Introduction
Logical results can often be considered at various levels, depending on whether one has a model-theoretic or proof-theoretic perspective. The basic focus of logic, the study of logical consequence, can indeed be considered from the point of view (i) of models – considering validity statement – or (ii) of provability – existence of a proof in a given deductive system – possibly (iii) of more specific provability statements – existence of cut-free or focused proofs typically – but also at the level (iv) of proofs themselves or even (v) of the proofs’ dynamics. Such perspectives are typical in type theory but also prove to be common in proof theory. In particular, when focusing on the deductive side, those distinctions partly amount to whether one consider a proof-relevant or proof-irrelevant versions of the results under consideration.
Interpolation. Craig’s Interpolation theorem [8 , 9 ] – a central result of first-order logic which found many applications both in mathematical logic [17 ] and in computer science [20 , 28 , 16 ] – is typically such a result that can be considered at various levels, with semantical or deductive versions of the statement of interpolation (often connected when one has a completeness theorem) and even a cut-free provability version that is the heart of Maehara’s celebrated method. Indeed, Maehara’s method [26 ] happens to turn Craig’s interpolation as a direct consequence of cut-admissibility, and is thus applicable to a wide range of contexts [29 , 35 , 23 , 19 , 24 , 2 , 36 , 17 , 39 ] which may explain why this method is popular among proof-theorists.
From Maehara to Proof-relevant interpolation. The idea behind Maehara’s method is to consider a cut-free proof of the entailment to interpolate and to show that each sequent of the derivation can be interpolated, generating an interpolant for the conclusion sequent. Maehara’s method was recently revisited by Saurin [34 ] who gave a series of proof-relevant interpolation theorems based on cut-introduction, by identifying the structure of proof-relevant interpolation situation, PRIS. (See § 2.2 in the background.) Those results prove, for several logics, that if π \pi π is a cut-free sequent proof of A ⊢ B A \vdash B A ⊢ B , one can find an interpolant formula I I I in the common vocabulary of A A A and B B B and interpolating proofs π 1 , π 2 \pi_1,\pi_2 π 1 , π 2 of A ⊢ I A \vdash I A ⊢ I and I ⊢ B I \vdash B I ⊢ B respectively such that π 1 \pi_1 π 1 composed with π 2 \pi_2 π 2 cut-reduces to π \pi π . Saurin’s result can precisely be viewed as the proof-relevant version of Maehara’s method and it generalizes previous results by Čubrić [10 , 11 ] . In the following we shall indistinctly speak of proof-relevant or Maehara-Čubrić interpolation for this style of interpolation results. Similarly to Saurin’s approach, Čubrić’ result is a proof-relevant version of Prawitz’ proof of the interpolation theorem that adapted Maehara’s method for natural deduction [32 ] . In this way, Čubrić’ result provides an interpolation result for simply typed λ \lambda λ -terms in normal form1 ^{1} 1 .
We stress this fundamental limitation of Čubrić’ approach to λ \lambda λ -terms in normal forms. The assumptions of Maehara. Indeed, despite being general and flexible, Maehara’s original method strongly relies on several structural assumptions:
the inductive character of sequent proofs, to reason by induction on the subderivations.
cut-admissibility, to reason on cut-free proofs and benefit from the subformula property,
The restriction to cut-free proofs is unsatisfactory from both a logical and a computational point of view, indeed (i) the logic under study may not satisfy a full cut-elimination theorem (e.g. cut-admissibility up to analytic cuts), cut-elimination induces a size blow-up of proofs, resulting in much larger interpolants. (ii) from the computational perspective of Čubrić and Saurin’s works, this restriction is a severe limitation: for instance, Čubrić result applies to normal forms only and therefore requires to first normalise λ \lambda λ -terms before interpolating. Why interpolating circular proofs with cuts. In the present paper we are interested in strengthening interpolation results for fixed-point logics by proving their proof-relevant counterpart. In order to do that, we will show how one can weaken the two assumptions mentioned in the previous paragraph in order to do proof-relevant interpolation in presence of cuts as well as for circular proofs. Additionally, proof-relevant interpolation over circular proofs must also preserve the validity condition specified by the proof system, which ensures the soundness of circular proofs. This is not automatic, and addressing it is one of the main technical points of this paper. The focus on circular proofs adds a further motivation for interpolating proofs with cuts. Indeed, while μ L L ∞ \mu\mathrm{LL}^{\infty} μ LL ∞ has a cut-elimination theorem, regularity of a proof is not preserved by cut-elimination in general2 ^{2} 2 .
Related works. To the best of our knowledge, we present the first results on (i) proof-relevant interpolation (ii) for circular proofs (iii) with cuts altogether. On the other hand, each of these three topics has been investigated independently in various prior works:
(i) Proof-relevant interpolation: the most closely related work in this direction is of course Saurin’s proof-relevant interpolation theorems for a variety of sequent calculi (and term-calculi for sequent calculus) [34 ] that we extend here. In the early 90s, Čubrić exploited Prawitz’ interpolation theorem in natural deduction [32 ] to prove an interpolation result for the simply typed λ \lambda λ -calculus as well as a corresponding factorisation result for bicartesian closed categories [10 , 11 ] later extended by Matthes [27 ] to a natural deduction with general elimination rules. Recently, Veltri and Wan proved a proof-relevant interpolation theorem for the non-associative Lambek-calculus, formalized in Agda [41 ] . (ii) Interpolation for fixed-point logics and circular proofs: After the seminar results by Agostino et al [14 , 15 ] on interpolation for the modal μ \mu μ -calculus, several works contributed to interpolation for the modal μ \mu μ -calculus using circular, or cyclic, proof systems both for Craig and Lyndon’s interpolation [2 , 22 ] and Uniform interpolation [3 ] . The techniques extending Maehara’s method to circular proofs are due to Shamkanov in the setting of Gödel-Löb provability logic [36 ] . Compared to our work, the absence of proof-relevance allow quite drastic modifications on proofs, for instance regularisation.
(iii) Interpolation for proofs with cuts: While several works aimed at extending interpolation theorems to incorporate some form of cuts [6 , 21 , 23 , 31 , 40 , 25 , 7 ] , typically to prove interpolation in presence of analytic cuts and semi-analytic rules [38 ] , most did not directly investigate how the cut inference itself impacts directly the interpolation process itself (that is the synthesis of the interpolant formula). Among the works on interpolation theorems in the presence of cuts, we should refer to Hetzl and Jalali’s work on completeness of interpolation algorithms [21 ] , van der Giessen’s lecture notes on interpolation which prove interpolation in the presence of analytic cuts [40 ] , a short abstract by Lolic that reports some results interpolation with cuts [25 ] and of course Fitting’s book that investigate interpolation for modal and intuitionistic logic and cuts [18 ] .
Contributions and organisation of the paper. Section 2 provides some background on μ \mu μ LL circular proofs and on proof-relevant interpolation. Then, we extend LL proof-relevant interpolation to proofs with cuts in Section 3 thus providing the first proof-relevant treatment of proofs with cuts and generalising several Craig’s interpolation methods with cuts; we also transfer this result to a computational system L generalising Saurin’s result on normal system L terms. Last, we consider in Section 4 the problem of proof-relevant interpolation for circular proofs with cuts in propositional μ \mu μ LL∞ ^\infty ∞ . We restrict to a better behaved (but less general) class of circular proofs called strongly valid , which we show to be closed under our interpolation procedure. This result provides a first step towards an interpolation theorem for all valid proofs in first order μ \mu μ LL∞ ^\infty ∞ . We conclude in Section 5.
Background
Circular proofs for linear logic with least and greatest fixed-points
For simplifying the background presentation, we first introduce a common framework, μ \mu μ LLP ∞ ^\infty_{\mathrm{P}} P ∞ , in which our background notions are set. The following sections will focus on two fragments of this logic: the fixed-point-free fragment, LL, and the FO quantifier free, μ \mu μ LLP ∞ ^\infty_{\mathrm{P}} P ∞ respectively.
We assume a first-order language L \mathcal{L} L without equality nor function symbols for simplicity. The language of first-order linear logic formulas with least and greatest fixed-points (or μ \mu μ LL formulas) and its deduction system, are defined in the following way:
Definition 1 (μ \mu μ LL formulas). We assume to be given countable sets of (i) atomic formulas equipped with an involution inducing a partitioning between positive atoms, written a ( x ⃗ ) a(\vec{x}) a ( x ) (or just a a a ), and negative atoms, written a ⊥ ( x ⃗ ) a^\bot(\vec{x}) a ⊥ ( x ) (or just a ⊥ a^\bot a ⊥ ) and of (ii) fixed-point variables, written X , Y , Z , … X,Y,Z,\ldots X , Y , Z , … such that X ⊥ = X X^\bot=X X ⊥ = X . μ \mu μ LL formulas are defined inductively as:
F : : = a ⊥ ( x ⃗ ) ∣ ⊥ ∣ ⊤ ∣ F & F ∣ F & F ∣ ? F ∣ ∀ x F ∣ a ( x ⃗ ) ∣ 1 ∣ 0 ∣ F ⊗ F ∣ F ⊕ F ∣ ! F ∣ ∃ x F ∣ X ∣ μ X . F ∣ ν X . F (1) \begin{aligned}
F &::= a^\bot(\vec{x}) \mid\bot\mid\top\mid F \mathbin{\&} F \mid F \mathbin{\&} F \mid?F \mid\forall xF \\
&\quad\mid a(\vec{x}) \mid1 \mid0 \mid F \mathbin{\otimes} F \mid F \mathbin{\oplus} F \mid!F \mid\exists xF \\
&\quad\mid X \mid\mu X.F \mid\nu X.F \tag*{(1)}
\end{aligned} F ::= a ⊥ ( x ) ∣ ⊥ ∣ ⊤ ∣ F & F ∣ F & F ∣ ? F ∣ ∀ x F ∣ a ( x ) ∣ 1 ∣ 0 ∣ F ⊗ F ∣ F ⊕ F ∣ ! F ∣ ∃ x F ∣ X ∣ μ X . F ∣ ν X . F (1)
A LL formula is a formula build using only the first two lines of the above grammar. Linear negation is defined as extending the LL involution on LL formulas (changing a ⊥ a^{\perp} a ⊥ , ⊥ \bot ⊥ , ⊤ \top ⊤ , & \mathbin{\&} & , ⊥ \mathbin{\bot} ⊥ , ? ? ? into a a a , 1 1 1 , 0 0 0 , ⊗ \otimes ⊗ , ⊕ \oplus ⊕ , ∃ \exists ∃ , ! ! ! respectively) to fixed-points constructs: ( μ X . F ) ⊥ = ν X . F ⊥ (\mu X.F)^{\perp} = \nu X.F^{\perp} ( μ X . F ) ⊥ = ν X . F ⊥ (note that μ \mu μ , ν \nu ν bind fixed-point variables). A one-sided µLL sequent is an ordered list of µLL formulas usually written ⊢ Γ \vdash\Gamma ⊢ Γ .
Definition 2 (µLL∞ and LL sequent calculus). The inference rules of µLL∞ (resp. LL) sequent calculus are given by the rows a, b and d in Figure 1 (resp. rows a, b and c in Figure 1 ). The colored formulas in each inference rule define the ancestor (or subocurrence ) relation between the conclusion and the premises of the rules.
Figure 1. µLL∞ inferences: (a) Propositional MALL inferences (no rule for 0); (b) LL exponential inferences; (c) First-order inferences (in (∀), x ∉ FV ( Γ ) x \notin\operatorname{FV}(\Gamma) x ∈ / FV ( Γ ) ); (d) Fixed-point inferences.
Remark 3. For µLL∞ proofs, we restrict ourselves to the propositional fragment µLL∞, since the cut-elimination procedure for circular proofs in [33 ] , that is used in the Theorem 37 , considers only propositional formulas. For a complete treatment of quantifiers, one should use parametrized fixed-point variables and respective unfolding rules, as in [1 ] . We leave this for future work.
We represent the tree structure of µLL∞ derivations using derivation graphs , after [5 ] .
Definition 4 (µLL∞ derivation graph). A µLL∞ derivation graph is a tuple D = ( V , s , r , p ) \mathcal{D}=(V,s,r,p) D = ( V , s , r , p ) where V V V is a set of vertices, s : V → S e q s s:V\to\mathit{Seqs} s : V → Seqs is a total function labelling vertices with sequents in µLL, r : V → R u l e s r:V\to\mathit{Rules} r : V → Rules is a partial function labelling vertices with inference rules in µLL∞, p : N × V → V p:\mathbb{N}\times V\to V p : N × V → V is a partial function that maps a number n n n and a vertex v v v , with a vertex v ′ v' v ′ , such that v ′ v' v ′ is labeled by the n t h n^{\mathrm{th}} n th premise of the rule r ( v ) r(v) r ( v ) over the sequent s ( v ) s(v) s ( v ) , and for every v ∈ V v\in V v ∈ V , ⊢ s ( p 1 ( v ) ) … ⊢ s ( p m ( v ) ) ⊢ s ( v ) ( r ( v ) ) \frac{\vdash s(p_1(v))\ \ldots\ \vdash s(p_m(v))}{\vdash s(v)}(r(v)) ⊢ s ( v ) ⊢ s ( p 1 ( v )) … ⊢ s ( p m ( v )) ( r ( v )) is an instance of the rule r ( v ) r(v) r ( v ) in µLL∞, with m m m premises. A vertex B ∈ V B\in V B ∈ V such that r ( B ) r(B) r ( B ) is undefined is called a bud node . The set of bud nodes of D \mathcal{D} D , is written as Bud ( D ) \operatorname{Bud}(\mathcal{D}) Bud ( D ) . If V V V is finite, we say that D \mathcal{D} D is a finite derivation graph.
A µLL∞ derivation graph D \mathcal{D} D can be seen as a directed graph ( V , E ) (V,E) ( V , E ) where the set of edges is defined as E = { ( v , p j ( v ) ) ∣ v ∈ V and p j ( v ) is defined } E=\{(v,p_j(v))\mid v\in V\text{ and }p_j(v)\text{ is defined}\} E = {( v , p j ( v )) ∣ v ∈ V and p j ( v ) is defined } . Moreover, D \mathcal{D} D is said to be a derivation tree , if its directed graph is a tree. We name the root of the tree as root ( D ) ∈ V \operatorname{root}(\mathcal{D})\in V root ( D ) ∈ V and s ( root ( D ) ) s(\operatorname{root}(\mathcal{D})) s ( root ( D )) the conclusion of the tree. Note that the ancestor relation can be extended by transitivity to formulas of non-consecutive sequents of a derivation graph.
Definition 5 (LL proof). An LL open proof π \pi π of a sequent ⊢ Γ \vdash\Gamma ⊢ Γ (written as π ⊢ Γ \pi\vdash\Gamma π ⊢ Γ ) is a finite derivation tree with conclusion ⊢ Γ \vdash\Gamma ⊢ Γ such that all its formulas (resp. inferences) are LL formulas (resp. inferences). An LL proof π \pi π is an open proof such that Bud ( π ) = ∅ \operatorname{Bud}(\pi)=\varnothing Bud ( π ) = ∅ .
Due to the recursive nature of the fixed-point rules, it is possible to construct infinite derivation trees even in the absence of cuts and contractions. Such infinite derivation trees are referred to as μ L L P ∞ \mu\mathrm{LL}^{\infty}_{P} μ LL P ∞ pre-proofs. In some cases, these infinite trees contain only finitely many distinct subtrees. When this occurs, we call the derivation trees regular. Such regular infinite trees can be finitely represented as trees with back-edges that form cycles, yielding what are known as μ L L P ω \mu\mathrm{LL}^{\omega}_{P} μ LL P ω pre-proofs.
Definition 6 (μ L L ω \mu\mathrm{LL}^{\omega} μ LL ω pre-proof graph). A μ L L ω \mu\mathrm{LL}^{\omega} μ LL ω pre-proof graph (or simply pre-proof) of the sequent ⊢ Γ \vdash\Gamma ⊢ Γ , is a tuple P = ( D , R ) \mathcal{P}=(\mathcal{D},\mathcal{R}) P = ( D , R ) , where D = ( V , s , r , p ) \mathcal{D}=(V,s,r,p) D = ( V , s , r , p ) is a μ L L ∞ \mu\mathrm{LL}^{\infty} μ LL ∞ finite derivation tree and R : Bud ( D ) → V \mathcal{R}:\operatorname{Bud}(\mathcal{D})\to V R : Bud ( D ) → V is a total function, that maps each one of the bud nodes B ∈ Bud ( D ) B\in\operatorname{Bud}(\mathcal{D}) B ∈ Bud ( D ) to an internal node on the path from root ( D ) \operatorname{root}(\mathcal{D}) root ( D ) to B B B , called a companion, satisfying that if R ( B ) = C \mathcal{R}(B)=C R ( B ) = C , then r ( C ) r(C) r ( C ) is defined and s ( B ) = s ( C ) s(B)=s(C) s ( B ) = s ( C ) . We say that P \mathcal{P} P is an open pre-proof if R \mathcal{R} R is not defined for all B ∈ Bud ( D ) B\in\operatorname{Bud}(\mathcal{D}) B ∈ Bud ( D ) .
The graph of a pre-proof P = ( D , R ) \mathcal{P}=(\mathcal{D},\mathcal{R}) P = ( D , R ) , written as G P \mathcal{G}_{\mathcal{P}} G P , is the graph obtained after adding the edges induced by R \mathcal{R} R and deleting the bud vertices. Specifically, G P = ( V ′ , s , r , p ′ ) \mathcal{G}_{\mathcal{P}}=(V',s,r,p') G P = ( V ′ , s , r , p ′ ) where, V ′ = V ∖ Bud ( D ) V'=V\setminus\operatorname{Bud}(\mathcal{D}) V ′ = V ∖ Bud ( D ) and p ′ p' p ′ is defined for any j j j as: p j ′ ( v ) = R ( p j ( v ) ) p'_j(v)=\mathcal{R}(p_j(v)) p j ′ ( v ) = R ( p j ( v )) if p j ( v ) ∈ Bud ( D ) p_j(v)\in\operatorname{Bud}(\mathcal{D}) p j ( v ) ∈ Bud ( D ) , or p j ′ ( v ) = p j ( v ) p'_j(v)=p_j(v) p j ′ ( v ) = p j ( v ) otherwise. Soundness of μ L L ω \mu\mathrm{LL}^{\omega} μ LL ω pre-proofs is ensured by imposing a correctness criterion known as the progress or validity or global trace condition :
Definition 7 (Traces and validity). Let P \mathcal{P} P be a μ L L ω \mu\mathrm{LL}^{\omega} μ LL ω pre-proof. A trace over a (possibly infinite) path ( v i ) i ∈ ω (v_i)_{i\in\omega} ( v i ) i ∈ ω in the graph G P \mathcal{G}_{\mathcal{P}} G P , is defined as a sequence of formulas ( F i ) i ∈ ω (F_i)_{i\in\omega} ( F i ) i ∈ ω , such that for each index i i i , we have F i ∈ s ( v i ) F_i\in s(v_i) F i ∈ s ( v i ) and either F i → F i + 1 F_i\to F_{i+1} F i → F i + 1 or F i = F i + 1 F_i=F_{i+1} F i = F i + 1 , where the relation → \to → denotes the Fischer-Ladner subformula relation , defined as follows:
( F ⋆ G ) → F ( F ⋆ G ) → G ( Δ F ) → F ( F ⊥ ) → F ( σ X . F ) → F [ σ X . F / X ] \begin{aligned}
(F\mathbin{\star}G)&\to F & (F\mathbin{\star}G)&\to G & (\Delta F)&\to F & (F^{\perp})&\to F & (\sigma X.F)&\to F[\sigma X.F/X]
\end{aligned} ( F ⋆ G ) → F ( F ⋆ G ) → G ( Δ F ) → F ( F ⊥ ) → F ( σ X . F ) → F [ σ X . F / X ]
With ⋆ ∈ { ⊗ , & , ⊗ , ⊕ } \star\in\{\mathbin{\otimes},\mathbin{\&},\mathbin{\otimes},\mathbin{\oplus}\} ⋆ ∈ { ⊗ , & , ⊗ , ⊕ } , Δ ∈ { ∀ x , ∃ x , ? , ! } \Delta\in\{\forall x,\exists x,?,!\} Δ ∈ { ∀ x , ∃ x , ? , !} , and σ ∈ { μ , ν } \sigma\in\{\mu,\nu\} σ ∈ { μ , ν } . Let F = ν X . F ′ F=\nu X.F' F = ν X . F ′ . A F F F -progress point of a trace, corresponds to the vertex v j v_j v j in ( v i ) i ∈ ω (v_i)_{i\in\omega} ( v i ) i ∈ ω such that F j = F F_j=F F j = F , F j + 1 = F ′ [ ν X . F ′ / X ] F_{j+1}=F'[\nu X.F'/X] F j + 1 = F ′ [ ν X . F ′ / X ] and r ( v j ) = ν r(v_j)=\nu r ( v j ) = ν . We say that an infinite path in G P \mathcal{G}_{\mathcal{P}} G P is valid if it has a trace τ = ( F i ) i ∈ ω \tau=(F_i)_{i\in\omega} τ = ( F i ) i ∈ ω such that the minimal formula in the set of infinitely recurring formulas3 , denoted min ( Inf ( τ ) ) \min(\operatorname{Inf}(\tau)) min ( Inf ( τ )) , is a ν \nu ν -formula, i.e. of the form F ′ = ν X . F ′ F'=\nu X.F' F ′ = ν X . F ′ , where F F F is minimal with respect to the usual subformula ordering, and τ \tau τ infinitely often visits a F F F -progress point. P \mathcal{P} P is said to be a valid proof or just a proof if and only if every infinite path in G P \mathcal{G}_{\mathcal{P}} G P is valid.
In addition, we consider a stronger notion of validity, known as strong validity .
Definition 8 (Strong validity). Let P \mathcal{P} P be a μ L L ω \mu\mathrm{LL}^{\omega} μ LL ω pre-proof. An infinite path ( v i ) (v_i) ( v i ) in G P \mathcal{G}_{\mathcal{P}} G P is said to be strongly valid if there exists a valid trace τ = ( F i ) \tau=(F_i) τ = ( F i ) over ( v i ) (v_i) ( v i ) , and there exists an index k k k such that for all h , i ≥ k h,i\geq k h , i ≥ k , whenever v h = v i v_h=v_i v h = v i , it follows that F h = F i F_h=F_i F h = F i (as formula occurrences of the considered sequent). A pre-proof P \mathcal{P} P is said to be a strongly valid proof if and only if every infinite path in G P \mathcal{G}_{\mathcal{P}} G P is strongly valid.
This ensures that τ \tau τ exhibits some stability or convergence along the path, and can be represented as a function τ ( v ) \tau(v) τ ( v ) mapping vertices v v v of the pre-proof to occurrences in s ( v ) s(v) s ( v ) .
Proof-relevant interpolation for cut-free proofs in linear logic
We now recall Saurin’s proof-relevant interpolation result for cut-free linear logic [34 ] . To state the result, we first need the definition of vocabulary of a formula:
Definition 9 (Vocabulary of a formula). Let F F F be a LL formula. Voc ( F ) \operatorname{Voc}(F) Voc ( F ) is defined as the set of all atoms occurring in F F F , together with their linear negations, while Voc + ( F ) \operatorname{Voc}^{+}(F) Voc + ( F ) (resp. Voc − ( F ) \operatorname{Voc}^{-}(F) Voc − ( F ) ) is the set of positive (resp. negative) atoms occurring in F F F . This notion is naturally lifted to list of formulas Γ \Gamma Γ , written as Voc ( Γ ) \operatorname{Voc}(\Gamma) Voc ( Γ ) .
