entire carrier. The one-sided cases give (L6).
Attained coefficient comparisons
Proof of Lemma 5.3. For any old v v v ,
v X J ≤ X J ⟺ v ≤ s J . (64) vX_J \leq X_J \Longleftrightarrow v \leq s_J. \tag*{(64)} v X J ≤ X J ⟺ v ≤ s J . (64)
if v ≤ e v\leq e v ≤ e the inequality is automatic, while for v ≥ e v\geq e v ≥ e it is equivalent to equality. The action law shows s J 2 s_J^2 s J 2 fixes X J X_J X J ; since s J 2 ≥ s J ≥ e s_J^2\geq s_J\geq e s J 2 ≥ s J ≥ e , maximality gives idempotence.
Suppose u X J = X K uX_J=X_K u X J = X K and put v = u − v=u^{-} v = u − . With h = u b h=ub h = u b the activity criterion gives v h = b vh=b v h = b , and u v h = h uvh=h uv h = h . Thus u u u is an admissible return for v v v at K K K , and v X K = X J vX_K=X_J v X K = X J . Hence, for any old w w w ,
w X J ≤ X K ⟺ ( w v ) X K ≤ X K ⟺ w v ≤ s K ⟺ w ≤ s K / v . wX_J \leq X_K \Longleftrightarrow(wv)X_K \leq X_K \Longleftrightarrow wv \leq s_K \Longleftrightarrow w \leq s_K/v. w X J ≤ X K ⟺ ( w v ) X K ≤ X K ⟺ w v ≤ s K ⟺ w ≤ s K / v .
Also v Z ≤ X J vZ\leq X_J v Z ≤ X J if and only if Z ≤ X K Z\leq X_K Z ≤ X K : the reverse implication is monotonicity, and if Z > X K Z>X_K Z > X K then Z ≥ h Z\geq h Z ≥ h , so v Z ≥ v h = b > X J vZ\geq vh=b>X_J v Z ≥ v h = b > X J . It follows that the same cut is { w : v w ≤ s J } = ↓ ( v \ s J ) \{w:vw\leq s_J\}=\downarrow(v\backslash s_J) { w : v w ≤ s J } =↓ ( v \ s J ) . Its maximum is at least u u u and therefore transports X J X_J X J to X K X_K X K . This proves the fourth line.
For old y y y , the anchor formula gives w X ≤ y wX\leq y w X ≤ y exactly when w r ( X ) ≤ y wr(X)\leq y w r ( X ) ≤ y . For old x x x and new X K X_K X K , the product w x wx w x is old, so w x ≤ X K wx\leq X_K w x ≤ X K exactly when w x ≤ c wx\leq c w x ≤ c . If both arguments are new and there is no transport, w X J ≤ X K wX_J\leq X_K w X J ≤ X K means w X J ≤ c wX_J\leq c w X J ≤ c , equivalently w b ≤ c wb\leq c w b ≤ c . These observations prove the other three lines.
Cyclic selection and old-factor equations
Proof of Lemma 5.4. The old target r ( X ) r ( Y ) r(X)r(Y) r ( X ) r ( Y ) is eligible for (L10), by old cyclic adjunction and monotonicity of the old actions. Thus k ( X , Y ) ≤ r ( X ) r ( Y ) k(X,Y)\leq r(X)r(Y) k ( X , Y ) ≤ r ( X ) r ( Y ) , and
k ( X , Y ) ≤ z ⟺ z ′ X ≤ Y ′ ( z ∈ A ) . (65) k(X,Y)\leq z \Longleftrightarrow z'X\leq Y' \qquad(z\in A). \tag*{(65)} k ( X , Y ) ≤ z ⟺ z ′ X ≤ Y ′ ( z ∈ A ) . (65)
If either upper-action bound r ( X ) Y r(X)Y r ( X ) Y or X r ( Y ) Xr(Y) X r ( Y ) is at most an old z z z , its cyclic adjunction implies z ′ X ≤ Y ′ z'X\leq Y' z ′ X ≤ Y ′ . Hence an upper-action bound below k ( X , Y ) k(X,Y) k ( X , Y ) can only be the midpoint immediately below k ( X , Y ) k(X,Y) k ( X , Y ) ; a smaller anchor would be an eligible old target below k k k . This proves (L12). When an argument is old, one upper-action bound is the prescribed product, the other is above it, and k k k is its anchor. All prescribed products and the unit are therefore retained. The cuts in (L10) show k k k is monotone in both arguments; the anchors and old actions are monotone too, proving monotonicity of P P P .
The formula for P P P gives the following exact ternary comparison:
P ( X , Y ) ≤ Z ′ ⟺ ( r ( X ) Y ≤ Z ′ ) or ( r ( Y ) Z ≤ X ′ ) or ( r ( Z ) X ≤ Y ′ ) . (66) P(X,Y) \le Z' \Longleftrightarrow(r(X)Y \le Z') \text{ or } (r(Y)Z \le X') \text{ or } (r(Z)X \le Y'). \tag*{(66)} P ( X , Y ) ≤ Z ′ ⟺ ( r ( X ) Y ≤ Z ′ ) or ( r ( Y ) Z ≤ X ′ ) or ( r ( Z ) X ≤ Y ′ ) . (66)
Indeed the term k ( X , Y ) k(X,Y) k ( X , Y ) is old, so k ( X , Y ) ≤ Z ′ k(X,Y) \le Z' k ( X , Y ) ≤ Z ′ is k ( X , Y ) ≤ r ( Z ′ ) k(X,Y) \le r(Z') k ( X , Y ) ≤ r ( Z ′ ) , which is the third clause by (L16). The second upper-action bound yields the second clause by old cyclic adjunction; the other gives the first. Cyclic rotation permutes the clauses. Since prime is an antitone involution, this proves both adjunctions in (L13). In particular the residual into d = e ′ d=e' d = e ′ is the prescribed prime.
For completeness the old-factor equations follow from the cut, not from an assumption of associativity. The action laws give r ( u X ) = u r ( X ) r(uX)=ur(X) r ( u X ) = u r ( X ) and r ( X u ) = r ( X ) u r(Xu)=r(X)u r ( X u ) = r ( X ) u , while comparisons with any old z z z give
k ( u X , Y ) = u k ( X , Y ) , k ( X , Y u ) = k ( X , Y ) u , k ( X u , Y ) = k ( X , u Y ) . (67) k(uX,Y)=uk(X,Y), \qquad k(X,Yu)=k(X,Y)u, \qquad k(Xu,Y)=k(X,uY). \tag*{(67)} k ( u X , Y ) = u k ( X , Y ) , k ( X , Y u ) = k ( X , Y ) u , k ( X u , Y ) = k ( X , u Y ) . (67)
For the first identity, its left side is at most z z z exactly when ( z ′ u ) X ≤ Y ′ (z'u)X \le Y' ( z ′ u ) X ≤ Y ′ , equivalently k ( X , Y ) ≤ ( z ′ u ) ′ k(X,Y) \le(z'u)' k ( X , Y ) ≤ ( z ′ u ) ′ , equivalently u k ( X , Y ) ≤ z uk(X,Y) \le z u k ( X , Y ) ≤ z . For the second, z ′ X ≤ ( Y u ) ′ z'X \le(Yu)' z ′ X ≤ ( Y u ) ′ is equivalent to u ( z ′ X ) ≤ Y ′ u(z'X) \le Y' u ( z ′ X ) ≤ Y ′ , and hence to k ( X , Y ) u ≤ z k(X,Y)u \le z k ( X , Y ) u ≤ z . For the third, ( z ′ X ) u ≤ Y ′ (z'X)u \le Y' ( z ′ X ) u ≤ Y ′ is equivalent to z ′ X ≤ ( u Y ) ′ z'X \le(uY)' z ′ X ≤ ( u Y ) ′ . A monotone map of a chain preserves finite minima. Applying these identities to the three terms in (L11), and using commutation of left and right old actions, proves (L14).
Finally the comparison of k ( P ( X , Y ) , Z ) k(P(X,Y),Z) k ( P ( X , Y ) , Z ) with old w w w is w ′ P ( X , Y ) ≤ Z ′ w'P(X,Y) \le Z' w ′ P ( X , Y ) ≤ Z ′ . By (L14) and cyclicity this is equivalent to P ( w ′ X , Y ) ≤ Z ′ P(w'X,Y) \le Z' P ( w ′ X , Y ) ≤ Z ′ , then to P ( Y , Z ) ≤ ( w ′ X ) ′ P(Y,Z) \le(w'X)' P ( Y , Z ) ≤ ( w ′ X ) ′ , and finally to w ′ X ≤ P ( Y , Z ) ′ w'X \le P(Y,Z)' w ′ X ≤ P ( Y , Z ) ′ . By (L16) this says k ( X , P ( Y , Z ) ) ≤ w k(X,P(Y,Z)) \le w k ( X , P ( Y , Z )) ≤ w . The two old elements are equal, and (L12) proves (L15).
Transfer of all finite histories
Proof of Lemma 5.8. The unit, paired involution and actual residuals were proved in Lemma 5.4. The coefficient d d d has no active domain by Lemma 5.6, so on every new point X X X both d X dX d X and X d Xd X d equal the old element d r ( X ) = r ( X ) d dr(X)=r(X)d d r ( X ) = r ( X ) d . Thus d d d remains central. The residual formulas give X \ e = ( d X ) ′ = ( X d ) ′ = X\backslash e=(dX)'=(Xd)'= X \ e = ( d X ) ′ = ( X d ) ′ = e / X e/X e / X , proving e e e -cyclicity. The old products, prime, constants and order are retained; the residual formulas then retain both old residuals, as well as the dual product and lattice operations. This proves fullness.
We verify every original finite history. An arbitrary central floor power c = q j c=q^j c = q j has local inverse c ˉ = q − j \bar{c}=q^{-j} c ˉ = q − j . It satisfies
c c ˉ = c ˉ c = p . (68) c\bar{c}=\bar{c}c=p. \tag*{(68)} c c ˉ = c ˉ c = p . (68)
Indeed c ˉ c = p ≤ e \bar{c}c=p\le e c ˉ c = p ≤ e gives c ˉ ≤ c − \bar{c}\le c^{-} c ˉ ≤ c − and hence p ≤ c c ˉ ≤ e p\le c\bar{c}\le e p ≤ c c ˉ ≤ e . Since c c ˉ ∈ p A c\bar{c}\in pA c c ˉ ∈ p A , multiplying the upper inequality by p p p gives c c ˉ ≤ p c\bar{c}\le p c c ˉ ≤ p ; centrality gives the other identity. The left and right activity tests for c c c consequently coincide, both being p b = b pb=b p b = b at an upper endpoint b b b . Their upper values agree, so every q j q^j q j remains central in D ( A ) D(A) D ( A ) . In particular p Z ≤ Z pZ\le Z pZ ≤ Z for every point.
Fix a finite history w w w with caps C = c w C=c_w C = c w , H = h w H=h_w H = h w , and put j = # S ( w ) − # T ( w ) j=\#S(w)-\#T(w) j = # S ( w ) − # T ( w ) . Fullness of the old projection and the history recursions in the local group give
c = p C = q j , h = p H = q − j , c h = p . (69) c=pC=q^j,\qquad h=pH=q^{-j},\qquad ch=p. \tag*{(69)} c = pC = q j , h = p H = q − j , c h = p . (69)
Suppose X Y ≤ C XY\le C X Y ≤ C , U V ≤ H UV\le H U V ≤ H , and write B = Y X B=YX B = Y X , E = V U E=VU E = V U . Multiplying the first premise by the old central local inverse c ˉ \bar{c} c ˉ gives c X Y c ˉ ≤ p ≤ e cXY\bar{c}\le p\le e c X Y c ˉ ≤ p ≤ e . Unit cyclicity and centrality rotate this to c c ˉ Y X ≤ e c\bar{c}YX\le e c c ˉ Y X ≤ e ; multiplication by c c c gives p B ≤ c pB\le c pB ≤ c . Similarly p E ≤ h pE\le h pE ≤ h .
If w w w ends in S S S , then C = d C 0 C=dC_0 C = d C 0 . Centrality of d d d and d 2 = q 2 d^2=q^2 d 2 = q 2 give C 2 = ( p C ) 2 = c 2 C^2=(pC)^2=c^2 C 2 = ( pC ) 2 = c 2 . Therefore
B 3 = Y ( X Y ) 2 X ≤ Y C 2 X = c 2 B = c 2 ( p B ) ≤ c 3 . B^3=Y(XY)^2X\le YC^2X=c^2B=c^2(pB)\le c^3. B 3 = Y ( X Y ) 2 X ≤ Y C 2 X = c 2 B = c 2 ( pB ) ≤ c 3 .
For the monotone map F ( Z ) = B Z E F(Z)=BZE F ( Z ) = B Z E , associativity and centrality give
F 3 ( Z ) = B 3 Z E 3 ≤ c 3 Z E 3 = c 3 Z ( p E ) 3 ≤ c 3 Z h 3 = p Z ≤ Z . F^3(Z)=B^3ZE^3\le c^3ZE^3=c^3Z(pE)^3\le c^3Zh^3=pZ\le Z. F 3 ( Z ) = B 3 Z E 3 ≤ c 3 Z E 3 = c 3 Z ( pE ) 3 ≤ c 3 Z h 3 = pZ ≤ Z .
If F ( Z ) > Z F(Z)>Z F ( Z ) > Z , monotonicity implies F 3 ( Z ) ≥ F ( Z ) > Z F^3(Z)\ge F(Z)>Z F 3 ( Z ) ≥ F ( Z ) > Z , a contradiction. Hence Y X Z V U ≤ Z YXZVU\le Z Y X Z V U ≤ Z , the required history. If w w w ends in T T T , then H 2 = h 2 H^2=h^2 H 2 = h 2 ; the same proof uses E 3 ≤ h 3 E^3\le h^3 E 3 ≤ h 3 and p B ≤ c pB\le c pB ≤ c , without changing the order of factors. For the empty history, e e e -cyclicity gives B , E ≤ e B,E\le e B , E ≤ e immediately. Since w w w was arbitrary, the entire history scheme holds in the extension.
Preservation of the insertion hypotheses
Proof of Lemma 5.9. For every central idempotent p ≤ e p\le e p ≤ e one has p − p = p p − = p p^{-}p=pp^{-}=p p − p = p p − = p . Indeed p − ≥ e p^{-}\ge e p − ≥ e gives p ≤ p p − ≤ e p\le pp^{-}\le e p ≤ p p − ≤ e . If v = p p − v=pp^{-} v = p p − , then p v = v pv=v p v = v by idempotence, while v ≤ e v \leq e v ≤ e gives p v ≤ p pv \leq p p v ≤ p ; thus v = p v = p v = p . Consequently the two activity tests for p p p at a gap with upper endpoint b b b are p b = b pb = b p b = b . Internal gaps make them inactive, so
p X = p r ( X ) = X p ∈ G (70) pX = pr(X) = Xp \in G \tag*{(70)} pX = p r ( X ) = X p ∈ G (70)
for every new X X X . Hence p p p stays central and p D ( A ) = G pD(A) = G p D ( A ) = G . Fullness retains the closed support term, q = p d q = pd q = p d , and d 2 = q 2 d^2 = q^2 d 2 = q 2 . Thus the new chain is a lower complete core with this exact group.
Every gap of D ( A ) D(A) D ( A ) is one of the halves ( a , X ) (a, X) ( a , X ) or ( X , b ) (X, b) ( X , b ) of an old gap ( a , b ) (a, b) ( a , b ) . The upper half has p b < b pb < b p b < b as before. For the lower half, p X pX pX is old and p X ≤ X pX \leq X pX ≤ X , so p X < X pX < X pX < X . Thus all gaps remain internal.
To prove (PC), let J = ( a , b ) J = (a, b) J = ( a , b ) be an old gap, X = X J X = X_J X = X J , and let Y Y Y be the formal midpoint of either half H H H in D ( D ( A ) ) D(D(A)) D ( D ( A )) . Put
E J = { z ∈ A : z ′ X ≤ X } , E H = { z ∈ D ( A ) : z ′ Y ≤ Y } . E_J = \{z \in A : z'X \leq X\}, \qquad E_H = \{z \in D(A) : z'Y \leq Y\}. E J = { z ∈ A : z ′ X ≤ X } , E H = { z ∈ D ( A ) : z ′ Y ≤ Y } .
We claim E H ∩ A = E J E_H \cap A = E_J E H ∩ A = E J . For z ∈ A z \in A z ∈ A put u = z ′ u = z' u = z ′ and v = u − u v = u^{-}u v = u − u , whose values are retained by fullness. If u ≤ e u \leq e u ≤ e , both comparisons hold. For u ≥ e u \geq e u ≥ e , comparison is fixation. At the old gap this means v b = b vb = b v b = b and u b = b ub = b u b = b . For the upper half the criterion is identical. For the lower half it means v X = X vX = X v X = X and u X = X uX = X u X = X . The latter equality already implies the former: the active inverse u − u^{-} u − returns the fixed midpoint to itself, so v X = u − ( u X ) = X vX = u^{-}(uX) = X v X = u − ( u X ) = X . This proves the claim.
Let k = min E J k = \min E_J k = min E J , which exists by (PC) and prime reversal. The set E H E_H E H is an upper set and contains k k k . If w ∈ E H w \in E_H w ∈ E H with w < k w < k w < k , the claim forces w w w to be new relative to A A A . Its old upper anchor belongs to E H ∩ A E_H \cap A E H ∩ A and is at most k k k , hence equals k k k . There is at most one such midpoint. Thus E H E_H E H has a least member, either k k k or that midpoint. By prime reversal its self-comparison cut has a maximum, which is at least e e e and fixes Y Y Y . This is (PC) for each new gap.
Canonical completion and support attainment
We next attain (PC) on a changed carrier. We use canonical extension only for bounded chains, after applying the full MacNeille completion theorem. For a bounded residuated chain A A A , let B = A δ B = A^\delta B = A δ be the canonical extension of its lattice reduct. Write
K ( B ) = { ⋀ F : F ⊆ A } , O ( B ) = { ⋁ I : I ⊆ A } . \mathcal{K}(B) = \{\bigwedge F : F \subseteq A\}, \qquad\mathcal{O}(B) = \{\bigvee I : I \subseteq A\}. K ( B ) = { ⋀ F : F ⊆ A } , O ( B ) = { ⋁ I : I ⊆ A } .
Every point is a join of closed points and a meet of open points. Compactness says that ⋀ F ≤ ⋁ I \bigwedge F \leq\bigvee I ⋀ F ≤ ⋁ I already follows from finite subsets. The product is its σ \sigma σ -extension and the residuals their π \pi π -extensions. These form a residuated monoid, the embedding is full, and a named cyclic dualizer remains cyclic and dualizing. The precise standard results used here are Gehrke, Topological Duality and Algebraic Completions , Proposition 5, Corollary 2, and Propositions 7–8 (including the identification of the extended linear negation in the proof of Proposition 8) [18 ] ; and Kowalski–Metcalfe, Uniform Interpolation and Coherence , 12 January 2019 manuscript, Lemmas 5.9–5.10 [19 ] .2 In particular, for closed k , l k,l k , l ,
k l = ⋀ { a b : a , b ∈ A , k ≤ a , l ≤ b } , (71) kl = \bigwedge\{ab : a,b \in A,\, k \leq a,\, l \leq b\}, \tag*{(71)} k l = ⋀ { ab : a , b ∈ A , k ≤ a , l ≤ b } , (71)
and multiplication preserves arbitrary joins separately.