The interpolation theorem for LL is stated as:
Theorem 10 (Saurin [34 ] , Theorem 5). Let Γ , Δ \Gamma,\Delta Γ , Δ be lists of LL formulas and π ⊢ Γ , Δ \pi\vdash\Gamma,\Delta π ⊢ Γ , Δ a cut-free proof. There exists a LL formula I I I such that Voc + ( I ) ⊆ Voc − ( Γ ) ∩ Voc + ( Δ ) \operatorname{Voc}^{+}(I) \subseteq\operatorname{Voc}^{-}(\Gamma) \cap\operatorname{Voc}^{+}(\Delta) Voc + ( I ) ⊆ Voc − ( Γ ) ∩ Voc + ( Δ ) , Voc − ( I ) ⊆ Voc + ( Γ ) ∩ Voc − ( Δ ) \operatorname{Voc}^{-}(I) \subseteq\operatorname{Voc}^{+}(\Gamma) \cap\operatorname{Voc}^{-}(\Delta) Voc − ( I ) ⊆ Voc + ( Γ ) ∩ Voc − ( Δ ) , and two cut-free proofs π L , π R \pi^{L},\pi^{R} π L , π R of ⊢ Γ , I \vdash\Gamma,I ⊢ Γ , I and ⊢ I ⊥ , Δ \vdash I^{\perp},\Delta ⊢ I ⊥ , Δ respectively such that
π L ⊢ Γ , I π R ⊢ I ⊥ , Δ ( Cut ) ⟶ cut ∗ π \frac{\pi^{L}}{\vdash\Gamma,I}\quad\frac{\pi^{R}}{\vdash I^{\perp},\Delta}\;(\operatorname{Cut})\longrightarrow_{\operatorname{cut}}^{*}\pi ⊢ Γ , I π L ⊢ I ⊥ , Δ π R ( Cut ) ⟶ cut ∗ π ⊢ Γ , Δ \vdash\Gamma,\Delta ⊢ Γ , Δ .
The proof of this theorem amounts to a cut-introduction process synthesizing the interpolant in two phases, called the ascending and descending phase.
Ascending phase. Interpolation works with decorated proofs where the formulas of each sequents are colored or split. Given a split of the conclusion ⊢ Γ \vdash\Gamma ⊢ Γ of π \pi π into a pair of sublists ( Γ ′ , Γ ′ ′ ) (\Gamma^{\prime},\Gamma^{\prime\prime}) ( Γ ′ , Γ ′′ ) (often represented by coloring the formulas of Γ \Gamma Γ red or blue), the ascending phase consists in traversing π \pi π upward from conclusion to axioms, while splitting each sequent ⊢ Δ \vdash\Delta ⊢ Δ into a split ( Δ ′ , Δ ′ ′ ) (\Delta^{\prime},\Delta^{\prime\prime}) ( Δ ′ , Δ ′′ ) using the ancestor relation, propagating the split of the conclusion. A proof decorated in coherence with the ancestor relation is called coherently decorated .
Descending phase. Once every sequent in the proof has been decorated during the ascending phase, the descending phase consists in applying cut-introductions rules , starting from the leaves of the derivation tree, until one reaches the conclusion. This cut-introduction process is described by Saurin [34 ] by using PRIS, or proof-relevant interpolation situation that are represented in Figure 2 . A PRIS for ( Γ , Δ ) (\Gamma,\Delta) ( Γ , Δ ) is the data of (i) The goal , a cut-free LL decorated proof π \pi π of split conclusion ⊢ Γ , Δ \vdash\Gamma,\Delta ⊢ Γ , Δ , with open (split) premises ( ⊢ Γ i , Δ i ) 1 ≤ i ≤ n (\vdash\Gamma_i,\Delta_i)_{1\leq i\leq n} ( ⊢ Γ i , Δ i ) 1 ≤ i ≤ n , (ii) The partial interpolants , that is for each open premise ⊢ Γ i , Δ i \vdash\Gamma_i,\Delta_i ⊢ Γ i , Δ i , a formula I i I_i I i st. Voc + ( I i ) ⊆ Voc − ( Γ i ) ∩ Voc + ( Δ i ) \operatorname{Voc}^{+}(I_i) \subseteq\operatorname{Voc}^{-}(\Gamma_i) \cap\operatorname{Voc}^{+}(\Delta_i) Voc + ( I i ) ⊆ Voc − ( Γ i ) ∩ Voc + ( Δ i ) and Voc − ( I i ) ⊆ Voc + ( Γ i ) ∩ Voc − ( Δ i ) \operatorname{Voc}^{-}(I_i) \subseteq\operatorname{Voc}^{+}(\Gamma_i) \cap\operatorname{Voc}^{-}(\Delta_i) Voc − ( I i ) ⊆ Voc + ( Γ i ) ∩ Voc − ( Δ i ) and (iii) The partial solutions , that is for each open premise ⊢ Γ i , Δ i \vdash\Gamma_i,\Delta_i ⊢ Γ i , Δ i , derivations π i L \pi_i^{L} π i L (resp. π i R \pi_i^{R} π i R ) of conclusions ⊢ Γ i , I i \vdash\Gamma_i,I_i ⊢ Γ i , I i (resp. ⊢ I i ⊥ , Δ i \vdash I_i^{\perp},\Delta_i ⊢ I i ⊥ , Δ i ). Saurin singles out three types of PRIS: initial, solved and elementary PRIS, which are represented in Figure 3 .
Figure 2. A PRIS− Proof-relevant Interpolation Situation.
Figure 3. (a) Initial PRIS, (b) Solved PRIS, (c) Elementary PRIS.
The interpolation problem is then rephrased as follows: Given an initial PRIS, is it possible to produce a solved PRIS using only cut-introduction rules?
The key ingredient is elementary PRIS where there remains a unique inference rule to solve the interpolation problem: Theorem 10 follows directly from solving that case.
Remark 11. In the following we will focus on Craig interpolation rather than Lyndon as it is more natural to consider with cuts; from now on we neglect the polarity of the vocabulary.
Interpolation with Cuts
In the present section, we extend Saurin’s result on proof-relevant interpolation for LL sequent proofs [34 ] to non-cut-free proofs, meaning that we aim at performing interpolation directly on proofs containing cuts. Note that the results of this section readily apply to intuitionistic and classical sequent calculi as well. Before proceeding, a number of observations is in order:
Since Saurin’s proof-relevant interpolation procedure only relies on cut-introduction rules, the presence of cuts should not prevent the dynamics of the cut-elimination process to work: indeed, only the vocabulary of the synthesized candidate-interpolant formula depends on the cut-freeness assumption, a key condition for interpolation to hold though.
Instead of designing an ad hoc constraint on the cut-formulas, one can consider a proof-relevant weak interpolation process that handles every proof with cuts, at the cost of relaxing the control on the interpolant vocabulary (since the cut-formulas can contribute to the interpolant vocabulary): the cut-introduction performed in solving PRIS can indeed be extended to handle the cut-case, as long as the cut formulas are split themselves and the interpolant is allowed to use a vocabulary exploiting the language of the cut-formulas.
Last, our strategy will be to analyze conditions over the cut-formulas, that ensure an actual interpolation results. We will therefore obtain sufficient conditions so that the synthesized formula is an interpolant of the conclusion sequent.
Remark 12. There is a structuring principle, often exploited in the theory of linear logic proof nets correctness criteria, that states that a cut inference on C C C , C ⊥ C^\perp C ⊥ has the same geometry as a ⊗ \otimes ⊗ rule introducing C ⊗ C ⊥ C \otimes C^\perp C ⊗ C ⊥ (or the corresponding conjunction when working in classical or intuitionistic logic), followed by some form of hiding operation. This works fine when considering MLL (where proof nets best behave) but it gets more complex in other fragments of LL (with additives, exponentials, etc.) While one could carry formally the above simulation of the cut inference on C C C , C ⊥ C^\perp C ⊥ for the various fragments of LL4 , we will instead introduce a special treatment of the cuts in PRIS, for two reasons. First, our treatment will be more uniform than considering specific encoding for each specific logical fragment/system that one considers. The second and deeper reason amounts to the fact that the crucial part of our analysis (detailed in the following section) relies in identifying the cut-vocabulary and allowing to express adequate constraints on the cut vocabulary.
CPRIS: Proof-relevant Interpolation Situations with Cuts
First, one needs the following lemma that is straightforwardly proved by induction that all LL proofs (with cuts) can be coherently decorated, given an initial splitting of their conclusion, as well as a splitting of every cut-inference occurrence, i.e. for each cut inference i i i , we assign an arbitrary color for the cut formulas C i C_i C i and C i ⊥ C_i^\perp C i ⊥ , that we use in the ascending phase.
Lemma 13. For any LL proof π \pi π and any splittings of its conclusion sequent and of its cut inference occurrences, the ascending phase terminates with a coherently decorated proof.
One can now propose the adequate generalisation of the structures of [34 ] allowing to solve PRIS in presence of cuts and allowing to take into account how cut formulas can contribute to the vocabulary of the interpolant: we introduce CPRIS for that purpose that generalise PRIS with the notion of cut vocabularies.5
Definition 14 (PRIS with Cuts). A CPRIS for ( Γ , Δ ) (\Gamma, \Delta) ( Γ , Δ ) is the data of:
the goal , that is, an LL coherently decorated proof π \pi π of ⊢ Γ , Δ \vdash\Gamma, \Delta ⊢ Γ , Δ with n ≥ 0 n \ge0 n ≥ 0 open premises ( ⊢ Γ i , Δ i ) 1 ≤ i ≤ n (\vdash\Gamma_i, \Delta_i)_{1 \le i \le n} ( ⊢ Γ i , Δ i ) 1 ≤ i ≤ n such that for each 1 ≤ i ≤ n 1 \le i \le n 1 ≤ i ≤ n the formulas in Γ i \Gamma_i Γ i (resp. Δ i \Delta_i Δ i ) are ancestors of formulas in Γ \Gamma Γ (resp. Δ \Delta Δ ) or ancestors of a cut-formula;
the cut vocabularies : for 1 ≤ i ≤ n 1 \le i \le n 1 ≤ i ≤ n , a pair ( C i L , C i R ) (C_i^L, C_i^R) ( C i L , C i R ) of sets of literals;
the partial interpolants , that is for each 1 ≤ i ≤ n 1 \le i \le n 1 ≤ i ≤ n , a formula I i I_i I i st. Voc ( I i ) ⊆ ( Voc ( Γ i ) ∪ C i L ) ∩ ( Voc ( Δ i ) ∪ C i R ) \operatorname{Voc}(I_i) \subseteq(\operatorname{Voc}(\Gamma_i) \cup C_i^L) \cap(\operatorname{Voc}(\Delta_i) \cup C_i^R) Voc ( I i ) ⊆ ( Voc ( Γ i ) ∪ C i L ) ∩ ( Voc ( Δ i ) ∪ C i R ) and;
the partial solutions , that is, for each 1 ≤ i ≤ n 1 \le i \le n 1 ≤ i ≤ n , derivations π i L \pi_i^L π i L (resp. π i R \pi_i^R π i R ) of conclusion ⊢ Γ i , I i \vdash\Gamma_i, I_i ⊢ Γ i , I i (resp. ⊢ I i ⊥ , Δ i \vdash I_i^\perp, \Delta_i ⊢ I i ⊥ , Δ i ) such that if C L C^L C L (resp. C R C^R C R ) is the set of cut-formulas in π i L \pi_i^L π i L (resp. in π i R \pi_i^R π i R ), then ⋃ C ∈ C ϵ Voc ( C ) = C i ϵ \bigcup_{C \in C^\epsilon} \operatorname{Voc}(C) = C_i^\epsilon ⋃ C ∈ C ϵ Voc ( C ) = C i ϵ , ϵ ∈ { L , R } \epsilon\in\{L, R\} ϵ ∈ { L , R } .
With the above definition, the notions of initial, solved, and elementary CPRIS are trivially adapted from the previous section.
Weak proof-relevant interpolation theorem with cuts
Lemma 15. For any elementary CPRIS π \pi π there exist I I I , π L \pi^L π L , π R \pi^R π R such that π ′ = ⊢ Γ , I ⊢ I ⊥ , Δ \pi' = \vdash\Gamma, I \quad\vdash I^\perp, \Delta π ′ =⊢ Γ , I ⊢ I ⊥ , Δ and ⊢ Γ , I ⊢ I ⊥ , Δ ⊢ Γ , Δ ( C u t ) \frac{\vdash\Gamma, I \qquad\vdash I^\perp, \Delta}{\vdash\Gamma, \Delta}\;(\mathrm{Cut}) ⊢ Γ , Δ ⊢ Γ , I ⊢ I ⊥ , Δ ( Cut ) is a solved CPRIS and π ← ⋆ c u t π ′ \pi\mathrel{\xleftarrow{\star}_{\mathrm{cut}}} \pi' π ⋆ cut π ′ .
Proof sketch. By case analysis on the last rule of the elementary CPRIS. The extension is trivial for all inferences but the cut rule. In the case of an elementary CPRIS concluded with a cut-inference, the extension directly follows for the case of the ⊗ \otimes ⊗ rule, the only difference is that instead of being part of the vocabulary of the context formulas, the cut-formulas contribute to the cut vocabularies. (Details are provided in Section B.1.) ◀
Lemma 15 implies by an easy induction the following theorem, using a CPRIS obtained by splittings consistent with Γ , Δ \Gamma, \Delta Γ , Δ and a partitioning of cut formulas as { ( C i , C i ⊥ ) , i ∈ L } \{(C_i, C_i^\perp), i \in L\} {( C i , C i ⊥ ) , i ∈ L } , { ( C i , C i ⊥ ) , i ∈ R } \{(C_i, C_i^\perp), i \in R\} {( C i , C i ⊥ ) , i ∈ R } . Indeed, each application of Lemma 15 decreases the size of the goal of the CPRIS by one which straightforwardly ensures the result.
Theorem 16 (Weak interpolation theorem with cuts). Let π \pi π be an LL proof of ⊢ Γ , Δ \vdash\Gamma, \Delta ⊢ Γ , Δ . Let C = { ( C 1 , C 1 ⊥ ) , … , ( C n , C n ⊥ ) } C = \{(C_1, C_1^\perp), \ldots, (C_n, C_n^\perp)\} C = {( C 1 , C 1 ⊥ ) , … , ( C n , C n ⊥ )} be the cut formulas of π \pi π , let L , R L, R L , R be any partitioning of { 1 , … , n } \{1, \ldots, n\} { 1 , … , n } and C L = Voc ( C i , i ∈ L ) C^L = \operatorname{Voc}(C_i, i \in L) C L = Voc ( C i , i ∈ L ) and C R = Voc ( C i , i ∈ R ) C^R = \operatorname{Voc}(C_i, i \in R) C R = Voc ( C i , i ∈ R ) . Then there exists a LL formula I I I such that Voc ( I ) ⊆ ( Voc ( Γ ) ∪ C L ) ∩ ( Voc ( Δ ) ∪ C R ) \operatorname{Voc}(I) \subseteq(\operatorname{Voc}(\Gamma) \cup C^L) \cap(\operatorname{Voc}(\Delta) \cup C^R) Voc ( I ) ⊆ ( Voc ( Γ ) ∪ C L ) ∩ ( Voc ( Δ ) ∪ C R ) and two LL proofs π 1 , π 2 \pi_1, \pi_2 π 1 , π 2 of ⊢ Γ , I \vdash\Gamma, I ⊢ Γ , I and ⊢ I ⊥ , Δ \vdash I^\perp, \Delta ⊢ I ⊥ , Δ respectively such that ⊢ Γ , I ⊢ I ⊥ , Δ ⊢ Γ , Δ ( C u t ) → ⋆ c u t π \frac{\vdash\Gamma, I \qquad\vdash I^\perp, \Delta}{\vdash\Gamma, \Delta}\;(\mathrm{Cut}) \mathrel{\xrightarrow{\star}_{\mathrm{cut}}} \pi ⊢ Γ , Δ ⊢ Γ , I ⊢ I ⊥ , Δ ( Cut ) ⋆ cut π .
As such, the above result does not satisfy the vocabulary constraints required by Craig’s interpolation theorem as the interpolant formula I I I may resort to the cut-vocabulary which is why we speak of a weak interpolation result. On the other hand, the above results provide us a toolkit to analyse situations where interpolation may be achieved in the presence of cuts: indeed, in Theorem 16 , we have the choice of the partitioning of the set of cut-formulas C \mathcal{C} C .
Proof-relevant interpolation with cuts
In the present section, we want to express conditions on the cut inference occurring in a proof that allow to get a full proof-relevant interpolation result from the above weak interpolation theorem. To motivate this, let us consider the simpler case where the proof π \pi π contains a single occurrence of the cut inference on formula C C C that is split to the left.
If Voc ( C ) ⊆ Voc ( Γ ) \operatorname{Voc}(C) \subseteq\operatorname{Voc}(\Gamma) Voc ( C ) ⊆ Voc ( Γ ) , then Voc ( I ) ⊆ ( Voc ( Γ ) ∪ Voc ( C ) ) ∩ Voc ( Δ ) = Voc ( Γ ) ∩ Voc ( Δ ) \operatorname{Voc}(I) \subseteq(\operatorname{Voc}(\Gamma) \cup\operatorname{Voc}(C)) \cap\operatorname{Voc}(\Delta) = \operatorname{Voc}(\Gamma) \cap\operatorname{Voc}(\Delta) Voc ( I ) ⊆ ( Voc ( Γ ) ∪ Voc ( C )) ∩ Voc ( Δ ) = Voc ( Γ ) ∩ Voc ( Δ ) : I I I is indeed a Craig interpolant (same if C C C is on the right and Voc ( C ) ⊆ Voc ( Δ ) \operatorname{Voc}(C) \subseteq\operatorname{Voc}(\Delta) Voc ( C ) ⊆ Voc ( Δ ) ).
If Voc ( C ) ∩ Voc ( Δ ) = ∅ \operatorname{Voc}(C) \cap\operatorname{Voc}(\Delta) = \varnothing Voc ( C ) ∩ Voc ( Δ ) = ∅ , then Voc ( I ) ⊆ ( Voc ( Γ ) ∪ Voc ( C ) ) ∩ Voc ( Δ ) = Voc ( Γ ) ∩ Voc ( Δ ) \operatorname{Voc}(I) \subseteq(\operatorname{Voc}(\Gamma) \cup\operatorname{Voc}(C)) \cap\operatorname{Voc}(\Delta) = \operatorname{Voc}(\Gamma) \cap\operatorname{Voc}(\Delta) Voc ( I ) ⊆ ( Voc ( Γ ) ∪ Voc ( C )) ∩ Voc ( Δ ) = Voc ( Γ ) ∩ Voc ( Δ ) : I I I is also a Craig interpolant (same if C C C is on the right and Voc ( C ) ∩ Voc ( Γ ) = ∅ \operatorname{Voc}(C) \cap\operatorname{Voc}(\Gamma) = \varnothing Voc ( C ) ∩ Voc ( Γ ) = ∅ ).
Exploiting the second remark of the above analysis, one easily sees that if the set of cut formulas C \mathcal{C} C can be split in C L \mathcal{C}^{L} C L and C R \mathcal{C}^{R} C R (of respective vocabulary C L \mathcal{C}^{L} C L and C R \mathcal{C}^{R} C R ) such that ( C L ∖ Voc ( Γ ) ) ∩ ( Voc ( Δ ) ∪ C R ) = ∅ (\mathcal{C}^{L} \setminus\operatorname{Voc}(\Gamma)) \cap(\operatorname{Voc}(\Delta) \cup\mathcal{C}^{R}) = \varnothing ( C L ∖ Voc ( Γ )) ∩ ( Voc ( Δ ) ∪ C R ) = ∅ and ( C R ∖ Voc ( Δ ) ) ∩ ( Voc ( Γ ) ∪ C L ) = ∅ (\mathcal{C}^{R} \setminus\operatorname{Voc}(\Delta)) \cap(\operatorname{Voc}(\Gamma) \cup\mathcal{C}^{L}) = \varnothing ( C R ∖ Voc ( Δ )) ∩ ( Voc ( Γ ) ∪ C L ) = ∅ , then the interpolation vocabulary constraint is satisfied in Theorem 16 .
This allows us to state the following proof-relevant interpolation theorem with cuts:
Theorem 17 (Interpolation theorem with cuts). Let π \pi π be an LL proof of ⊢ Γ , Δ \vdash\Gamma, \Delta ⊢ Γ , Δ . Let C = { ( C 1 , C 1 ⊥ ) , … , ( C n , C n ⊥ ) } \mathcal{C} = \{(C_{1}, C_{1}^{\perp}), \ldots, (C_{n}, C_{n}^{\perp})\} C = {( C 1 , C 1 ⊥ ) , … , ( C n , C n ⊥ )} be the cut formulas of π \pi π , let L , R L, R L , R be any partitioning of { 1 , … , n } \{1, \ldots, n\} { 1 , … , n } and C L = Voc ( C i , i ∈ L ) \mathcal{C}^{L} = \operatorname{Voc}(C_{i}, i \in L) C L = Voc ( C i , i ∈ L ) and C R = Voc ( C i , i ∈ R ) \mathcal{C}^{R} = \operatorname{Voc}(C_{i}, i \in R) C R = Voc ( C i , i ∈ R ) such that ( C L ∖ Voc ( Γ ) ) ∩ ( Voc ( Δ ) ∪ C R ) = ∅ (\mathcal{C}^{L} \setminus\operatorname{Voc}(\Gamma)) \cap(\operatorname{Voc}(\Delta) \cup\mathcal{C}^{R}) = \varnothing ( C L ∖ Voc ( Γ )) ∩ ( Voc ( Δ ) ∪ C R ) = ∅ and ( C R ∖ Voc ( Δ ) ) ∩ ( Voc ( Γ ) ∪ C L ) = ∅ (\mathcal{C}^{R} \setminus\operatorname{Voc}(\Delta)) \cap(\operatorname{Voc}(\Gamma) \cup\mathcal{C}^{L}) = \varnothing ( C R ∖ Voc ( Δ )) ∩ ( Voc ( Γ ) ∪ C L ) = ∅ .
Then there exists an LL formula I I I such that Voc ( I ) ⊆ Voc ( Γ ) ∩ Voc ( Δ ) \operatorname{Voc}(I) \subseteq\operatorname{Voc}(\Gamma) \cap\operatorname{Voc}(\Delta) Voc ( I ) ⊆ Voc ( Γ ) ∩ Voc ( Δ ) and two LL proofs π 1 , π 2 \pi_{1}, \pi_{2} π 1 , π 2 of ⊢ Γ , I \vdash\Gamma, I ⊢ Γ , I and ⊢ I ⊥ , Δ \vdash I^{\perp}, \Delta ⊢ I ⊥ , Δ respectively such that
⊢ Γ , I ⊢ I ⊥ , Δ ⊢ Γ , Δ ( C u t ) ⟶ c u t π . \frac{\vdash\Gamma, I \qquad\vdash I^{\perp}, \Delta}{\vdash\Gamma, \Delta}\;(\mathrm{Cut}) \longrightarrow_{\mathrm{cut}} \pi. ⊢ Γ , Δ ⊢ Γ , I ⊢ I ⊥ , Δ ( Cut ) ⟶ cut π .
Of course, when each cut inference with cut formula C C C is (semi-)analytic, then either Voc ( C ) ⊆ Voc ( Γ ) \operatorname{Voc}(C) \subseteq\operatorname{Voc}(\Gamma) Voc ( C ) ⊆ Voc ( Γ ) or Voc ( C ) ⊆ Voc ( Δ ) \operatorname{Voc}(C) \subseteq\operatorname{Voc}(\Delta) Voc ( C ) ⊆ Voc ( Δ ) , the condition Theorem 17 is met, resulting in the following corollary:
Corollary 19 (Proof-relevant interpolation theorem with (semi-)analytic cuts). Let A , B A, B A , B be LL formulas and π \pi π be an LL proof of ⊢ A , B \vdash A, B ⊢ A , B such that every cut-inference is (semi-)analytic. Then there exists a LL formula I I I such that Voc ( I ) ⊆ Voc ( A ) ∩ Voc ( B ) \operatorname{Voc}(I) \subseteq\operatorname{Voc}(A) \cap\operatorname{Voc}(B) Voc ( I ) ⊆ Voc ( A ) ∩ Voc ( B ) and two LL proofs π 1 , π 2 \pi_{1}, \pi_{2} π 1 , π 2 of ⊢ A , I \vdash A, I ⊢ A , I and ⊢ I ⊥ , B \vdash I^{\perp}, B ⊢ I ⊥ , B respectively such that
⊢ A , I ⊢ I ⊥ , B ⊢ A , B ( C u t ) ⟶ c u t π . \frac{\vdash A, I \qquad\vdash I^{\perp}, B}{\vdash A, B}\;(\mathrm{Cut}) \longrightarrow_{\mathrm{cut}} \pi. ⊢ A , B ⊢ A , I ⊢ I ⊥ , B ( Cut ) ⟶ cut π .
in axiom rules: only the subformulas of cut formulas involved in axioms need to be considered to check whether the generated interpolant falls in the adequate language.
A first benefit of the above results is that performing cut elimination can induce a blow up in the size of the proof and therefore of the interpolant and it directly affects the efficiency of the procedure for computing an interpolant. Another one is that one can hope to apply this proof-relevant interpolation technique to proof systems having only a partial cut-elimination result (for instance when cuts may be stuck on non logical axioms for instance). Last, the extension of the results to non normal proof is important for the computational interpretation as already stressed in the introduction: we shall discuss this in what follows.
Computational interpretations
While we leave the task of extending Čubrić’ result to non-evaluated λ \lambda λ -terms to future work, we explain below how our results already provide computational results in System L, which is a term calculus due to Curien and Herbelin [12 , 13 , 30 ] with a typing system in correspondence with the sequent calculus system, instead of the natural deduction system. Our results are indeed readily applicable to Saurin’s computational analysis of interpolation in System L [34, Section 5] where he shows an analogous to Čubrić’ result for the multiplicative and additive typing of System L but terms must be reduced to normal forms for the procedure to be applicable6 . Theorem 17 readily allows to strengthen the above result (due to space constraints, we refer the reading to [34, Section 5] for the notations and definitions that we adopt below). The following theorem is proved in Section B.2:
Theorem 21. Let c : ( Γ 1 , Γ 2 ⊢ Δ 1 , Δ 2 ) c : (\Gamma_1, \Gamma_2 \vdash\Delta_1, \Delta_2) c : ( Γ 1 , Γ 2 ⊢ Δ 1 , Δ 2 ) be a decorated L L L -command and let C C C be the set of types of all the t , e ′ t, e' t , e ′ such that ⟨ t ′ ∣ e ′ ⟩ \langle t' \mid e' \rangle ⟨ t ′ ∣ e ′ ⟩ is a subcommand of c c c . Let us assume that C C C is split into C L , C R C^{L}, C^{R} C L , C R and C L = Voc ( C L ) C^{L} = \operatorname{Voc}(C^{L}) C L = Voc ( C L ) and C R = Voc ( C R ) C^{R} = \operatorname{Voc}(C^{R}) C R = Voc ( C R ) such that ( C L ∖ Voc ( Γ 1 , Δ 1 ) ) ∩ ( Voc ( Γ 2 , Δ 2 ) ∪ C R ) = ∅ (C^{L} \setminus\operatorname{Voc}(\Gamma_1, \Delta_1)) \cap(\operatorname{Voc}(\Gamma_2, \Delta_2) \cup C^{R}) = \varnothing ( C L ∖ Voc ( Γ 1 , Δ 1 )) ∩ ( Voc ( Γ 2 , Δ 2 ) ∪ C R ) = ∅ and ( C R ∖ Voc ( Γ 2 , Δ 2 ) ) ∩ ( Voc ( Γ 1 , Δ 1 ) ∪ C L ) = ∅ (C^{R} \setminus\operatorname{Voc}(\Gamma_2, \Delta_2)) \cap(\operatorname{Voc}(\Gamma_1, \Delta_1) \cup C^{L}) = \varnothing ( C R ∖ Voc ( Γ 2 , Δ 2 )) ∩ ( Voc ( Γ 1 , Δ 1 ) ∪ C L ) = ∅ . Then, there exist a type I ∈ Voc ( Γ 1 , Δ 1 ) ∩ Voc ( Γ 2 , Δ 2 ) I \in\operatorname{Voc}(\Gamma_1, \Delta_1) \cap\operatorname{Voc}(\Gamma_2, \Delta_2) I ∈ Voc ( Γ 1 , Δ 1 ) ∩ Voc ( Γ 2 , Δ 2 ) and t , e t, e t , e such that Γ 1 ⊢ t : I ∣ Δ 1 \Gamma_1 \vdash t : I \mid\Delta_1 Γ 1 ⊢ t : I ∣ Δ 1 and Γ 2 ∣ e : I ⊢ Δ 2 \Gamma_2 \mid e : I \vdash\Delta_2 Γ 2 ∣ e : I ⊢ Δ 2 , and ⟨ t ∣ e ⟩ → ∗ c \langle t \mid e \rangle\mathrel{\to^{*}} c ⟨ t ∣ e ⟩ → ∗ c .
Proof Sketch. While the proof by Saurin [34 ] relies on a long case analysis on both terms, evaluation contexts and commands, it is sufficient to adapt the case of commands as the other cases are straightforward by a simple check that the propagation of the cut-vocabulary constraint works. Regarding the case of commands, while Saurin considers only cases ⟨ x ∣ e ⟩ \langle x \mid e \rangle ⟨ x ∣ e ⟩ and ⟨ t ∣ α ⟩ \langle t \mid\alpha\rangle ⟨ t ∣ α ⟩ (due to the assumption on normal terms), we now have full fledged commands which are interpolated in a way close to the tensor case in [34, Thm 20].
Proof-relevant interpolation for circular proofs
In this section, we extend the proof-relevant interpolation procedure for LL proofs presented in the Section 2.2, for μ \mu μ LLω ^{\omega} ω proofs (restricted to the propositional setting for simplicity). We will allow derivations with back-edges and cuts. While the cuts will be subject to conditions inherited from the previous section, the structure of the back-edges may or may not have an impact on the interpolation result. Indeed, while any pre-proof can be interpolated in a proof-relevant way (this is an easy corollary of the construction of the following section), it is meaningful to wonder whether a derivation π \pi π that satisfies the progress condition can be interpolated with two proofs π 1 , π 2 \pi_1, \pi_2 π 1 , π 2 which also satisfy the progress condition. Intuitively, given the structure of proof-relevant interpolation, the reader should have in mind that each cycle of π \pi π will give rise to both a cycle in π 1 \pi_1 π 1 and in π 2 \pi_2 π 2 , while the inferences of π \pi π will get dispatched between π 1 \pi_1 π 1 and π 2 \pi_2 π 2 depending on whether they correspond to a left-split or right-split formula. A consequence is that if a given cycle is validated by a trace on a formula F F F that is left-split, π 1 \pi_1 π 1 will inherited the progress of π \pi π but not π 2 \pi_2 π 2 : the interpolant must therefore be built in such a way as to restore progress to the corresponding cycle in π 2 \pi_2 π 2 . The following paragraphs explain why and how this is possible.
To accomplish this goal, we first need to introduce some additional notions on circular proofs, adapting to μ L L ω \mu\mathrm{LL}^{\omega} μ LL ω proofs various notions due to Brotherston [5 ] and then we will respectively extend the ascending and descending phase for circular pre-proofs. Without loss of generality, we will always assume that the bud-companion relation R \mathcal{R} R of a pre-proof P \mathcal{P} P is injective, as explained in the Appendix C.2.
(Ordered) trace manifolds for circular pre-proofs
Intuitively, (ordered) trace manifolds are finite structures containing finite segments of valid infinite traces that can be stitched together to generate every progressing trace for each infinite path in the pre-proof. We first introduce some preliminary definitions and lemmas.
Definition 22 (Induced subgraph). Let G = ( V , s , r , p ) G = (V,s,r,p) G = ( V , s , r , p ) be a derivation graph and let V ′ ⊆ V V' \subseteq V V ′ ⊆ V . Then the subgraph of G G G induced by V ′ V' V ′ is defined as S = ( V ′ , s , r , p ′ ) S = (V',s,r,p') S = ( V ′ , s , r , p ′ ) where for some j ∈ N j \in\mathbb{N} j ∈ N , p j ′ ( v ) = p j ( v ) p'_j(v) = p_j(v) p j ′ ( v ) = p j ( v ) if v ∈ V ′ v \in V' v ∈ V ′ and p j ( v ) ∈ V ′ p_j(v) \in V' p j ( v ) ∈ V ′ , and undefined otherwise. We write S ⊆ G S \subseteq G S ⊆ G to say that S S S is a subgraph of G G G . If V ′ = ∅ V' = \varnothing V ′ = ∅ we say that the subgraph is trivial.
Every strongly connected subgraph of a circular pre-proof can be characterized by the cycles obtained from the bud-companion relationships. These are basic cycles .
Definition 23 (Basic cycle). Let P = ( D , R ) \mathcal{P} = (\mathcal{D},\mathcal{R}) P = ( D , R ) be a μ L L ω \mu\mathrm{LL}^{\omega} μ LL ω pre-proof and let B ∈ Bud ( D ) B \in\operatorname{Bud}(\mathcal{D}) B ∈ Bud ( D ) . Then the basic cycle C B C_B C B in G P \mathcal{G}_{\mathcal{P}} G P is obtained from the unique path from R ( B ) \mathcal{R}(B) R ( B ) to B B B by replacing the unique edge ( v , B ) (v,B) ( v , B ) in D \mathcal{D} D for the edge ( v , R ( B ) ) (v,\mathcal{R}(B)) ( v , R ( B )) in G P \mathcal{G}_{\mathcal{P}} G P . Since C B C_B C B is a cycle in G P \mathcal{G}_{\mathcal{P}} G P , it is also a non-trivial strongly connected subgraph in G P \mathcal{G}_{\mathcal{P}} G P .
We can characterize how basic cycles intersect with each other by extracting the structural connectivity of the bud nodes.
Definition 24 (Structural connectivity). Let P = ( D , R ) \mathcal{P} = (\mathcal{D},\mathcal{R}) P = ( D , R ) be a μ L L ω \mu\mathrm{LL}^{\omega} μ LL ω pre-proof. The relation ⪯ P \preceq_{\mathcal{P}} ⪯ P on Bud ( D ) \operatorname{Bud}(\mathcal{D}) Bud ( D ) is defined as: B 2 ⪯ P B 1 B_2 \preceq_{\mathcal{P}} B_1 B 2 ⪯ P B 1 iff R ( B 2 ) \mathcal{R}(B_2) R ( B 2 ) appears on the basic cycle C B 1 ∈ G P C_{B_1} \in\mathcal{G}_{\mathcal{P}} C B 1 ∈ G P .
The union of basic cycles, whose bud nodes are weakly ⪯ P \preceq_{\mathcal{P}} ⪯ P -connected, give rise to strongly connected subgraphs in a pre-proof.
Lemma 25 (Brotherston [5 ] , Lemma 7.1.7). Let P = ( D , R ) \mathcal{P} = (\mathcal{D},\mathcal{R}) P = ( D , R ) be a μ L L ω \mu\mathrm{LL}^{\omega} μ LL ω pre-proof and let S S S be a non-trivial subgraph of G P \mathcal{G}_{\mathcal{P}} G P . Then S S S is strongly connected iff there exists a non-empty set of buds B S ⊆ Bud ( D ) B_S \subseteq\operatorname{Bud}(\mathcal{D}) B S ⊆ Bud ( D ) such that B S B_S B S is weakly ⪯ P \preceq_{\mathcal{P}} ⪯ P -connected and S = ⋃ B ∈ B S C B S = \bigcup_{B \in B_S} C_B S = ⋃ B ∈ B S C B .
Using the previous constructions, we define trace manifolds in the setting of valid μ L L ω \mu\mathrm{LL}^{\omega} μ LL ω pre-proofs. We therefore adapt Brotherston’s original definition of trace manifolds ([5 ] , Definition 7.2.1) to accommodate to the requirements of valid proofs in our system.
Definition 26 (Trace manifold). A trace manifold for a μ L L ω \mu\mathrm{LL}^{\omega} μ LL ω pre-proof P = ( D , R ) \mathcal{P}=(\mathcal{D},\mathcal{R}) P = ( D , R ) is a set of traces { τ S , B i ∣ S = ⋃ B ∈ B S C B where B S ⊆ B u d ( D ) weakly ≤ P -connected , B i ∈ B S } \{\tau_{S,B_i}\mid S=\bigcup_{B\in\mathcal{B}_S}\mathcal{C}_B\text{ where }\mathcal{B}_S\subseteq Bud(\mathcal{D})\text{ weakly }\leq_{\mathcal{P}}\text{-connected},\ B_i\in\mathcal{B}_S\} { τ S , B i ∣ S = ⋃ B ∈ B S C B where B S ⊆ B u d ( D ) weakly ≤ P -connected , B i ∈ B S } s.t.:
For all S S S and B i ∈ B S B_i\in\mathcal{B}_S B i ∈ B S , τ S , B i \tau_{S,B_i} τ S , B i is a finite trace following the basic cycle C B i \mathcal{C}_{B_i} C B i in G P \mathcal{G}_{\mathcal{P}} G P and furthermore, it uses the same formula at both instances of R ( B i ) \mathcal{R}(B_i) R ( B i ) on C B i \mathcal{C}_{B_i} C B i . (Equivalently, τ S , B i \tau_{S,B_i} τ S , B i is a trace following the unique path in D \mathcal{D} D from R ( B i ) \mathcal{R}(B_i) R ( B i ) to B i B_i B i and uses the same formula at R ( B i ) \mathcal{R}(B_i) R ( B i ) and B i B_i B i ).
For all S S S , if B i , B j ∈ B S B_i,B_j\in\mathcal{B}_S B i , B j ∈ B S and B j ≤ P B i B_j\leq_{\mathcal{P}}B_i B j ≤ P B i then τ S , B j ( R ( B j ) ) = τ S , B i ( R ( B j ) ) \tau_{S,B_j}(\mathcal{R}(B_j))=\tau_{S,B_i}(\mathcal{R}(B_j)) τ S , B j ( R ( B j )) = τ S , B i ( R ( B j )) .
For every S S S , let τ S , B i 1 , τ S , B i 2 , … τ S , B i n \tau_{S,B_{i_1}},\tau_{S,B_{i_2}},\ldots\tau_{S,B_{i_n}} τ S , B i 1 , τ S , B i 2 , … τ S , B i n be the traces associated to S S S in the manifold. Then there is a k k k such that τ S , B i k \tau_{S,B_{i_k}} τ S , B i k has at leat one F F F -progress point and F F F is minimal in ⋃ j = 1 n Form ( τ S , B i j ) \bigcup_{j=1}^{n}\operatorname{Form}(\tau_{S,B_{i_j}}) ⋃ j = 1 n Form ( τ S , B i j ) with respect to the subformula relation, where Form ( τ ) \operatorname{Form}(\tau) Form ( τ ) is the set of formulas that appear in τ \tau τ .
The notion of ordered trace manifolds defined below allows us to group further the valid traces of a circular pre-proof, according to an induction order .
Definition 27 (Induction order). Let P = ( D , R ) \mathcal{P}=(\mathcal{D},\mathcal{R}) P = ( D , R ) be a μ L L ω \mu\mathrm{LL}^{\omega} μ LL ω pre-proof. A (non-strict) partial order ( B u d ( D ) , ⊲ ) (Bud(\mathcal{D}),\lhd) ( B u d ( D ) , ⊲ ) is said to be an induction order for P \mathcal{P} P if:
it is forest-like : B ⊲ B 1 B\lhd B_1 B ⊲ B 1 and B ⊲ B 2 B\lhd B_2 B ⊲ B 2 implies B 1 = B 2 B_1=B_2 B 1 = B 2 or B 1 ⊲ B 2 B_1\lhd B_2 B 1 ⊲ B 2 or B 2 ⊲ B 1 B_2\lhd B_1 B 2 ⊲ B 1
Every weakly ≤ P \leq_{\mathcal{P}} ≤ P -connected set B ⊆ B u d ( D ) \mathcal{B}\subseteq Bud(\mathcal{D}) B ⊆ B u d ( D ) has a ⊲ \lhd ⊲ -greatest element, i.e. an element B max ∈ B B_{\max}\in\mathcal{B} B m a x ∈ B such that B ⊲ B max B\lhd B_{\max} B ⊲ B m a x for all B ∈ B B\in\mathcal{B} B ∈ B . (Note that in particular we then have B 1 ≤ P B 2 B_1\leq_{\mathcal{P}}B_2 B 1 ≤ P B 2 implies B 1 ⊲ B 2 B_1\lhd B_2 B 1 ⊲ B 2 or $B_2\lhd B_1)
Again, we modify Brotherston’s original definition of ordered trace manifold ([5 ] , Definition 7.2.6) to accommodate to our notion of validity.
Definition 28 (Ordered trace manifold). Let P = ( D , R ) \mathcal{P}=(\mathcal{D},\mathcal{R}) P = ( D , R ) be a μ L L ω \mu\mathrm{LL}^{\omega} μ LL ω pre-proof and let ⊲ \lhd ⊲ be an induction order for P \mathcal{P} P . An ordered trace manifold with respect to ⊲ \lhd ⊲ is a set of traces { τ B j , B i ∣ B j , B i ∈ B u d ( D ) , B i ⊲ B j } \{\tau_{B_j,B_i}\mid B_j,B_i\in Bud(\mathcal{D}),B_i\lhd B_j\} { τ B j , B i ∣ B j , B i ∈ B u d ( D ) , B i ⊲ B j } satisfying the following:
For all B i , B j ∈ B u d ( D ) B_i,B_j\in Bud(\mathcal{D}) B i , B j ∈ B u d ( D ) with B i ⊲ B j B_i\lhd B_j B i ⊲ B j , τ B j , B i \tau_{B_j,B_i} τ B j , B i is a trace following the basic cycle C B i \mathcal{C}_{B_i} C B i in G P \mathcal{G}_{\mathcal{P}} G P and furthermore, it uses the same formula at both instances of R ( B i ) \mathcal{R}(B_i) R ( B i ) on C B i \mathcal{C}_{B_i} C B i . (Equivalently, τ B j , B i \tau_{B_j,B_i} τ B j , B i is a trace following the unique path in D \mathcal{D} D from R ( B i ) \mathcal{R}(B_i) R ( B i ) to B i B_i B i and uses the same formula at R ( B i ) \mathcal{R}(B_i) R ( B i ) and B i B_i B i ).
For all B i , B j , B k ∈ B u d ( D ) B_i,B_j,B_k\in Bud(\mathcal{D}) B i , B j , B k ∈ B u d ( D ) , if B j , B i ⊲ B k B_j,B_i\lhd B_k B j , B i ⊲ B k and B j ≤ P B i B_j\leq_{\mathcal{P}}B_i B j ≤ P B i , then τ B k , B j ( R ( B j ) ) = τ B k , B i ( R ( B j ) ) \tau_{B_k,B_j}(\mathcal{R}(B_j))=\tau_{B_k,B_i}(\mathcal{R}(B_j)) τ B k , B j ( R ( B j )) = τ B k , B i ( R ( B j )) .
For every B k ∈ B u d ( D ) B_k\in Bud(\mathcal{D}) B k ∈ B u d ( D ) , let τ k , B i 1 , τ k , B i 2 , … τ k , B i n \tau_{k,B_{i_1}},\tau_{k,B_{i_2}},\ldots\tau_{k,B_{i_n}} τ k , B i 1 , τ k , B i 2 , … τ k , B i n be the traces associated to B k B_k B k in the manifold. Then τ B k , B k \tau_{B_k,B_k} τ B k , B k has at least one F F F -progress point and F F F is minimal in ⋃ j = 1 n Form ( τ B k , B i j ) \bigcup_{j=1}^{n}\operatorname{Form}(\tau_{B_k,B_{i_j}}) ⋃ j = 1 n Form ( τ B k , B i j ) with respect to the subformula relation.
Following Brotherston’s proof ideas ([5 ] , Proposition 7.2.3 and Lemma 7.2.8), we can prove that, for a given pre-proof, having a trace manifold implies validity, and that having a trace manifold is equivalent to having an ordered trace manifold. The proofs of Lemmas 29 and 30 are included in the Appendix C.1.
Lemma 29. A μ L L ω \mu\mathrm{LL}^{\omega} μ LL ω pre-proof is valid if it has a trace manifold.
Lemma 30. A μ L L ω \mu\mathrm{LL}^{\omega} μ LL ω pre-proof P \mathcal{P} P has a trace manifold if and only if there exists an induction order ⊲ \lhd ⊲ for P \mathcal{P} P , and P \mathcal{P} P has an ordered trace manifold with respect to ⊲ \lhd ⊲ .
Ascending phase for circular proofs
First, we adapt the definition of decorated and coherently decorated proofs to circular pre-proofs:
Definition 31 (Decorated and coherent μ L L ω \mu\mathrm{LL}^{\omega} μ LL ω pre-proofs). Let P = ( D , R ) \mathcal{P} = (\mathcal{D},\mathcal{R}) P = ( D , R ) be a μ L L ω \mu\mathrm{LL}^{\omega} μ LL ω pre-proof. P \mathcal{P} P is a decorated pre-proof if its derivation tree D \mathcal{D} D is decorated. P \mathcal{P} P is a coherently decorated pre-proof if (i) its derivation tree D \mathcal{D} D is coherent and (ii) for all bud node B ∈ Bud ( D ) B \in\operatorname{Bud}(\mathcal{D}) B ∈ Bud ( D ) , then the split of s ( B ) s(B) s ( B ) is the same split of its companion node s ( R ( B ) ) s(\mathcal{R}(B)) s ( R ( B )) .
It is not immediate to generalise Lemma 13 to circular proofs since the coherence property is not necessarily maintained by the relation R \mathcal{R} R (see example in Appendix C.3). Up to some unfoldings of back-edges, this can however be restored by preserving validity:
Lemma 32. Let P 0 = ( D 0 , R 0 ) \mathcal{P}_0 = (\mathcal{D}_0,\mathcal{R}_0) P 0 = ( D 0 , R 0 ) be a valid μ L L ω \mu\mathrm{LL}^{\omega} μ LL ω proof with a trace manifold, with a split conclusion ⊢ Γ 0 , Δ 0 \vdash\Gamma_0,\Delta_0 ⊢ Γ 0 , Δ 0 , and injective R \mathcal{R} R . There exists a μ L L ω \mu\mathrm{LL}^{\omega} μ LL ω proof P = ( D , R ) \mathcal{P} = (\mathcal{D},\mathcal{R}) P = ( D , R ) with a trace manifold, having the same infinite unfolding as P 0 \mathcal{P}_0 P 0 , such that the split conclusion ⊢ Γ 0 , Δ 0 \vdash\Gamma_0,\Delta_0 ⊢ Γ 0 , Δ 0 produces a coherently decorated proof after the ascending phase.
Proof sketch. The proof of the lemma amounts to the finiteness of the number of splittings of a given sequent: by unfolding one can ensure that every bud can be associated with a companion decorated similarly. In Appendix C.4.1 the proof goes by introducing an algorithm (Algorithm 1) that actually builds for any μ L L ω \mu\mathrm{LL}^{\omega} μ LL ω pre-proof P = ( D , R ) \mathcal{P} = (\mathcal{D},\mathcal{R}) P = ( D , R ) with injective R \mathcal{R} R , an equivalent pre-proof via unfolding, that maintains coherence after the ascending phase.
Remark 33. The previous lemma can be generalised to ensure the existence of a valid μ L L ω \mu\mathrm{LL}^{\omega} μ LL ω proof P \mathcal{P} P that can be coherently decorated w.r.t. to any split.
Tree-compatibility
Before the definition of the descending phase for circular proofs, we need to make sure that the circular proof that we want to interpolate satisfies the tree-compatibility property.
Definition 34 (Tree-Compatibility). An induction order ( Bud ( D ) , ⊲ ) (\operatorname{Bud}(\mathcal{D}), \lhd) ( Bud ( D ) , ⊲ ) is said to be tree-compatible if B 1 ≤ P B 2 B_1 \leq_{\mathcal{P}} B_2 B 1 ≤ P B 2 and B 2 ≰ P B 1 B_2 \nleq_{\mathcal{P}} B_1 B 2 ≰ P B 1 implies B 1 ⊲ B 2 B_1 \lhd B_2 B 1 ⊲ B 2 , for any B 1 , B 2 ∈ Bud ( D ) B_1, B_2 \in\operatorname{Bud}(\mathcal{D}) B 1 , B 2 ∈ Bud ( D ) .
Fortunately, we can transform any trace manifold of a valid circular proof that can be coherently decorated, into an ordered trace manifold using Lemma 30. Additionally, this proof can be furthered transformed into an equivalent proof that can be also coherently decorated, with a new ordered trace manifold that has induction order that is tree-compatible.
Lemma 35. Let P 0 = ( D 0 , R 0 ) \mathcal{P}_0 = (\mathcal{D}_0,\mathcal{R}_0) P 0 = ( D 0 , R 0 ) be a valid μ L L ω \mu\mathrm{LL}^{\omega} μ LL ω proof with an ordered trace manifold with induction order ( Bud ( D 0 ) , ⊲ 0 ) (\operatorname{Bud}(\mathcal{D}_0), \lhd_0) ( Bud ( D 0 ) , ⊲ 0 ) that can be coherently decorated. There is a valid μ L L ω \mu\mathrm{LL}^{\omega} μ LL ω proof P \mathcal{P} P with an ordered trace manifold with a tree-compatible induction order ( Bud ( D ) , ⊲ ) (\operatorname{Bud}(\mathcal{D}), \lhd) ( Bud ( D ) , ⊲ ) , that can be coherently decorated.