Proof of Lemma 5.11. First, the closed points are linearly ordered, since filters of an old chain are linearly ordered by inclusion. If x ≰ y x \not\leq y x ≤ y , join density supplies a closed k ≤ x k \leq x k ≤ x with k ≰ y k \not\leq y k ≤ y . Every closed point below y y y is then at most k k k , so y ≤ k ≤ x y \leq k \leq x y ≤ k ≤ x . Thus B B B is a chain. The quoted completion results supply associativity, residuals and involution. The old unit can also be checked directly: (L25) gives e k = k = k e ek = k = ke e k = k = k e for closed k k k , by taking the old approximant e e e in that coordinate; join density extends this to all points.
We prove the additional defining conditions of K R \mathcal{K}_R K R directly. If x y ≤ e xy \leq e x y ≤ e and closed k ≤ x k \leq x k ≤ x , l ≤ y l \leq y l ≤ y are given, (L25) and compactness give old a ≥ k a \geq k a ≥ k , b ≥ l b \geq l b ≥ l with a b ≤ e ab \leq e ab ≤ e . Indeed the finite list of products supplied by compactness can be replaced by the product of the least first and least second coordinates. If that finite list is empty, the target cap is top, and we may take both old coordinates equal to top. Old e e e -cyclicity gives b a ≤ e ba \leq e ba ≤ e , hence l k ≤ e lk \leq e l k ≤ e . Join density gives y x ≤ e yx \leq e y x ≤ e .
Fix any finite history w w w and suppose x y ≤ c w xy \leq c_w x y ≤ c w , u v ≤ h w uv \leq h_w uv ≤ h w in B B B . The caps are old by fullness. For closed approximants k x ≤ x k_x \leq x k x ≤ x , k y ≤ y k_y \leq y k y ≤ y , k u ≤ u k_u \leq u k u ≤ u , k v ≤ v k_v \leq v k v ≤ v , k z ≤ z k_z \leq z k z ≤ z and an open o ≥ z o \geq z o ≥ z , the same compactness argument supplies old a ≥ k x a \geq k_x a ≥ k x , b ≥ k y b \geq k_y b ≥ k y , s ≥ k u s \geq k_u s ≥ k u , t ≥ k v t \geq k_v t ≥ k v with a b ≤ c w ab \leq c_w ab ≤ c w , s t ≤ h w st \leq h_w s t ≤ h w . Compactness between the closed k z k_z k z and open o o o supplies old r r r with k z ≤ r ≤ o k_z \leq r \leq o k z ≤ r ≤ o . The old history gives
k y k x k z k v k u ≤ b a r t s ≤ r ≤ o . k_y k_x k_z k_v k_u \leq barts \leq r \leq o. k y k x k z k v k u ≤ ba r t s ≤ r ≤ o .
Taking joins of the five closed lower approximants and then the meet of the open upper approximants proves y x z v u ≤ z yxzvu \le z y x z v u ≤ z . This works for each finite w w w without a length bound.
We record the compactness facts needed for attainment. If a ≺ b a \prec b a ≺ b in B B B , join density makes b b b closed and meet density makes a a a open. If x < y < z x < y < z x < y < z , choose closed k k k and open o o o with x < k ≤ y ≤ o < z x < k \le y \le o < z x < k ≤ y ≤ o < z ; compactness gives an old point t t t with x < t < z x < t < z x < t < z . Finally, for nonempty R ⊆ A R \subseteq A R ⊆ A and closed b b b ,
( ⋀ R ) b = ⋀ r ∈ R r b , b ( ⋀ R ) = ⋀ r ∈ R b r . (72) \left(\bigwedge R\right)b = \bigwedge_{r\in R} rb,\qquad b\left(\bigwedge R\right) = \bigwedge_{r\in R} br. \tag*{(72)} ( ⋀ R ) b = r ∈ R ⋀ r b , b ( ⋀ R ) = r ∈ R ⋀ b r . (72)
For the nontrivial inequality in the first identity, put v = ⋀ R v = \bigwedge R v = ⋀ R . Given old a , c a,c a , c with v ≤ a v \le a v ≤ a , b ≤ c b \le c b ≤ c , compactness yields some r ∈ R r \in R r ∈ R with r ≤ a r \le a r ≤ a (if the finite subset is empty, any r r r works). Thus ⋀ r ∈ R r b ≤ r b ≤ r c ≤ a c \bigwedge_{r\in R} rb \le rb \le rc \le ac ⋀ r ∈ R r b ≤ r b ≤ r c ≤ a c . Meet over all such a , c a,c a , c and use (L25). The second identity is proved with the factors in reversed order.
For a gap a ≺ b a \prec b a ≺ b of B B B , its left self-support set S = { u ≥ e : u b = b , u − b = b } S = \{u \ge e : ub = b,\ u^{-}b = b\} S = { u ≥ e : u b = b , u − b = b } contains e e e and is an initial interval of B ≥ e B_{\ge e} B ≥ e . In fact if e ≤ v ≤ u ∈ S e \le v \le u \in S e ≤ v ≤ u ∈ S , then b ≤ v b ≤ u b = b b \le vb \le ub = b b ≤ v b ≤ u b = b and b = u − b ≤ v − b ≤ b b = u^{-}b \le v^{-}b \le b b = u − b ≤ v − b ≤ b . If S S S had no maximum, for each x ∈ S x \in S x ∈ S choose x < y < z x < y < z x < y < z in S S S . The old-point interpolation just proved puts some old t t t strictly between x x x and z z z ; initiality puts t t t in S S S . Thus T = S ∩ A T = S \cap A T = S ∩ A is nonempty and cofinal in S S S . Put s = ⋁ T s = \bigvee T s = ⋁ T . Complete join preservation and residuation give s b = b sb = b s b = b and s − = ⋀ u ∈ T u − s^{-} = \bigwedge_{u\in T}u^{-} s − = ⋀ u ∈ T u − . The entries u − u^{-} u − are old and b b b is closed, so (L26) gives
s − b = ⋀ u ∈ T u − b = b . s^{-}b = \bigwedge_{u\in T}u^{-}b = b. s − b = u ∈ T ⋀ u − b = b .
Hence s ∈ S s \in S s ∈ S is its maximum, a contradiction. The right-support proof uses the other identity of (L26). The maximum is idempotent as well: supports are closed under multiplication because v − u − ≤ ( u v ) − v^{-}u^{-} \le(uv)^{-} v − u − ≤ ( uv ) − , while u v ≥ e uv \ge e uv ≥ e ; thus ( u v ) b = b (uv)b = b ( uv ) b = b and b = v − u − b ≤ ( u v ) − b ≤ b b = v^{-}u^{-}b \le(uv)^{-}b \le b b = v − u − b ≤ ( uv ) − b ≤ b . The maximum s s s therefore has s 2 ≤ s s^{2} \le s s 2 ≤ s , and s ≥ e s \ge e s ≥ e gives the reverse inequality.
Representation above the closed support
We retain the closed terms p p p , q q q , Q Q Q of Theorem 4.2. The case p = e p = e p = e has a central invertible dualizer and is covered by the support densification theorem. Throughout the construction below p > e p > e p > e .
Lemma D.1 (The upper corner and its two possibilities). *Put I = p C I = pC I = pC , f = p ′ f = p' f = p ′ , and n = q ′ n = q' n = q ′ . The set I I I is closed under product and the original prime and has local unit p p p , local dualizer f f f , and the actual restricted residuals. The elements f , n f,n f , n are central in C C C , and
n = p \ e = Q f < e < p , q n = f < d < q . n=p\backslash e=Qf<e<p,\qquad qn=f<d<q. n = p \ e = Q f < e < p , q n = f < d < q .
The corner is unit cyclic, n < p n<p n < p is a cover in I I I , and exactly one of the following holds: n 2 = n n^2=n n 2 = n , or n n n is invertible in I I I . Every historical guard and its prime belongs to the central set
{ e , d } ∪ { q j , q j f : j ∈ Z } , q 0 = p . \{e,d\}\cup\{q^j,q^jf:j\in\mathbb{Z}\},\qquad q^0=p. { e , d } ∪ { q j , q j f : j ∈ Z } , q 0 = p .
Proof. If x = p x x=px x = p x , then x ( p x ′ ) = x x ′ ≤ d x(px')=xx'\le d x ( p x ′ ) = x x ′ ≤ d , so p x ′ ≤ x ′ px'\le x' p x ′ ≤ x ′ by residuation; the reverse inequality follows from p ≥ e p\ge e p ≥ e . Thus I I I is prime closed. Products are closed by centrality and idempotence of p p p . The involutive residual formulas prove both restricted residual assertions and identify the local prime with the original prime, with local dualizer p ′ = f p'=f p ′ = f . Residual currying gives q ′ = p \ e q'=p\backslash e q ′ = p \ e ; taking the complement of the local central unit q q q gives n = Q f n=Qf n = Q f and q n = f qn=f q n = f . Normalizing the local dualizer f f f by the central local unit q q q makes n n n another cyclic dualizer in I I I . For z ∈ I z\in I z ∈ I , z ≤ e z\le e z ≤ e is equivalent to z ≤ n z\le n z ≤ n , since p z = z pz=z p z = z .
We verify unit cyclicity of I I I from an actual history, rather than assuming it follows merely from prime closure. Suppose x , y ∈ I x,y\in I x , y ∈ I and x y ≤ p xy\le p x y ≤ p . The S S SS S S history has guards c S S = d 2 = q 2 c_{SS}=d^2=q^2 c S S = d 2 = q 2 and h S S = ( d 3 ) ′ = Q 3 f h_{SS}=(d^3)'=Q^3f h S S = ( d 3 ) ′ = Q 3 f . Apply it to ( q 2 x ) y ≤ q 2 (q^2x)y\le q^2 ( q 2 x ) y ≤ q 2 , to h S S e ≤ h S S h_{SS}e\le h_{SS} h S S e ≤ h S S , and to middle input e e e . It yields
y x q 2 h S S = y x n ≤ e . yxq^2h_{SS}=yxn\le e. y x q 2 h S S = y x n ≤ e .
This product lies in I I I , so y x n ≤ n yxn\le n y x n ≤ n . Since n n n is a cyclic dualizer of the local monoid, its residual into itself is p p p ; therefore y x ≤ p yx\le p y x ≤ p . Unit cyclicity makes f f f central in I I I by Lemma 4.1. Any element a a a central in I I I is central in C C C , because a x = a ( p x ) = ( p x ) a = x a ax=a(px)=(px)a=xa a x = a ( p x ) = ( p x ) a = x a . Thus f f f and n = Q f n=Qf n = Q f are globally central.
No point of I I I lies strictly between n n n and p p p : a point z ≤ e z\le e z ≤ e satisfies z ≤ p \ e = n z\le p\backslash e=n z ≤ p \ e = n , and a point z ≥ e z\ge e z ≥ e satisfies z = p z ≥ p e = p z=pz\ge pe=p z = p z ≥ p e = p . Since e ∉ I e\notin I e ∈ / I , prime closure gives d ∉ I d\notin I d ∈ / I , so n < e < p n<e<p n < e < p and f < d < q f<d<q f < d < q .
Set s = n \ p = ( f n ) ′ ∈ I s=n\backslash p=(fn)'\in I s = n \ p = ( f n ) ′ ∈ I . We have p ≤ s p\le s p ≤ s , n s ≤ p ns\le p n s ≤ p , and n \ n = ( q n ) ′ = f ′ = p n\backslash n=(qn)'=f'=p n \ n = ( q n ) ′ = f ′ = p . The cover implies either n s = p ns=p n s = p or n s ≤ n ns\le n n s ≤ n . In the first case centrality gives s n = p sn=p s n = p . In the second, residuation gives s ≤ p s\le p s ≤ p , so s = p s=p s = p ; taking primes in ( q n 2 ) ′ = s = p = ( q n ) ′ (qn^2)'=s=p=(qn)' ( q n 2 ) ′ = s = p = ( q n ) ′ and multiplying by Q Q Q yields n 2 = n n^2=n n 2 = n . Since n < p n<p n < p , an idempotent n n n cannot also be invertible.
Finally, on I I I multiplication by d d d is multiplication by q q q , and d \ x = Q x d\backslash x=Qx d \ x = Q x : if d y ≤ x ∈ I dy\le x\in I d y ≤ x ∈ I , then q ( p y ) ≤ x q(py)\le x q ( p y ) ≤ x and y ≤ p y ≤ Q x y\le py\le Qx y ≤ p y ≤ Q x , with converse immediate.
Outside these two local orbits the guard recursion only uses d e = d de=d d e = d , d d = q 2 dd=q^{2} dd = q 2 , d \ e = Q 2 f d\backslash e=Q^{2}f d \ e = Q 2 f , and d \ d = e d\backslash d=e d \ d = e . Since e + x = d \ x e+x=d\backslash x e + x = d \ x , induction on the history word gives exactly the displayed central guard set; prime exchanges the two local orbits and exchanges e , d e,d e , d .
Lemma D.2 (The exact protected bands). Let G = U ( I ) G=U(I) G = U ( I ) . For x ∉ I x\notin I x ∈ / I put a = p x a=px a = p x , b = p x ′ b=px' b = p x ′ . Then
a b = b a = q , b = q a − 1 , ( p x ′ ) ′ = a n , a n < x < a . ab=ba=q,\qquad b=qa^{-1},\qquad(px')'=an,\qquad an<x<a. ab = ba = q , b = q a − 1 , ( p x ′ ) ′ = an , an < x < a . The endpoints a n < a an<a an < a are consecutive in I I I , and
C ∖ I = ⋃ ˙ a ∈ G ( a n , a ) C . C\setminus I=\mathop{\dot{\bigcup}}_{a\in G}(an,a)_{C}. C ∖ I = ⋃ ˙ a ∈ G ( an , a ) C . Prime sends ( a n , a ) (an,a) ( an , a ) to ( b n , b ) (bn,b) ( bn , b ) for b = q a − 1 b=qa^{-1} b = q a − 1 . Moreover I I I is directly finite. If n n n is invertible, G G G is discretely ordered, with central predecessor n n n of p p p , and
K = { x : p x ∈ G } = G ∪ ( C ∖ I ) K=\{x:px\in G\}=G\cup(C\setminus I) K = { x : p x ∈ G } = G ∪ ( C ∖ I ) is a full subalgebra, called the upper-invertible complete core. Its fibre at g g g is ( g n , g ] (gn,g] ( g n , g ] .
Proof. The closure x ↦ p x x\mapsto px x ↦ p x gives the least point of I I I above x x x ; by prime, ( p x ′ ) ′ (px')' ( p x ′ ) ′ is the greatest point of I I I below x x x . For x ∉ I x\notin I x ∈ / I the two inequalities are strict and these two support endpoints are consecutive. From x < a x<a x < a we obtain a x ′ > d ax'>d a x ′ > d by cyclic duality. On the other hand a b = p ( x x ′ ) ≤ p d = q ab=p(xx')\le pd=q ab = p ( x x ′ ) ≤ p d = q . As a b ∈ I ab\in I ab ∈ I and q q q is the least point of I I I above d d d , it follows that a b = q ab=q ab = q . Cyclicity gives b a > d ba>d ba > d , and the same upper bound gives b a = q ba=q ba = q . Thus a , b a,b a , b are local units and b = q a − 1 b=qa^{-1} b = q a − 1 . Complementation of a local unit gives ( p x ′ ) ′ = b ′ = a q ′ = a n (px')'=b'=aq'=an ( p x ′ ) ′ = b ′ = a q ′ = an . Translation by a unit carries the protected cover n < p n<p n < p to a n < a an<a an < a , proving the converse and disjointness. Prime gives the stated reversal of bands.
If x y = p xy=p x y = p in I I I , unit cyclicity gives y x ≤ p yx\le p y x ≤ p . Were y x < p yx<p y x < p , the cover would give y x ≤ n yx\le n y x ≤ n , whence p = ( x y ) 2 = x ( y x ) y ≤ x n y = n x y = n < p p=(xy)^{2}=x(yx)y\le xny=nxy=n<p p = ( x y ) 2 = x ( y x ) y ≤ x n y = n x y = n < p , a contradiction. Therefore y x = p yx=p y x = p . If n n n is a unit, translate its cover and its inverse cover to obtain the discrete group order and all protected covers. Products of core points have group projection, while the prime of a top g g g is f g − 1 fg^{-1} f g − 1 and the prime of an interior point has projection q g − 1 qg^{-1} q g − 1 . Thus K K K is product and prime closed, contains e , d e,d e , d , and is full. The preceding endpoint calculation identifies its fibres.
Lemma D.3 (The interior projection in the idempotent case). If n 2 = n n^2=n n 2 = n , then x ↦ n x x\mapsto nx x ↦ n x is the full interior projection of Proposition 4.3. Its local dualizer is n d = q n = f nd=qn=f n d = q n = f , a central unit of n C nC n C with inverse Q n Qn Q n . Hence its complete core K = { x : n x ∈ U ( n C ) } K=\{x:nx\in U(nC)\} K = { x : n x ∈ U ( n C )} and its nonunit support complement have the form and gluing interface of Lemma 4.4.
Proof. The preceding lemmas give centrality, idempotence and n < e n<e n < e . Also p n = n pn=n p n = n , so n d = n p d = n q = f nd=npd=nq=f n d = n p d = n q = f , and ( n d ) ( Q n ) = n 2 d Q = n p = n (nd)(Qn)=n^2dQ=np=n ( n d ) ( Q n ) = n 2 d Q = n p = n in both orders. The generic interior projection and complete-core lemmas now apply with r = n r=n r = n .
A useful history check will be used repeatedly. If a constructed cyclic involutive chain has central d d d , and every guard prime is central, then a b ≤ c w ab\le c_w ab ≤ c w implies b a ≤ c w ba\le c_w ba ≤ c w : the first inequality is equivalent to a b c w ′ ≤ d abc'_w\le d ab c w ′ ≤ d , which rotates cyclically and uses centrality of c w ′ c'_w c w ′ . Rotating both premises and using Lemma 4.1 yields b a z v u ≤ c w z h w ≤ z bazvu\le c_wzh_w\le z ba z v u ≤ c w z h w ≤ z . This proves all histories simultaneously.
Dense coefficient and action extensions
An ordered residuated monoid U U U is unit cyclic if u \ e = e / u u\backslash e=e/u u \ e = e / u for every u u u . Write
u ˉ = u \ e = e / u , p u = u ˉ u , q u = u u ˉ . \bar{u}=u\backslash e=e/u,\qquad p_u=\bar{u}u,\qquad q_u=u\bar{u}. u ˉ = u \ e = e / u , p u = u ˉ u , q u = u u ˉ .
A cyclic action chain over U U U is a chain L L L with an order-reversing involution x ↦ x † x\mapsto x^\dagger x ↦ x † , commuting unital left and right actions λ u x = u x \lambda_u x=ux λ u x = ux and ρ u x = x u \rho_u x=xu ρ u x = xu , monotone in both variables, and satisfying
u x ≤ y † ⟺ y u ≤ x † . ux\le y^\dagger\qquad\Longleftrightarrow\qquad yu\le x^\dagger. ux ≤ y † ⟺ y u ≤ x † .
Thus the right adjoint of λ u \lambda_u λ u is † ρ u † {}^\dagger\rho_u{}^\dagger † ρ u † , and conversely.
Lemma D.4 (Return identities). For u , v ∈ U u,v\in U u , v ∈ U ,
p u v ≤ p v , v p u v ≤ p u v , v ˉ p u v ≤ p u v . p_{uv}\le p_v,\qquad vp_{uv}\le p_{uv},\qquad\bar{v}\,p_{uv}\le p_{uv}. p uv ≤ p v , v p uv ≤ p uv , v ˉ p uv ≤ p uv . Consequently, in every ordered left action and at every point b b b ,
p u v b = b ⟺ p v b = b and p u ( v b ) = v b . p_{uv}b=b\qquad\Longleftrightarrow\qquad p_vb=b\ \text{ and }\ p_u(vb)=vb. p uv b = b ⟺ p v b = b and p u ( v b ) = v b . The right-handed statements follow by reversing multiplication.
Proof. All p u , q u p_u,q_u p u , q u are at most e e e , and u ≤ u − − u \le u^{--} u ≤ u −− . Put t = ( u v ) − t=(uv)^- t = ( uv ) − . From t u v , u v t ≤ e tuv, uvt \le e t uv , uv t ≤ e one obtains t u ≤ v − tu \le v^- t u ≤ v − and v t ≤ u − vt \le u^- v t ≤ u − , which give the first two inequalities. Also u v v − u − ≤ u u − ≤ e uvv^-u^- \le uu^- \le e uv v − u − ≤ u u − ≤ e , whence v − u − ≤ ( u v ) − v^-u^- \le(uv)^- v − u − ≤ ( uv ) − and the third inequality. If p u v b = b p_uvb=b p u v b = b , contractivity and the first two inequalities give both displayed fixed-point conditions. Conversely those conditions give b = v − v b = v − p u v b ≤ p u v b ≤ b b=v^-vb=v^-p_uvb \le p_uvb \le b b = v − v b = v − p u v b ≤ p u v b ≤ b . □
Lemma D.5 (Filling action gaps). Every cyclic action chain L L L over a unit-cyclic residuated monoid has a dense cyclic action extension obtained by inserting a copy of Q \mathbb{Q} Q in each adjacent old pair. Every new point has old lower and upper bounds. Countability is preserved.
The same assertion holds for a selected collection of adjacent pairs provided it is closed under involution and under the transports defined in the proof.
Proof. Write an inserted point as ( a , b , r ) (a,b,r) ( a , b , r ) , where a < b a<b a < b is old and adjacent and r ∈ Q r\in\mathbb{Q} r ∈ Q , and set
( a , b , r ) † = ( b † , a † , − r ) . (a,b,r)^\dagger=(b^\dagger,a^\dagger,-r). ( a , b , r ) † = ( b † , a † , − r ) .
Retain the old actions and prescribe
u ( a , b , r ) = { ( pred ( u b ) , u b , r ) , p u b = b , u b , p u b ≠ b , u(a,b,r)=
\begin{cases}
(\operatorname{pred}(ub),ub,r), & p_ub=b,\\
ub, & p_ub\ne b,
\end{cases} u ( a , b , r ) = { ( pred ( u b ) , u b , r ) , u b , p u b = b , p u b = b ,
where, in the active case p u b = b p_ub=b p u b = b , pred ( u b ) \operatorname{pred}(ub) pred ( u b ) denotes the immediate old predecessor of u b ub u b whose existence is verified in the next paragraph. On the right use the identical prescription with b q u = b bq_u=b b q u = b .
Suppose p u b = b p_ub=b p u b = b and put d = u b d=ub d = u b . For every old y y y ,
y < d ⟺ u − y < b . y<d \Longleftrightarrow u^-y<b. y < d ⟺ u − y < b .
Indeed y < d y<d y < d and u − y ≥ b u^-y\ge b u − y ≥ b would imply y ≥ u u − y ≥ u b = d y\ge uu^-y\ge ub=d y ≥ u u − y ≥ u b = d ; the reverse implication follows from u − d = b u^-d=b u − d = b . If g − g_- g − is the right adjoint of λ u − \lambda_{u^-} λ u − , adjacency gives y < d ⟺ u − y ≤ a ⟺ y ≤ g − ( a ) y<d \Longleftrightarrow u^-y\le a \Longleftrightarrow y\le g_-(a) y < d ⟺ u − y ≤ a ⟺ y ≤ g − ( a ) . Thus c = g − ( a ) c=g_-(a) c = g − ( a ) is the required predecessor of d d d . Moreover u a < d ua<d u a < d , since otherwise p u a ≥ b > a p_ua\ge b>a p u a ≥ b > a , contradicting contractivity. The new value lies between the old endpoint images, so the action is monotone in its point argument. Finally q u d = d q_ud=d q u d = d and p u − d = u − − u − d ≥ u u − d = d p_{u^-}d=u^{--}u^-d\ge uu^-d=d p u − d = u −− u − d ≥ u u − d = d ; contractivity gives equality. Therefore u − u^- u − transports the target gap back to the source gap. The opposite monoid proves these assertions on the right.
By Lemma D.4, u v uv uv transports a gap exactly when v v v transports it and u u u transports its image. If both transport, either composite has upper endpoint u v b uvb uv b and coordinate r r r ; otherwise both collapse to that old point. This proves the left action law, and the opposite monoid gives the right action law. For cross commutation, if u u u transports at b b b and H = ρ q v H=\rho_{q_v} H = ρ q v , old commutation and the local inverse give H ( u b ) = u b ⟺ H b = b H(ub)=ub \Longleftrightarrow Hb=b H ( u b ) = u b ⟺ H b = b : apply u − u^{-} u − for the forward implication. The analogous assertion holds with left and right interchanged. Both orders of the two actions consequently transport to the same gap with the same coordinate, or collapse to the same old point. The unit transports every gap identically.
For coefficient monotonicity let u ≤ v u\leq v u ≤ v . If the v v v -action collapses, its old value bounds the u u u -action above. If it transports and u b < v b ub<vb u b < v b , then u b ub u b lies below the predecessor of v b vb v b , giving the same conclusion. Otherwise u b = v b ub=vb u b = v b and p v b = b p_vb=b p v b = b ; hence b = v − v b ≤ u − u b = p u b ≤ b b=v^{-}vb\leq u^{-}ub=p_ub\leq b b = v − v b ≤ u − u b = p u b ≤ b . Both actions then transport to the same gap with the same coordinate. The right proof is identical.
It remains to verify cyclic adjunction. For k ≤ e k\leq e k ≤ e and an old adjacent pair a < b a<b a < b , the adjunction between F = λ k ≤ id F=\lambda_k\leq\operatorname{id} F = λ k ≤ id and G = † ρ k † ≥ id G={\dagger}\rho_k{\dagger}\geq\operatorname{id} G = † ρ k † ≥ id gives
F ( b ) = b ⟺ F ( b ) > a ⟺ b > G ( a ) ⟺ G ( a ) = a . F(b)=b \Longleftrightarrow F(b)>a \Longleftrightarrow b>G(a) \Longleftrightarrow G(a)=a. F ( b ) = b ⟺ F ( b ) > a ⟺ b > G ( a ) ⟺ G ( a ) = a .
Equivalently, k b = b ⟺ a † k = a † kb=b\Longleftrightarrow a^{\dagger}k=a^{\dagger} k b = b ⟺ a † k = a † . If h = λ u h=\lambda_u h = λ u transports a < b a<b a < b to c < d c<d c < d , its old right adjoint g g g satisfies g ( c ) = a g(c)=a g ( c ) = a . Since q u d = d q_ud=d q u d = d , the preceding equivalence says c † q u = c † c^{\dagger}q_u=c^{\dagger} c † q u = c † . The right u u u -action therefore transports ( d † , c † ) (d^{\dagger},c^{\dagger}) ( d † , c † ) back to ( b † , a † ) (b^{\dagger},a^{\dagger}) ( b † , a † ) . Conversely every such right transport arises this way. Conjugating by dagger gives inverse transports for the extended h h h and g g g .
For new X = ( a , b , r ) X=(a,b,r) X = ( a , b , r ) and Z = ( c , d , t ) Z=(c,d,t) Z = ( c , d , t ) , the inequality h X ≤ Z hX\leq Z h X ≤ Z holds exactly when h ( b ) ≤ c h(b)\leq c h ( b ) ≤ c , or when h h h transports the source gap to the target gap and r ≤ t r\leq t r ≤ t . The inequality X ≤ g Z X\leq gZ X ≤ g Z has the corresponding alternatives b ≤ g ( c ) b\leq g(c) b ≤ g ( c ) and the inverse transport with r ≤ t r\leq t r ≤ t . Old adjunction and the preceding paragraph identify these alternatives. If Z Z Z is old the two inequalities reduce to h ( b ) ≤ Z ⟺ b ≤ g ( Z ) h(b)\leq Z\Longleftrightarrow b\leq g(Z) h ( b ) ≤ Z ⟺ b ≤ g ( Z ) ; if X X X is old they reduce to h ( X ) ≤ c ⟺ X ≤ g ( c ) h(X)\leq c\Longleftrightarrow X\leq g(c) h ( X ) ≤ c ⟺ X ≤ g ( c ) . Thus adjunction holds everywhere. All gaps have been filled densely. The selected-gap variant uses only the indicated transports and their duals, so the same verification applies. □
Lemma D.6 (Completing cyclic actions). The Dedekind–MacNeille completion L ‾ \overline{L} L of a cyclic action chain carries extending cyclic actions and involution. Its actions preserve arbitrary joins. If L L L is dense, so is L ‾ . \overline{L}. L .
Proof. Old points are join and meet dense in the completion. Set
X † = ⋁ x ∈ L , x ≤ X x † , u X = ⋁ x ∈ L , x ≤ X u x , X u = ⋁ x ∈ L , x ≤ X x u . X^{\dagger}=\bigvee_{\substack{x\in L,\,x\le X}}x^{\dagger},\qquad uX=\bigvee_{\substack{x\in L,\,x\le X}}ux,\qquad Xu=\bigvee_{\substack{x\in L,\,x\le X}}xu. X † = x ∈ L , x ≤ X ⋁ x † , u X = x ∈ L , x ≤ X ⋁ ux , X u = x ∈ L , x ≤ X ⋁ xu .
In the lower-cut presentation, applying dagger twice is exactly MacNeille closure; thus it is an antitone involution. Join density and old cyclicity give
u X ≤ Y † ⟺ u x ≤ y † ( x ≤ X , y ≤ Y old ) uX\le Y^{\dagger}\Longleftrightarrow ux\le y^{\dagger}\quad(x\le X,y\le Y\text{ old}) u X ≤ Y † ⟺ ux ≤ y † ( x ≤ X , y ≤ Y old )
⟺ y u ≤ x † ( x ≤ X , y ≤ Y old ) \Longleftrightarrow yu\le x^{\dagger}\quad(x\le X,y\le Y\text{ old}) ⟺ y u ≤ x † ( x ≤ X , y ≤ Y old )
⟺ Y u ≤ X † . \Longleftrightarrow Yu\le X^{\dagger}. ⟺ Y u ≤ X † .
Consequently both actions have right adjoints and preserve all joins. Their laws, commutation, coefficient monotonicity and unit law now follow by expanding arbitrary points as joins of old points. Every old value is retained since its defining join includes that point. If X < Y X<Y X < Y were adjacent in the completion, join density would force Y Y Y old and meet density would force X X X old, contrary to old density. □ \square □
Lemma D.7 (Dense graded coefficient extension). Let U U U be a unit-cyclic residuated chain and let μ : U → H \mu:U\to H μ : U → H preserve multiplication, both residuals, order and unit, where H H H is a biordered group. There is a full unit-residuated extension U ↪ V U\hookrightarrow V U ↪ V and an extending full onto degree map V → H V\to H V → H whose degree fibres are dense and have no least points. An old greatest point in a fibre is retained as its greatest point, and no greatest point is introduced in a fibre which had none. Countable data admit a countable such extension.
If the neutral fibre of U U U has a greatest point, every fibre does. If it has no greatest point, no fibre has either endpoint.
Proof. Put J = Q ∩ ( 0 , 1 ] J=\mathbb{Q}\cap(0,1] J = Q ∩ ( 0 , 1 ] and order F = U × J F=U\times J F = U × J lexicographically. Write i ( u ) = ( u , 1 ) i(u)=(u,1) i ( u ) = ( u , 1 ) and π ( u , r ) = u \pi(u,r)=u π ( u , r ) = u . Define
L ( u , v ) ⟺ p u v = v , R ( u , v ) ⟺ u q v = u , L(u,v)\Longleftrightarrow p_uv=v,\qquad R(u,v)\Longleftrightarrow uq_v=u, L ( u , v ) ⟺ p u v = v , R ( u , v ) ⟺ u q v = u ,
( u , r ) ( v , s ) = ( u v , min ( { 1 } ∪ { r : R ( u , v ) } ∪ { s : L ( u , v ) } ) ) . (u,r)(v,s)=\left(uv,\min\left(\{1\}\cup\{r:R(u,v)\}\cup\{s:L(u,v)\}\right)\right). ( u , r ) ( v , s ) = ( uv , min ( { 1 } ∪ { r : R ( u , v )} ∪ { s : L ( u , v )} ) ) .
The return identities and their opposites give
L ( u v , w ) ⟺ L ( v , w ) ∧ L ( u , v w ) , L(uv,w)\Longleftrightarrow L(v,w)\land L(u,vw), L ( uv , w ) ⟺ L ( v , w ) ∧ L ( u , v w ) ,
R ( u , v w ) ⟺ R ( u , v ) ∧ R ( u v , w ) . R(u,vw)\Longleftrightarrow R(u,v)\land R(uv,w). R ( u , v w ) ⟺ R ( u , v ) ∧ R ( uv , w ) .
Also L ( u , v ) ∧ R ( u v , w ) ⟺ R ( v , w ) ∧ L ( u , v w ) L(u,v) \wedge R(uv,w) \Longleftrightarrow R(v,w) \wedge L(u,vw) L ( u , v ) ∧ R ( uv , w ) ⟺ R ( v , w ) ∧ L ( u , v w ) . For the forward implication, multiply u v q w = u v uvq_w=uv uv q w = uv on the left by u − u^{-} u − , using p u v = v p_uv=v p u v = v ; the other assertion is immediate. For the reverse implication, multiply p u v w = v w p_uv w=vw p u v w = v w on the right by w − w^{-} w − , using v q w = v vq_w=v v q w = v . Hence each bracketing of a triple retains exactly the same subset of its three rational coordinates, and multiplication is associative. Its unit is i ( e ) i(e) i ( e ) , and i , π i,\pi i , π are monoid maps.
For monotonicity, the rational coordinates cause no difficulty. If u < v u<v u < v and u w = v w uw=vw u w = v w , the condition R ( v , w ) R(v,w) R ( v , w ) is impossible because it would imply u ≥ u q w = u w w − = v w w − = v u \geq uq_w=uww^{-}=vww^{-}=v u ≥ u q w = u w w − = v w w − = v . If L ( v , w ) L(v,w) L ( v , w ) holds, then w = v − v w ≤ u − u w = p u w ≤ w w=v^{-}vw \leq u^{-}uw=p_uw \leq w w = v − v w ≤ u − u w = p u w ≤ w , so L ( u , w ) L(u,w) L ( u , w ) holds. Thus the product at the smaller base has rational coordinate no larger than the product at the larger base. The opposite argument handles the other variable. Moreover
( u , r ) ( v , s ) ≤ i ( e ) ⟺ u v ≤ e ⟺ v u ≤ e , π − 1 ( ↓ u ) = ↓ i ( u ) . (u,r)(v,s) \leq i(e) \Longleftrightarrow uv \leq e \Longleftrightarrow vu \leq e,\qquad\pi^{-1}(\downarrow u)=\downarrow i(u). ( u , r ) ( v , s ) ≤ i ( e ) ⟺ uv ≤ e ⟺ v u ≤ e , π − 1 ( ↓ u ) =↓ i ( u ) .
Thus F F F is unit cyclic. Each projection fibre has no least point, and F F F is dense: between different bases decrease the upper parameter.
Write j j j for MacNeille closure of lower subsets of F F F . Every translation preimage of a principal lower ideal is closed. Indeed, in each projection fibre its section is empty, whole, or of the form { ( v , s ) : s ≤ t } \{(v,s):s\leq t\} {( v , s ) : s ≤ t } . Every excluded point therefore has a smaller excluded point in the same fibre; that point bounds the whole lower set and excludes the original point from its MacNeille closure. Preimages of arbitrary closed lower sets are closed by intersection. If z z z bounds A B AB A B , first apply this fact to each left translation by a point of A A A , then to each right translation by a point of j ( B ) j(B) j ( B ) . It follows that z z z bounds j ( A ) j ( B ) j(A)j(B) j ( A ) j ( B ) , and hence
j ( A ) j ( B ) ⊆ j ( ↓ A B ) . j(A)j(B) \subseteq j(\downarrow AB). j ( A ) j ( B ) ⊆ j ( ↓ A B ) .
Consequently closed lower sets have associative product I ⊗ K = j ( ↓ I K ) I \otimes K=j(\downarrow IK) I ⊗ K = j ( ↓ I K ) , principal unit ↓ i ( e ) \downarrow i(e) ↓ i ( e ) , and residuals
I \ K = { b : I b ⊆ K } , K / I = { b : b I ⊆ K } . I\backslash K=\{b:Ib\subseteq K\},\qquad K/I=\{b:bI\subseteq K\}. I \ K = { b : I b ⊆ K } , K / I = { b : b I ⊆ K } .
The residuals are closed by the preimage result, and their defining equivalences prove residuation. Associativity follows by removing each interior closure using the displayed nucleus inequality and expansivity.
Give ( u , r ) (u,r) ( u , r ) degree μ ( u ) \mu(u) μ ( u ) , and retain precisely the cuts
V h = { I = j ( I ) : ∃ a , b ∈ F h ( a ∈ I ⊆ ↓ b ) } , V = ⋃ h ∈ H V h . V_h=\{I=j(I):\exists a,b\in F_h\ (a\in I\subseteq\downarrow b)\},\qquad V=\bigcup_{h\in H}V_h. V h = { I = j ( I ) : ∃ a , b ∈ F h ( a ∈ I ⊆↓ b )} , V = h ∈ H ⋃ V h .