Proof sketch. The proof (see Appendix C.4.2) generalises to our setting a similar property by Sprenger and Dam: we introduce Algorithm 2 that performs the required unfoldings of back-edges to the original pre-proof in order to make it tree-compatible. After executing Algorithm 2, relation R \mathcal{R} R may not be injective anymore but P \mathcal{P} P can be transformed to maintain tree-compatibility while restoring injectivity of R \mathcal{R} R , as explained in Appendix C.2.
Descending phase for circular proofs
We extend the descending phase to work on μ \mu μ LLω ^\omega ω pre-proofs P = ( D , R ) \mathcal{P}=(\mathcal{D},\mathcal{R}) P = ( D , R ) with injective R \mathcal{R} R . Our general approach follows two steps: the first step is to complete the treatment of the missing derivations cases for the elementary CPRIS by including the bud nodes and the fixed-point rules, so we can use the descending phase over the derivation tree D \mathcal{D} D :
Bud nodes: Let Bud ( D ) = { B 1 , … , B n } \operatorname{Bud}(\mathcal{D})=\{B_1,\ldots,B_n\} Bud ( D ) = { B 1 , … , B n } . To each bud node B i B_i B i with split sequent ⊢ Γ , Δ \vdash\Gamma,\Delta ⊢ Γ , Δ , we associate the fresh fixed-point variable X B i X_{B_i} X B i , and transform the leaf in D \mathcal{D} D , as done in the subfigure (a) of Figure 4 .
Figure 4. Elementary and solved CPRIS (a) for bud nodes, (b) for the μ , ν \mu,\nu μ , ν rules.
Fixed-point rules: Without loss of generality, consider a split ( Γ , Δ ) (\Gamma,\Delta) ( Γ , Δ ) with Γ = Γ ′ ∪ { σ X . F } \Gamma=\Gamma'\cup\{\sigma X.F\} Γ = Γ ′ ∪ { σ X . F } . The elementary CPRIS for the σ \sigma σ -rules, where σ ∈ { μ , ν } \sigma\in\{\mu,\nu\} σ ∈ { μ , ν } , can be solved as shown in the subfigure (b) of Figure 4 .
The second step is to maintain the bud-companion relations of the original proof by transforming the companion nodes during the descending phase, while appropriately pairing each new bud with its corresponding companion:
Companion nodes: Suppose we reach a companion node C C C with a split sequent ⊢ Γ , Δ \vdash\Gamma,\Delta ⊢ Γ , Δ , during the descending phase. Since R \mathcal{R} R is injective, there is a unique bud node B ∈ Bud ( D ) B\in\operatorname{Bud}(\mathcal{D}) B ∈ Bud ( D ) such that R ( B ) = C \mathcal{R}(B)=C R ( B ) = C , with an associated fixed-point variable X B X_B X B , as explained before. After solving the elementary CPRIS for the node C C C with rule r ( C ) r(C) r ( C ) , we are left with the solved CPRIS
P l ⊢ Γ , I P r ⊢ Δ , I ⊥ C u t ⊢ Γ , Δ ‾ \frac{\mathcal{P}^{l}}{\vdash\Gamma,I}
\qquad
\frac{\mathcal{P}^{r}}{\vdash\Delta,I^\bot}
\mathrel{\mathrm{Cut}}
\overline{\vdash\Gamma,\Delta} ⊢ Γ , I P l ⊢ Δ , I ⊥ P r Cut ⊢ Γ , Δ
where P l = ( D l , R l ) \mathcal{P}^{l}=(\mathcal{D}^{l},\mathcal{R}^{l}) P l = ( D l , R l ) and P r = ( D r , R r ) \mathcal{P}^{r}=(\mathcal{D}^{r},\mathcal{R}^{r}) P r = ( D r , R r ) are partial pre-proofs, and X B X_B X B is a free variable in I I I . We then transform the mentioned solved CPRIS into:
P l [ σ X B . I / X B ] ⊢ Γ , I [ σ X B . I / X B ] ⊢ Γ , σ X B . I ( σ ) P r [ σ ⊥ X B . I ⊥ / X B ] ⊢ Δ , I ⊥ [ σ ⊥ X B . I ⊥ / X B ] ⊢ Δ , σ ⊥ X B . I ⊥ ( σ ⊥ ) \frac{\mathcal{P}^{l}[\sigma X_B.I/X_B]}{\vdash\Gamma,I[\sigma X_B.I/X_B]}
\quad
\frac{\vdash\Gamma,\sigma X_B.I}{}
\quad(\sigma)
\qquad
\frac{\mathcal{P}^{r}[\sigma^\bot X_B.I^\bot/X_B]}{\vdash\Delta,I^\bot[\sigma^\bot X_B.I^\bot/X_B]}
\quad
\frac{\vdash\Delta,\sigma^\bot X_B.I^\bot}{}
\quad(\sigma^\bot) ⊢ Γ , I [ σ X B . I / X B ] P l [ σ X B . I / X B ] ⊢ Γ , σ X B . I ( σ ) ⊢ Δ , I ⊥ [ σ ⊥ X B . I ⊥ / X B ] P r [ σ ⊥ X B . I ⊥ / X B ] ⊢ Δ , σ ⊥ X B . I ⊥ ( σ ⊥ )
⊢ Γ , σ X B . I ⊢ Δ , σ ⊥ X B . I ⊥ ⊢ Γ , Δ ( C u t ) \frac{\vdash\Gamma,\sigma X_B.I \qquad\vdash\Delta,\sigma^\bot X_B.I^\bot}{\vdash\Gamma,\Delta}
\quad(\mathrm{Cut}) ⊢ Γ , Δ ⊢ Γ , σ X B . I ⊢ Δ , σ ⊥ X B . I ⊥ ( Cut )
where P l [ σ X B . I / X B ] \mathcal{P}^{l}[\sigma X_B.I/X_B] P l [ σ X B . I / X B ] (resp. P r [ σ ⊥ X B . I ⊥ / X B ] \mathcal{P}^{r}[\sigma^\bot X_B.I^\bot/X_B] P r [ σ ⊥ X B . I ⊥ / X B ] ) updates all occurrences of X B X_B X B in P l \mathcal{P}^{l} P l (resp. P r \mathcal{P}^{r} P r ) with σ X B . I \sigma X_B.I σ X B . I (resp. σ ⊥ X B . I ⊥ \sigma^\bot X_B.I^\bot σ ⊥ X B . I ⊥ ). Hence, this substitution also updates the bud node B l B^{l} B l labelled with ⊢ Γ , X B \vdash\Gamma,X_B ⊢ Γ , X B (resp. B r B^{r} B r labelled with ⊢ Δ , X B \vdash\Delta,X_B ⊢ Δ , X B ) in D l \mathcal{D}^{l} D l (resp. D r \mathcal{D}^{r} D r ) into ⊢ Γ , σ X B . I \vdash\Gamma,\sigma X_B.I ⊢ Γ , σ X B . I (resp. ⊢ Δ , σ ⊥ X B . I ⊥ \vdash\Delta,\sigma^\bot X_B.I^\bot ⊢ Δ , σ ⊥ X B . I ⊥ ). Consequently, in the function R l \mathcal{R}^{l} R l (resp. R r \mathcal{R}^{r} R r ), we pair the bud node B l B^{l} B l (resp. B r B^{r} B r ) with the left (resp. right) premise of the cut. Additionally, σ ∈ { μ , ν } \sigma\in\{\mu,\nu\} σ ∈ { μ , ν } with μ ⊥ = ν \mu^\bot=\nu μ ⊥ = ν and ν ⊥ = μ \nu^\bot=\mu ν ⊥ = μ . While the choice of the value to assign for σ \sigma σ is irrelevant if we were simply interpolating pre-proof (disregarding validity), it will on the contrary happen to be crucial to interpolate while preserving progress as mentioned at the start of the section. This condition will be studied in the proof of Theorem 37 .
Remark 36. Two crucial remarks are in order at this point: (i) while the process described above (that does not cope yet with the question of progress) does a global update of the proof when companions are encountered during the descending phase – via the substitution of the interpolating formula – the sequence of proofs corresponds locally to a cut-introduction: as a consequence, once all the substitution are introduced when the descending phase terminates, reversing the proof sequence provides a cut-elimination sequence . Moreover, (ii) the above-mentioned cut-elimination sequence happens to synchronise on the buds: this is the reason why circularity is preserved by cut-elimination in the next theorem.
The following theorem provides a detailed description of the properties of the resulting pre-proof after the interpolation process.
Theorem 37 (Proof-relevant interpolation with (semi-)analytic cuts for μ L L ω \mu\mathrm{LL}^{\omega} μ LL ω ). Let P 0 = ( D 0 , R 0 ) P_0 = (D_0, R_0) P 0 = ( D 0 , R 0 ) be a valid μ L L ω \mu\mathrm{LL}^{\omega} μ LL ω proof with a trace manifold and a split conclusion ⊢ Γ , Δ \vdash\Gamma, \Delta ⊢ Γ , Δ , such that every cut-inference is (semi-)analytic. After the ascending and descending phase on P 0 P_0 P 0 , we obtain an interpolated pre-proof
P = P l P r ⊢ Γ , Δ ( C u t ) \mathcal{P} = \frac{\mathcal{P}^{l}\quad\mathcal{P}^{r}}{\vdash\Gamma,\Delta}\;(\mathrm{Cut}) P = ⊢ Γ , Δ P l P r ( Cut ) satisfying that (i) P l = ( D l , R l ) \mathcal{P}^{l} = (D^{l}, R^{l}) P l = ( D l , R l ) and P r = ( D r , R r ) \mathcal{P}^{r} = (D^{r}, R^{r}) P r = ( D r , R r ) are two pre-proofs with conclusions ⊢ Γ , I \vdash\Gamma, I ⊢ Γ , I and ⊢ Δ , I ⊥ \vdash\Delta, I^{\perp} ⊢ Δ , I ⊥ respectively, where R l R^{l} R l (resp. R r R^{r} R r ) is injective and Voc ( I ) ⊆ Voc ( Γ ) ∩ Voc ( Δ ) \operatorname{Voc}(I) \subseteq\operatorname{Voc}(\Gamma) \cap\operatorname{Voc}(\Delta) Voc ( I ) ⊆ Voc ( Γ ) ∩ Voc ( Δ ) , (ii) P \mathcal{P} P is a valid proof, (iii)
P → C u t ω P 0 . \mathcal{P} \mathrel{\to_{\mathrm{Cut}}^{\omega}} P_0. P → Cut ω P 0 .
Proof. The full proof is given in Appendix C.5. We transform the trace manifold for P 0 P_0 P 0 into an ordered trace manifold using the Lemma 30 . Then using Algorithm 1 and Algorithm 2, we obtain a new proof that can be coherently decorated, and has an ordered trace manifold with a tree-compatible induction order. After the descending phase, the structural connectivity and induction order of P 0 P_0 P 0 are duplicated on both sides P l \mathcal{P}^{l} P l and P r \mathcal{P}^{r} P r , while each trace of its ordered trace manifold is transferred to exactly one side (Lemma 51 ). The remaining copies of the cycles are validated by traces following the interpolant. It suffices to choose each σ i \sigma_i σ i (resp. σ i ⊥ \sigma_i^{\perp} σ i ⊥ ) as ν \nu ν (resp. μ \mu μ ) opposite to the transfer side, since tree-compatibility guarantees the minimality condition (Theorem 52 ).
For the last part of the theorem one can perform an (infinite) cut-elimination reduction. Remark 36.(i) ensures that it consists in a cut-elimination reduction sequence while Remark 36.(ii) ensures that the cut-reduction can be performed by unfolding the circular proofs and preserving the synchronisation of the two circular proofs so that at the limit, the proof object is regular and is the infinite unfolding of the original circular proof P 0 P_0 P 0 . ◀
▶ Remark 38. Note that while usually μ L L ω \mu\mathrm{LL}^{\omega} μ LL ω circular proofs are not closed by cut-elimination [4 ] (the limit of the cut-reduction sequence may be non-regular), the structure of interpolants and interpolating proofs is such that this property holds. Intuitively, this comes from the fact that the constructions we perform essentially works by "infinitary" cut-introduction.
Conclusion
Summary of the paper. In this paper we established Maehara-Čubrić style interpolation lemmas and the respective interpolation theorems first for LL proofs with cuts (under adequate assumptions on the vocabularies used by the cut formulas) and for μ L L ω \mu\mathrm{LL}^{\omega} μ LL ω circular proofs (under the assumption that the circular proof has a trace manifold). The proof-relevance requirement forced us to be extra careful in the management of the proofs during interpolation. For instance, one cannot exploit a general cut-free completeness theorem coupled with a non-wellfounded proof regularisation result to consider only cut-free circular proofs. This is the reason why our result is subject to some assumptions on the class of proofs that we can interpolate, providing a further motivation for considering cuts in proofs during interpolation.
To our knowledge, our management of cuts extends the various existing (non proof-relevant) treatments of interpolation with cuts. Moreover, we discussed that proof-relevance allows to foresee generalisations of the results that exploit more precise information on how the cut-formulas are actually used in the proof.
Future works. First, we plan to analyse under which condition on the use of cut-formulas in axioms we can preserve the interpolation theorems of this paper.
Second, a crucial future work is about achieving this refinement of Čubrić’ results. In § 3.4 , we discussed the computational meaning of our results and how our treatment of cuts could allow extending Čubrić’ results in the simply-typed λ \lambda λ -calculus to non normal terms. We conjecture that our results can be used to strengthen Čubrić results for the simply typed λ \lambda λ -calculus. Still, extending our analysis of proof-relevant interpolation for non cut-free proofs to the simply-typed λ \lambda λ -calculus and natural deduction is not direct due to the fact that the notion of redexes lives at the meta-level which makes it more complex to handle7 .
Still, to support and illustrate our conjecture we show how it works on a simple example. Assume that we are given types A , B , C A, B, C A , B , C such that the atomic types of C C C all occurs in B B B , and assume that we have x : A ⊢ λ y . u : C → B x : A \vdash\lambda y.u : C \to B x : A ⊢ λ y . u : C → B and x : A ⊢ v : C x : A \vdash v : C x : A ⊢ v : C both normal. We easily see that one can interpolate ( λ y . u ) v (\lambda y.u)v ( λ y . u ) v without needing to normalize first: by Čubrić result there are (i) types I I I with atomic types occurring both in A A A and B B B and J J J with atomic types occurring both in A A A and C C C , (ii) normal terms x : A ⊢ u 1 : I x : A \vdash u_1 : I x : A ⊢ u 1 : I and z ′ : I ⊢ u 2 : C → B z' : I \vdash u_2 : C \to B z ′ : I ⊢ u 2 : C → B such that u 2 { u 1 / z ′ } → β ∗ u u_2\{u_1/z'\} \mathrel{\to_\beta^*} u u 2 { u 1 / z ′ } → β ∗ u and (iii) normal terms x : A ⊢ v 1 : J x : A \vdash v_1 : J x : A ⊢ v 1 : J and z ′ ′ : J ⊢ v 2 : C z'' : J \vdash v_2 : C z ′′ : J ⊢ v 2 : C such that v 2 { v 1 / z ′ ′ } → β ∗ v v_2\{v_1/z''\} \mathrel{\to_\beta^*} v v 2 { v 1 / z ′′ } → β ∗ v . By setting K = I × J K = I \times J K = I × J , t 1 = ( u 1 , v 1 ) t_1 = (u_1,v_1) t 1 = ( u 1 , v 1 ) and t 2 = λ z K . ( u 2 { π 1 ( z ) / z ′ } ) v 2 { π 2 ( z ) / z ′ ′ } t_2 = \lambda z^K.(u_2\{\pi_1(z)/z'\})v_2\{\pi_2(z)/z''\} t 2 = λ z K . ( u 2 { π 1 ( z ) / z ′ }) v 2 { π 2 ( z ) / z ′′ } , one easily sees that (i) the atomic types of K K K occur both in A A A and B B B , (ii) x : A ⊢ t 1 : K x : A \vdash t_1 : K x : A ⊢ t 1 : K , (iii) ⊢ t 2 : K → B \vdash t_2 : K \to B ⊢ t 2 : K → B and (iv) ( t 2 ) t 1 → β ∗ ( λ y . u ) v (t_2)t_1 \mathrel{\to_\beta^*} (\lambda y.u)v ( t 2 ) t 1 → β ∗ ( λ y . u ) v . The recent work of Lennon-Bertrand and Saurin who revisited Čubrić’ results using bidirectional typing systems to control the subformula property and better state the inductive assumption at the core of Čubrić’ proof may be a useful tool in this direction in order to control the language of the types involved in redexes.
Appendix on Section 2
LL cut-elimination reduction
Definition 39. Cut commutations or External reductions are defined in Figure 5 .
Figure 5. Cut-commutations rules, or External reduction rules, where r = ( e x t , F ) r = (\mathrm{ext}, F) r = ( ext , F ) and F F F is the principal occurrence.
Definition 40. LL cut-reduction are the reduction rules given in Figure 5 , Figure 6 together with the usual cut-elimination rule for first-order quantifiers and the following two reductions (together with their symmetrical cases):
Figure 6. Key cut-reduction rules, or Principal reductions, where r = ( p r i n c , { F , F ′ ⊥ } ) r = (\mathit{princ}, \{F,F'^{\perp}\}) r = ( princ , { F , F ′ ⊥ }) with { F , F ′ ⊥ } \{F,F'^{\perp}\} { F , F ′ ⊥ } the principal occurrences that have been reduced.
⊢ Σ , G ⊢ Δ , F , G ⊥ ⊢ Γ , F ⊥ ⊢ G ⊥ , Δ , Γ ( C u t ) ⊢ Σ , Δ , Γ ( C u t ) ⟶ r ⊢ Σ , G ⊢ Δ , F , G ⊥ ⊢ Σ , Δ , F ( C u t ) ⊢ Γ , F ⊥ ⊢ Σ , Δ , Γ ( C u t ) \frac{ \vdash\Sigma,G \qquad\frac{ \vdash\Delta,F,G^{\perp} \qquad\vdash\Gamma,F^{\perp}}{ \vdash G^{\perp},\Delta,\Gamma}\ (\mathrm{Cut})}{ \vdash\Sigma,\Delta,\Gamma}\ (\mathrm{Cut})
\;\longrightarrow_{r}\;
\frac{ \frac{ \vdash\Sigma,G \qquad\vdash\Delta,F,G^{\perp}}{ \vdash\Sigma,\Delta,F}\ (\mathrm{Cut}) \qquad\vdash\Gamma,F^{\perp}}{ \vdash\Sigma,\Delta,\Gamma}\ (\mathrm{Cut}) ⊢ Σ , Δ , Γ ⊢ Σ , G ⊢ G ⊥ , Δ , Γ ⊢ Δ , F , G ⊥ ⊢ Γ , F ⊥ ( Cut ) ( Cut ) ⟶ r ⊢ Σ , Δ , Γ ⊢ Σ , Δ , F ⊢ Σ , G ⊢ Δ , F , G ⊥ ( Cut ) ⊢ Γ , F ⊥ ( Cut )
⊢ F , F ⊥ ( A x ) π ⊢ F , Γ ⊢ F , Γ ( C u t ) ⟶ r π ⊢ F , Γ \frac{ \vdash F,F^{\perp}\ (\mathrm{Ax}) \qquad\frac{\pi}{\vdash F,\Gamma}}{ \vdash F,\Gamma}\ (\mathrm{Cut})
\;\longrightarrow_{r}\;
\frac{\pi}{\vdash F,\Gamma} ⊢ F , Γ ⊢ F , F ⊥ ( Ax ) ⊢ F , Γ π ( Cut ) ⟶ r ⊢ F , Γ π
Appendix on Section 3
Details on the Proof of the Main Lemma
We now provide the full proof of Lemma 15.
Axiom case
If π = ⊢ F , F ⊥ ‾ ( A x ) \pi= \overline{\vdash F,F^{\perp}}\ (\mathrm{Ax}) π = ⊢ F , F ⊥ ( Ax ) , one simply takes I = F ⊥ I = F^{\perp} I = F ⊥ , π 1 l = ⊢ F , F ⊥ ‾ ( A x ) \pi^{l}_{1} = \overline{\vdash F,F^{\perp}}\ (\mathrm{Ax}) π 1 l = ⊢ F , F ⊥ ( Ax ) and π 1 r = ⊢ F , F ⊥ ‾ ( A x ) \pi^{r}_{1} = \overline{\vdash F,F^{\perp}}\ (\mathrm{Ax}) π 1 r = ⊢ F , F ⊥ ( Ax ) . The cut between π 1 l \pi^{l}_{1} π 1 l and π 1 r \pi^{r}_{1} π 1 r reduces to π \pi π by one cut-axiom reduction step.
If π = ⊢ F , F ⊥ ‾ ( A x ) \pi= \overline{\vdash F,F^{\perp}}\ (\mathrm{Ax}) π = ⊢ F , F ⊥ ( Ax ) , the case is symmetrical taking I = F I = F I = F .
If π = ⊢ F , F ⊥ ‾ ( A x ) \pi= \overline{\vdash F,F^{\perp}}\ (\mathrm{Ax}) π = ⊢ F , F ⊥ ( Ax ) , one takes I = ⊥ I = \bot I = ⊥ , π 1 l = π ⊢ F , F ⊥ , ⊥ ( ⊥ ) \pi^{l}_{1} = \frac{\pi}{\vdash F,F^{\perp},\bot}\ (\bot) π 1 l = ⊢ F , F ⊥ , ⊥ π ( ⊥ ) and π 1 r = ⊢ 1 ‾ ( 1 ) \pi^{r}_{1} = \overline{\vdash1}\ (1) π 1 r = ⊢ 1 ( 1 ) . The cut of π 1 l \pi^{l}_{1} π 1 l and π 1 r \pi^{r}_{1} π 1 r reduces to π \pi π by a key 1 / ⊥ 1/\bot 1/⊥ case.
If π = ⊢ F , F ⊥ ‾ ( A x ) \pi= \overline{\vdash F,F^{\perp}}\ (\mathrm{Ax}) π = ⊢ F , F ⊥ ( Ax ) , the case is symmetrical to the previous one, taking I = 1 I = 1 I = 1 : π 1 l = ⊢ 1 ‾ ( 1 ) \pi^{l}_{1} = \overline{\vdash1}\ (1) π 1 l = ⊢ 1 ( 1 ) and π 1 r = π ⊢ F , F ⊥ , ⊥ ( ⊥ ) \pi^{r}_{1} = \frac{\pi}{\vdash F,F^{\perp},\bot}\ (\bot) π 1 r = ⊢ F , F ⊥ , ⊥ π ( ⊥ ) .