Every such cut has a unique degree. Lower and upper same-degree bounds show that products from V h V_h V h , V k V_k V k lie in V h k V_{hk} V hk . For the residual, bound I ∈ V h I \in V_h I ∈ V h above by i ( u + ) i(u_+) i ( u + ) and choose ( v , t ) ∈ K ∩ F k (v,t) \in K \cap F_k ( v , t ) ∈ K ∩ F k . If v v v is not least in U k U_k U k , choose w < v w < v w < v in U k U_k U k ; then i ( u + \ w ) i(u_+\backslash w) i ( u + \ w ) is eligible in I \ K I\backslash K I \ K and has degree h − 1 k h^{-1}k h − 1 k . If v v v is least, put w = u − v w = u_-v w = u − v . Minimality gives u + w = v u_+w = v u + w = v , and thus p u + w = w p_{u_+}w = w p u + w = w and i ( u + ) ( w , t ) = ( v , t ) i(u_+)(w,t) = (v,t) i ( u + ) ( w , t ) = ( v , t ) . This gives an eligible point of the same required degree. Conversely, choose ( a , r ) ∈ I ∩ F h (a,r) \in I \cap F_h ( a , r ) ∈ I ∩ F h and bound K K K above by i ( k + ) i(k_+) i ( k + ) . Every eligible b b b satisfies b ≤ i ( a \ k + ) b \le i(a\backslash k_+) b ≤ i ( a \ k + ) , an upper bound of degree h − 1 k h^{-1}k h − 1 k . Therefore I \ K ∈ V h − 1 k I\backslash K \in V_{h^{-1}k} I \ K ∈ V h − 1 k . For K / I K/I K / I , the witnesses are i ( w / u + ) i(w/u_+) i ( w / u + ) or ( v u + − 1 , t ) (vu_+^{-1},t) ( v u + − 1 , t ) , and the upper bound is i ( k + / a ) i(k_+/a) i ( k + / a ) ; its degree is k h − 1 kh^{-1} k h − 1 . Thus V V V is residuated with the stated full degree map. Its unit cyclicity follows by testing all raw pairs against i ( e ) i(e) i ( e ) .
For old coefficients,
{ b : i ( u ) b ≤ i ( w ) } = π − 1 ( ↓ ( u \ w ) ) = ↓ i ( u \ w ) , \{b : i(u)b \le i(w)\} = \pi^{-1}(\mathord{\downarrow}(u\backslash w)) = \mathord{\downarrow}i(u\backslash w), { b : i ( u ) b ≤ i ( w )} = π − 1 ( ↓ ( u \ w )) = ↓ i ( u \ w ) ,
and likewise on the right. The map u ↦ ↓ i ( u ) u \mapsto\mathord{\downarrow}i(u) u ↦ ↓ i ( u ) therefore retains both residuals, as well as product, order and unit. Within each V h V_h V h the principal raw cuts are order dense: if I ⊊ K I \subsetneq K I ⊊ K , choose b ∈ K ∖ I b \in K \setminus I b ∈ K ∖ I in that degree and then c < b c < b c < b of the same degree outside I I I ; such c c c exists by closedness of I I I and density of F h F_h F h . Then I ⊊ ↓ c ⊊ K I \subsetneq\mathord{\downarrow}c \subsetneq K I ⊊ ↓ c ⊊ K . There is no least point, because each raw fibre has none. The greatest-point assertions follow from the same-degree upper bound in the definition of V h V_h V h . In the countable case, the subalgebra generated by all principal raw cuts is countable, full, and dense in each fibre, with the same endpoint properties.
Finally, if T T T is greatest in U 1 U_1 U 1 and v ∈ U h v \in U_h v ∈ U h , then for x ∈ U h x \in U_h x ∈ U h one has x v − ≤ T xv^{-} \le T x v − ≤ T and therefore x ≤ T / v − x \le T/v^{-} x ≤ T / v − . This last point has degree h h h and is greatest in that fibre. Conversely a greatest T h T_h T h bounds every u v uv uv with u ∈ U 1 u \in U_1 u ∈ U 1 , so T h / v ∈ U 1 T_h/v \in U_1 T h / v ∈ U 1 bounds the neutral fibre. If b b b is least in U h U_h U h , every left multiplication by u ∈ U 1 u \in U_1 u ∈ U 1 preserves it: the map has a right adjoint on that fibre and hence preserves its least point. Thus u b = b ub = b u b = b for all such u u u , and b / b ∈ U 1 b/b \in U_1 b / b ∈ U 1 is greatest. Absence of a neutral maximum excludes both endpoints in every fibre.
Arbitrary multiplication in upper-support bands
We use the following explicit form of the upper-invertible support structure. The support points are t g t_g t g for a biordered group G G G , whose identity is 1 1 1 . There are central δ , ν ∈ G \delta,\nu\in G δ , ν ∈ G , with ν \nu ν the immediate predecessor of 1 1 1 , such that
K = ∏ g ∈ G ( t g ν , t g ] , μ ( x ) = g ⟺ t g ν < x ≤ t g , K=\prod_{g\in G}(t_{g\nu},t_g],\qquad\mu(x)=g\Longleftrightarrow t_{g\nu}<x\leq t_g, K = g ∈ G ∏ ( t g ν , t g ] , μ ( x ) = g ⟺ t g ν < x ≤ t g ,
μ ( x y ) = μ ( x ) μ ( y ) , t g x = t g μ ( x ) , x t g = t μ ( x ) g , \mu(xy)=\mu(x)\mu(y),\qquad t_gx=t_{g\mu(x)},\qquad xt_g=t_{\mu(x)g}, μ ( x y ) = μ ( x ) μ ( y ) , t g x = t g μ ( x ) , x t g = t μ ( x ) g ,
t g ′ = t g − 1 δ ν , μ ( x ′ ) = μ ( x ) − 1 δ ( x ∉ { t g : g ∈ G } ) , t'_g=t_{g^{-1}\delta\nu},\qquad\mu(x')=\mu(x)^{-1}\delta\quad(x\notin\{t_g:g\in G\}), t g ′ = t g − 1 δ ν , μ ( x ′ ) = μ ( x ) − 1 δ ( x ∈ / { t g : g ∈ G }) ,
t ν < e < t 1 , t δ ν < d < t δ , e ′ = d , d 2 = t δ 2 . t_\nu<e<t_1,\qquad t_{\delta\nu}<d<t_\delta,\qquad e'=d,\qquad d^2=t_{\delta^2}. t ν < e < t 1 , t δ ν < d < t δ , e ′ = d , d 2 = t δ 2 .
Here d d d is central. Write P = { t g : g ∈ G } P=\{t_g:g\in G\} P = { t g : g ∈ G } and N = K ∖ P N=K\setminus P N = K ∖ P . All references below to a support-band structure mean these identities, with the actual cyclic involutive residuated operations of K K K .
For the application to the upper-invertible branch, take the full complete core K K K from Lemma D.2 and identify the degree group with its actual support group G = p K = U ( p K ) G=pK=U(pK) G = p K = U ( p K ) . In this identification the data are
1 G = p , t g = g , μ ( x ) = p x , δ = p d , ν = ( p d ) ′ . 1_G=p,\qquad t_g=g,\qquad\mu(x)=px,\qquad\delta=pd,\qquad\nu=(pd)'. 1 G = p , t g = g , μ ( x ) = p x , δ = p d , ν = ( p d ) ′ .
Thus t 1 = p t_1=p t 1 = p and t ν = ( p d ) ′ t_\nu=(pd)' t ν = ( p d ) ′ ; the group identity 1 G 1_G 1 G is the local unit p p p , not the ambient monoid unit e e e of K K K . The group products and inverses are taken with respect to this local unit. Lemmas D.1 and D.2 give the displayed support-band identities for these data. Since K K K is a full subalgebra of the original member of K R K_R K R , all its history conditions are inherited. The greatest neutral coefficient required in Theorem D.11 is supplied by the extension of Theorem D.15 . It is therefore a property of a constructed extension, not an additional hypothesis on the input of Theorem 7.1 .
Lemma D.8 (Central guards imply every history). Let L L L be a cyclic involutive residuated chain with central d d d . If a product- and prime-closed set of central elements contains e , d e,d e , d , then L L L satisfies unit cyclicity and every finite history law.
Proof. The identities ( d x ) ′ = x \ e (dx)'=x\backslash e ( d x ) ′ = x \ e and ( x d ) ′ = e / x (xd)'=e/x ( x d ) ′ = e / x give unit cyclicity. Also e + x = ( d x ′ ) ′ = d \ x e+x=(dx')'=d\backslash x e + x = ( d x ′ ) ′ = d \ x . All guards and their primes therefore belong to the indicated central set. For a guard c c c with central c ′ c' c ′ , cyclicity at d d d gives a b ≤ c ⟺ c ′ a b ≤ d ⟺ b c ′ a ≤ d ⟺ b a ≤ c ab\leq c\Longleftrightarrow c'ab\leq d\Longleftrightarrow bc'a\leq d\Longleftrightarrow ba\leq c ab ≤ c ⟺ c ′ ab ≤ d ⟺ b c ′ a ≤ d ⟺ ba ≤ c . Induction on the history word gives c w z h w ≤ z c_wzh_w\leq z c w z h w ≤ z : the empty case is equality, and the two steps use
( d c w ) z ( d \ h w ) ≤ c w z h w , ( d \ c w ) z ( d h w ) ≤ c w z h w , (dc_w)z(d\backslash h_w)\leq c_wzh_w,\qquad(d\backslash c_w)z(dh_w)\leq c_wzh_w, ( d c w ) z ( d \ h w ) ≤ c w z h w , ( d \ c w ) z ( d h w ) ≤ c w z h w ,
by centrality of d d d and residuation. Thus a b ≤ c w ab\leq c_w ab ≤ c w and u v ≤ h w uv\leq h_w uv ≤ h w imply b a z v u ≤ c w z h w ≤ z bazvu\leq c_wzh_w\leq z ba z v u ≤ c w z h w ≤ z . ∎
Lemma D.9 (Intrinsic coefficient and middle pieces). In a support-band structure put x − = x \ e = e / x x^{-}=x\backslash e=e/x x − = x \ e = e / x and
U = { x ∈ N : x − ∈ N } , W = U ′ , M = N ∖ ( U ∪ W ) . U=\{x\in N:x^{-}\in N\},\qquad W=U^{\prime},\qquad M=N\setminus(U\cup W). U = { x ∈ N : x − ∈ N } , W = U ′ , M = N ∖ ( U ∪ W ) . Then U U U is a full unit-cyclic unit-residuated subchain. Its degree map is a full onto map to a subgroup H ≤ G H\leq G H ≤ G . The sets U , W U,W U , W are disjoint and respectively initial and final in each band, and M ′ = M M^{\prime}=M M ′ = M . Their product types are summarized below. An entry contains every product with its left factor in the row class and its right factor in the column class; it does not assert equality of these sets.
Table 4.
The entries involving P P P use the support-factor identities from the definition of a support-band structure. This table describes types, not commutativity: throughout, the degree of x y xy x y is the ordered product μ ( x ) μ ( y ) \mu(x)\mu(y) μ ( x ) μ ( y ) . Every product of a coefficient with a nonsupport point is nonsupport. For u , v ∈ U u,v\in U u , v ∈ U one has u v ′ = ( v / u ) ′ uv^{\prime}=(v/u)^{\prime} u v ′ = ( v / u ) ′ and v ′ u = ( u \ v ) ′ v^{\prime}u=(u\backslash v)^{\prime} v ′ u = ( u \ v ) ′ .
Proof. For x ∈ N x\in N x ∈ N of degree g g g , t g − 1 ν t_{g-1}\nu t g − 1 ν is eligible in x \ e x\backslash e x \ e , while t g − 1 t_{g-1} t g − 1 and all higher degrees are not. Consequently x − x^{-} x − either is nonsupport of degree g − 1 g-1 g − 1 or equals t g − 1 ν t_{g-1}\nu t g − 1 ν . If x ∈ U x\in U x ∈ U , then x ≤ x − − x\leq x^{--} x ≤ x −− and the ineligible upper support point show that x − ∈ U x^{-}\in U x − ∈ U . If u ∈ U u\in U u ∈ U , z ∈ N z\in N z ∈ N and u z uz u z were support, then u − ( u z ) u^{-}(uz) u − ( u z ) would be the support top of z z z ’s degree, whereas u − u z ≤ z u^{-}uz\leq z u − u z ≤ z . The right-handed argument is the same. Thus x y xy x y is nonsupport for x , y ∈ U x,y\in U x , y ∈ U , and the eligible nonsupport point y − x − y^{-}x^{-} y − x − proves ( x y ) − (xy)^{-} ( x y ) − nonsupport. Hence U U ⊆ U UU\subseteq U U U ⊆ U .
For x , y ∈ U x,y\in U x , y ∈ U of degrees g , h g,h g , h , the nonsupport point x − y x^{-}y x − y is eligible in x \ y x\backslash y x \ y , at degree g − 1 h g^{-1}h g − 1 h . The support top there is ineligible since its product with x x x is t h > y t_h>y t h > y . Thus z = x \ y z=x\backslash y z = x \ y is nonsupport of that degree. From x z y − ≤ e xzy^{-}\leq e x z y − ≤ e , unit cyclicity gives z ( y − x ) ≤ e z(y^{-}x)\leq e z ( y − x ) ≤ e ; the eligible nonsupport second factor proves z ∈ U z\in U z ∈ U . For y / x y/x y / x use y x − yx^{-} y x − and the inequality ( y / x ) ( x y − ) ≤ e (y/x)(xy^{-})\leq e ( y / x ) ( x y − ) ≤ e . The unit belongs to U U U , so this proves fullness and the full group degree formulas, and H = μ [ U ] H=\mu[U] H = μ [ U ] is a subgroup.
Within a band, x ≤ y ∈ U x\leq y\in U x ≤ y ∈ U implies x − ≥ y − x^{-}\geq y^{-} x − ≥ y − , and the same residual bounds force x − ∈ N x^{-}\in N x − ∈ N . Thus U U U is initial and W W W final. The prime–product formulas
and full residual closure give the two reflection identities and U W , W U ⊆ W UW, WU \subseteq W U W , W U ⊆ W . If a ∈ U ∩ W a \in U \cap W a ∈ U ∩ W , then a − a ∈ W a^{-}a \in W a − a ∈ W and a − a ≤ e a^{-}a \leq e a − a ≤ e in the neutral band. Finality gives e ∈ W e \in W e ∈ W and hence d ∈ U d \in U d ∈ U , contradicting d 2 ∈ P d^{2} \in P d 2 ∈ P . Thus U ∩ W = ∅ U \cap W = \varnothing U ∩ W = ∅ and M ′ = M M' = M M ′ = M .
If x y ∈ U xy \in U x y ∈ U for x , y ∈ N x,y \in N x , y ∈ N , then y ( x y ) − y(xy)^{-} y ( x y ) − is nonsupport of inverse degree to x x x and x [ y ( x y ) − ] ≤ e x[y(xy)^{-}] \leq e x [ y ( x y ) − ] ≤ e , so x ∈ U x \in U x ∈ U . Rotating at e e e gives y ∈ U y \in U y ∈ U as well. Next, take w = u ′ ∈ W w = u' \in W w = u ′ ∈ W and z ∈ N ∖ U z \in N \setminus U z ∈ N ∖ U . If w z wz w z were nonsupport, then r = ( w z ) ′ = z \ u r = (wz)' = z\backslash u r = ( w z ) ′ = z \ u would be nonsupport and z r ≤ u zr \leq u z r ≤ u . The nonsupport point r u − ru^{-} r u − would give z ( r u − ) ≤ e z(ru^{-}) \leq e z ( r u − ) ≤ e , forcing z ∈ U z \in U z ∈ U , a contradiction. For z w zw z w use r = ( z w ) ′ = u / z r = (zw)' = u/z r = ( z w ) ′ = u / z and u − r u^{-}r u − r , together with unit cyclicity. Finally u m , m u um,mu u m , m u for u ∈ U , m ∈ M u \in U,m \in M u ∈ U , m ∈ M are nonsupport and not in U U U . If u m ∈ W um \in W u m ∈ W , then u − ( u m ) ∈ W u^{-}(um) \in W u − ( u m ) ∈ W and u − ( u m ) ≤ m u^{-}(um) \leq m u − ( u m ) ≤ m at the same degree, contrary to finality and m ∉ W m \notin W m ∈ / W . The right proof uses ( m u ) u − (mu)u^{-} ( m u ) u − . All asserted types follow.
Lemma D.10 (Completing bands and retaining coefficient maxima). Every support-band structure has a full extension with the same support in which each closed band [ t g ν , t g ] [t_{g\nu},t_g] [ t g ν , t g ] is complete. If U 1 U_1 U 1 has a greatest point, all coefficient fibres have greatest points T h T_h T h , and the extension’s coefficient fibre at h h h is exactly ( t h ν , T h ] (t_{h\nu},T_h] ( t h ν , T h ] . No new coefficient degree appears.
Proof. Take the MacNeille completion of the whole chain. Its multiplication is the closed lower ideal generated by old products, and its prime is the antitone extension of old prime. Here is the needed algebraic verification. Preimages of old principal lower ideals under old translations are principal, by residuation; hence preimages of closed lower sets are closed. The two-successive-preimage argument in Lemma D.7 proves that MacNeille closure is a multiplicative nucleus. It gives associative multiplication and preserves every old product and residual. For completed X , Y X,Y X , Y ,
X Y ≤ d ⟺ a b ≤ d ( a ≤ X , b ≤ Y old ) ⟺ Y ≤ X ′ . XY \leq d \Longleftrightarrow ab \leq d \quad(a \leq X,b \leq Y\ \text{old}) \Longleftrightarrow Y \leq X'. X Y ≤ d ⟺ ab ≤ d ( a ≤ X , b ≤ Y old ) ⟺ Y ≤ X ′ .
so prime is cyclic dualizing and supplies both actual residuals. Old central elements stay central by the join definition of product. The old set { e , d } ∪ { t r ν s : r , s ∈ Z } \{e,d\} \cup\{t_{r\nu s}:r,s \in\mathbb{Z}\} { e , d } ∪ { t r ν s : r , s ∈ Z } is product- and prime-closed and central, so Lemma D.8 gives all histories in the completion.
Retain only K 0 = ⋃ g ( t g ν , t g ] K_0 = \bigcup_g(t_{g\nu},t_g] K 0 = ⋃ g ( t g ν , t g ] . Join density gives an old same-band lower witness below every point of K 0 K_0 K 0 ; hence p x = x p = t g px = xp = t_g p x = x p = t g there. For x , y x,y x , y in bands g , h g,h g , h , old such witnesses give t g h ν < x y ≤ t g h t_{gh\nu} < xy \leq t_{gh} t g h ν < x y ≤ t g h . For interior x x x , old interior witnesses above and below x x x put x ′ x' x ′ in the required interior band; at a support top use its old prime. Thus K 0 K_0 K 0 is product- and prime-closed, contains e e e , d d d , and is full. Its closed bands inherit completeness and its support is exactly the old support.
Lemma D.7 gives the old maxima T h T_h T h . For t h ν < x ≤ T h t_{h\nu}<x\leq T_h t h ν < x ≤ T h , its unit residual is at least T h − T_h^{-} T h − , so it is nonsupport and x x x is a coefficient. If x > T h x>T_h x > T h , join density supplies an old same-band point a a a with T h < a ≤ x T_h<a\leq x T h < a ≤ x . Its supported unit residual bounds x − x^{-} x − above, excluding x x x from the coefficient piece. In an old band with no coefficient, any old same-band lower witness gives the same argument. This proves the claims about coefficients and their maxima. □ \square □
Theorem D.11 (Dense bands over a greatest neutral coefficient). Every countable support-band structure whose intrinsic U 1 U_1 U 1 has a greatest point has a full countable dense extension, satisfying every finite history, with exactly its old support group. Every protected closed band becomes densely ordered.
Formalization scope. The companion Lean development verifies the concrete closed-support core instance used in Theorem 7.1, including the preparation of a greatest neutral coefficient. An independent adapter from arbitrary support-band data in the formulation above to that concrete formal interface is not included; see Section E.
Proof. Apply Lemma D.10; the old algebra used during the construction may now be uncountable. Denote its coefficient subgroup by H H H and its coefficient maxima by T h T_h T h . In degree g g g the old middle is the open interval between
ℓ g = { T g , g ∈ H , t g ν , g ∉ H , r g = { T g − 1 δ ′ , g − 1 δ ∈ H , t g , g − 1 δ ∉ H . \ell_g =
\begin{cases}
T_g, & g \in H,\\
t_{g\nu}, & g \notin H,
\end{cases}
\qquad
r_g =
\begin{cases}
T'_{g^{-1}\delta}, & g^{-1}\delta\in H,\\
t_g, & g^{-1}\delta\notin H.
\end{cases} ℓ g = { T g , t g ν , g ∈ H , g ∈ / H , r g = { T g − 1 δ ′ , t g , g − 1 δ ∈ H , g − 1 δ ∈ / H .
These endpoints are distinct. Form an auxiliary complete interval M ‾ g = [ ℓ g , r g ] \overline{M}_g=[\ell_g,r_g] M g = [ ℓ g , r g ] , keeping separate endpoint copies where different intervals share an actual support endpoint. Prime pairs the intervals at g g g and g − 1 δ g^{-1}\delta g − 1 δ and exchanges their two endpoint types. For h ∈ H h\in H h ∈ H , membership of g g g in H H H is equivalent to membership of either h g hg h g or g h gh g h in H H H . For the upper endpoint types, centrality of δ \delta δ gives
( h g ) − 1 δ = ( g − 1 δ ) h − 1 , ( g h ) − 1 δ = h − 1 ( g − 1 δ ) . (hg)^{-1}\delta=(g^{-1}\delta)h^{-1},\qquad(gh)^{-1}\delta=h^{-1}(g^{-1}\delta). ( h g ) − 1 δ = ( g − 1 δ ) h − 1 , ( g h ) − 1 δ = h − 1 ( g − 1 δ ) .
Membership of these degrees in H H H is therefore equivalent to g − 1 δ ∈ H g^{-1}\delta\in H g − 1 δ ∈ H . These translations use only the subgroup property of H H H and centrality of $\delta.
For u ∈ U h u\in U_h u ∈ U h , retain its old interior actions, map lower endpoints to lower endpoints and upper endpoints to upper endpoints, at degrees h g hg h g on the left. and g h gh g h on the right. The upper-endpoint values are their actual old products, since
T a / u = T a h − 1 , u \ T a = T h − 1 a . T_a/u = T_{ah^{-1}}, \qquad u\backslash T_a = T_{h^{-1}a}. T a / u = T a h − 1 , u \ T a = T h − 1 a .
Indeed all coefficients of the indicated degree are eligible, and the full degree map determines the residual’s degree. At support endpoints use the support product rule. Interior actions remain in the middle by Lemma D.9 . Hence clamping only the lower endpoints respects both action laws and commutation. Cyclic adjunction on interiors is inherited. At matching degrees, if either tested point is a lower endpoint, both comparisons hold; if one is an upper endpoint and the other is not lower, both fail, because actions preserve interiors and endpoint types. These exhaust the endpoint cases. Distinct degrees decide both comparisons in the group order. These are thus cyclic action chains, with the indicated graded actions. The set of occupied middle degrees is invariant under left and right H H H -translation: act by a coefficient of degree h h h and then one of degree h − 1 h^{-1} h − 1 .
Apply Lemma D.5 only to adjacent pairs within one auxiliary interval. This family is invariant under prime and transport: a transported upper endpoint remains interior or upper, never lower, so its predecessor remains in the same target interval. Complete each enlarged interval using Lemma D.6 , whose proof applies separately between the corresponding domain and target grades. Delete its two auxiliary endpoints and call the resulting dense interval N g N_g N g . The old complete interval retains all its joins and meets. Every new point lies in one old adjacent gap: its old floor and ceiling otherwise coincide, by old completeness, making it old. Its ceiling
a ( X ) = min { m ∈ M ‾ g : X ≤ m } , X ∈ N g , a(X) = \min\{m \in\overline{M}_g : X \leq m\}, \qquad X \in N_g, a ( X ) = min { m ∈ M g : X ≤ m } , X ∈ N g ,
therefore belongs to M g ∪ { r g } M_g \cup\{r_g\} M g ∪ { r g } . Endpoints used as anchors always mean their actual old values. The transport prescription gives
a ( u X ) = u a ( X ) , a ( X u ) = a ( X ) u , X ≤ m ⟺ a ( X ) ≤ m ( m ∈ M ‾ g ) . (73) \begin{aligned}
a(uX) &= u\,a(X), \qquad a(Xu) = a(X)u,\\
X &\leq m \iff a(X) \leq m \qquad(m \in\overline{M}_g). \tag*{(73)}
\end{aligned} a ( u X ) X = u a ( X ) , a ( X u ) = a ( X ) u , ≤ m ⟺ a ( X ) ≤ m ( m ∈ M g ) . (73)
These assertions survive completion within each inserted gap. Actions retain N g N_g N g : interior upper anchors have interior images, while every degree-one return fixes an upper boundary, so its gap is transported and never collapsed to the deleted endpoint. This also handles empty old middle intervals. The ceiling preserves joins, being left adjoint to the inclusion of the old complete interval.
Use the raw clone F F F and the graded closed-cut coefficient extension V = ⋃ h ∈ H V h V=\bigcup_{h\in H}V_h V = ⋃ h ∈ H V h from Lemma D.7. For A ∈ V h A\in V_h A ∈ V h define
λ A X = ⋁ f ∈ A ∩ F h π ( f ) X , ρ A X = ⋁ f ∈ A ∩ F h X π ( f ) . (74) \lambda_A X=\bigvee_{f\in A\cap F_h}\pi(f)X,\qquad\rho_A X=\bigvee_{f\in A\cap F_h}X\pi(f). \tag*{(74)} λ A X = f ∈ A ∩ F h ⋁ π ( f ) X , ρ A X = f ∈ A ∩ F h ⋁ X π ( f ) . (74)
These joins are taken in the complete target auxiliary intervals. They are nonempty and lie between one interior action value and the interior value for T h T_h T h , so belong to the target N N N interval. For X ∈ N g X\in N_g X ∈ N g , Z ∈ N j Z\in N_j Z ∈ N j and b = j g − 1 ∈ H b=jg^{-1}\in H b = j g − 1 ∈ H , let
E b ( X , Z ) = { f ∈ F b : π ( f ) X ≤ Z } . E_b(X,Z)=\{f\in F_b:\pi(f)X\le Z\}. E b ( X , Z ) = { f ∈ F b : π ( f ) X ≤ Z } .
For the later comparison with Z = Y ′ Z=Y' Z = Y ′ and Y ∈ N k Y\in N_k Y ∈ N k , its degree is j = k − 1 δ j=k^{-1}\delta j = k − 1 δ , so centrality of δ \delta δ gives
j g − 1 = k − 1 δ g − 1 = ( g k ) − 1 δ . jg^{-1}=k^{-1}\delta g^{-1}=(gk)^{-1}\delta. j g − 1 = k − 1 δ g − 1 = ( g k ) − 1 δ .
Adjoin all lower-degree fibres to this boundary section. The resulting lower cut is MacNeille closed: each old projection fibre is whole or empty, and every excluded point has a smaller excluded point in its own fibre. If E b E_b E b is nonempty, the cut has a witness in F b F_b F b and upper bound i ( T b ) i(T_b) i ( T b ) , so is an actual element κ b ( X , Z ) ∈ V b \kappa_b(X,Z)\in V_b κ b ( X , Z ) ∈ V b . An empty boundary is recorded separately, and is not replaced by a coefficient of a smaller degree.
The maps in (47) preserve joins. To verify their action law, test λ A λ B X ≤ Z \lambda_A\lambda_BX\le Z λ A λ B X ≤ Z in the target grade. Expanding joins tests every raw boundary product from ( A ∩ F h ) ( B ∩ F k ) (A\cap F_h)(B\cap F_k) ( A ∩ F h ) ( B ∩ F k ) against the full comparison cut just described. Lower-degree products are automatically eligible. Closedness makes this equivalent to the same test on the MacNeille closure of the product, namely λ A ⊗ B X ≤ Z \lambda_{A\otimes B}X\le Z λ A ⊗ B X ≤ Z . The right law is identical. Old commutation and independent joins give cross commutation. Testing each raw coefficient and using old cyclic adjunction gives λ A X ≤ Y ′ ⟺ ρ A Y ≤ X ′ \lambda_A X\le Y'\Longleftrightarrow\rho_A Y\le X' λ A X ≤ Y ′ ⟺ ρ A Y ≤ X ′ . The unit and the old actions are retained because the principal cut of i ( u ) i(u) i ( u ) contains i ( u ) i(u) i ( u ) and all its other same-degree projections are at most u u u . For an old middle point these joins are old: they are bounded within its old target middle by one image and the T h T_h T h image, and old joins are retained. On an upper anchor they are its actual old action. Join preservation of the ceiling consequently gives
a ( λ A X ) = λ A a ( X ) , a ( ρ A X ) = ρ A a ( X ) . (75) a(\lambda_A X)=\lambda_A a(X),\qquad a(\rho_A X)=\rho_A a(X). \tag*{(75)} a ( λ A X ) = λ A a ( X ) , a ( ρ A X ) = ρ A a ( X ) . (75)
In every degree g g g now put the ordered pieces
[ V g if g ∈ H ] + N g + [ V g − 1 δ ∗ if g − 1 δ ∈ H ] + { t g } . [V_g\text{ if }g\in H]+N_g+[V^*_{g^{-1}\delta}\text{ if }g^{-1}\delta\in H]+\{t_g\}. [ V g if g ∈ H ] + N g + [ V g − 1 δ ∗ if g − 1 δ ∈ H ] + { t g } .
Prime exchanges the two coefficient copies, retains interval prime, and has the old support carry. Keep coefficient products, the two actions, support products, and the reflection rules
A B ∗ = ( B / A ) ∗ , A ∗ B = ( B \ A ) ∗ . AB^*=(B/A)^*,\qquad A^*B=(B\backslash A)^*. A B ∗ = ( B / A ) ∗ , A ∗ B = ( B \ A ) ∗ .
Products of a reverse coefficient with a middle or reverse coefficient are the support top of the product degree. For X ∈ N g , Y ∈ N k X\in N_g,Y\in N_k X ∈ N g , Y ∈ N k put b = ( g k ) − 1 δ b=(gk)^{-1}\delta b = ( g k ) − 1 δ and define
X ∘ Y = { a ( X ) a ( Y ) , a ( X ) a ( Y ) ∈ M , κ b ( X , Y ′ ) ∗ , otherwise , b ∈ H , E b ( X , Y ′ ) ≠ ∅ , t g k , otherwise . (76) X\mathbin{\circ}Y=
\begin{cases}
a(X)a(Y), & a(X)a(Y)\in M,\\
\kappa_b(X,Y')^*, & \text{otherwise},\ b\in H,\ E_b(X,Y')\ne\varnothing,\\
t_{gk}, & \text{otherwise}.
\end{cases} \tag*{(76)} X ∘ Y = ⎩ ⎨ ⎧ a ( X ) a ( Y ) , κ b ( X , Y ′ ) ∗ , t g k , a ( X ) a ( Y ) ∈ M , otherwise , b ∈ H , E b ( X , Y ′ ) = ∅ , otherwise . (76)
All comparison values in this rule are attained members of V b V_b V b .
If either input in (49) is an old middle point, the output is exactly the embedded old product of its two anchors. Indeed (46) gives u X ≤ m ′ ⟺ u a ( X ) ≤ m ′ uX\le m'\Longleftrightarrow ua(X)\le m' u X ≤ m ′ ⟺ u a ( X ) ≤ m ′ for old m m m , and, by prime, u m ≤ Y ′ ⟺ u m ≤ a ( Y ) ′ um\le Y'\Longleftrightarrow um\le a(Y)' u m ≤ Y ′ ⟺ u m ≤ a ( Y ) ′ . These tests also hold for upper anchors. A reverse old output gives precisely the principal lift of its old prime as comparison cut; a support output gives an empty boundary. The internal case is already the first line of (49). In particular every old product is retained.
We verify associativity, including the nonunit coefficient cases. For three middle factors, an internal inner product is an old middle point. If both inner products are internal, both bracketings are the old three-anchor product. If one inner product is reverse or support, its final product is support; the corresponding old anchor product has that same type. If the other inner product is internal, its bracketing therefore gives the same old support value; if not, both bracketings are support with the same degree.
For an outer coefficient A ∈ V h A\in V_h A ∈ V h , put z = a ( X ) a ( Y ) z=a(X)a(Y) z = a ( X ) a ( Y ) and choose f 0 ∈ A ∩ F h f_0\in A\cap F_h f 0 ∈ A ∩ F h , which is nonempty by the construction of V h V_h V h . Every f ∈ A ∩ F h f\in A\cap F_h f ∈ A ∩ F h has π ( f ) ∈ U h \pi(f)\in U_h π ( f ) ∈ U h and π ( f ) ≤ T h \pi(f)\le T_h π ( f ) ≤ T h . Equations (47) and (48), together with monotonicity of old multiplication, give the two-sided bounds
π ( f 0 ) z ≤ a ( λ A X ) a ( Y ) ≤ T h z , \pi(f_0)z\le a(\lambda_A X)a(Y)\le T_hz, π ( f 0 ) z ≤ a ( λ A X ) a ( Y ) ≤ T h z ,
z π ( f 0 ) ≤ a ( X ) a ( ρ A Y ) ≤ z T h . z\pi(f_0)\le a(X)a(\rho_A Y)\le zT_h. z π ( f 0 ) ≤ a ( X ) a ( ρ A Y ) ≤ z T h .
For X ∈ N g , Y ∈ N k X\in N_g,Y\in N_k X ∈ N g , Y ∈ N k , all three terms in the first line have degree h g k hgk h g k , and those in the second have degree g k h gkh g k h . If z ∈ M z\in M z ∈ M , both bounds in each line lie in the middle piece by Lemma D.9; hence the intervening anchor product is internal. If z ∈ W z \in W z ∈ W , the bounds lie in the reverse piece by U W ⊆ W UW \subseteq W U W ⊆ W and W U ⊆ W WU \subseteq W W U ⊆ W . Convexity of that piece in its degree therefore keeps the anchor product in W W W . In particular the lower witness excludes a fall into the middle, while the upper bound excludes the support top. If z z z is support, both bounds equal the support point of the target degree, so the anchor product is support as well. Thus outer coefficient actions preserve the internal, reverse, and support types. In the internal case, (47)–(48) and preservation of old bounded joins by multiplication give the required associativity. For a coefficient between two middle factors, the two anchor products agree without assuming their type in advance:
a ( ρ A X ) a ( Y ) = ⋁ f ∈ A ∩ F h ( a ( X ) π ( f ) ) a ( Y ) = ⋁ f ∈ A ∩ F h a ( X ) ( π ( f ) a ( Y ) ) = a ( X ) a ( λ A Y ) . \begin{aligned}
a(\rho_A X)a(Y) &= \bigvee_{f\in A\cap F_h}(a(X)\pi(f))a(Y) \\
&= \bigvee_{f\in A\cap F_h}a(X)(\pi(f)a(Y)) = a(X)a(\lambda_A Y).
\end{aligned} a ( ρ A X ) a ( Y ) = f ∈ A ∩ F h ⋁ ( a ( X ) π ( f )) a ( Y ) = f ∈ A ∩ F h ⋁ a ( X ) ( π ( f ) a ( Y )) = a ( X ) a ( λ A Y ) .
If this common product is internal, the equality settles the case. In all noninternal cases the required identities are
κ b h − 1 ( λ A X , Y ′ ) = κ b ( X , Y ′ ) / A , \kappa_{bh^{-1}}(\lambda_A X,Y')=\kappa_b(X,Y')/A, κ b h − 1 ( λ A X , Y ′ ) = κ b ( X , Y ′ ) / A ,
κ h − 1 b ( X , ( ρ A Y ) ′ ) = A \ κ b ( X , Y ′ ) , \kappa_{h^{-1}b}(X,(\rho_A Y)')=A\backslash\kappa_b(X,Y'), κ h − 1 b ( X , ( ρ A Y ) ′ ) = A \ κ b ( X , Y ′ ) ,
κ c ( ρ A X , Y ′ ) = κ c ( X , ( λ A Y ) ′ ) , c = ( g h k ) − 1 δ . \kappa_c(\rho_A X,Y')=\kappa_c(X,(\lambda_A Y)'),\qquad c=(ghk)^{-1}\delta. κ c ( ρ A X , Y ′ ) = κ c ( X , ( λ A Y ) ′ ) , c = ( g hk ) − 1 δ .
Each follows by testing an arbitrary coefficient in its displayed boundary degree, using the action laws and cyclic adjunction. If a boundary is empty, an eligible product would give an eligible coefficient in that boundary, a contradiction. Thus the identities also cover support outputs. If b ∉ H b \notin H b ∈ / H , neither translated boundary exists; the middle placement has the same boundary on both sides. Together with the anchor-type preservation established above, this completes all one-coefficient cases. Three coefficients use associativity in V V V . Two coefficients on a middle use the action laws. Two coefficients on a reverse use ( C / B ) / A = C / ( A B ) (C/B)/A=C/(AB) ( C / B ) / A = C / ( A B ) , ( A B ) \ C = B \ ( A \ C ) (AB)\backslash C=B\backslash(A\backslash C) ( A B ) \ C = B \ ( A \ C ) , and B \ ( C / A ) = ( B \ C ) / A B\backslash(C/A)=(B\backslash C)/A B \ ( C / A ) = ( B \ C ) / A , all immediate from residuation. Every remaining triple has support output in both bracketings, with the same degree. The old unit is still a unit.
For monotonicity within a middle fibre, the anchor product is monotone and its internal-output region is initial. In that region use old monotonicity; outside it, comparison sets decrease with either factor, so their reversed values increase, with an empty boundary represented by the support top. Crossing the region boundary increases the type from middle to reverse or support. At a coefficient–middle interface, if g ∈ H g \in H g ∈ H , A ∈ V g A \in V_g A ∈ V g , and X ∈ N g X \in N_g X ∈ N g , an internal output satisfies
A Y ≤ T g Y ≤ T g a ( Y ) ≤ a ( X ) a ( Y ) = X ∘ Y . AY \le T_gY \le T_ga(Y) \le a(X)a(Y) = X \circ Y. A Y ≤ T g Y ≤ T g a ( Y ) ≤ a ( X ) a ( Y ) = X ∘ Y .
The right interface is identical. Noninternal outputs lie above the middle action values. At a middle–reverse interface multiplication by a middle has support upper values; coefficient multiplication preserves the order of the pieces. The reflection and support rules settle the other interfaces. Different degrees are strictly compared by the biorder. Thus multiplication is monotone everywhere.