Cut rules
If the last rule is (Cut), with cut formulas in the right splitting ie. if
⊢ C ⊥ , Δ , F ⊢ C ⊥ , Δ , ? F ( ? d ) , ⊢ Σ , C ⊢ C ⊥ , Δ , ? F ⊢ Σ , Δ , ? F ( C u t ) ⟶ r ⊢ Σ , C ⊢ C ⊥ , Δ , F ⊢ Σ , Δ , F ( C u t ) , ⊢ Σ , Δ , F ⊢ Σ , Δ , ? F ( ? d ) ⊢ C ⊥ , Δ , ? F , ? F ⊢ C ⊥ , Δ , ? F ( ? c ) , ⊢ Σ , C ⊢ C ⊥ , Δ , ? F , ? F ⊢ Σ , Δ , ? F , ? F ( C u t ) ⟶ r ⊢ Σ , C ⊢ C ⊥ , Δ , ? F , ? F ⊢ Σ , Δ , ? F , ? F ( C u t ) , ⊢ Σ , Δ , ? F , ? F ⊢ Σ , Δ , ? F ( ? c ) ⊢ C ⊥ , Δ ⊢ C ⊥ , Δ , ? F ( ? w ) , ⊢ Σ , C ⊢ C ⊥ , Δ , ? F ⊢ Σ , Δ , ? F ( C u t ) ⟶ r ⊢ Σ , C ⊢ C ⊥ , Δ ⊢ Σ , Δ ( C u t ) , ⊢ Σ , Δ ⊢ Σ , ? F ( ? w ) ⊢ F , ? Γ , ? C ⊢ ! F , ? Γ , ? C ( ! p ) , ⊢ ! F , ? Γ , ? C ⊢ ! C ⊥ , ? Σ ⊢ ! F , ? Γ , ? Σ ( C u t ) ⟶ r ⊢ F , ? Γ ⊢ ! C ⊥ , ? Σ ⊢ F , ? Γ , ? Σ ( C u t ) , ⊢ F , ? Γ , ? Σ ⊢ ! F , ? Γ , ? Σ ( ! p ) ⊢ C , Δ , F ⊢ Γ , G ⊢ C , Δ , Γ , F ⊗ G ( ⊗ ) , ⊢ Σ , C ⊥ ⊢ C , Δ , Γ , F ⊗ G ⊢ Σ , Δ , Γ , F ⊗ G ( C u t ) ⟶ r ⊢ Σ , C ⊥ ⊢ C , Δ , F ⊢ Σ , Δ , F ( C u t ) , ⊢ Γ , G ⊢ Δ , Γ , F ⊗ G ( ⊗ ) ⊢ C , Δ , F , G ⊢ C , Δ , F ⊗ G ( ⊗ ) , ⊢ Σ , C ⊥ ⊢ C , Δ , F ⊗ G ⊢ Σ , Δ , F ⊗ G ( C u t ) ⟶ r ⊢ Σ , C ⊥ ⊢ C , Δ , F , G ⊢ Σ , Δ , F , G ( C u t ) , ⊢ Σ , Δ , F , G ⊢ Σ , Δ , F ⊗ G ( ⊗ ) ⊢ C , Δ , F ⊢ C , Δ , G ⊢ C , Δ , F & G ( & ) , ⊢ Σ , C ⊥ ⊢ C , Δ , F & G ⊢ Σ , Δ , F & G ( C u t ) ⟶ r ⊢ Σ , C ⊥ ⊢ C , Δ , F ⊢ Σ , Δ , F ( C u t ) , ⊢ Σ , C ⊥ ⊢ C , Δ , G ⊢ Σ , Δ , G ( C u t ) , ⊢ Σ , Δ , F ⊢ Σ , Δ , G ⊢ Σ , Δ , F & G ( & ) ⊢ C , Δ , F i ⊢ C , Δ , F 1 ⊕ F 2 ( ⊕ i ) , ⊢ Σ , C ⊥ ⊢ C , Δ , F 1 ⊕ F 2 ⊢ Σ , Δ , F 1 ⊕ F 2 ( C u t ) ⟶ r ⊢ Σ , C ⊥ ⊢ C , Δ , F i ⊢ Σ , Δ , F i ( C u t ) , ⊢ Σ , Δ , F i ⊢ Σ , Δ , F 1 ⊕ F 2 ( ⊕ i ) ⊢ C , Δ , ⊤ ⊢ Σ , Δ , ⊤ ( C u t ) ⟶ r ⊢ Σ , Δ , ⊤ ( ⊤ ) , ⊢ Σ , C ⊥ ⊢ C , Δ ⊢ Σ , Δ , ⊥ ( C u t ) ⟶ r ⊢ Σ , C ⊥ ⊢ C , Δ ⊢ Σ , Δ ( C u t ) , ⊢ Σ , Δ , ⊥ ⊢ Σ , Δ , ⊥ ( ⊥ ) \begin{aligned}
&\frac{\vdash C^{\perp},\Delta,F}{\vdash C^{\perp},\Delta,?F}\ (?d),\quad
\frac{\vdash\Sigma,C\qquad\vdash C^{\perp},\Delta,?F}{\vdash\Sigma,\Delta,?F}\ (\mathrm{Cut})
\ \longrightarrow_{r}\
\frac{\vdash\Sigma,C\qquad\vdash C^{\perp},\Delta,F}{\vdash\Sigma,\Delta,F}\ (\mathrm{Cut}),\quad
\frac{\vdash\Sigma,\Delta,F}{\vdash\Sigma,\Delta,?F}\ (?d)\\
&\frac{\vdash C^{\perp},\Delta,?F,?F}{\vdash C^{\perp},\Delta,?F}\ (?c),\quad
\frac{\vdash\Sigma,C\qquad\vdash C^{\perp},\Delta,?F,?F}{\vdash\Sigma,\Delta,?F,?F}\ (\mathrm{Cut})
\ \longrightarrow_{r}\
\frac{\vdash\Sigma,C\qquad\vdash C^{\perp},\Delta,?F,?F}{\vdash\Sigma,\Delta,?F,?F}\ (\mathrm{Cut}),\quad
\frac{\vdash\Sigma,\Delta,?F,?F}{\vdash\Sigma,\Delta,?F}\ (?c)\\
&\frac{\vdash C^{\perp},\Delta}{\vdash C^{\perp},\Delta,?F}\ (?w),\quad
\frac{\vdash\Sigma,C\qquad\vdash C^{\perp},\Delta,?F}{\vdash\Sigma,\Delta,?F}\ (\mathrm{Cut})
\ \longrightarrow_{r}\
\frac{\vdash\Sigma,C\qquad\vdash C^{\perp},\Delta}{\vdash\Sigma,\Delta}\ (\mathrm{Cut}),\quad
\frac{\vdash\Sigma,\Delta}{\vdash\Sigma,?F}\ (?w)\\
&\frac{\vdash F,? \Gamma,? C}{\vdash!F,? \Gamma,? C}\ (!p),\quad
\frac{\vdash!F,? \Gamma,? C\qquad\vdash!C^{\perp},?\Sigma}{\vdash!F,? \Gamma,?\Sigma}\ (\mathrm{Cut})
\ \longrightarrow_{r}\
\frac{\vdash F,? \Gamma\qquad\vdash!C^{\perp},?\Sigma}{\vdash F,? \Gamma,?\Sigma}\ (\mathrm{Cut}),\quad
\frac{\vdash F,? \Gamma,?\Sigma}{\vdash!F,? \Gamma,?\Sigma}\ (!p)\\
&\frac{\vdash C,\Delta,F\qquad\vdash\Gamma,G}{\vdash C,\Delta,\Gamma,F\otimes G}\ (\otimes),\quad
\frac{\vdash\Sigma,C^{\perp}\qquad\vdash C,\Delta,\Gamma,F\otimes G}{\vdash\Sigma,\Delta,\Gamma,F\otimes G}\ (\mathrm{Cut})
\ \longrightarrow_{r}\
\frac{\vdash\Sigma,C^{\perp}\qquad\vdash C,\Delta,F}{\vdash\Sigma,\Delta,F}\ (\mathrm{Cut}),\quad
\frac{\vdash\Gamma,G}{\vdash\Delta,\Gamma,F\otimes G}\ (\otimes)\\
&\frac{\vdash C,\Delta,F,G}{\vdash C,\Delta,F\otimes G}\ (\otimes),\quad
\frac{\vdash\Sigma,C^{\perp}\qquad\vdash C,\Delta,F\otimes G}{\vdash\Sigma,\Delta,F\otimes G}\ (\mathrm{Cut})
\ \longrightarrow_{r}\
\frac{\vdash\Sigma,C^{\perp}\qquad\vdash C,\Delta,F,G}{\vdash\Sigma,\Delta,F,G}\ (\mathrm{Cut}),\quad
\frac{\vdash\Sigma,\Delta,F,G}{\vdash\Sigma,\Delta,F\otimes G}\ (\otimes)\\
&\frac{\vdash C,\Delta,F\qquad\vdash C,\Delta,G}{\vdash C,\Delta,F\mathbin{\&}G}\ (\mathbin{\&}),\quad
\frac{\vdash\Sigma,C^{\perp}\qquad\vdash C,\Delta,F\mathbin{\&}G}{\vdash\Sigma,\Delta,F\mathbin{\&}G}\ (\mathrm{Cut})
\ \longrightarrow_{r}\
\frac{\vdash\Sigma,C^{\perp}\qquad\vdash C,\Delta,F}{\vdash\Sigma,\Delta,F}\ (\mathrm{Cut}),\quad
\frac{\vdash\Sigma,C^{\perp}\qquad\vdash C,\Delta,G}{\vdash\Sigma,\Delta,G}\ (\mathrm{Cut}),\quad
\frac{\vdash\Sigma,\Delta,F\qquad\vdash\Sigma,\Delta,G}{\vdash\Sigma,\Delta,F\mathbin{\&}G}\ (\mathbin{\&})\\
&\frac{\vdash C,\Delta,F_i}{\vdash C,\Delta,F_1\oplus F_2}\ (\oplus_i),\quad
\frac{\vdash\Sigma,C^{\perp}\qquad\vdash C,\Delta,F_1\oplus F_2}{\vdash\Sigma,\Delta,F_1\oplus F_2}\ (\mathrm{Cut})
\ \longrightarrow_{r}\
\frac{\vdash\Sigma,C^{\perp}\qquad\vdash C,\Delta,F_i}{\vdash\Sigma,\Delta,F_i}\ (\mathrm{Cut}),\quad
\frac{\vdash\Sigma,\Delta,F_i}{\vdash\Sigma,\Delta,F_1\oplus F_2}\ (\oplus_i)\\
&\frac{\vdash C,\Delta,\top}{\vdash\Sigma,\Delta,\top}\ (\mathrm{Cut})
\ \longrightarrow_{r}\
\vdash\Sigma,\Delta,\top\ (\top),\qquad
\frac{\vdash\Sigma,C^{\perp}\qquad\vdash C,\Delta}{\vdash\Sigma,\Delta,\bot}\ (\mathrm{Cut})
\ \longrightarrow_{r}\
\frac{\vdash\Sigma,C^{\perp}\qquad\vdash C,\Delta}{\vdash\Sigma,\Delta}\ (\mathrm{Cut}),\quad
\frac{\vdash\Sigma,\Delta,\bot}{\vdash\Sigma,\Delta,\bot}\ (\bot)
\end{aligned} ⊢ C ⊥ , Δ , ? F ⊢ C ⊥ , Δ , F ( ? d ) , ⊢ Σ , Δ , ? F ⊢ Σ , C ⊢ C ⊥ , Δ , ? F ( Cut ) ⟶ r ⊢ Σ , Δ , F ⊢ Σ , C ⊢ C ⊥ , Δ , F ( Cut ) , ⊢ Σ , Δ , ? F ⊢ Σ , Δ , F ( ? d ) ⊢ C ⊥ , Δ , ? F ⊢ C ⊥ , Δ , ? F , ? F ( ? c ) , ⊢ Σ , Δ , ? F , ? F ⊢ Σ , C ⊢ C ⊥ , Δ , ? F , ? F ( Cut ) ⟶ r ⊢ Σ , Δ , ? F , ? F ⊢ Σ , C ⊢ C ⊥ , Δ , ? F , ? F ( Cut ) , ⊢ Σ , Δ , ? F ⊢ Σ , Δ , ? F , ? F ( ? c ) ⊢ C ⊥ , Δ , ? F ⊢ C ⊥ , Δ ( ? w ) , ⊢ Σ , Δ , ? F ⊢ Σ , C ⊢ C ⊥ , Δ , ? F ( Cut ) ⟶ r ⊢ Σ , Δ ⊢ Σ , C ⊢ C ⊥ , Δ ( Cut ) , ⊢ Σ , ? F ⊢ Σ , Δ ( ? w ) ⊢ ! F , ? Γ , ? C ⊢ F , ? Γ , ? C ( ! p ) , ⊢ ! F , ? Γ , ? Σ ⊢ ! F , ? Γ , ? C ⊢ ! C ⊥ , ? Σ ( Cut ) ⟶ r ⊢ F , ? Γ , ? Σ ⊢ F , ? Γ ⊢ ! C ⊥ , ? Σ ( Cut ) , ⊢ ! F , ? Γ , ? Σ ⊢ F , ? Γ , ? Σ ( ! p ) ⊢ C , Δ , Γ , F ⊗ G ⊢ C , Δ , F ⊢ Γ , G ( ⊗ ) , ⊢ Σ , Δ , Γ , F ⊗ G ⊢ Σ , C ⊥ ⊢ C , Δ , Γ , F ⊗ G ( Cut ) ⟶ r ⊢ Σ , Δ , F ⊢ Σ , C ⊥ ⊢ C , Δ , F ( Cut ) , ⊢ Δ , Γ , F ⊗ G ⊢ Γ , G ( ⊗ ) ⊢ C , Δ , F ⊗ G ⊢ C , Δ , F , G ( ⊗ ) , ⊢ Σ , Δ , F ⊗ G ⊢ Σ , C ⊥ ⊢ C , Δ , F ⊗ G ( Cut ) ⟶ r ⊢ Σ , Δ , F , G ⊢ Σ , C ⊥ ⊢ C , Δ , F , G ( Cut ) , ⊢ Σ , Δ , F ⊗ G ⊢ Σ , Δ , F , G ( ⊗ ) ⊢ C , Δ , F & G ⊢ C , Δ , F ⊢ C , Δ , G ( & ) , ⊢ Σ , Δ , F & G ⊢ Σ , C ⊥ ⊢ C , Δ , F & G ( Cut ) ⟶ r ⊢ Σ , Δ , F ⊢ Σ , C ⊥ ⊢ C , Δ , F ( Cut ) , ⊢ Σ , Δ , G ⊢ Σ , C ⊥ ⊢ C , Δ , G ( Cut ) , ⊢ Σ , Δ , F & G ⊢ Σ , Δ , F ⊢ Σ , Δ , G ( & ) ⊢ C , Δ , F 1 ⊕ F 2 ⊢ C , Δ , F i ( ⊕ i ) , ⊢ Σ , Δ , F 1 ⊕ F 2 ⊢ Σ , C ⊥ ⊢ C , Δ , F 1 ⊕ F 2 ( Cut ) ⟶ r ⊢ Σ , Δ , F i ⊢ Σ , C ⊥ ⊢ C , Δ , F i ( Cut ) , ⊢ Σ , Δ , F 1 ⊕ F 2 ⊢ Σ , Δ , F i ( ⊕ i ) ⊢ Σ , Δ , ⊤ ⊢ C , Δ , ⊤ ( Cut ) ⟶ r ⊢ Σ , Δ , ⊤ ( ⊤ ) , ⊢ Σ , Δ , ⊥ ⊢ Σ , C ⊥ ⊢ C , Δ ( Cut ) ⟶ r ⊢ Σ , Δ ⊢ Σ , C ⊥ ⊢ C , Δ ( Cut ) , ⊢ Σ , Δ , ⊥ ⊢ Σ , Δ , ⊥ ( ⊥ )
In the promotion commutation case, the premisse in ⊢ ! C ⊥ , ? Σ \vdash!C^{\perp},?\Sigma ⊢ ! C ⊥ , ? Σ is assumed to end with a promotion rule.
\pi=
\frac{\frac{\vdash\Gamma_1,I_1}{\vdash\Gamma_1,\Delta_1,C}\;(\mathrm{Cut})\qquad\frac{\vdash I_1^\perp,\Delta_1,C}{\vdash\Gamma_1,\Delta_1,C}}
⊢ Γ 2 , I 2 ⊢ I 2 ⊥ , Δ 2 , C ⊥ ⊢ Γ 2 , Δ 2 , C ⊥ ( C u t ) ⊢ Γ 1 , Γ 2 , Δ 1 , Δ 2 ⊢ Γ 1 , Δ 1 , C ⊢ Γ 2 , Δ 2 , C ⊥ ( C u t ) \frac{\vdash\Gamma_2,I_2\qquad\vdash I_2^\perp,\Delta_2,C^\perp}{\vdash\Gamma_2,\Delta_2,C^\perp}\;(\mathrm{Cut})
\qquad
\frac{\vdash\Gamma_1,\Gamma_2,\Delta_1,\Delta_2}{\vdash\Gamma_1,\Delta_1,C\qquad\vdash\Gamma_2,\Delta_2,C^\perp}\;(\mathrm{Cut}) ⊢ Γ 2 , Δ 2 , C ⊥ ⊢ Γ 2 , I 2 ⊢ I 2 ⊥ , Δ 2 , C ⊥ ( Cut ) ⊢ Γ 1 , Δ 1 , C ⊢ Γ 2 , Δ 2 , C ⊥ ⊢ Γ 1 , Γ 2 , Δ 1 , Δ 2 ( Cut )
then setting I = I 1 ⊗ I 2 I=I_1\otimes I_2 I = I 1 ⊗ I 2 , π L = \pi^L= π L =
⊢ Γ 1 , I 1 ⊢ Γ 2 , I 2 ⊢ Γ 1 , Γ 2 , I 1 ⊗ I 2 ( ⊗ ) \frac{\vdash\Gamma_1,I_1\qquad\vdash\Gamma_2,I_2}{\vdash\Gamma_1,\Gamma_2,I_1\otimes I_2}\;(\otimes) ⊢ Γ 1 , Γ 2 , I 1 ⊗ I 2 ⊢ Γ 1 , I 1 ⊢ Γ 2 , I 2 ( ⊗ )
and π R = \pi^R= π R =
⊢ I 1 ⊥ , Δ 1 , C ⊢ I 2 ⊥ , Δ 2 , C ⊥ ⊢ I 1 ⊥ , I 2 ⊥ , Δ 1 , Δ 2 ( C u t ) ⊢ I 1 ⊥ , I 2 ⊥ , Δ 1 , Δ 2 ⊢ ( I 1 ⊗ I 2 ) ⊥ , Δ 1 , Δ 2 ( ⊗ ) \frac{\vdash I_1^\perp,\Delta_1,C\qquad\vdash I_2^\perp,\Delta_2,C^\perp}{\vdash I_1^\perp,I_2^\perp,\Delta_1,\Delta_2}\;(\mathrm{Cut})
\qquad
\frac{\vdash I_1^\perp,I_2^\perp,\Delta_1,\Delta_2}{\vdash(I_1\otimes I_2)^\perp,\Delta_1,\Delta_2}\;(\otimes) ⊢ I 1 ⊥ , I 2 ⊥ , Δ 1 , Δ 2 ⊢ I 1 ⊥ , Δ 1 , C ⊢ I 2 ⊥ , Δ 2 , C ⊥ ( Cut ) ⊢ ( I 1 ⊗ I 2 ) ⊥ , Δ 1 , Δ 2 ⊢ I 1 ⊥ , I 2 ⊥ , Δ 1 , Δ 2 ( ⊗ )
one gets a solved CPRIS π ′ \pi' π ′ (indeed, the constraint on the interpolant vocabulary is preserved as formulas C , C ⊥ C,C^\perp C , C ⊥ move from the right context to the right cut vocabulary C R C^R C R ) such that π ← c u t ∗ π ′ \pi\mathrel{\xleftarrow[\mathrm{cut}]{*}}\pi' π ∗ cut π ′ by a commutative reduction of (Cut) and a key ( ⊗ ) / ( ≯ ) (\otimes)/(\ngtr) ( ⊗ ) / ( ≯ ) case. The case when the cut formulas are colored left is symmetrical:
if π = ⊢ Γ 1 , C , I 1 ⊢ I 1 ⊥ , Δ 1 ⊢ Γ 1 , C , Δ 1 ( C u t ) ⊢ Γ 2 , C ⊥ , I 2 ⊢ I 2 ⊥ , Δ 2 ⊢ Γ 2 , C ⊥ , Δ 2 ( C u t ) ⊢ Γ 1 , Γ 2 , Δ 1 , Δ 2 ( C u t ) \text{if }\pi=
\frac{\frac{\vdash\Gamma_1,C,I_1\qquad\vdash I_1^\perp,\Delta_1}{\vdash\Gamma_1,C,\Delta_1}\;(\mathrm{Cut})
\qquad
\frac{\vdash\Gamma_2,C^\perp,I_2\qquad\vdash I_2^\perp,\Delta_2}{\vdash\Gamma_2,C^\perp,\Delta_2}\;(\mathrm{Cut})}
{\vdash\Gamma_1,\Gamma_2,\Delta_1,\Delta_2}\;(\mathrm{Cut}) if π = ⊢ Γ 1 , Γ 2 , Δ 1 , Δ 2 ⊢ Γ 1 , C , Δ 1 ⊢ Γ 1 , C , I 1 ⊢ I 1 ⊥ , Δ 1 ( Cut ) ⊢ Γ 2 , C ⊥ , Δ 2 ⊢ Γ 2 , C ⊥ , I 2 ⊢ I 2 ⊥ , Δ 2 ( Cut ) ( Cut )
then setting I = I 1 ≯ I 2 I=I_1\ngtr I_2 I = I 1 ≯ I 2 , π L = \pi^L= π L =
⊢ Γ 1 , C , I 1 ⊢ Γ 2 , C ⊥ , I 2 ⊢ Γ 1 , Γ 2 , I 1 , I 2 ( C u t ) ⊢ Γ 1 , Γ 2 , I 1 , I 2 ⊢ Δ 1 , Δ 2 , I 1 ≯ I 2 ( ≯ ) \frac{\vdash\Gamma_1,C,I_1\qquad\vdash\Gamma_2,C^\perp,I_2}{\vdash\Gamma_1,\Gamma_2,I_1,I_2}\;(\mathrm{Cut})
\qquad
\frac{\vdash\Gamma_1,\Gamma_2,I_1,I_2}{\vdash\Delta_1,\Delta_2,I_1\ngtr I_2}\;(\ngtr) ⊢ Γ 1 , Γ 2 , I 1 , I 2 ⊢ Γ 1 , C , I 1 ⊢ Γ 2 , C ⊥ , I 2 ( Cut ) ⊢ Δ 1 , Δ 2 , I 1 ≯ I 2 ⊢ Γ 1 , Γ 2 , I 1 , I 2 ( ≯ )
and π R = \pi^R= π R =
⊢ I 1 ⊥ , Δ 1 ⊢ I 2 ⊥ , Δ 2 ⊢ ( I 1 ≯ I 2 ) ⊥ , Δ 1 , Δ 2 ( ⊗ ) \frac{\vdash I_1^\perp,\Delta_1\qquad\vdash I_2^\perp,\Delta_2}{\vdash(I_1\ngtr I_2)^\perp,\Delta_1,\Delta_2}\;(\otimes) ⊢ ( I 1 ≯ I 2 ) ⊥ , Δ 1 , Δ 2 ⊢ I 1 ⊥ , Δ 1 ⊢ I 2 ⊥ , Δ 2 ( ⊗ )
one gets a solved CPRIS π ′ \pi' π ′ (indeed, the constraint on the interpolant vocabulary is preserved as formulas C , C ⊥ C,C^\perp C , C ⊥ move from the left context to the left cut vocabulary C L C^L C L ) such that π ← c u t ∗ π ′ \pi\mathrel{\xleftarrow[\mathrm{cut}]{*}}\pi' π ∗ cut π ′ by a commutative reduction of (Cut) and a key ( ⊗ ) / ( ≯ ) (\otimes)/(\ngtr) ( ⊗ ) / ( ≯ ) case.
Logical rules
We analyze the possible cases for a logical rule involved in an elementary CPRIS. Note that in each case, the principal formula may be part of the left or right part of the splitting; we treat only one case each time since the other is symmetrical be taking the dual interpolant and exchanging π L \pi^L π L and π R \pi^R π R .
If the last rule is ( ≯ ) (\ngtr) ( ≯ ) , that is if π = \pi= π =
⊢ Γ , I ⊢ I ⊥ , Δ , A , B ⊢ Γ , Δ , A , B ( C u t ) ⊢ Γ , Δ , A , B ⊢ Γ , Δ , A ≯ B ( ≯ ) \frac{\vdash\Gamma,I\qquad\vdash I^\perp,\Delta,A,B}{\vdash\Gamma,\Delta,A,B}\;(\mathrm{Cut})
\qquad
\frac{\vdash\Gamma,\Delta,A,B}{\vdash\Gamma,\Delta,A\ngtr B}\;(\ngtr) ⊢ Γ , Δ , A , B ⊢ Γ , I ⊢ I ⊥ , Δ , A , B ( Cut ) ⊢ Γ , Δ , A ≯ B ⊢ Γ , Δ , A , B ( ≯ )
then taking I ′ = I I'=I I ′ = I , π L = π L \pi^L=\pi^L π L = π L
⊢ Γ , I ⊢ Γ , I \frac{\vdash\Gamma,I}{\vdash\Gamma,I} ⊢ Γ , I ⊢ Γ , I
and π R = \pi^R= π R =
⊢ I ⊥ , Δ , A , B ⊢ I ⊥ , Δ , A ≯ B ( ≯ ) \frac{\vdash I^\perp,\Delta,A,B}{\vdash I^\perp,\Delta,A\ngtr B}\;(\ngtr) ⊢ I ⊥ , Δ , A ≯ B ⊢ I ⊥ , Δ , A , B ( ≯ )
we obtain a solved CPRIS π ′ \pi' π ′ such that π ← c u t π ′ \pi\mathrel{\xleftarrow[\mathrm{cut}]{}\pi'} π cut π ′ by a commutative reduction of (Cut) (Indeed, C L , C R C^L,C^R C L , C R are unchanged).