For the cyclic cut it suffices to consider total degree δ \delta δ . An internal middle anchor product is below d d d , and hence X ≤ a ( X ) ≤ a ( Y ) ′ ≤ Y ′ X \le a(X) \le a(Y)' \le Y' X ≤ a ( X ) ≤ a ( Y ) ′ ≤ Y ′ . Otherwise (49) gives
X ∘ Y ≤ d ⟺ e V ≤ κ 1 ( X , Y ′ ) ⟺ X ≤ Y ′ . X \circ Y \le d \Longleftrightarrow e_V \le\kappa_1(X,Y') \Longleftrightarrow X \le Y'. X ∘ Y ≤ d ⟺ e V ≤ κ 1 ( X , Y ′ ) ⟺ X ≤ Y ′ .
An empty boundary fails both conditions. For coefficient–reverse pairs the same cut is ordinary residuation in V V V ; coefficient–coefficient and coefficient–middle pairs satisfy both comparisons, while reverse–middle, reverse–reverse and pairs with a support factor fail both. At other total degrees the group order decides the cut. Thus prime is cyclic dualizing, and associativity supplies both actual residuals by the prime–product formulas. All old operations, constants and prime were retained, so the old inclusion is full.
Finally d = e ∗ d = e^* d = e ∗ is central: on coefficients this follows from unit cyclicity of V V V , and on middle, reverse and support pieces both products are support with equal degree, since δ \delta δ is central. Every t δ r ν s t_{\delta}^{r\nu s} t δ r ν s is central by the support product rule. The set { e , d } ∪ { t δ r ν s : r , s ∈ Z } \{e,d\} \cup\{t_{\delta}^{r\nu s}: r,s \in\mathbb{Z}\} { e , d } ∪ { t δ r ν s : r , s ∈ Z } is closed under prime and product. Lemma D.8 therefore proves every finite history in the constructed algebra. Multiplication by the retained closed support p = t 1 p = t_1 p = t 1 fixes precisely the old support tops, so its support group is exactly G G G .
Every N g N_g N g is dense with neither endpoint. Each coefficient fibre is dense without a least point and retains T h T_h T h as greatest; its reverse has no greatest point. These facts fill every interface and make every closed protected band dense. They also make the whole chain dense: between points of different degrees one may interpolate in the upper point’s band, using the protected predecessor there. Starting with the original countable source, alternately take full subalgebra closure and choose one interpolant in each strict pair. The union is countable and dense, full in the constructed algebra, and has exactly G G G , since all degrees are old and all old support points were retained. This proves the theorem. □ \square □
A greatest neutral coefficient from proper filters
Lemma D.12 (A bounded full extension before filter completion). Every nontrivial cyclic involutive residuated chain K ∈ K R K \in\mathcal{K}_{R} K ∈ K R has a full bounded extension C ∈ K R C \in\mathcal{K}_{R} C ∈ K R obtained by adjoining a new least point 0 0 0 and a new greatest point 1 1 1 . All original finite histories hold simultaneously in C C C . The extension has no nonzero zero divisors, regardless of whether the old chain has an absorbing point in its own order.
Proof. On C = { 0 } + K + { 1 } C = \{0\} + K + \{1\} C = { 0 } + K + { 1 } retain the old product, unit and dualizer; make 0 0 0 absorbing and 1 1 1 absorb every nonzero point. Retain old prime and put 0 ′ = 1 0' = 1 0 ′ = 1 , 1 ′ = 0 1' = 0 1 ′ = 0 . These rules are associative: a zero factor makes either bracketing zero; without a zero, a factor 1 1 1 makes either bracketing 1 1 1 ; otherwise old associativity applies. They are monotone and retain the old unit. For old factors the cyclic cut is unchanged. A zero factor and the cases involving 1 1 1 give directly
x y ≤ d ⟺ y ≤ x ′ . xy \le d \iff y \le x'. x y ≤ d ⟺ y ≤ x ′ . Thus prime is cyclic dualizing, and the prime–product formulas supply both residuals. Old products and prime are retained, so the old inclusion is full in the original signature. The new bottom cannot be the product of two nonzero points.
Unit cyclicity follows by the same cases at the old threshold e e e . Fix any finite history w w w . Its guards c w c_w c w , h w h_w h w are old closed terms. If a premise product at either guard is zero, one of its two factors is zero, and the rotated five-factor conclusion is zero. Otherwise both premise products are at most old guard values and are nonzero; neither can involve 1 1 1 , so all four premise factors belong to K K K . For an old middle factor z z z , use the old history. For z = 0 z = 0 z = 0 the conclusion is zero, and for z = 1 z = 1 z = 1 the conclusion is 1 ≤ 1 1 \le1 1 ≤ 1 . This proves every history, with no bound on its length. □ \square □
Lemma D.13 (A proper-filter completion retaining every finite history). Let C ∈ K R C \in\mathcal{K}_{R} C ∈ K R be nontrivial and bounded. It has a full bounded embedding into a complete chain D ∈ K R D \in\mathcal{K}_{R} D ∈ K R . If C C C has no nonzero zero divisors, the same is true of D D D . In particular this one completion retains every original finite history simultaneously. The embedding need not preserve infinite joins or meets.
Proof. Write C + = C ∖ { 0 } C^{+}=C\setminus\{0\} C + = C ∖ { 0 } . Let P P P be the set of proper lattice filters of C C C : these are exactly its nonempty upward closed subsets not containing 0 0 0 . They are linearly ordered by inclusion. Let D D D be the chain of all upward closed subsets of ( P , ⊆ ) (P,\subseteq) ( P , ⊆ ) , including the empty set, ordered by inclusion. Arbitrary joins and meets in D D D are unions and intersections. Set
i ( a ) = { F ∈ P : a ∈ F } , j ( F ) = { H ∈ P : F ⊆ H } . i(a)=\{F\in P:a\in F\},\qquad j(F)=\{H\in P:F\subseteq H\}. i ( a ) = { F ∈ P : a ∈ F } , j ( F ) = { H ∈ P : F ⊆ H } .
Then i i i is an injective bounded lattice homomorphism, and every X ∈ D X\in D X ∈ D is ⋃ F ∈ X j ( F ) \bigcup_{F\in X}j(F) ⋃ F ∈ X j ( F ) . Injectivity follows, for example, because a < b a<b a < b is separated by the proper filter ↑ b \uparrow b ↑ b .
For F , G ∈ P F,G\in P F , G ∈ P define
F ∗ G = ↑ { a b : a ∈ F , b ∈ G } , X ⊙ Y = ⋃ F ∈ X , G ∈ Y j ( F ∗ G ) . F*G=\mathord{\uparrow}\{ab:a\in F,\ b\in G\},\qquad X\mathbin{\odot}Y=\bigcup_{\substack{F\in X,\ G\in Y}}j(F*G). F ∗ G = ↑ { ab : a ∈ F , b ∈ G } , X ⊙ Y = F ∈ X , G ∈ Y ⋃ j ( F ∗ G ) .
The filter F ∗ G F*G F ∗ G is allowed to be the improper filter C C C . For this one additional filter put j ( C ) = ∅ j(C)=\varnothing j ( C ) = ∅ . On the set P ∪ { C } P\cup\{C\} P ∪ { C } , the improper filter is absorbing for ∗ * ∗ , because the old bottom is absorbing. Multiplication ∗ * ∗ is associative: taking upward closure before or after a product gives the same result, by monotonicity of the old multiplication. Its unit is ↑ e \uparrow e ↑ e . Also
j ( F ) ⊙ j ( G ) = j ( F ∗ G ) , i ( a ) ⊙ i ( b ) = i ( a b ) . j(F)\mathbin{\odot}j(G)=j(F*G),\qquad i(a)\mathbin{\odot}i(b)=i(ab). j ( F ) ⊙ j ( G ) = j ( F ∗ G ) , i ( a ) ⊙ i ( b ) = i ( ab ) .
For the second identity when a , b > 0 a,b>0 a , b > 0 , use the principal filters ↑ a , ↑ b \uparrow a,\uparrow b ↑ a , ↑ b for one inclusion; for the other, a ∈ F , b ∈ G a\in F,b\in G a ∈ F , b ∈ G implies a b ∈ F ∗ G ab\in F*G ab ∈ F ∗ G . Zero factors give empty unions. The product definition preserves arbitrary unions in each coordinate. Consequently the principal-set product identities and associativity of ∗ * ∗ prove associativity on all of D D D , with unit i ( e ) i(e) i ( e ) .
Define
τ F = { a ∈ C : a ′ ∉ F } , X † = P ∖ τ − 1 ( X ) . \tau F=\{a\in C:a'\notin F\},\qquad X^\dagger=P\setminus\tau^{-1}(X). τ F = { a ∈ C : a ′ ∈ / F } , X † = P ∖ τ − 1 ( X ) .
The map τ \tau τ is an inclusion-reversing involution of P P P : its values are nonempty upward closed sets containing 1 1 1 and omitting 0 0 0 , and τ 2 F = F \tau^2F=F τ 2 F = F . Thus † \dagger † is an order-reversing involution of D D D and i ( a ) † = i ( a ′ ) i(a)^\dagger=i(a') i ( a ) † = i ( a ′ ) . For principal closed sets one has the exact equivalences
j ( F ) ⊙ j ( G ) ⊆ i ( d ) ⟺ d ∈ F ∗ G ⟺ ( ∃ a ∈ F ) ( ∃ b ∈ G ) a b ≤ d ⟺ ( ∃ b ∈ G ) b ′ ∈ F ⟺ j ( F ) ⊆ j ( G ) † . \begin{aligned}
j(F)\mathbin{\odot}j(G)\subseteq i(d)\iff d\in F*G \\
&\iff(\exists a\in F)(\exists b\in G)\ ab\le d \\
&\iff(\exists b\in G)\ b'\in F \\
&\iff j(F)\subseteq j(G)^\dagger.
\end{aligned} j ( F ) ⊙ j ( G ) ⊆ i ( d ) ⟺ d ∈ F ∗ G ⟺ ( ∃ a ∈ F ) ( ∃ b ∈ G ) ab ≤ d ⟺ ( ∃ b ∈ G ) b ′ ∈ F ⟺ j ( F ) ⊆ j ( G ) † .
These equivalences also hold when F ∗ G = C F \ast G = C F ∗ G = C : its principal set is empty, and the definition of upward closure supplies a zero-product witness. The third equivalence uses old cyclic duality and upward closure of F F F . Taking unions in the two arguments gives
X ⊙ Y ⊆ i ( d ) ⟺ X ⊆ Y † . X \odot Y \subseteq i(d) \Longleftrightarrow X \subseteq Y^{\dagger}. X ⊙ Y ⊆ i ( d ) ⟺ X ⊆ Y † .
Associativity and the cyclic cut give the full residuals X \ Z = ( Z † ⊙ X ) † X\backslash Z = (Z^{\dagger} \odot X)^{\dagger} X \ Z = ( Z † ⊙ X ) † and Z / Y = ( Y ⊙ Z † ) † Z/Y = (Y \odot Z^{\dagger})^{\dagger} Z / Y = ( Y ⊙ Z † ) † . Thus i i i preserves both residuals as well as prime, product, order and all constants.
Unit cyclicity is verified separately and exactly. On principal closed sets,
j ( F ) ⊙ j ( G ) ⊆ i ( e ) ⟺ ( ∃ a ∈ F ) ( ∃ b ∈ G ) a b ≤ e ⟺ j ( G ) ⊙ j ( F ) ⊆ i ( e ) , j(F) \odot j(G) \subseteq i(e) \Longleftrightarrow(\exists a \in F)(\exists b \in G)\;ab \le e \Longleftrightarrow j(G) \odot j(F) \subseteq i(e), j ( F ) ⊙ j ( G ) ⊆ i ( e ) ⟺ ( ∃ a ∈ F ) ( ∃ b ∈ G ) ab ≤ e ⟺ j ( G ) ⊙ j ( F ) ⊆ i ( e ) ,
using old unit cyclicity. Arbitrary unions give the same equivalence for all X , Y X,Y X , Y .
We verify all histories directly, without a square-top assumption. Fix an arbitrary finite history w w w . Its two guard constants c w , h w c_w,h_w c w , h w are old terms, so their new values are i ( c w ) , i ( h w ) i(c_w),i(h_w) i ( c w ) , i ( h w ) . Suppose X Y ⊆ i ( c w ) XY \subseteq i(c_w) X Y ⊆ i ( c w ) and R S ⊆ i ( h w ) RS \subseteq i(h_w) R S ⊆ i ( h w ) . For arbitrary F ∈ X , G ∈ Y , H ∈ R , J ∈ S F \in X,G \in Y,H \in R,J \in S F ∈ X , G ∈ Y , H ∈ R , J ∈ S , the filter-product formulas supply a ∈ F , b ∈ G , r ∈ H , s ∈ J a \in F,b \in G,r \in H,s \in J a ∈ F , b ∈ G , r ∈ H , s ∈ J with a b ≤ c w ab \le c_w ab ≤ c w and r s ≤ h w rs \le h_w r s ≤ h w . For any K ∈ Z K \in Z K ∈ Z and k ∈ K k \in K k ∈ K , the old history gives b a k s r ≤ k baksr \le k bak sr ≤ k . Thus
G ∗ F ∗ K ∗ J ∗ H ⊇ K , j ( G ∗ F ∗ K ∗ J ∗ H ) ⊆ j ( K ) ⊆ Z . G \ast F \ast K \ast J \ast H \supseteq K,\qquad j(G \ast F \ast K \ast J \ast H) \subseteq j(K) \subseteq Z. G ∗ F ∗ K ∗ J ∗ H ⊇ K , j ( G ∗ F ∗ K ∗ J ∗ H ) ⊆ j ( K ) ⊆ Z .
Taking the independent unions over these five filters gives Y X Z S R ⊆ Z YXZSR \subseteq Z Y X Z S R ⊆ Z . If one of the relevant sets is empty, the product is empty and the same conclusion is immediate. This proves the chosen history, and w w w was arbitrary. If the old chain has no nonzero zero divisors, every product F ∗ G F \ast G F ∗ G of proper filters is proper. A product of two nonempty sets then contains a nonempty j ( F ∗ G ) j(F \ast G) j ( F ∗ G ) and is nonempty. This proves the optional zero-freeness assertion.
Lemma D.14 (The successor of an old point in the filter completion). Let C ∈ K R C \in K_R C ∈ K R be nontrivial and bounded, and let D D D be its proper-filter completion from Lemma D.13 , with embedding i : C → D i:C \to D i : C → D . For every a < 1 a < 1 a < 1 in C C C , put
F a = { c ∈ C : a < c } , B a = j ( F a ) . F_a = \{c \in C : a < c\},\qquad B_a = j(F_a). F a = { c ∈ C : a < c } , B a = j ( F a ) . Then B a B_a B a is the immediate successor of i ( a ) i(a) i ( a ) in D D D .
Proof. Use the proper-filter notation P P P , i i i , j j j of Lemma D.13. The set F a F_a F a is nonempty, upward closed and omits 0 0 0 , so it is a member of P P P . Every filter containing a a a contains F a F_a F a , whereas a ∉ F a a \notin F_a a ∈ / F a . Hence i ( a ) ⊊ B a i(a) \subsetneq B_a i ( a ) ⊊ B a . If X ∈ D X \in D X ∈ D strictly contains i ( a ) i(a) i ( a ) , choose H ∈ X ∖ i ( a ) H \in X \setminus i(a) H ∈ X ∖ i ( a ) . Since a ∉ H a \notin H a ∈ / H , upward closure of H H H gives H ⊆ F a H \subseteq F_a H ⊆ F a . Upward closure of X X X in the filter-inclusion order now gives F a ∈ X F_a \in X F a ∈ X and B a ⊆ X B_a \subseteq X B a ⊆ X . Thus no point lies strictly between i ( a ) i(a) i ( a ) and B a B_a B a .
Theorem D.15 (Every upper-invertible complete core acquires a neutral maximum). Let K ∈ K R K \in\mathcal{K}_{R} K ∈ K R be a countable upper-invertible complete group core, with its closed support p > e p > e p > e , support group G = p K G = pK G = p K , q = p d q = pd q = p d , and n = q ′ < e n = q' < e n = q ′ < e .
There is a full countable extension K ↪ A ∈ K R K \hookrightarrow A \in\mathcal{K}_{R} K ↪ A ∈ K R such that
p A = G , the intrinsic neutral coefficient fibre of A has a greatest point. pA = G,\qquad\text{the intrinsic neutral coefficient fibre of } A \text{ has a greatest point.} p A = G , the intrinsic neutral coefficient fibre of A has a greatest point. No condition is imposed on the original coefficient chain, its support subgroup, its actions, or the middle multiplication.
Proof. First apply Lemma D.12, and then the proper-filter completion of Lemma D.13. This gives full embeddings K ⊆ C → i D ∈ K R K \subseteq C \xrightarrow{i} D \in\mathcal{K}_{R} K ⊆ C i D ∈ K R . The completion preserves all old operations and every finite history, although it may break old infinite suprema. Identify old points with their images when writing algebraic products in D D D .
We use the principal-set product identity from Lemma D.13:
j ( F ) ⊙ j ( H ) = j ( F ∗ H ) , F ∗ H = ↑ { x y : x ∈ F , y ∈ H } , j(F) \odot j(H) = j(F * H),\qquad F * H = \uparrow\{xy : x \in F,\ y \in H\}, j ( F ) ⊙ j ( H ) = j ( F ∗ H ) , F ∗ H =↑ { x y : x ∈ F , y ∈ H } , with j ( C ) = ∅ j(C) = \varnothing j ( C ) = ∅ for an improper product filter.
Set B = B n = j ( F n ) B = B_n = j(F_n) B = B n = j ( F n ) , where F n = ( n , 1 ] C F_n = (n,1]_C F n = ( n , 1 ] C . Lemma D.14 gives
Here e ∈ F n e \in F_n e ∈ F n proves B ≤ i ( e ) B \le i(e) B ≤ i ( e ) . We claim
p B = B p = p . pB = Bp = p. pB = B p = p . Indeed, for every old c > n c > n c > n in K K K , one has p c ≥ c > n pc \ge c > n p c ≥ c > n , while p c pc p c belongs to the old support group. The protected support cover n < p n < p n < p therefore gives p c ≥ p pc \ge p p c ≥ p ; the same holds on the right. For the new point c = 1 c = 1 c = 1 the product is 1 ≥ p 1 \ge p 1 ≥ p . Taking c = e c = e c = e gives equality p e = p pe = p p e = p . Hence ↑ ( p F n ) = ↑ p \uparrow(pF_n) = \uparrow p ↑ ( p F n ) =↑ p and likewise ↑ ( F n p ) = ↑ p \uparrow(F_np) = \uparrow p ↑ ( F n p ) =↑ p in C C C . The principal-set identity of Lemma D.13, with i ( p ) = j ( ↑ p ) i(p) = j(\uparrow p) i ( p ) = j ( ↑ p ) , now gives p B = B p = p pB = Bp = p pB = B p = p .