If the last rule is ( ⊗ ) (\otimes) ( ⊗ ) , that is if
π = ⊢ Γ 1 , I 1 ⊢ I 1 ⊥ , Δ 1 , A ⊢ Γ 1 , Δ 1 , A ( C u t ) ⊢ Γ 2 , I 2 ⊢ I 2 ⊥ , Δ 2 , B ⊢ Γ 2 , Δ 2 , B ( C u t ) ⊢ Γ 1 , Γ 2 , Δ 1 , Δ 2 , A ⊗ B ( ⊗ ) \pi=
\frac{\frac{\vdash\Gamma_1,I_1\qquad\vdash I_1^\perp,\Delta_1,A}{\vdash\Gamma_1,\Delta_1,A}\;(\mathrm{Cut})
\qquad
\frac{\vdash\Gamma_2,I_2\qquad\vdash I_2^\perp,\Delta_2,B}{\vdash\Gamma_2,\Delta_2,B}\;(\mathrm{Cut})}
{\vdash\Gamma_1,\Gamma_2,\Delta_1,\Delta_2,A\otimes B}\;(\otimes) π = ⊢ Γ 1 , Γ 2 , Δ 1 , Δ 2 , A ⊗ B ⊢ Γ 1 , Δ 1 , A ⊢ Γ 1 , I 1 ⊢ I 1 ⊥ , Δ 1 , A ( Cut ) ⊢ Γ 2 , Δ 2 , B ⊢ Γ 2 , I 2 ⊢ I 2 ⊥ , Δ 2 , B ( Cut ) ( ⊗ )
then setting I = I 1 ⊗ I 2 I=I_1\otimes I_2 I = I 1 ⊗ I 2 , π L = \pi^L= π L =
⊢ Γ 1 , I 1 ⊢ Γ 2 , I 2 ⊢ Γ 1 , Γ 2 , I 1 ⊗ I 2 ( ⊗ ) \frac{\vdash\Gamma_1,I_1\qquad\vdash\Gamma_2,I_2}{\vdash\Gamma_1,\Gamma_2,I_1\otimes I_2}\;(\otimes) ⊢ Γ 1 , Γ 2 , I 1 ⊗ I 2 ⊢ Γ 1 , I 1 ⊢ Γ 2 , I 2 ( ⊗ )
and π R = \pi^R= π R =
⊢ I 1 ⊥ , Δ 1 , A ⊢ I 2 ⊥ , Δ 2 , B ⊢ I 1 ⊥ , I 2 ⊥ , Δ 1 , Δ 2 , A ⊗ B ( ⊗ ) ⊢ I 1 ⊥ , I 2 ⊥ , Δ 1 , Δ 2 , A ⊗ B ⊢ ( I 1 ⊗ I 2 ) ⊥ , Δ 1 , Δ 2 , A ⊗ B ( ≯ ) \frac{\vdash I_1^\perp,\Delta_1,A\qquad\vdash I_2^\perp,\Delta_2,B}{\vdash I_1^\perp,I_2^\perp,\Delta_1,\Delta_2,A\otimes B}\;(\otimes)
\qquad
\frac{\vdash I_1^\perp,I_2^\perp,\Delta_1,\Delta_2,A\otimes B}{\vdash(I_1\otimes I_2)^\perp,\Delta_1,\Delta_2,A\otimes B}\;(\ngtr) ⊢ I 1 ⊥ , I 2 ⊥ , Δ 1 , Δ 2 , A ⊗ B ⊢ I 1 ⊥ , Δ 1 , A ⊢ I 2 ⊥ , Δ 2 , B ( ⊗ ) ⊢ ( I 1 ⊗ I 2 ) ⊥ , Δ 1 , Δ 2 , A ⊗ B ⊢ I 1 ⊥ , I 2 ⊥ , Δ 1 , Δ 2 , A ⊗ B ( ≯ )
one gets a solved CPRIS π ′ \pi' π ′ such that π ← c u t ∗ π ′ \pi\mathrel{\xleftarrow[\mathrm{cut}]{*}}\pi' π ∗ cut π ′ by a commutative reduction of (Cut) and a key ( ⊗ ) / ( ≯ ) (\otimes)/(\ngtr) ( ⊗ ) / ( ≯ ) case (Indeed, C L = C 1 L ∪ C 2 L C^{L}=C^{L}_{1}\cup C^{L}_{2} C L = C 1 L ∪ C 2 L and C R = C 1 R ∪ C 2 R C^{R}=C^{R}_{1}\cup C^{R}_{2} C R = C 1 R ∪ C 2 R ).
If the last rule is ( ⊥ ) (\perp) ( ⊥ ) , that is if
π = π 1 L π 1 R ⊢ Γ , I 1 ⊢ I 1 ⊥ , Δ ( C u t ) ⊢ Γ , Δ ⊢ ⊥ , Γ , Δ ( ⊥ ) , \pi=\frac{\pi^{L}_{1}\quad\pi^{R}_{1}}{\vdash\Gamma,I_{1}\quad\vdash I_{1}^{\perp},\Delta\;(\mathrm{Cut})}\quad\frac{\vdash\Gamma,\Delta}{\vdash\perp,\Gamma,\Delta}\;(\perp), π = ⊢ Γ , I 1 ⊢ I 1 ⊥ , Δ ( Cut ) π 1 L π 1 R ⊢⊥ , Γ , Δ ⊢ Γ , Δ ( ⊥ ) ,
by setting I = I 1 I=I_{1} I = I 1 , π L = π 1 L ⊢ Γ , I 1 ⊢ Γ , Δ ⊢ ⊥ , Γ , Δ ( ⊥ ) \pi^{L}=\frac{\pi^{L}_{1}}{\vdash\Gamma,I_{1}}\frac{\vdash\Gamma,\Delta}{\vdash\perp,\Gamma,\Delta}\;(\perp) π L = ⊢ Γ , I 1 π 1 L ⊢⊥ , Γ , Δ ⊢ Γ , Δ ( ⊥ ) and π R = π 1 R \pi^{R}=\pi^{R}_{1} π R = π 1 R , one gets a solved CPRIS π ′ \pi' π ′ = ⊢ ⊥ , Γ , I ⊢ I ⊥ , Δ ⊢ ⊥ , Γ , Δ ( C u t ) =\frac{\vdash\perp,\Gamma,I\quad\vdash I^{\perp},\Delta}{\vdash\perp,\Gamma,\Delta}\;(\mathrm{Cut}) = ⊢⊥ , Γ , Δ ⊢⊥ , Γ , I ⊢ I ⊥ , Δ ( Cut ) such that
π ′ ⟶ c u t π \pi'\longrightarrow_{\mathrm{cut}}\pi π ′ ⟶ cut π
by a cut-commutation case (Indeed, C L , C R C^{L},C^{R} C L , C R are unchanged).
If the last rule is ( 1 ) (1) ( 1 ) , that is if π = ⊢ 1 ‾ \pi=\overline{\vdash1} π = ⊢ 1 (which is indeed both an initial CPRIS and an elementary CPRIS), let I = 1 I=1 I = 1 and π 1 L = ⊢ 1 ‾ \pi^{L}_{1}=\overline{\vdash1} π 1 L = ⊢ 1 and π 1 R = ⊢ ⊥ , 1 ‾ ( A x ) \pi^{R}_{1}=\overline{\vdash\perp,1}\;(\mathrm{Ax}) π 1 R = ⊢⊥ , 1 ( Ax ) . One gets a solved CPRIS π ′ \pi' π ′ such that π ⟵ c u t ⋆ π ′ \pi\longleftarrow^{\star}_{\mathrm{cut}}\pi' π ⟵ cut ⋆ π ′ by a key 1 / ⊥ 1/\perp 1/ ⊥ case.
If the last rule is ( ⊕ i ) (\oplus_i) ( ⊕ i ) , i ∈ { 1 , 2 } i\in\{1,2\} i ∈ { 1 , 2 } , that is if
π = ⊢ F i , Γ , I 1 ⊢ I 1 ⊥ , Δ ⊢ F i , Γ , Δ ( C u t ) ⊢ F 1 ⊕ F 2 , Γ , Δ ( ⊕ i ) , \pi=\frac{\frac{\vdash F_i,\Gamma,I_1\quad\vdash I_1^\perp,\Delta}{\vdash F_i,\Gamma,\Delta}\;(\mathrm{Cut})}{\vdash F_1\oplus F_2,\Gamma,\Delta}\;(\oplus_i), π = ⊢ F 1 ⊕ F 2 , Γ , Δ ⊢ F i , Γ , Δ ⊢ F i , Γ , I 1 ⊢ I 1 ⊥ , Δ ( Cut ) ( ⊕ i ) ,
then setting I = I 1 I=I_1 I = I 1 , π L = π 1 L ⊢ F i , Γ , I 1 ⊢ F 1 ⊕ F 2 , Γ , I 1 ⊢ F 1 ⊕ F 2 , Γ , I 1 ( ⊕ i ) \pi^{L}=\frac{\pi^{L}_{1}}{\vdash F_i,\Gamma,I_1}\frac{\vdash F_1\oplus F_2,\Gamma,I_1}{\vdash F_1\oplus F_2,\Gamma,I_1}\;(\oplus_i) π L = ⊢ F i , Γ , I 1 π 1 L ⊢ F 1 ⊕ F 2 , Γ , I 1 ⊢ F 1 ⊕ F 2 , Γ , I 1 ( ⊕ i ) and π R = π 1 R \pi^{R}=\pi^{R}_{1} π R = π 1 R , one gets a solved CPRIS π ′ \pi' π ′ such that π ⟵ c u t ⋆ π ′ \pi\longleftarrow^{\star}_{\mathrm{cut}}\pi' π ⟵ cut ⋆ π ′ by a commutative reduction of (Cut) (Indeed, C L , C R C^{L},C^{R} C L , C R are unchanged).
If the last rule is ( & ) (\&) ( & ) , that is if
π = ⊢ F , Γ , I 1 ⊢ I 1 ⊥ , Δ ⊢ F , Γ , Δ ( C u t ) ⊢ G , Γ , I 2 ⊢ I 2 ⊥ , Δ ⊢ G , Γ , Δ ( C u t ) ⊢ F & G , Γ , Δ ( & ) \pi=\frac{\frac{\vdash F,\Gamma,I_1\quad\vdash I_1^\perp,\Delta}{\vdash F,\Gamma,\Delta}\;(\mathrm{Cut})\qquad\frac{\vdash G,\Gamma,I_2\quad\vdash I_2^\perp,\Delta}{\vdash G,\Gamma,\Delta}\;(\mathrm{Cut})}{\vdash F\&G,\Gamma,\Delta}\;(\&) π = ⊢ F & G , Γ , Δ ⊢ F , Γ , Δ ⊢ F , Γ , I 1 ⊢ I 1 ⊥ , Δ ( Cut ) ⊢ G , Γ , Δ ⊢ G , Γ , I 2 ⊢ I 2 ⊥ , Δ ( Cut ) ( & )
then setting I = I 1 ⊕ I 2 I=I_1\oplus I_2 I = I 1 ⊕ I 2 and
π L = ⊢ F , Γ , I 1 ⊢ F , Γ , I 1 ⊕ I 2 ( ⊕ 0 ) ⊢ G , Γ , I 2 ⊢ G , Γ , I 1 ⊕ I 2 ( ⊕ 1 ) ⊢ F & G , Γ , I ( & ) \pi^{L}=\frac{\frac{\vdash F,\Gamma,I_1}{\vdash F,\Gamma,I_1\oplus I_2}\;(\oplus_0)\qquad\frac{\vdash G,\Gamma,I_2}{\vdash G,\Gamma,I_1\oplus I_2}\;(\oplus_1)}{\vdash F\&G,\Gamma,I}\;(\&) π L = ⊢ F & G , Γ , I ⊢ F , Γ , I 1 ⊕ I 2 ⊢ F , Γ , I 1 ( ⊕ 0 ) ⊢ G , Γ , I 1 ⊕ I 2 ⊢ G , Γ , I 2 ( ⊕ 1 ) ( & )
and
π R = ⊢ I 1 ⊥ , Δ ⊢ I 2 ⊥ , Δ ⊢ I 1 ⊥ ≯ I 2 ⊥ , Δ ( & ) , \pi^{R}=\frac{\frac{\vdash I_1^\perp,\Delta\qquad\vdash I_2^\perp,\Delta}{\vdash I_1^\perp\mathbin{\ngtr}I_2^\perp,\Delta}\;(\&)}, π R = , ⊢ I 1 ⊥ ≯ I 2 ⊥ , Δ ⊢ I 1 ⊥ , Δ ⊢ I 2 ⊥ , Δ ( & )
one observes that:
π L π R ⊢ Γ , I ⊢ I ⊥ , Δ ⊢ Γ , Δ ( C u t ) ⟶ c u t ⊢ F , Γ , I 1 ⊢ F , Γ , I ( ⊕ 0 ) ⊢ I ⊥ , Δ ⊢ F , Γ , Δ ( C u t ) ⊢ G , Γ , I 2 ⊢ G , Γ , I ( ⊕ 1 ) ⊢ I ⊥ , Δ ⊢ G , Γ , Δ ( C u t ) ⊢ F & G , Γ , Δ ( & ) \frac{\pi^{L}\qquad\pi^{R}}{\vdash\Gamma,I\qquad\vdash I^\perp,\Delta}{\vdash\Gamma,\Delta}\;(\mathrm{Cut})
\longrightarrow_{\mathrm{cut}}
\frac{\frac{\vdash F,\Gamma,I_1}{\vdash F,\Gamma,I}\;(\oplus_0)\qquad\frac{\vdash I^\perp,\Delta}{\vdash F,\Gamma,\Delta}\;(\mathrm{Cut})\qquad\frac{\vdash G,\Gamma,I_2}{\vdash G,\Gamma,I}\;(\oplus_1)\qquad\frac{\vdash I^\perp,\Delta}{\vdash G,\Gamma,\Delta}\;(\mathrm{Cut})}{\vdash F\&G,\Gamma,\Delta}\;(\&) ⊢ Γ , I ⊢ I ⊥ , Δ π L π R ⊢ Γ , Δ ( Cut ) ⟶ cut ⊢ F & G , Γ , Δ ⊢ F , Γ , I ⊢ F , Γ , I 1 ( ⊕ 0 ) ⊢ F , Γ , Δ ⊢ I ⊥ , Δ ( Cut ) ⊢ G , Γ , I ⊢ G , Γ , I 2 ( ⊕ 1 ) ⊢ G , Γ , Δ ⊢ I ⊥ , Δ ( Cut ) ( & )
⟶ c u t ⊢ F , Γ , I 1 ⊢ I 1 ⊥ , Δ ⊢ F , Γ , Δ ( C u t ) ⊢ G , Γ , I 2 ⊢ G , Γ , I ( ⊕ 1 ) ⊢ I ⊥ , Δ ⊢ G , Γ , Δ ( C u t ) ⊢ F & G , Γ , Δ ( & ) \longrightarrow_{\mathrm{cut}}
\frac{\frac{\vdash F,\Gamma,I_1\quad\vdash I_1^\perp,\Delta}{\vdash F,\Gamma,\Delta}\;(\mathrm{Cut})\qquad\frac{\vdash G,\Gamma,I_2}{\vdash G,\Gamma,I}\;(\oplus_1)\qquad\frac{\vdash I^\perp,\Delta}{\vdash G,\Gamma,\Delta}\;(\mathrm{Cut})}{\vdash F\&G,\Gamma,\Delta}\;(\&) ⟶ cut ⊢ F & G , Γ , Δ ⊢ F , Γ , Δ ⊢ F , Γ , I 1 ⊢ I 1 ⊥ , Δ ( Cut ) ⊢ G , Γ , I ⊢ G , Γ , I 2 ( ⊕ 1 ) ⊢ G , Γ , Δ ⊢ I ⊥ , Δ ( Cut ) ( & )
⟶ c u t ⊢ F , Γ , I 1 ⊢ I 1 ⊥ , Δ ⊢ F , Γ , Δ ( C u t ) ⊢ G , Γ , I 2 ⊢ I 2 ⊥ , Δ ⊢ G , Γ , Δ ( C u t ) ⊢ F & G , Γ , Δ ( & ) = π \longrightarrow_{\mathrm{cut}}
\frac{\frac{\vdash F,\Gamma,I_1\quad\vdash I_1^\perp,\Delta}{\vdash F,\Gamma,\Delta}\;(\mathrm{Cut})\qquad\frac{\vdash G,\Gamma,I_2\quad\vdash I_2^\perp,\Delta}{\vdash G,\Gamma,\Delta}\;(\mathrm{Cut})}{\vdash F\&G,\Gamma,\Delta}\;(\&)=\pi ⟶ cut ⊢ F & G , Γ , Δ ⊢ F , Γ , Δ ⊢ F , Γ , I 1 ⊢ I 1 ⊥ , Δ ( Cut ) ⊢ G , Γ , Δ ⊢ G , Γ , I 2 ⊢ I 2 ⊥ , Δ ( Cut ) ( & ) = π
so one gets a solved CPRIS (Indeed, C L = C 1 L ∪ C 2 L C^{L}=C^{L}_{1}\cup C^{L}_{2} C L = C 1 L ∪ C 2 L and C R = C 1 R ∪ C 2 R C^{R}=C^{R}_{1}\cup C^{R}_{2} C R = C 1 R ∪ C 2 R ) such that π \pi π indeed cut-expands to π ′ \pi' π ′ .
If the last rule is ( ⊤ ) (\top) ( ⊤ ) , that is if π = ⊢ ⊤ , Γ , Δ ‾ ( ⊤ ) \pi=\overline{\vdash\top,\Gamma,\Delta}\;(\top) π = ⊢ ⊤ , Γ , Δ ( ⊤ ) , by setting I = 0 I=0 I = 0 , π L = ⊢ ⊤ , Γ , 0 ‾ \pi^{L}=\overline{\vdash\top,\Gamma,0} π L = ⊢ ⊤ , Γ , 0 and π R = ⊢ ⊤ , Δ ‾ \pi^{R}=\overline{\vdash\top,\Delta} π R = ⊢ ⊤ , Δ , then
π ′ = π L π R ⊢ ⊤ , Γ , Δ ( C u t ) \pi'=\frac{\pi^{L}\qquad\pi^{R}}{\vdash\top,\Gamma,\Delta}\;(\mathrm{Cut}) π ′ = ⊢ ⊤ , Γ , Δ π L π R ( Cut )
is a solved CPRIS to which π \pi π cut-expands by a ( ⊤ ) (\top) ( ⊤ ) commutation case.
**If the last rule is ( ? d ) (?d) ( ? d ) , that is π = \pi= π =
π 1 L π 1 R ⊢ F , Γ , I 1 ⊢ I 1 ⊥ , Δ ( C u t ) ⊢ F , Γ , Δ ( ? d ) \frac{\pi_1^L \quad\pi_1^R}{\frac{\vdash F,\Gamma,I_1 \quad\vdash I_1^\perp,\Delta\;(\mathrm{Cut})}{\vdash F,\Gamma,\Delta}\;(?d)} ⊢ F , Γ , Δ ⊢ F , Γ , I 1 ⊢ I 1 ⊥ , Δ ( Cut ) ( ? d ) π 1 L π 1 R
then setting I = I 1 I=I_1 I = I 1 , π L = \pi^L= π L =
π 1 L ⊢ F , Γ , I 1 ⊢ ? F , Γ , I 1 ( ? d ) \frac{\pi_1^L}{\frac{\vdash F,\Gamma,I_1}{\vdash?F,\Gamma,I_1}\;(?d)} ⊢ ? F , Γ , I 1 ⊢ F , Γ , I 1 ( ? d ) π 1 L
and π R = π 1 R \pi^R=\pi_1^R π R = π 1 R one gets:
π 1 L ⊢ ? F , Γ , I 1 π 1 R ⊢ I 1 ⊥ , Δ ( C u t ) ⊢ ? F , Γ , Δ ⟶ c u t π 1 L ⊢ F , Γ , I 1 π 1 R ⊢ I 1 ⊥ , Δ ( C u t ) ⊢ F , Γ , Δ ⊢ ? F , Γ , Δ ( ? d ) = π \frac{\frac{\pi_1^L}{\vdash?F,\Gamma,I_1}\quad\frac{\pi_1^R}{\vdash I_1^\perp,\Delta}\;(\mathrm{Cut})}{\vdash?F,\Gamma,\Delta}
\quad\longrightarrow_{\mathrm{cut}}\quad
\frac{\frac{\pi_1^L}{\vdash F,\Gamma,I_1}\quad\frac{\pi_1^R}{\vdash I_1^\perp,\Delta}\;(\mathrm{Cut})}{\frac{\vdash F,\Gamma,\Delta}{\vdash?F,\Gamma,\Delta}\;(?d)}
=\pi ⊢ ? F , Γ , Δ ⊢ ? F , Γ , I 1 π 1 L ⊢ I 1 ⊥ , Δ π 1 R ( Cut ) ⟶ cut ⊢ ? F , Γ , Δ ⊢ F , Γ , Δ ( ? d ) ⊢ F , Γ , I 1 π 1 L ⊢ I 1 ⊥ , Δ π 1 R ( Cut ) = π
so one gets a solved CPRIS (indeed, C L , C R C^L,C^R C L , C R are unchanged) such that π \pi π indeed cut-expands to π ′ \pi' π ′ .
**If the last rule is ( ! p ) (!p) ( ! p ) , that is π = \pi= π =
π 1 L π 1 R ⊢ F , ? Γ , I 1 ⊢ I 1 ⊥ , ? Δ ( C u t ) ⊢ F , ? Γ , ? Δ ( ! p ) \frac{\pi_1^L \quad\pi_1^R}{\frac{\vdash F,? \Gamma,I_1 \quad\vdash I_1^\perp,? \Delta\;(\mathrm{Cut})}{\vdash F,? \Gamma,? \Delta}\;(!p)} ⊢ F , ? Γ , ? Δ ⊢ F , ? Γ , I 1 ⊢ I 1 ⊥ , ? Δ ( Cut ) ( ! p ) π 1 L π 1 R
then by setting I = ? I 1 I=?I_1 I = ? I 1 , π L = \pi^L= π L =
π 1 L ⊢ F , ? Γ , I 1 ( ? d ) ⊢ F , ? Γ , ? I 1 ⊢ ! F , ? Γ , ? I 1 ( ! p ) \frac{\frac{\pi_1^L}{\vdash F,? \Gamma,I_1}\;(?d)}{\frac{\vdash F,? \Gamma,? I_1}{\vdash!F,? \Gamma,? I_1}\;(!p)} ⊢ ! F , ? Γ , ? I 1 ⊢ F , ? Γ , ? I 1 ( ! p ) ⊢ F , ? Γ , I 1 π 1 L ( ? d )
and π R = \pi^R= π R =
π 1 R ⊢ I 1 ⊥ , ? Δ ( ! p ) \frac{\pi_1^R}{\vdash I_1^\perp,? \Delta}\;(!p) ⊢ I 1 ⊥ , ? Δ π 1 R ( ! p )
one gets:
⊢ ! F , ? Γ , ? I 1 ⊢ ! I 1 ⊥ , ? Δ ( C u t ) ⊢ ! F , ? Γ , ? Δ \frac{\frac{\vdash!F,? \Gamma,? I_1}{\vdash!I_1^\perp,? \Delta}\;(\mathrm{Cut})}{\vdash!F,? \Gamma,? \Delta} ⊢ ! F , ? Γ , ? Δ ⊢ ! I 1 ⊥ , ? Δ ⊢ ! F , ? Γ , ? I 1 ( Cut )
⟶ c u t π 1 L ⊢ F , ? Γ , I 1 ( ? d ) ⊢ F , ? Γ , ? I 1 π 1 R ⊢ I 1 ⊥ , ? Δ ( ! p ) ⊢ I ⊥ , ? Δ ( C u t ) ⊢ F , ? Γ , ? Δ ⊢ ! F , ? Γ , ? Δ ( ! p ) \longrightarrow_{\mathrm{cut}}
\frac{\frac{\pi_1^L}{\vdash F,? \Gamma,I_1}\;(?d)}{\vdash F,? \Gamma,? I_1}
\quad
\frac{\frac{\pi_1^R}{\vdash I_1^\perp,? \Delta}\;(!p)}{\vdash I^\perp,? \Delta}\;(\mathrm{Cut})
\quad
\frac{\vdash F,? \Gamma,? \Delta}{\vdash!F,? \Gamma,? \Delta}\;(!p) ⟶ cut ⊢ F , ? Γ , ? I 1 ⊢ F , ? Γ , I 1 π 1 L ( ? d ) ⊢ I ⊥ , ? Δ ⊢ I 1 ⊥ , ? Δ π 1 R ( ! p ) ( Cut ) ⊢ ! F , ? Γ , ? Δ ⊢ F , ? Γ , ? Δ ( ! p )
⟶ c u t π 1 L ⊢ F , ? Γ , I 1 π 1 R ⊢ I 1 ⊥ , ? Δ ( C u t ) ⊢ F , ? Γ , ? Δ ⊢ ! F , ? Γ , ? Δ ( ! p ) = π \longrightarrow_{\mathrm{cut}}
\frac{\frac{\pi_1^L}{\vdash F,? \Gamma,I_1}\quad\frac{\pi_1^R}{\vdash I_1^\perp,? \Delta}\;(\mathrm{Cut})}{\frac{\vdash F,? \Gamma,? \Delta}{\vdash!F,? \Gamma,? \Delta}\;(!p)}
=\pi ⟶ cut ⊢ ! F , ? Γ , ? Δ ⊢ F , ? Γ , ? Δ ( ! p ) ⊢ F , ? Γ , I 1 π 1 L ⊢ I 1 ⊥ , ? Δ π 1 R ( Cut ) = π
so one gets a solved CPRIS (indeed, C L , C R C^L,C^R C L , C R are unchanged) such that π \pi π indeed cut-expands to π ′ \pi' π ′ .
**If the last rule is ( ? w ) (?w) ( ? w ) , that is π = \pi= π =
π 1 L π 1 R ⊢ Γ , I 1 ⊢ I 1 ⊥ , Δ ( C u t ) ⊢ Γ , Δ ( ? w ) \frac{\pi_1^L \quad\pi_1^R}{\frac{\vdash\Gamma,I_1 \quad\vdash I_1^\perp,\Delta\;(\mathrm{Cut})}{\vdash\Gamma,\Delta}\;(?w)} ⊢ Γ , Δ ⊢ Γ , I 1 ⊢ I 1 ⊥ , Δ ( Cut ) ( ? w ) π 1 L π 1 R
then by setting I = I 1 I=I_1 I = I 1 , π L = \pi^L= π L =
π 1 L ⊢ Γ , I 1 ⊢ ? F , Γ , I 1 ( ? w ) \frac{\pi_1^L}{\frac{\vdash\Gamma,I_1}{\vdash?F,\Gamma,I_1}\;(?w)} ⊢ ? F , Γ , I 1 ⊢ Γ , I 1 ( ? w ) π 1 L
and π R = π 1 R \pi^R=\pi_1^R π R = π 1 R , we ensure that, since Voc ( I ) ⊆ ( Voc ( Γ ) ∪ C L ) ∩ ( Voc ( Δ ) ∪ C R ) \operatorname{Voc}(I)\subseteq(\operatorname{Voc}(\Gamma)\cup C^L)\cap(\operatorname{Voc}(\Delta)\cup C^R) Voc ( I ) ⊆ ( Voc ( Γ ) ∪ C L ) ∩ ( Voc ( Δ ) ∪ C R ) ,
\pi'=\frac{\frac{\vdash?F,\Gamma,I\quad\vdash I^\perp,\Delta\;(\mathrm{Cut})}{\vdash?F,\Gamma,\Delta}}
is indeed a solved CPRIS (indeed, C L , C R C^L,C^R C L , C R are unchanged) and that
π ′ ⟶ c u t ⊢ Γ , I 1 ⊢ I 1 ⊥ , Δ ( C u t ) ⊢ Γ , Δ ( ? w ) ⊢ ? F , Γ , Δ = π . \pi'\longrightarrow_{\mathrm{cut}}
\frac{\frac{\vdash\Gamma,I_1\quad\vdash I_1^\perp,\Delta\;(\mathrm{Cut})}{\vdash\Gamma,\Delta}\;(?w)}
{\vdash?F,\Gamma,\Delta}
=\pi. π ′ ⟶ cut ⊢ ? F , Γ , Δ ⊢ Γ , Δ ⊢ Γ , I 1 ⊢ I 1 ⊥ , Δ ( Cut ) ( ? w ) = π .
by a cut commutation rule so that π \pi π indeed cut-expands to π ′ \pi' π ′ as expected.
If the last rule is ( ? c ) (?c) ( ? c ) , that is π = \pi= π =
⊢ ? F , ? F , Γ , I 1 ⊢ I 1 ⊥ , Δ ⊢ ? F , ? F , Γ , Δ ( C u t ) ⊢ ? F , Γ , Δ ( ? c ) \frac{\frac{\vdash?F,?F,\Gamma,I_1 \qquad\vdash I_1^\perp,\Delta}{\vdash?F,?F,\Gamma,\Delta}\ (\mathrm{Cut})}{\vdash?F,\Gamma,\Delta}\ (?c) ⊢ ? F , Γ , Δ ⊢ ? F , ? F , Γ , Δ ⊢ ? F , ? F , Γ , I 1 ⊢ I 1 ⊥ , Δ ( Cut ) ( ? c )
by setting I = I 1 I=I_1 I = I 1 , π L = \pi^L= π L =
⊢ ? F , ? F , Γ , I 1 ⊢ ? F , Γ , I ( ? c ) \frac{\vdash?F,?F,\Gamma,I_1}{\vdash?F,\Gamma,I}\ (?c) ⊢ ? F , Γ , I ⊢ ? F , ? F , Γ , I 1 ( ? c )
and π R = π 1 R \pi^R=\pi^R_1 π R = π 1 R one gets:
π ′ = ⊢ ? F , ? F , Γ , I 1 ⊢ I 1 ⊥ , Δ ⊢ ? F , ? F , Γ , Δ ( C u t ) ⊢ ? F , Γ , Δ ( ? c ) ⟶ c u t ⊢ ? F , ? F , Γ , I 1 ⊢ I 1 ⊥ , Δ ⊢ ? F , ? F , Γ , Δ ( C u t ) ⊢ ? F , Γ , Δ ( ? c ) = π \pi'=\frac{\frac{\vdash?F,?F,\Gamma,I_1 \qquad\vdash I_1^\perp,\Delta}{\vdash?F,?F,\Gamma,\Delta}\ (\mathrm{Cut})}{\vdash?F,\Gamma,\Delta}\ (?c)
\longrightarrow_{\mathrm{cut}}
\frac{\frac{\vdash?F,?F,\Gamma,I_1 \qquad\vdash I_1^\perp,\Delta}{\vdash?F,?F,\Gamma,\Delta}\ (\mathrm{Cut})}{\vdash?F,\Gamma,\Delta}\ (?c)
=\pi π ′ = ⊢ ? F , Γ , Δ ⊢ ? F , ? F , Γ , Δ ⊢ ? F , ? F , Γ , I 1 ⊢ I 1 ⊥ , Δ ( Cut ) ( ? c ) ⟶ cut ⊢ ? F , Γ , Δ ⊢ ? F , ? F , Γ , Δ ⊢ ? F , ? F , Γ , I 1 ⊢ I 1 ⊥ , Δ ( Cut ) ( ? c ) = π
so that π ′ \pi' π ′ is indeed a solved CPRIS (Indeed, C L , C R C^L,C^R C L , C R are unchanged) to which π \pi π cut-expands.
If the last rule is ( ∀ ) (\forall) ( ∀ ) , that is π = \pi= π =
⊢ F , Γ , I 1 ⊢ I 1 ⊥ , Δ ⊢ F , Γ , Δ ( C u t ) ⊢ ∀ x F , Γ , Δ ( ∀ ) x ∉ FV ( Γ , Δ ) \frac{\frac{\vdash F,\Gamma,I_1 \qquad\vdash I_1^\perp,\Delta}{\vdash F,\Gamma,\Delta}\ (\mathrm{Cut})}{\vdash\forall xF,\Gamma,\Delta}\ (\forall)\quad x\notin\operatorname{FV}(\Gamma,\Delta) ⊢ ∀ x F , Γ , Δ ⊢ F , Γ , Δ ⊢ F , Γ , I 1 ⊢ I 1 ⊥ , Δ ( Cut ) ( ∀ ) x ∈ / FV ( Γ , Δ )
with Voc ( I 1 ) ⊆ ( Voc ( F , Γ ) ∪ C L ) ∩ ( Voc ( Δ ) ∪ C R ) \operatorname{Voc}(I_1)\subseteq(\operatorname{Voc}(F,\Gamma)\cup C^L)\cap(\operatorname{Voc}(\Delta)\cup C^R) Voc ( I 1 ) ⊆ ( Voc ( F , Γ ) ∪ C L ) ∩ ( Voc ( Δ ) ∪ C R ) , by setting I = ∃ x I 1 I=\exists xI_1 I = ∃ x I 1 , π L = \pi^L= π L =
⊢ F , Γ , I 1 ⊢ F , Γ , ∃ x I 1 ( ∃ ) ⊢ ∀ x F , Γ , ∃ x . I 1 ( ∀ ) \frac{\frac{\vdash F,\Gamma,I_1}{\vdash F,\Gamma,\exists xI_1}\ (\exists)}{\vdash\forall xF,\Gamma,\exists x.I_1}\ (\forall) ⊢ ∀ x F , Γ , ∃ x . I 1 ⊢ F , Γ , ∃ x I 1 ⊢ F , Γ , I 1 ( ∃ ) ( ∀ )
and π R = \pi^R= π R =
⊢ I 1 ⊥ , Δ ⊢ ∀ x I 1 ⊥ , Δ ( ∀ ) \frac{\vdash I_1^\perp,\Delta}{\vdash\forall xI_1^\perp,\Delta}\ (\forall) ⊢ ∀ x I 1 ⊥ , Δ ⊢ I 1 ⊥ , Δ ( ∀ )
one gets that:
π ′ = ⊢ ∀ x F , Γ , ∃ x . I 1 ⊢ ∀ x I 1 ⊥ , Δ ⊢ ∀ x F , Γ , Δ ( C u t ) \pi'=\frac{\frac{\vdash\forall xF,\Gamma,\exists x.I_1 \qquad\vdash\forall xI_1^\perp,\Delta}{\vdash\forall xF,\Gamma,\Delta}\ (\mathrm{Cut})}{} π ′ = ⊢ ∀ x F , Γ , Δ ⊢ ∀ x F , Γ , ∃ x . I 1 ⊢ ∀ x I 1 ⊥ , Δ ( Cut )
is a solved CPRIS (indeed, C L , C R C^L,C^R C L , C R are unchanged) to which π \pi π cut-expands:
π ′ ⟶ c u t ⊢ F , Γ , I 1 ⊢ F , Γ , ∃ x . I 1 ( ∃ ) ⊢ I 1 ⊥ , Δ ⊢ ∀ x I 1 ⊥ , Δ ( ∀ ) ⊢ F , Γ , Δ ( C u t ) ⊢ ∀ x F , Γ , Δ ( ∀ ) ⟶ c u t π . \pi'\longrightarrow_{\mathrm{cut}}
\frac{\frac{\frac{\vdash F,\Gamma,I_1}{\vdash F,\Gamma,\exists x.I_1}\ (\exists)\qquad\frac{\vdash I_1^\perp,\Delta}{\vdash\forall xI_1^\perp,\Delta}\ (\forall)}{\vdash F,\Gamma,\Delta}\ (\mathrm{Cut})}{\vdash\forall xF,\Gamma,\Delta}\ (\forall)
\longrightarrow_{\mathrm{cut}}\pi. π ′ ⟶ cut ⊢ ∀ x F , Γ , Δ ⊢ F , Γ , Δ ⊢ F , Γ , ∃ x . I 1 ⊢ F , Γ , I 1 ( ∃ ) ⊢ ∀ x I 1 ⊥ , Δ ⊢ I 1 ⊥ , Δ ( ∀ ) ( Cut ) ( ∀ ) ⟶ cut π .
If the last rule is ( ∃ ) (\exists) ( ∃ ) , that is π = \pi= π =
⊢ F { y / x } , Γ , I 1 ⊢ I 1 ⊥ , Δ ⊢ F { y / x } , Γ , Δ ( C u t ) ⊢ ∃ x F , Γ , Δ ( ∃ ) \frac{\frac{\vdash F\{y/x\},\Gamma,I_1 \qquad\vdash I_1^\perp,\Delta}{\vdash F\{y/x\},\Gamma,\Delta}\ (\mathrm{Cut})}{\vdash\exists xF,\Gamma,\Delta}\ (\exists) ⊢ ∃ x F , Γ , Δ ⊢ F { y / x } , Γ , Δ ⊢ F { y / x } , Γ , I 1 ⊢ I 1 ⊥ , Δ ( Cut ) ( ∃ )
with Voc ( I 1 ) ⊆ ( Voc ( F { y / x } , Γ ) ∪ C L ) ∩ ( Voc ( Δ ) ∪ C R ) \operatorname{Voc}(I_1)\subseteq(\operatorname{Voc}(F\{y/x\},\Gamma)\cup C^L)\cap(\operatorname{Voc}(\Delta)\cup C^R) Voc ( I 1 ) ⊆ ( Voc ( F { y / x } , Γ ) ∪ C L ) ∩ ( Voc ( Δ ) ∪ C R ) .
In this case, note that we treat only the case of a FO language containing no function symbols.
In this case, we reason by case on whether y y y occurs in Γ , Δ \Gamma,\Delta Γ , Δ :
If y y y occurs in both, then we simply take I = I 1 I=I_1 I = I 1 as interpolant, π L = ⊢ ∃ x F , Γ , I 1 ⊢ ∃ x F , Γ , I 1 ( ∃ ) \pi^L=\frac{\vdash\exists xF,\Gamma,I_1}{\vdash\exists xF,\Gamma,I_1}\ (\exists) π L = ⊢ ∃ x F , Γ , I 1 ⊢ ∃ x F , Γ , I 1 ( ∃ ) and π R = π 1 R \pi^R=\pi^R_1 π R = π 1 R . Since Voc ( I ) = Voc ( I 1 ) ⊆ ( Voc ( F , Γ ) ∪ C L ) ∩ ( Voc ( Δ ) ∪ C R ) = ( Voc ( ∃ x F , Γ ) ∪ C L ) ∩ ( Voc ( Δ ) ∪ C R ) \operatorname{Voc}(I)=\operatorname{Voc}(I_1)\subseteq(\operatorname{Voc}(F,\Gamma)\cup C^L)\cap(\operatorname{Voc}(\Delta)\cup C^R)=(\operatorname{Voc}(\exists xF,\Gamma)\cup C^L)\cap(\operatorname{Voc}(\Delta)\cup C^R) Voc ( I ) = Voc ( I 1 ) ⊆ ( Voc ( F , Γ ) ∪ C L ) ∩ ( Voc ( Δ ) ∪ C R ) = ( Voc ( ∃ x F , Γ ) ∪ C L ) ∩ ( Voc ( Δ ) ∪ C R ) , we have that π ′ = ⊢ ∃ x F , Γ , I ⊢ I ⊥ , Δ ⊢ ∃ x F , Γ , Δ ( C u t ) \pi'=\frac{\frac{\vdash\exists xF,\Gamma,I \qquad\vdash I^\perp,\Delta}{\vdash\exists xF,\Gamma,\Delta}\ (\mathrm{Cut})}{} π ′ = ⊢ ∃ x F , Γ , Δ ⊢ ∃ x F , Γ , I ⊢ I ⊥ , Δ ( Cut ) is a solved CPRIS to which π \pi π cut-expands via a cut-commutation rule.
If y y y occurs in Γ \Gamma Γ but not in Δ \Delta Δ , then we set I = ∃ y I 1 I=\exists yI_1 I = ∃ y I 1 , π L = ⊢ ∃ x F , Γ , I 1 ⊢ ∃ x F , Γ , ∃ y I 1 ( ∃ ) ⊢ ∃ x F , Γ , ∃ y I 1 ( ∃ ) \pi^L=\frac{\frac{\vdash\exists xF,\Gamma,I_1}{\vdash\exists xF,\Gamma,\exists yI_1}\ (\exists)}{\vdash\exists xF,\Gamma,\exists yI_1}\ (\exists) π L = ⊢ ∃ x F , Γ , ∃ y I 1 ⊢ ∃ x F , Γ , ∃ y I 1 ⊢ ∃ x F , Γ , I 1 ( ∃ ) ( ∃ ) and
π R = π 1 R ⊢ I 1 ⊥ , Δ ⊢ ∀ y I 1 ⊥ , Δ ( ∀ ) \pi^{R} =
\frac{\pi^{R}_{1}}{\vdash I_{1}^{\perp},\Delta\atop\vdash\forall y I_{1}^{\perp},\Delta}
\quad(\forall) π R = ⊢ ∀ y I 1 ⊥ , Δ ⊢ I 1 ⊥ , Δ π 1 R ( ∀ )
we have that π ′ = π 1 L π R ⊢ ∃ x F , Γ , I ⊢ I ⊥ , Δ ⊢ ∃ x F , Γ , Δ ( C u t ) \pi' = \frac{\pi^{L}_{1}\quad\pi^{R}}{\vdash\exists xF,\Gamma,I\quad\vdash I^{\perp},\Delta\over\vdash\exists xF,\Gamma,\Delta}\;(\mathop{\mathrm{Cut}}) π ′ = ⊢ ∃ x F , Γ , Δ ⊢ ∃ x F , Γ , I ⊢ I ⊥ , Δ π 1 L π R ( Cut ) is a solved CPRIS to which π \pi π cut-expands via a cut-commutation rule and a key ( ∃ ) / ( ∀ ) (\exists)/(\forall) ( ∃ ) / ( ∀ ) rule:
π ′ ⟶ c u t ⊢ F { y / x } , Γ , I 1 ⊢ ∃ x F , Γ , I 1 ( ∃ ) π 1 R ⊢ ∃ x F , Γ , Δ ( C u t ) \pi'
\longrightarrow_{\mathrm{cut}}
\frac{\dfrac{\vdash F\{y/x\},\Gamma,I_{1}}{\vdash\exists xF,\Gamma,I_{1}}\;(\exists)
\quad
\pi^{R}_{1}}
{\vdash\exists xF,\Gamma,\Delta}\;(\mathop{\mathrm{Cut}}) π ′ ⟶ cut ⊢ ∃ x F , Γ , Δ ⊢ ∃ x F , Γ , I 1 ⊢ F { y / x } , Γ , I 1 ( ∃ ) π 1 R ( Cut )
⟶ c u t ⊢ F { y / x } , Γ , I 1 π 1 R ⊢ I 1 ⊥ , Δ ⊢ F { y / x } , Γ , Δ ( C u t ) ⊢ ∃ x F , Γ , Δ ( ∃ ) = π . \longrightarrow_{\mathrm{cut}}
\frac{\dfrac{\vdash F\{y/x\},\Gamma,I_{1}\quad\pi^{R}_{1}\quad\vdash I_{1}^{\perp},\Delta}{\vdash F\{y/x\},\Gamma,\Delta}\;(\mathop{\mathrm{Cut}})}
{\vdash\exists xF,\Gamma,\Delta}\;(\exists)
= \pi. ⟶ cut ⊢ ∃ x F , Γ , Δ ⊢ F { y / x } , Γ , Δ ⊢ F { y / x } , Γ , I 1 π 1 R ⊢ I 1 ⊥ , Δ ( Cut ) ( ∃ ) = π .
If y y y occurs in Δ \Delta Δ but not in Γ \Gamma Γ , then we set I = ∀ y I 1 I=\forall yI_{1} I = ∀ y I 1 , π L = ⊢ F { y / x } , Γ , I 1 ⊢ ∃ x F , Γ , I 1 ( ∃ ) \pi^{L}=\dfrac{\vdash F\{y/x\},\Gamma,I_{1}}{\vdash\exists xF,\Gamma,I_{1}}\;(\exists) π L = ⊢ ∃ x F , Γ , I 1 ⊢ F { y / x } , Γ , I 1 ( ∃ ) and
π R = π 1 R ⊢ I 1 ⊥ , Δ ⊢ ∃ y I 1 ⊥ , Δ ( ∃ ) . \pi^{R}=
\frac{\pi^{R}_{1}}{\vdash I_{1}^{\perp},\Delta\atop\vdash\exists yI_{1}^{\perp},\Delta}
\quad(\exists). π R = ⊢ ∃ y I 1 ⊥ , Δ ⊢ I 1 ⊥ , Δ π 1 R ( ∃ ) .
One gets that π ′ = π L π R ⊢ ∃ x F , Γ , ∀ y I 1 ⊢ ∃ y I 1 ⊥ , Δ ⊢ ∃ x F , Γ , Δ ( C u t ) \pi'=\dfrac{\pi^{L}\quad\pi^{R}}{\vdash\exists xF,\Gamma,\forall yI_{1}\quad\vdash\exists yI_{1}^{\perp},\Delta\over\vdash\exists xF,\Gamma,\Delta}\;(\mathop{\mathrm{Cut}}) π ′ = ⊢ ∃ x F , Γ , Δ ⊢ ∃ x F , Γ , ∀ y I 1 ⊢ ∃ y I 1 ⊥ , Δ π L π R ( Cut ) is a solved CPRIS to which π \pi π cut-expands:
π ′ ⟶ c u t ⊢ F { y / x } , Γ , I 1 ⊢ ∃ x F , Γ , I 1 ( ∃ ) π 1 R ⊢ ∃ x F , Γ , Δ ( C u t ) \pi'
\longrightarrow_{\mathrm{cut}}
\frac{\dfrac{\vdash F\{y/x\},\Gamma,I_{1}}{\vdash\exists xF,\Gamma,I_{1}}\;(\exists)
\quad
\pi^{R}_{1}}
{\vdash\exists xF,\Gamma,\Delta}\;(\mathop{\mathrm{Cut}}) π ′ ⟶ cut ⊢ ∃ x F , Γ , Δ ⊢ ∃ x F , Γ , I 1 ⊢ F { y / x } , Γ , I 1 ( ∃ ) π 1 R ( Cut )
⟶ c u t ⊢ F { y / x } , Γ , I 1 π 1 R ⊢ I 1 ⊥ , Δ ⊢ F { y / x } , Γ , Δ ( C u t ) ⊢ ∃ x F , Γ , Δ ( ∃ ) = π . \longrightarrow_{\mathrm{cut}}
\frac{\dfrac{\vdash F\{y/x\},\Gamma,I_{1}\quad\pi^{R}_{1}\quad\vdash I_{1}^{\perp},\Delta}{\vdash F\{y/x\},\Gamma,\Delta}\;(\mathop{\mathrm{Cut}})}
{\vdash\exists xF,\Gamma,\Delta}\;(\exists)
= \pi. ⟶ cut ⊢ ∃ x F , Γ , Δ ⊢ F { y / x } , Γ , Δ ⊢ F { y / x } , Γ , I 1 π 1 R ⊢ I 1 ⊥ , Δ ( Cut ) ( ∃ ) = π .
Appendix on Section 3.4
As usual now, the proof of Theorem 21 follows straightforwardly from a weak interpolation theorem and checking that the conditions on the cut formulas satisfy the interpolation constraint on the vocabularies:
Theorem 41. In what follows, t t t (resp. e e e , resp. c c c ) is an L L L -term (resp. normal L L L -context, resp. normal L L L -command). Let C C C be the set of types of all the t ′ t' t ′ , e ′ e' e ′ such that ⟨ t ′ ∣ e ′ ⟩ \langle t' \mid e'\rangle ⟨ t ′ ∣ e ′ ⟩ is a subcommand of t t t (resp. e e e , resp. c c c ). Let us assume that C C C is split into C L , C R C^{L},C^{R} C L , C R and C L = Voc ( C L ) C^{L}=\operatorname{Voc}(C^{L}) C L = Voc ( C L ) and C R = Voc ( C R ) C^{R}=\operatorname{Voc}(C^{R}) C R = Voc ( C R ) . The following weak interpolating results hold:
If c : ( Γ 1 , Γ 2 ⊢ Δ 1 , Δ 2 ) c:(\Gamma_{1},\Gamma_{2}\vdash\Delta_{1},\Delta_{2}) c : ( Γ 1 , Γ 2 ⊢ Δ 1 , Δ 2 ) , there exist a type I ∈ Voc ( Γ 1 , Δ 1 , C L ) ∩ Voc ( Γ 2 , Δ 2 , C R ) I\in\operatorname{Voc}(\Gamma_{1},\Delta_{1},C^{L})\cap\operatorname{Voc}(\Gamma_{2},\Delta_{2},C^{R}) I ∈ Voc ( Γ 1 , Δ 1 , C L ) ∩ Voc ( Γ 2 , Δ 2 , C R ) and t , e t,e t , e such that Γ 1 ⊢ t : I ∣ Δ 1 \Gamma_{1}\vdash t:I\mid\Delta_{1} Γ 1 ⊢ t : I ∣ Δ 1 and Γ 2 ∣ e : I ⊢ Δ 2 \Gamma_{2}\mid e:I\vdash\Delta_{2} Γ 2 ∣ e : I ⊢ Δ 2 , and ⟨ t ∣ e ⟩ ⟶ ∗ c \langle t\mid e\rangle\longrightarrow^{\ast}c ⟨ t ∣ e ⟩ ⟶ ∗ c .