All closed support terms keep their old values in D D D . In particular the old inverse equations make q q q and n n n local units in p D pD p D . The upper support theory and Lemma D.2 therefore apply in this ambient chain: n < p n < p n < p remains a cover in p D pD p D , and each old group point g ∈ G g \in G g ∈ G bounds its protected band ( g n , g ] D (gn,g]_{D} ( g n , g ] D . The subset
E = { x ∈ D : p x ∈ G } E = \{x \in D : px \in G\} E = { x ∈ D : p x ∈ G }
is a full original-signature subalgebra containing K K K , with p E = G pE = G pE = G . This uses only the ambient support identities, not density or standardness of D D D . Explicitly products stay in E E E because p p p is central and idempotent. For a support point g ∈ G g \in G g ∈ G , its prime is the old support point p ′ g − 1 ∈ G p'g^{-1} \in G p ′ g − 1 ∈ G ; for a nonsupport point over g g g , Lemma D.2 puts its prime over q g q − 1 ∈ G qgq^{-1} \in G q g q − 1 ∈ G . Thus E E E is prime closed. It contains e , d e,d e , d , so the prime–product residual formulas give full closure. Every old g ∈ G g \in G g ∈ G belongs to K K K , proving the exact support equality. By the preceding product equality, B ∈ E B \in E B ∈ E .
Use the actual residual in D D D to define
Since B ≤ e B \le e B ≤ e , the point e e e is eligible, so T ≥ e T \ge e T ≥ e . Since p B = p > e pB = p > e pB = p > e , the point p p p is ineligible; hence T < p T < p T < p . Monotonicity gives p T = p pT = p pT = p , so T ∈ E T \in E T ∈ E and it is nonsupport. Moreover T B ≤ e TB \le e T B ≤ e gives B ≤ T \ e = T − B \le T\backslash e = T^{-} B ≤ T \ e = T − . On the other hand T p = p > e Tp = p > e T p = p > e makes p p p ineligible in T \ e T\backslash e T \ e , so
n < B ≤ T − < p . n < B \le T^{-} < p. n < B ≤ T − < p .
There is no support point in ( n , p ) D (n,p)_{D} ( n , p ) D . Thus T − T^{-} T − is nonsupport and has neutral support degree. It follows that T T T belongs to the intrinsic neutral coefficient fibre of E E E .
Now let x x x be any intrinsic neutral coefficient in E E E . Its unit residual x − x^{-} x − is nonsupport and has neutral support degree, by Lemma D.9. Equivalently, n n n is eligible in x \ e x\backslash e x \ e , because x n = n xn = n x n = n , while p p p is ineligible because x p = p > e xp = p > e x p = p > e ; nonsupport excludes equality at n n n . Thus n < x − < p n < x^{-} < p n < x − < p . The successor property yields B ≤ x − B \le x^{-} B ≤ x − , and therefore x B ≤ e xB \le e x B ≤ e . Actual residuation in the definition of T T T gives x ≤ T x \le T x ≤ T . This proves that T T T is the greatest intrinsic neutral coefficient of E E E .
Finally let A A A be the full original-signature subalgebra of E E E generated by the countable old K K K together with B , T B,T B , T . It is countable, lies in K R K_R K R , contains every old support point, and has p A = G pA = G p A = G . Its unit residuals are inherited from E E E ; in particular T − T^{-} T − belongs to A A A and is nonsupport. The intrinsic neutral coefficient fibre of A A A therefore contains T T T , and every one of its elements is at most T T T by the preceding argument in E E E . This proves the assertion.
Restoring the entire upper support
Lemma D.16 (Protected top gluing). Let C ∈ K R C \in\mathcal{K}_R C ∈ K R be countable with p > e p > e p > e and invertible n n n in I = p C I = pC I = pC . Put G = U ( I ) G = U(I) G = U ( I ) and K = { x : p x ∈ G } K = \{x : px \in G\} K = { x : p x ∈ G } . Suppose K K K has a countable full extension K ^ ∈ K R \widehat{K} \in\mathcal{K}_R K ∈ K R with p K ^ = G p\widehat{K} = G p K = G and with every closed protected band [ g n , g ] [gn,g] [ g n , g ] densely ordered. Then C C C has a countable full dense extension in K R \mathcal{K}_R K R , and hence a full standard embedding.
Proof. The corner I I I has local central unit dualizer f = q n f = qn f = q n , is directly finite, and has the central unit predecessor n n n of p p p . The relative support extension of Corollary B.10 gives a countable full I ⊆ J I \subseteq J I ⊆ J with exactly the old unit group G G G , with q , n q,n q , n still central, and with the pairs g n < g gn < g g n < g as its only adjacent pairs. It is directly finite and has no endpoints. In a directly finite monoid the nonunits form a two-sided ideal: if a b ab ab is a unit with inverse c c c , then a ( b c ) = p a(bc) = p a ( b c ) = p makes a a a a unit, and then b = a − 1 ( a b ) b = a^{-1}(ab) b = a − 1 ( ab ) is a unit.
Set B = K ^ ∪ ˙ ( J ∖ G ) B = \widehat{K} \mathbin{\dot\cup} (J \setminus G) B = K ∪ ˙ ( J ∖ G ) , with projection Π : B → J \Pi: B \to J Π : B → J equal to x ↦ p x x \mapsto px x ↦ p x on K ^ \widehat{K} K and to the identity outside. Order first by Π \Pi Π , then within a fibre by its given order, identifying g ∈ G g \in G g ∈ G with the top of its fibre. Retain the product of K ^ \widehat{K} K and put
x y = Π ( x ) Π ( y ) if one factor is outside K ^ . xy = \Pi(x)\Pi(y) \qquad\text{if one factor is outside } \widehat{K}. x y = Π ( x ) Π ( y ) if one factor is outside K .
Such a mixed product is outside by the ideal property. For every top point a ∈ J a \in J a ∈ J and every x ∈ B x \in B x ∈ B , these rules give
a x = a Π ( x ) , x a = Π ( x ) a . ax = a\Pi(x), \qquad xa = \Pi(x)a. a x = a Π ( x ) , x a = Π ( x ) a .
For old unit tops this already holds in K ^ \widehat{K} K by centrality of p p p . Thus a triple with any outside factor has the same projected top product under either bracketing; triples in K ^ \widehat{K} K use its associativity. The old e e e is a unit. Multiplication by a top is monotone by the corresponding product in J J J . Any nontop multiplier has unit projection, whose translations strictly preserve inequalities between distinct bases; equal-base comparisons use monotonicity in K ^ \widehat{K} K . This proves both monotonicity laws.
Use prime from K ^ \widehat{K} K and the local prime from J J J elsewhere. The latter preserves nonunits since its dualizer f f f is a unit. Protected covers and Lemma D.2 make this an antitone involution. In particular, for y ∈ J ∖ G y \in J \setminus G y ∈ J ∖ G and x ∈ ( g n , g ] K ^ x \in(gn,g]_{\widehat{K}} x ∈ ( g n , g ] K ,
y ≤ x ′ ⟺ y ≤ g ′ . (77) y \leq x' \quad\Longleftrightarrow\quad y \leq g'. \tag*{(77)} y ≤ x ′ ⟺ y ≤ g ′ . (77)
A mixed product is a nonunit support point. The cover f < q f < q f < q contains d d d strictly, so such a point is at most d d d exactly when it is at most f f f . Therefore local cyclicity gives
x y ≤ d ⟺ Π ( x ) Π ( y ) ≤ f ⟺ Π ( y ) ≤ Π ( x ′ ) J ⟺ y ≤ x ′ . xy \le d \iff\Pi(x)\Pi(y) \le f \iff\Pi(y) \le\Pi(x')^{J} \iff y \le x'. x y ≤ d ⟺ Π ( x ) Π ( y ) ≤ f ⟺ Π ( y ) ≤ Π ( x ′ ) J ⟺ y ≤ x ′ .
The final equivalence uses the preceding protected-cover test if x ∈ K ^ x \in\widehat{K} x ∈ K ; otherwise x ′ x' x ′ is a singleton nonunit, whose comparisons are comparisons of bases. Two core inputs use their old cyclicity. Associativity and this cut give both actual residuals.
The embedding of C C C retains every product, because an old supported factor already satisfies a x = a ( p x ) ax = a(px) a x = a ( p x ) and x a = ( p x ) a xa = (px)a x a = ( p x ) a , and retains prime and constants. It is full. The old d d d is central on the core and acts on outside points through its central projection q q q . Every history guard and prime belongs to the central set in Lemma D.1; its elements remain central on B B B because their projections are powers of q q q and f f f . The central-guard argument following Lemma D.3 proves all histories. Thus B ∈ K R B \in K_R B ∈ K R and p B = J pB = J pB = J .
For density, equal-base strict pairs lie in a dense protected band. Different bases which do not form a protected cover have an intervening point of J J J . If the bases are g n < g gn < g g n < g , a lower-fibre point below its top is separated from the upper fibre by g n gn g n itself; pairs starting at g n gn g n are separated inside [ g n , g ] [gn, g] [ g n , g ] . These are all cases, since the protected covers are the only gaps in J J J . Its lack of endpoints also gives that property to B B B . The standard completion theorem now gives a full standard embedding.
Corollary D.17 (The upper-invertible branch). Every countable upper-invertible complete core has a countable full extension with exactly its old support group and with every protected closed band densely ordered. Every countable whole input in that branch has a countable full dense extension and a full standard embedding.
Proof. Theorem D.15 supplies a countable full core extension with the same support group and a greatest neutral coefficient. Theorem D.11 gives the required exact-group dense extension. Lemma D.16 restores the entire original upper support and fills its remaining gaps.
The idempotent upper branch
Throughout this subsection we write Un ( M ) = U ( M ) \operatorname{Un}(M) = U(M) Un ( M ) = U ( M ) for the group of units of a local monoid M M M , keeping the notation distinct from the coefficient chains denoted by U U U earlier in this section.
Lemma D.18 (Positive bands with separate group floors). Let A ∈ K R A \in\mathcal{K}_{R} A ∈ K R be countable, with p > e p > e p > e , n = ( p d ) ′ n = (pd)' n = ( p d ) ′ , n 2 = n n^{2} = n n 2 = n . Suppose H = Un ( p A ) H = \operatorname{Un}(pA) H = Un ( p A ) and
A = ⋃ a ∈ H [ n a , a ] , [ n a , a ] ∩ p A = { n a , a } . (78) A = \bigcup_{a \in H} [na,a], \qquad[na,a] \cap pA = \{na,a\}. \tag*{(78)} A = a ∈ H ⋃ [ na , a ] , [ na , a ] ∩ p A = { na , a } . (78) Then A A A has a full countable extension A + A^{+} A + in which every displayed closed band is densely ordered. Its p p p -corner is exactly the old p A = n H ∪ H pA = nH \cup H p A = n H ∪ H , and its n n n -floor group is exactly n H nH n H .
Proof. Put q = p d ∈ Z ( H ) q = pd \in Z(H) q = p d ∈ Z ( H ) and B a = ( n a , a ] B_a = (na,a] B a = ( na , a ] . The map a ↦ n a a \mapsto na a ↦ na is injective: if n a = n na = n na = n , multiplication by the local unit a a a sends the cover n < p n < p n < p in p A pA p A to the cover n < a n < a n < a . If a > p a > p a > p , the point p p p contradicts this cover; if a < p a < p a < p , then a > n a > n a > n (otherwise multiplying by a − 1 a^{-1} a − 1 gives p ≤ n p \le n p ≤ n ), and a a a contradicts the first cover. Thus a = p a = p a = p , proving the assertion by taking a quotient of two units. Consequently the closed bands are disjoint and ordered by the biordered group H H H .
For x ∈ B a x \in B_a x ∈ B a , y ∈ B b y \in B_b y ∈ B b , one has p ( x y ) = a b p(xy) = ab p ( x y ) = ab , so x y ∈ B a b xy \in B_{ab} x y ∈ B ab . Indeed a floor n a b nab nab is fixed by p p p and is different from the unit a b ab ab . A top acts by a y = a b = x b ay = ab = xb a y = ab = x b ; a floor acts by n a y = n a b = x n b na\,y = nab = x\,nb na y = nab = x nb . Prime maps the open band ( n a , a ) (na,a) ( na , a ) to ( n q a − 1 , q a − 1 ) (nqa^{-1},qa^{-1}) ( n q a − 1 , q a − 1 ) and exchanges a a a with n q a − 1 nqa^{-1} n q a − 1 .
Form the biordered group H ~ = H × Z \widetilde{H} = H \times\mathbb{Z} H = H × Z , with central predecessor ν = ( p , − 1 ) \nu= (p,-1) ν = ( p , − 1 ) of ( p , 0 ) (p,0) ( p , 0 ) . At degree ( a , k ) (a,k) ( a , k ) put a labelled copy of B a B_a B a . Write its top as t ( a , k ) t_{(a,k)} t ( a , k ) , and order these copies lexicographically. For positive old labels define
( a , k , x ) ( b , l , y ) = ( a b , k + l , x y ) . (79) (a,k,x)(b,l,y) = (ab,k+l,xy). \tag*{(79)} ( a , k , x ) ( b , l , y ) = ( ab , k + l , x y ) . (79)
The unit is ( p , 0 , e ) (p,0,e) ( p , 0 , e ) and the dualizer is ( q , 0 , d ) (q,0,d) ( q , 0 , d ) . Define prime on an interior point and on a top respectively by
( a , k , x ) ′ = ( q a − 1 , − k , x ′ ) ( x < a ) , t ( a , k ) ′ = t ( q a − 1 , − k − 1 ) . (80) (a,k,x)' = (qa^{-1},-k,x') \quad(x<a), \qquad t'_{(a,k)} = t_{(qa^{-1},-k-1)}. \tag*{(80)} ( a , k , x ) ′ = ( q a − 1 , − k , x ′ ) ( x < a ) , t ( a , k ) ′ = t ( q a − 1 , − k − 1 ) . (80)
Associativity, the unit and monotonicity follow from the old positive-band multiplication and the biorder. Prime is an antitone involution. For the cyclic cut, first compare a b ab ab with q q q ; at equality compare k + l k+l k + l with 0 0 0 . At equality of both, interior factors use precisely the old test x y ≤ d ⟺ y ≤ x ′ xy \le d \Longleftrightarrow y \le x' x y ≤ d ⟺ y ≤ x ′ . A top factor admits no positive point at that boundary degree, and the largest eligible point is the support top one integer level earlier, exactly as in (52). Thus prime is cyclic dualizing, and both residuals are the actual prime–product maxima.
The dualizer is central. The set consisting of e e e , d d d and t ( q j , k ) t_{(q_j,k)} t ( q j , k ) , for j , k ∈ Z j,k \in\mathbb{Z} j , k ∈ Z , is central and closed under product and prime. It contains all closed historical guards and their primes. Cyclic rotation at a guard with central prime gives x y ≤ c ⟺ y x ≤ c xy \le c \Longleftrightarrow yx \le c x y ≤ c ⟺ y x ≤ c ; Lemma 4.1 then proves every finite history. Hence the unfolded chain is in K R \mathcal{K}_{R} K R . Direct calculation gives d 2 = t ( q 2 , 0 ) d^2=t_{(q_2,0)} d 2 = t ( q 2 , 0 ) , d 3 \ d 2 = t ( q − 1 , 0 ) d^3\backslash d^2=t_{(q^{-1},0)} d 3 \ d 2 = t ( q − 1 , 0 ) , so its canonical support is t ( p , 0 ) t_{(p,0)} t ( p , 0 ) and its support group is exactly H ^ \widehat{H} H . Its ( p d ) ′ = t ( p , − 1 ) (pd)'=t_{(p,-1)} ( p d ) ′ = t ( p , − 1 ) is a local unit. The unfolded chain is therefore an actual upper-invertible complete group core.
Apply Corollary D.17 . In the resulting countable full extension E E E , the support is still exactly H ^ \widehat{H} H , and all protected bands are dense. Retain only the positive bands
P a = ( t ( a , − 1 ) , t ( a , 0 ) ] E ( a ∈ H ) . (81) P_a=(t_{(a,-1)},t_{(a,0)}]_{E}\qquad(a\in H). \tag*{(81)} P a = ( t ( a , − 1 ) , t ( a , 0 ) ] E ( a ∈ H ) . (81)
They are product closed with P a P b ⊆ P a b P_aP_b\subseteq P_{ab} P a P b ⊆ P ab , because their support degrees have integer coordinate 0. The prime of an interior point is interior in P q a − 1 P_{qa^{-1}} P q a − 1 . These assertions are the universal upper-support identities, and do not require that E E E be obtained by a prescribed equivariant construction.
Adjoin a separate floor s a s_a s a immediately below each open band P a P_a P a , retaining its top t a = t ( a , 0 ) t_a=t_{(a,0)} t a = t ( a , 0 ) . For x ∈ P b x\in P_b x ∈ P b put
s a x = s a s b = s a b , x s a = s b a , s a ′ = t q a − 1 , t a ′ = s q a − 1 . (82) s_ax=s_as_b=s_{ab},\qquad xs_a=s_{ba},\qquad s'_a=t_{qa^{-1}},\qquad t'_a=s_{qa^{-1}}. \tag*{(82)} s a x = s a s b = s ab , x s a = s ba , s a ′ = t q a − 1 , t a ′ = s q a − 1 . (82)
Keep products of positive points, their interior primes, and the old e e e , d d d from E E E . A floor factor makes either bracketing a floor of the same group degree; without a floor, associativity is that of E E E . Monotonicity follows from the biorder and the endpoint rules. At product degree q q q , a floor is below d d d , and the endpoint exchange in (54) gives exactly the cyclic cut. On positive pairs the old cut in E E E gives that test. At other degrees the biorder decides both sides. Thus this is a cyclic involutive residuated chain with actual residual maxima.
Centrality of d d d is retained, and the central product–prime closed set is now
{ e , d } ∪ { s q j , t q j : j ∈ Z } . (83) \{e,d\}\cup\{s_{q^j},t_{q^j}:j\in\mathbb{Z}\}. \tag*{(83)} { e , d } ∪ { s q j , t q j : j ∈ Z } . (83)
Centrality of the old q q q and its local inverse is retained in E E E ; all the tops in (55) are their old powers. The new floors at these central degrees are central by (54). The same guard argument proves every history. Mapping old positive points to their level-zero copies and n a na na to s a s_a s a retains product, prime, e e e , d d d , and hence the full signature. The old closed terms p p p , n n n are retained.
Multiplication by p = t p p=t_p p = t p fixes precisely the new floors and old tops, while multiplication by n = s p n=s_p n = s p projects each band to its floor. The corner and floor assertions follow. Finally every closed band [ s a , t a ] [s_a,t_a] [ s a , t a ] is dense, since (53) is a dense protected band with its lower endpoint renamed. All carriers used after applying the countable core theorem are countable.
Lemma D.19 (Deleting the new unit-degree fibres). Let D ∈ K R D \in\mathcal{K}_R D ∈ K R be countable, with central idempotent r < e D r < e_D r < e D , a full projection x ↦ r x x \mapsto rx x ↦ r x onto a group G G G , and dualizer d D = s δ ∈ G d_D=s_\delta\in G d D = s δ ∈ G , where s 1 = r s_1=r s 1 = r and δ ∈ Z ( G ) \delta\in Z(G) δ ∈ Z ( G ) . Write its fibres as D g = [ s g , t g ] D_g=[s_g,t_g] D g = [ s g , t g ] and suppose D 1 = { s 1 , t 1 } D_1=\{s_1,t_1\} D 1 = { s 1 , t 1 } , where t 1 = e D t_1=e_D t 1 = e D . Let H H H be the image in G G G of the global unit group of D D D . Suppose D D D has a full standard extension retaining a specified central global unit Q Q Q of degree δ \delta δ as a central global unit. Then D D D has a countable full extension D + D^+ D + with exactly the old floor group, unchanged fibres over H H H , dense fibres over G ∖ H G \setminus H G ∖ H , and unchanged global unit group. The element Q Q Q remains central.
Proof. Every global unit u u u of degree h h h induces an order isomorphism D 1 → D h D_1 \to D_h D 1 → D h by multiplication. Its two values are s h s_h s h and u u u ; thus D h = { s h , t h } D_h=\{s_h,t_h\} D h = { s h , t h } and u = t h u=t_h u = t h . In particular the degree map is injective on units and H H H is a subgroup. Since Q Q Q has degree δ \delta δ , we have δ ∈ H \delta\in H δ ∈ H .
For g ∉ H g \notin H g ∈ / H , the two products t g t g − 1 t_g t_{g^{-1}} t g t g − 1 and t g − 1 t g t_{g^{-1}}t_g t g − 1 t g belong to D 1 D_1 D 1 . Neither can equal t 1 t_1 t 1 . Indeed, if x y = t 1 xy=t_1 x y = t 1 and y x = s 1 yx=s_1 y x = s 1 , then floor absorption gives
t 1 = ( x y ) 2 = x ( y x ) y = x s 1 y = s 1 , t_1=(xy)^2=x(yx)y=xs_1y=s_1, t 1 = ( x y ) 2 = x ( y x ) y = x s 1 y = s 1 ,
a contradiction. Thus either equality to t 1 t_1 t 1 would give a two-sided inverse for t g t_g t g , contrary to g ∉ H g \notin H g ∈ / H . Hence
t g t g − 1 = s 1 = t g − 1 t g ( g ∉ H ) . (84) t_g t_{g^{-1}}=s_1=t_{g^{-1}}t_g \qquad(g \notin H). \tag*{(84)} t g t g − 1 = s 1 = t g − 1 t g ( g ∈ / H ) . (84)
If g , h ∉ H g,h \notin H g , h ∈ / H and k = g h ∈ H k=gh \in H k = g h ∈ H , the old global unit v = t k v=t_k v = t k carries the greatest point of one fibre to that of its translated fibre. Therefore t h v − 1 = t g − 1 t_hv^{-1}=t_{g^{-1}} t h v − 1 = t g − 1 , and
t g t h v − 1 = s 1 , t g t h = s k . (85) t_gt_hv^{-1}=s_1,\qquad t_gt_h=s_k. \tag*{(85)} t g t h v − 1 = s 1 , t g t h = s k . (85)
These are identities between old points, so they hold in every full extension. Identify D D D with its image in the specified standard extension S S S . We first check centrality of the floor in this target. Since r Q = s δ = d D rQ=s_\delta=d_D r Q = s δ = d D , the retained inverse equation gives r = d D Q − 1 r=d_DQ^{-1} r = d D Q − 1 . The target dualizer d D d_D d D is central because S ∈ K R S \in\mathcal{K}_R S ∈ K R , and Q Q Q remains a central global unit by hypothesis. Thus r r r is central in S S S ; its idempotence and inequality r < e D r < e_D r < e D are retained as well. The interior projection therefore applies in S S S . Pull back the old floor group:
E = { x ∈ S : r x ∈ G } . E = \{x \in S : rx \in G\}. E = { x ∈ S : r x ∈ G } .
The full interior projection makes this a full subalgebra. For completeness, the old group G G G is full in r S rS r S : multiplication by an old group element and by its old inverse are mutually inverse order isomorphisms, so its actual residuals are group division, and its local prime is g ↦ s δ g − 1 g \mapsto s_{\delta g^{-1}} g ↦ s δ g − 1 . Thus the inverse image is indeed closed under all original operations. Its fibre over g g g is the whole old-endpoint interval [ s g , t g ] S [s_g,t_g]_S [ s g , t g ] S .
Now retain only
E ∘ = D H ∪ ⋃ g ∈ G ∖ H [ s g , t g ] S , D H = ⋃ h ∈ H { s h , t h } . E^\circ= D_H \mathbin{\cup} \bigcup_{g\in G\setminus H} [s_g,t_g]_S,\qquad D_H = \bigcup_{h\in H}\{s_h,t_h\}. E ∘ = D H ∪ g ∈ G ∖ H ⋃ [ s g , t g ] S , D H = h ∈ H ⋃ { s h , t h } .
We check closure rather than assuming that deleting fibres commutes with completion. A product with a floor is the floor at its product degree. The old H H H -part is closed, and a product with exactly one factor of degree outside H H H again has degree outside H H H . For two outside factors x ∈ E g x \in E_g x ∈ E g , y ∈ E h y \in E_h y ∈ E h , the only case requiring a check is g h ∈ H gh \in H g h ∈ H . By monotonicity and (85) ,
s g h ≤ x y ≤ t g t h = s g h . s_{gh} \le xy \le t_g t_h = s_{gh}. s g h ≤ x y ≤ t g t h = s g h .
Thus this product is the retained old floor. Prime sends a fibre of degree g g g to the fibre of degree δ g − 1 \delta g^{-1} δ g − 1 ; since δ ∈ H \delta\in H δ ∈ H , it preserves the selected degrees, and it preserves D H D_H D H . The constants lie in D H D_H D H . Product and prime closure imply closure under both actual residuals by the involutive residual formulas. Consequently E ∘ E^\circ E ∘ is a full subalgebra of S S S , belongs to K R K_R K R , and retains every original finite history. Centrality of Q Q Q is inherited from S S S .
No new unit has been retained. If an outside point x x x were a unit, its inverse would have degree g − 1 g^{-1} g − 1 and be bounded by t g − 1 t_{g^{-1}} t g − 1 . Equation (84) would force x x − 1 ≤ s 1 < t 1 xx^{-1} \le s_1 < t_1 x x − 1 ≤ s 1 < t 1 , a contradiction. Over H H H only old points remain, so the old unit group is exact.
Each outside fibre is a closed interval of the dense standard order, so each of its strict pairs has an intermediate point in that same fibre. Starting with D D D , choose one such point for every strict outside-fibre pair in the current countable subalgebra, take full subalgebra closure inside E ∘ E^\circ E ∘ , and repeat countably many times. The union D + D^+ D + is countable and full. Every strict pair in any of its outside fibres appears at a finite stage and is split at the next. Its floor group and its entire H H H -part remain exact, and centrality of Q Q Q is retained. This proves all assertions.
Theorem D.20 (All upper-idempotent interior complete cores). Every countable C ∈ K R C \in K_R C ∈ K R with p > e p > e p > e and n = ( p d ) ′ n = (pd)' n = ( p d ) ′ satisfying n 2 = n n^2 = n n 2 = n has a full standard embedding. Its complete core K = { x : n x ∈ Un ( n C ) } K = \{x : nx \in\operatorname{Un}(nC)\} K = { x : n x ∈ Un ( n C )} has a full countable extension with exactly its old n n n -floor group and densely ordered internal floor fibres. All original operations, actual residuals and every finite history are retained.
Proof. Work first in the complete core K K K . Put G = n K G = nK G = n K , D = p K D = pK D = p K and f = n d = p ′ f = nd = p' f = n d = p ′ . The interior projection theorem gives floor fibres [ g , ( f g − 1 ) ′ ] [g, (fg^{-1})'] [ g , ( f g − 1 ) ′ ] . In D D D , the neutral floor fibre is exactly { n , p } \{n,p\} { n , p } , by the protected cover n < p n < p n < p . Its local unit is p p p and its dualizer is f = n q f = nq f = n q , with q = p d q = pd q = p d a central local unit. Its canonical floor is n n n : for its own constants, f 3 \ D f 2 = q − 1 f^3 \backslash Df^2 = q^{-1} f 3 \ D f 2 = q − 1 and f 2 ( q − 1 ) 2 = n f^2(q^{-1})^2 = n f 2 ( q − 1 ) 2 = n . The residual assertion follows from its floor degree f − 1 f^{-1} f − 1 : every point of that fibre is eligible, and its greatest point is the old unit q − 1 q^{-1} q − 1 ; higher degrees are ineligible. The old central set { q j , n q j : j ∈ Z } \{q^j, nq^j : j \in\mathbb{Z}\} { q j , n q j : j ∈ Z } is closed under the local product and prime and contains the local constants p , f p, f p , f . The central-guard argument and Lemma 4.1 verify every local history, so D ∈ K R D \in K_R D ∈ K R with these local constants. Thus D D D is an integral-neutral regular core of the type in the preceding lemma, with δ = f \delta= f δ = f as a degree of G G G and central unit q q q lifting that degree.
Let H ≤ G H \leq G H ≤ G be the image of Un ( D ) \operatorname{Un}(D) Un ( D ) under a ↦ n a a \mapsto na a ↦ na . The injectivity argument in Lemma D.18 applies. A unit a a a of D D D has floor fibre exactly [ n a , a ] [na,a] [ na , a ] , because ( f ( n a ) − 1 ) ′ = a (f(na)^{-1})' = a ( f ( na ) − 1 ) ′ = a . Its intersection with D D D consists only of its two endpoints, by the translated protected cover. Conversely every x ∈ K ∖ D x \in K \setminus D x ∈ K ∖ D has p x ∈ Un ( D ) px \in\operatorname{Un}(D) p x ∈ Un ( D ) and n x = n ( p x ) nx = n(px) n x = n ( p x ) , by Lemma D.2 . Hence the full inverse image
A = { x ∈ K : n x ∈ H } (86) A = \{x \in K : nx \in H\} \tag*{(86)} A = { x ∈ K : n x ∈ H } (86)
is precisely the union of these active closed bands, and every point outside A A A belongs to D D D . Fullness follows from the full floor projection and the subgroup H H H , which contains the local dualizer degree. Therefore K = A ∪ D K = A \cup D K = A ∪ D and A ∩ D = D H A \cap D = D_H A ∩ D = D H , with exactly two points in every H H H -fibre.
Apply Lemma D.18 to obtain A + A^+ A + with all its fibres dense and exactly the same corner D H D_H D H . Apply Theorem 5.14 and Corollary 5.15 , retaining the old central global unit q q q , to D D D . Lemma D.19 then gives D + D^+ D + , with unchanged D H D_H D H and dense fibres outside H H H . Glue them along this actual common subalgebra of the corner:
B = A + ∪ D H D + . (87) B = A^{+} \cup_{D_H} D^{+}. \tag*{(87)} B = A + ∪ D H D + . (87)
Order by G G G and by the retained orders within its fibres. Keep the product on A + A^{+} A + and on D + D^{+} D + . For a ∈ A + a \in A^{+} a ∈ A + and x ∈ D + ∖ D H x \in D^{+} \setminus D_H x ∈ D + ∖ D H use
a x = ( p a ) x , x a = x ( p a ) , (88) ax = (pa)x,\qquad xa = x(pa), \tag*{(88)} a x = ( p a ) x , x a = x ( p a ) , (88)
with the right sides evaluated in D + D^{+} D + . The map j ( a ) = p a j(a) = pa j ( a ) = p a is a multiplicative map from A + A^{+} A + onto D H D_H D H , with j ( e ) = p j(e) = p j ( e ) = p . For every k ∈ D H k \in D_H k ∈ D H , its endpoint multiplication in A + A^{+} A + gives a k = j ( a ) k ak = j(a)k ak = j ( a ) k and k a = k j ( a ) ka = kj(a) k a = k j ( a ) , and these are again points of D H D_H D H .
These facts verify associativity in every mixed case. With one outside factor use multiplicativity of j j j and associativity in D + D^{+} D + . With two outside factors, if their product returns to H H H , it is a point of the common D H D_H D H and the displayed endpoint identities apply; otherwise all mixed products are evaluated in D + D^{+} D + . With three outside factors, an intermediate return is fixed by j j j , so both bracketings remain those of D + D^{+} D + . The old unit is a unit, and monotonicity follows within a fibre from monotonicity of j j j and the two original products; different fibres are compared by the biorder.
Keep prime from A + A^{+} A + on H H H and from D + D^{+} D + elsewhere. It is an antitone involution, because it maps degree g g g to δ g − 1 \delta g^{-1} δ g − 1 and δ ∈ H \delta\in H δ ∈ H . If the total degree is δ \delta δ , either both factors lie on H H H or both lie outside it. The former case uses duality in A + A^{+} A + . In the latter the product is in D δ = { f , q } D_{\delta} = \{f,q\} D δ = { f , q } , and
k ≤ d ⟺ k ≤ f ( k ∈ D δ ) , (89) k \le d \Longleftrightarrow k \le f \qquad(k \in D_{\delta}), \tag*{(89)} k ≤ d ⟺ k ≤ f ( k ∈ D δ ) , (89)
because f < d < q f < d < q f < d < q . Duality in D + D^{+} D + at its actual dualizer f f f therefore gives exactly x y ≤ d ⟺ y ≤ x xy \le d \Longleftrightarrow y \le x x y ≤ d ⟺ y ≤ x . At other total degrees the group order gives this equivalence. Both residuals are consequently the actual prime–product maxima.
The old d d d is central: it is central on A + A^{+} A + , and on the other fibres its action is that of the retained central corner unit q q q . The product–prime closed set
{ e , d } ∪ { q j , n q j : j ∈ Z } (90) \{e,d\} \cup\{q^{j},nq^{j}: j \in\mathbb{Z}\} \tag*{(90)} { e , d } ∪ { q j , n q j : j ∈ Z } (90)
is central in B B B . On A + A^{+} A + this is (55); on outside fibres use centrality of q q q in D + D^{+} D + and the central floor degrees. It contains every old closed guard and its prime. Cyclic rotation at each guard and Lemma 4.1 prove every finite history, so B ∈ K R B \in\mathcal{K}_R B ∈ K R with no additional axiom. The original inclusion retains products on each old piece. Mixed products agree with (88) , since p p p was already central and every old point outside A A A was p p p -fixed. For old outside points x , y ∈ K ∖ A x,y\in K\setminus A x , y ∈ K ∖ A , the product x y xy x y lies in D = p K D=pK D = p K ; if its n n n -degree n ( x y ) n(xy) n ( x y ) belongs to H H H , then (86) also gives x y ∈ A xy\in A x y ∈ A , and hence
x y ∈ D ∩ A = p K ∩ A = D H . xy\in D\cap A=pK\cap A=D_H. x y ∈ D ∩ A = p K ∩ A = D H .
Thus the two descriptions agree even when an outside product returns to an H H H -fibre. Prime is retained on each old piece, as are e , d e,d e , d , so the inclusion preserves the entire original signature.
Let ι : K ↪ B \iota:K\hookrightarrow B ι : K ↪ B be the full embedding just constructed, and write
n K = ( p K d K ) ′ , n B = ( p B d B ) ′ . n_K=(p_Kd_K)^\prime,\qquad n_B=(p_Bd_B)^\prime. n K = ( p K d K ) ′ , n B = ( p B d B ) ′ .
Fullness preserves these closed terms, so n B = ι ( n K ) n_B=\iota(n_K) n B = ι ( n K ) and n B ι ( x ) = ι ( n K x ) n_B\iota(x)=\iota(n_Kx) n B ι ( x ) = ι ( n K x ) for every x ∈ K x\in K x ∈ K . The assertion that the floor group is unchanged is the set equality
{ n B z : z ∈ B } = ι [ { n K x : x ∈ K } ] . (91) \{n_Bz:z\in B\}=\iota[\{n_Kx:x\in K\}]. \tag*{(91)} { n B z : z ∈ B } = ι [{ n K x : x ∈ K }] . (91)
Indeed, the floor of every point of A + A^+ A + is an old floor over H H H by Lemma D.18 , while the floor of every point of D + D^+ D + is an old floor in G G G by Lemma D.19 . These old floor points are identified consistently in the gluing. Since B = A + ∪ D H D + B=A^+\cup_{D_H}D^+ B = A + ∪ D H D + , this proves the inclusion from left to right in (91) . Conversely, each point ι ( n K x ) \iota(n_Kx) ι ( n K x ) is n B ι ( x ) n_B\iota(x) n B ι ( x ) , which proves the reverse inclusion. Here n K D = n K K = G n_KD=n_KK=G n K D = n K K = G , since D = p K K D=p_KK D = p K K and n K p K = n K n_Kp_K=n_K n K p K = n K .
Thus this intermediate extension B B B is countable, has exactly the old n n n -floor group, and has dense fibres with their old endpoints. The equality (91) concerns B B B before the subsequent support densification. The central-unit support densification of Theorem 4.8 , followed by the interior gluing of Theorem 4.7 at r = n r=n r = n , fills all cross-fibre gaps while retaining these dense fibres. This gives a countable full dense extension of the complete core, and its standard completion is a full standard embedding. The whole-input reconstruction of Corollary 4.11 now restores C C C . Its applicability at r = n r=n r = n follows from Lemma D.3 , which also holds in the standard target because the closed terms and the retained equations p > e p>e p > e , n 2 = n n^2=n n 2 = n are unchanged. Thus the whole input has both a countable full dense extension and a full standard embedding. ∴
Scope of the Lean verification
The companion Lean 4 development v1.1.0 verifies the full standard representation theorem, the relative comparison construction, countable-chain completeness relative to arbitrary theories, and strong standard completeness for H R H_R H R . It also verifies effective Den elimination for finite proofs without additional assumption leaves. The definitions preserve the noncommutative signature and all finite histories. The standard carrier is [ 0 , 1 ] [0,1] [ 0 , 1 ] with its usual order, and the algebraic unit is not identified by definition with the upper endpoint. The transitive axiom audit permits only propext, Classical.choice, and Quot.sound.
For IUL and H p s U L ∗ \mathrm{HpsUL}^{*} HpsUL ∗ , the bounded embeddings and semantic transfers are verified, but ordinary chain completeness for arbitrary theories remains an explicit external parameter; it is not independently derived from the syntax of these calculi. The pointed pseudo-uninorm transfer includes a genuine additional constant f f f , interpreted at the image of its original value independently of the auxiliary involutive dualizer.
The graded construction verifies the concrete closed-support core, including its greatest neutral coefficient, used by Theorem 7.1 . An adapter from arbitrary support-band data in the general formulation of Theorem D.11 to this concrete interface is not included. The idempotent-branch equality (63) is verified for the intermediate gluing, before subsequent support densification.
The exported Den elimination functions are executable. They process ordinary rule nodes recursively and invoke verified enumeration at Den nodes. Decidable certificate checks, together with the proved existence of a successful code, establish termination on derivable targets. Ordinary Cut is retained; no complexity bound or decision procedure for underivability is claimed. Runtime tests of the exported functions and the transitive axiom audit are recorded separately from static source checks.
Development workflow and availability. OpenAI ChatGPT and API-based assistance were used to draft and revise Lean code and verification scripts. Candidate changes were checked with the pinned Lean toolchain and project verification scripts; compiler feedback was used in subsequent revision rounds. AI-generated suggestions were not treated as proof certificates. Kernel checking, transitive axiom auditing and correspondence with the paper’s mathematical statements are distinct verification tasks. The versioned source archive and reproduction instructions are available on Zenodo [13 ] .
Declaration of generative AI and AI-assisted technologies in the manuscript preparation process
During the preparation of this work, the author used OpenAI ChatGPT to assist with manuscript revision, organization, language editing and LaTeX preparation. AI assistance in the formalization workflow is described in Appendix E. The author takes responsibility for the final mathematical statements, proofs, references and correspondence between the manuscript and the formal development.