If Γ 1 , Γ 2 ⊢ t : A ∣ Δ 1 , Δ 2 \Gamma_{1},\Gamma_{2}\vdash t:A\mid\Delta_{1},\Delta_{2} Γ 1 , Γ 2 ⊢ t : A ∣ Δ 1 , Δ 2 , there exist a type I ∈ Voc ( Γ 1 , Δ 1 , A , C L ) ∩ Voc ( Γ 2 , Δ 2 , C R ) I\in\operatorname{Voc}(\Gamma_{1},\Delta_{1},A,C^{L})\cap\operatorname{Voc}(\Gamma_{2},\Delta_{2},C^{R}) I ∈ Voc ( Γ 1 , Δ 1 , A , C L ) ∩ Voc ( Γ 2 , Δ 2 , C R ) and α , t ′ , e ′ \alpha,t',e' α , t ′ , e ′ such that Γ 1 ⊢ t ′ : A ∣ α : I , Δ 1 \Gamma_{1}\vdash t':A\mid\alpha:I,\Delta_{1} Γ 1 ⊢ t ′ : A ∣ α : I , Δ 1 and Γ 2 ∣ e ′ : I ⊢ Δ 2 \Gamma_{2}\mid e':I\vdash\Delta_{2} Γ 2 ∣ e ′ : I ⊢ Δ 2 , and t ′ { e ′ / α } ⟶ ∗ t t'\{e'/\alpha\}\longrightarrow^{\ast}t t ′ { e ′ / α } ⟶ ∗ t .
If Γ 1 , Γ 2 ∣ e : A ⊢ Δ 1 , Δ 2 \Gamma_{1},\Gamma_{2}\mid e:A\vdash\Delta_{1},\Delta_{2} Γ 1 , Γ 2 ∣ e : A ⊢ Δ 1 , Δ 2 , there exist a type I ∈ Voc ( Γ 1 , Δ 1 , A , C L ) ∩ Voc ( Γ 2 , Δ 2 , C R ) I\in\operatorname{Voc}(\Gamma_{1},\Delta_{1},A,C^{L})\cap\operatorname{Voc}(\Gamma_{2},\Delta_{2},C^{R}) I ∈ Voc ( Γ 1 , Δ 1 , A , C L ) ∩ Voc ( Γ 2 , Δ 2 , C R ) and α , e ′ , e ′ ′ \alpha,e',e'' α , e ′ , e ′′ such that Γ 1 ∣ e ′ : A ∣ α : I , Δ 1 \Gamma_{1}\mid e':A\mid\alpha:I,\Delta_{1} Γ 1 ∣ e ′ : A ∣ α : I , Δ 1 and Γ 2 ∣ e ′ ′ : I ⊢ Δ 2 \Gamma_{2}\mid e'':I\vdash\Delta_{2} Γ 2 ∣ e ′′ : I ⊢ Δ 2 , and e ′ { e ′ ′ / α } ⟶ ∗ e e'\{e''/\alpha\}\longrightarrow^{\ast}e e ′ { e ′′ / α } ⟶ ∗ e .
If Γ 1 , Γ 2 ⊢ t : A ∣ Δ 1 , Δ 2 \Gamma_{1},\Gamma_{2}\vdash t:A\mid\Delta_{1},\Delta_{2} Γ 1 , Γ 2 ⊢ t : A ∣ Δ 1 , Δ 2 , there exist a type I ∈ Voc ( Γ 1 , Δ 1 , C L ) ∩ Voc ( Γ 2 , Δ 2 , A , C R ) I\in\operatorname{Voc}(\Gamma_{1},\Delta_{1},C^{L})\cap\operatorname{Voc}(\Gamma_{2},\Delta_{2},A,C^{R}) I ∈ Voc ( Γ 1 , Δ 1 , C L ) ∩ Voc ( Γ 2 , Δ 2 , A , C R ) and α , t ′ , t ′ ′ \alpha,t',t'' α , t ′ , t ′′ such that Γ 1 ⊢ t ′ ′ : I ∣ Δ 1 \Gamma_{1}\vdash t'':I\mid\Delta_{1} Γ 1 ⊢ t ′′ : I ∣ Δ 1 and Γ 2 , x : I ⊢ t ′ : A ∣ Δ 2 \Gamma_{2},x:I\vdash t':A\mid\Delta_{2} Γ 2 , x : I ⊢ t ′ : A ∣ Δ 2 , and t ′ { t ′ ′ / x } ⟶ ∗ t t'\{t''/x\}\longrightarrow^{\ast}t t ′ { t ′′ / x } ⟶ ∗ t .
If Γ 1 , Γ 2 ∣ e : A ⊢ Δ 1 , Δ 2 \Gamma_{1},\Gamma_{2}\mid e:A\vdash\Delta_{1},\Delta_{2} Γ 1 , Γ 2 ∣ e : A ⊢ Δ 1 , Δ 2 , there exist a type I ∈ Voc ( Γ 1 , Δ 1 , C L ) ∩ Voc ( Γ 2 , Δ 2 , A , C R ) I\in\operatorname{Voc}(\Gamma_{1},\Delta_{1},C^{L})\cap\operatorname{Voc}(\Gamma_{2},\Delta_{2},A,C^{R}) I ∈ Voc ( Γ 1 , Δ 1 , C L ) ∩ Voc ( Γ 2 , Δ 2 , A , C R ) and x , t ′ , e ′ x,t',e' x , t ′ , e ′ such that Γ 1 ⊢ t ′ : I ∣ Δ 1 \Gamma_{1}\vdash t':I\mid\Delta_{1} Γ 1 ⊢ t ′ : I ∣ Δ 1 and Γ 2 , x : I ∣ e ′ : A ⊢ Δ 2 \Gamma_{2},x:I\mid e':A\vdash\Delta_{2} Γ 2 , x : I ∣ e ′ : A ⊢ Δ 2 , and e ′ { t ′ / x } ⟶ ∗ e e'\{t'/x\}\longrightarrow^{\ast}e e ′ { t ′ / x } ⟶ ∗ e .
Proof. While the proof by Saurin [34 Thm 20] relies on a long case analysis on both terms, evaluation contexts and commands, it is sufficient to adapt the case of commands as the other cases are straightforward by a simple check that the propagation of the cut-vocabulary constraint works. Regarding the case of commands, while Saurin considers only cases ⟨ x ∣ e ⟩ \langle x \mid e\rangle ⟨ x ∣ e ⟩ and ⟨ t ∣ α ⟩ \langle t \mid\alpha\rangle ⟨ t ∣ α ⟩ (due to the assumption on normal terms), we now have full fledged commands which are interpolated in a way close to the tensor case in [34 Thm 20] .
Let c : ( Γ 1 , Γ 2 ⊢ Δ 1 , Δ 2 ) c : (\Gamma_1,\Gamma_2 \vdash\Delta_1,\Delta_2) c : ( Γ 1 , Γ 2 ⊢ Δ 1 , Δ 2 ) , be as in the theorem statement, with c = ⟨ t ∣ e ⟩ c = \langle t \mid e\rangle c = ⟨ t ∣ e ⟩ . We want to prove that there exist a type K K K such that Voc ( K ) ∈ Voc ( Γ 1 , Δ 1 , C L ) ∩ Voc ( Γ 2 , Δ 2 , C R ) \operatorname{Voc}(K) \in\operatorname{Voc}(\Gamma_1,\Delta_1,C^{\mathcal{L}}) \cap\operatorname{Voc}(\Gamma_2,\Delta_2,C^{\mathcal{R}}) Voc ( K ) ∈ Voc ( Γ 1 , Δ 1 , C L ) ∩ Voc ( Γ 2 , Δ 2 , C R ) and u , f u,f u , f such that Γ 1 ⊢ u : K ∣ Δ 1 \Gamma_1 \vdash u : K \mid\Delta_1 Γ 1 ⊢ u : K ∣ Δ 1 and Γ 2 ∣ f : K ⊢ Δ 2 \Gamma_2 \mid f : K \vdash\Delta_2 Γ 2 ∣ f : K ⊢ Δ 2 , and ⟨ u ∣ f ⟩ ⟶ ∗ c \langle u \mid f\rangle\longrightarrow^{\ast} c ⟨ u ∣ f ⟩ ⟶ ∗ c .
t t t and e e e must respectively be a term and a context of type C C C (that we assume is split to the left in the theorem assumption: C ∈ C L C \in C^{\mathcal{L}} C ∈ C L ; the other case is symmetrical). We then apply the induction hypothesis to Γ 1 ′ , Γ 2 ′ ⊢ t : C ∣ Δ 1 ′ , Δ 2 ′ \Gamma'_1,\Gamma'_2 \vdash t : C \mid\Delta'_1,\Delta'_2 Γ 1 ′ , Γ 2 ′ ⊢ t : C ∣ Δ 1 ′ , Δ 2 ′ and Γ 1 ′ ′ , Γ 2 ′ ′ ∣ e : C ⊢ Δ 1 ′ ′ , Δ 2 ′ ′ \Gamma''_1,\Gamma''_2 \mid e : C \vdash\Delta''_1,\Delta''_2 Γ 1 ′′ , Γ 2 ′′ ∣ e : C ⊢ Δ 1 ′′ , Δ 2 ′′ and we get (in what follows we can assume that α ∉ FV ( e ′ , e ′ ′ ) \alpha\notin\operatorname{FV}(e',e'') α ∈ / FV ( e ′ , e ′′ ) and that β ∉ FV ( t ′ , f ′ ) \beta\notin\operatorname{FV}(t',f') β ∈ / FV ( t ′ , f ′ ) ):
(i) Γ 1 ′ ⊢ t ′ : C ∣ α : I , Δ 1 ′ \Gamma'_1 \vdash t' : C \mid\alpha: I,\Delta'_1 Γ 1 ′ ⊢ t ′ : C ∣ α : I , Δ 1 ′ , (ii) Γ 2 ′ ∣ f ′ : I ⊢ Δ 2 ′ \Gamma'_2 \mid f' : I \vdash\Delta'_2 Γ 2 ′ ∣ f ′ : I ⊢ Δ 2 ′ and (iii) t ′ { f ′ / α } ⟶ ∗ t t'\{f'/\alpha\} \longrightarrow^{\ast} t t ′ { f ′ / α } ⟶ ∗ t ;
(i) Γ 1 ′ ′ ⊢ e ′ : C ⊢ β : J , Δ 1 ′ ′ \Gamma''_1 \vdash e' : C \vdash\beta: J,\Delta''_1 Γ 1 ′′ ⊢ e ′ : C ⊢ β : J , Δ 1 ′′ , (ii) Γ 2 ′ ′ ∣ e ′ ′ : J ⊢ Δ 2 ′ ′ \Gamma''_2 \mid e'' : J \vdash\Delta''_2 Γ 2 ′′ ∣ e ′′ : J ⊢ Δ 2 ′′ and (iii) e ′ { e ′ ′ / β } ⟶ ∗ e e'\{e''/\beta\} \longrightarrow^{\ast} e e ′ { e ′′ / β } ⟶ ∗ e .
With Voc ( I ) ⊆ Voc ( Γ 1 ′ , Δ 1 ′ , C , C L ) ∩ Voc ( Γ 2 ′ , Δ 2 ′ , C R ) \operatorname{Voc}(I) \subseteq\operatorname{Voc}(\Gamma'_1,\Delta'_1,C,C^{\mathcal{L}}) \cap\operatorname{Voc}(\Gamma'_2,\Delta'_2,C^{\mathcal{R}}) Voc ( I ) ⊆ Voc ( Γ 1 ′ , Δ 1 ′ , C , C L ) ∩ Voc ( Γ 2 ′ , Δ 2 ′ , C R ) and Voc ( J ) ⊆ Voc ( Γ 1 ′ ′ , Δ 1 ′ ′ , C , C L ) ∩ Voc ( Γ 2 ′ ′ , Δ 2 ′ ′ , C R ) \operatorname{Voc}(J) \subseteq\operatorname{Voc}(\Gamma''_1,\Delta''_1,C,C^{\mathcal{L}}) \cap\operatorname{Voc}(\Gamma''_2,\Delta''_2,C^{\mathcal{R}}) Voc ( J ) ⊆ Voc ( Γ 1 ′′ , Δ 1 ′′ , C , C L ) ∩ Voc ( Γ 2 ′′ , Δ 2 ′′ , C R ) .
Then, by setting: K = I ⊗ J K = I \otimes J K = I ⊗ J , u = μ [ α , β ] ⟨ t ′ ∣ e ′ ⟩ u = \mu[\alpha,\beta]\langle t' \mid e'\rangle u = μ [ α , β ] ⟨ t ′ ∣ e ′ ⟩ and f = [ f ′ , e ′ ′ ] f = [f',e''] f = [ f ′ , e ′′ ] We have Voc ( K ) ∈ Voc ( Γ 1 , Δ 1 , C L ) ∩ Voc ( Γ 2 , Δ 2 , C R ) \operatorname{Voc}(K) \in\operatorname{Voc}(\Gamma_1,\Delta_1,C^{\mathcal{L}}) \cap\operatorname{Voc}(\Gamma_2,\Delta_2,C^{\mathcal{R}}) Voc ( K ) ∈ Voc ( Γ 1 , Δ 1 , C L ) ∩ Voc ( Γ 2 , Δ 2 , C R ) and Γ 1 ⊢ u : K ∣ Δ 1 \Gamma_1 \vdash u : K \mid\Delta_1 Γ 1 ⊢ u : K ∣ Δ 1 and Γ 2 ∣ f : K ⊢ Δ 2 \Gamma_2 \mid f : K \vdash\Delta_2 Γ 2 ∣ f : K ⊢ Δ 2 as expected and ⟨ u ∣ f ⟩ ⟶ ⟨ t ′ ∣ e ′ ⟩ { f ′ / α , e ′ ′ / β } = ⟨ t ′ { f ′ / α } ∣ e ′ { e ′ ′ / β } ⟩ ⟶ ∗ c \langle u \mid f\rangle\longrightarrow\langle t' \mid e'\rangle\{f'/\alpha,e''/\beta\} = \langle t'\{f'/\alpha\} \mid e'\{e''/\beta\}\rangle\longrightarrow^{\ast} c ⟨ u ∣ f ⟩ ⟶ ⟨ t ′ ∣ e ′ ⟩ { f ′ / α , e ′′ / β } = ⟨ t ′ { f ′ / α } ∣ e ′ { e ′′ / β }⟩ ⟶ ∗ c by induction hypothesis. ◀
Appendix on Section 4
Proofs for (ordered) trace manifolds
Before the proof of Lemmas 29 and 30 , we introduce the following lemmas and definitions:
Lemma 42 (Brotherston [5 ] , Lemma 7.1.3). Let P = ( D , R ) \mathcal{P} = (\mathcal{D},\mathcal{R}) P = ( D , R ) be a μ L L ω \mu\mathrm{LL}^{\omega} μ LL ω pre-proof. Then for every infinite path π \pi π through G P \mathcal{G}_{\mathcal{P}} G P , there is a non-trivial strongly connected subgraph S π ⊆ G P S_{\pi} \subseteq\mathcal{G}_{\mathcal{P}} S π ⊆ G P and a tail π ′ \pi' π ′ of π \pi π such that π ′ \pi' π ′ is an infinite path through S π S_{\pi} S π on which each vertex of S π S_{\pi} S π occurs infinitely often.
Definition 43 (Path composition). Let G G G be a graph, let π 1 = v 0 . . . v m \pi_1 = v_0...v_m π 1 = v 0 ... v m be a finite path in G G G and let π 2 = v 0 ′ v 1 ′ v 2 ′ . . . \pi_2 = v'_0v'_1v'_2... π 2 = v 0 ′ v 1 ′ v 2 ′ ... be a (possibly infinite) path in G G G . We say that π 1 \pi_1 π 1 composes with π 2 \pi_2 π 2 if v m = v 0 ′ v_m = v'_0 v m = v 0 ′ and define the composed path π 1 π 2 \pi_1\pi_2 π 1 π 2 to be v 0 . . . v m − 1 v 0 ′ v 1 ′ v 2 ′ . . . v_0...v_{m-1}v'_0v'_1v'_2... v 0 ... v m − 1 v 0 ′ v 1 ′ v 2 ′ ... .
Lemma 44 (Brotherston [5 ] , Lemma 7.1.8). Let P = ( D , R ) \mathcal{P} = (\mathcal{D},\mathcal{R}) P = ( D , R ) be a μ L L ω \mu\mathrm{LL}^{\omega} μ LL ω pre-proof and let S S S be a non-trivial strongly connected subgraph of G P \mathcal{G}_{\mathcal{P}} G P . Note that S = ⋃ B ∈ B S C B S = \bigcup_{B \in\mathcal{B}_S} C_B S = ⋃ B ∈ B S C B by the Lemma 25 . Then, any infinite path in S S S , starting from the companion of a bud in B S \mathcal{B}_S B S , is composed of paths (of length > 0 > 0 > 0 ) each of the form R ( B i ) . . . R ( B j ) \mathcal{R}(B_i)...\mathcal{R}(B_j) R ( B i ) ... R ( B j ) , where B i , B j ∈ B S B_i,B_j \in\mathcal{B}_S B i , B j ∈ B S and R ( B i ) . . . R ( B j ) \mathcal{R}(B_i)...\mathcal{R}(B_j) R ( B i ) ... R ( B j ) is a subpath of some cycle C B C_B C B .
Proof of Lemma 29. We use Brotherston’s proof ([5 ] , Proposition 7.2.3) with slight modifications to accommodate for the modified properties of the validity condition. Let P = ( D , R ) \mathcal{P} = (\mathcal{D},\mathcal{R}) P = ( D , R ) be a μ L L ω \mu\mathrm{LL}^{\omega} μ LL ω pre-proof and consider any infinite path π \pi π through G P \mathcal{G}_{\mathcal{P}} G P . By Lemma 42 , π \pi π has a tail π ′ \pi' π ′ that is an infinite path through a non-trivial strongly connected subgraph S S S , where every vertex in S S S infinitely appears in π ′ \pi' π ′ . Moreover, by Lemma 25 , S = ⋃ B ∈ B S C B S = \bigcup_{B \in\mathcal{B}_S} C_B S = ⋃ B ∈ B S C B where B S ⊆ B u d ( D ) \mathcal{B}_S \subseteq Bud(\mathcal{D}) B S ⊆ B u d ( D ) and B S \mathcal{B}_S B S is weakly ≤ P \leq_{\mathcal{P}} ≤ P -connected. We then want to show that we can construct a valid trace on π ′ \pi' π ′ . Without loss of generality, assume that R ( B i 1 ) \mathcal{R}(B_{i_1}) R ( B i 1 ) is the first vertex in π ′ \pi' π ′ , for some B i 1 ∈ B S B_{i_1} \in\mathcal{B}_S B i 1 ∈ B S . Then by Lemma 44 , π ′ \pi' π ′ is composed of finite paths in S S S as follows:
π ′ = R ( B i 1 ) . . . R ( B i 2 ) . . . R ( B i 3 ) . . . (2) \pi' = \mathcal{R}(B_{i_1})...\mathcal{R}(B_{i_2})...\mathcal{R}(B_{i_3})... \tag*{(2)} π ′ = R ( B i 1 ) ... R ( B i 2 ) ... R ( B i 3 ) ... (2)
Where for all n ≥ 1 n \ge1 n ≥ 1 , B i n ∈ B S B_{i_n} \in\mathcal{B}_S B i n ∈ B S and R ( B i n ) … R ( B i n + 1 ) \mathcal{R}(B_{i_n}) \ldots\mathcal{R}(B_{i_{n+1}}) R ( B i n ) … R ( B i n + 1 ) is a subpath of some basic cycle C B j n C_{B_{j_n}} C B j n , with B j n ∈ B S B_{j_n} \in\mathcal{B}_S B j n ∈ B S . The idea is to form the trace that validates π ′ \pi' π ′ , by associating a segment of a trace τ S , B j n \tau_{S,B_{j_n}} τ S , B j n , to each subpath R ( B i n ) … R ( B i n + 1 ) \mathcal{R}(B_{i_n}) \ldots\mathcal{R}(B_{i_{n+1}}) R ( B i n ) … R ( B i n + 1 ) . Suppose we have the two composable subpaths R ( B i n ) … R ( B i n + 1 ) \mathcal{R}(B_{i_n}) \ldots\mathcal{R}(B_{i_{n+1}}) R ( B i n ) … R ( B i n + 1 ) in C B j n C_{B_{j_n}} C B j n and R ( B i n + 1 ) … R ( B i n + 2 ) \mathcal{R}(B_{i_{n+1}}) \ldots\mathcal{R}(B_{i_{n+2}}) R ( B i n + 1 ) … R ( B i n + 2 ) in C B j n + 1 C_{B_{j_{n+1}}} C B j n + 1 . We then associate a segment τ j n \tau_{j_n} τ j n of the trace τ S , B j n \tau_{S,B_{j_n}} τ S , B j n , that covers the subpath R ( B i n ) … R ( B i n + 1 ) \mathcal{R}(B_{i_n}) \ldots\mathcal{R}(B_{i_{n+1}}) R ( B i n ) … R ( B i n + 1 ) and a segment τ j n + 1 \tau_{j_{n+1}} τ j n + 1 of the trace τ S , B j n + 1 \tau_{S,B_{j_{n+1}}} τ S , B j n + 1 , that covers the subpath R ( B i n + 1 ) … R ( B i n + 2 ) \mathcal{R}(B_{i_{n+1}}) \ldots\mathcal{R}(B_{i_{n+2}}) R ( B i n + 1 ) … R ( B i n + 2 ) . Notice that R ( B i n + 1 ) \mathcal{R}(B_{i_{n+1}}) R ( B i n + 1 ) occurs both in C B j n C_{B_{j_n}} C B j n and C B j n + 1 C_{B_{j_{n+1}}} C B j n + 1 , hence B i n + 1 ≤ P B j n B_{i_{n+1}} \le_P B_{j_n} B i n + 1 ≤ P B j n and B i n + 1 ≤ P B j n + 1 B_{i_{n+1}} \le_P B_{j_{n+1}} B i n + 1 ≤ P B j n + 1 . Then by the property 2, we get that τ S , B j n ( R ( B i n + 1 ) ) = τ S , B i n + 1 ( R ( B i n + 1 ) ) = τ S , B j n + 1 ( R ( B i n + 1 ) ) \tau_{S,B_{j_n}}(\mathcal{R}(B_{i_{n+1}})) = \tau_{S,B_{i_{n+1}}}(\mathcal{R}(B_{i_{n+1}})) = \tau_{S,B_{j_{n+1}}}(\mathcal{R}(B_{i_{n+1}})) τ S , B j n ( R ( B i n + 1 )) = τ S , B i n + 1 ( R ( B i n + 1 )) = τ S , B j n + 1 ( R ( B i n + 1 )) , which then implies that τ j n \tau_{j_n} τ j n is composable with τ j n + 1 \tau_{j_{n+1}} τ j n + 1 . We then say that the trace τ \tau τ that covers π ′ \pi' π ′ is the trace formed by using for each subpath R ( B i n ) … R ( B i n + 1 ) \mathcal{R}(B_{i_n}) \ldots\mathcal{R}(B_{i_{n+1}}) R ( B i n ) … R ( B i n + 1 ) the trace segment τ j n \tau_{j_n} τ j n of the trace τ S , B j n \tau_{S,B_{j_n}} τ S , B j n . We would like to prove that the constructed trace τ \tau τ is a valid trace in the path π ′ \pi' π ′ . To do it, we prove 1) τ \tau τ infinitely visits a F F F -progress point, 2) min ( Inf ( τ ) ) = F \min(\operatorname{Inf}(\tau)) = F min ( Inf ( τ )) = F :
By the property 3, there is a B i k B_{i_k} B i k such that τ S , B i k \tau_{S,B_{i_k}} τ S , B i k has an F F F -progress point on the basic cycle C B i k C_{B_{i_k}} C B i k . Moreover, by Lemma 42 every vertex and edge in S S S infinitely appears in π ′ \pi' π ′ , which implies that we can find infinite instances of the subpath R ( B i n ) … R ( B i n + 1 ) \mathcal{R}(B_{i_n}) \ldots\mathcal{R}(B_{i_{n+1}}) R ( B i n ) … R ( B i n + 1 ) in π ′ \pi' π ′ , that includes the vertex with the F F F -progress point. Assume this subpath is contained in the basic cycles C B i k C_{B_{i_k}} C B i k and C B i h C_{B_{i_h}} C B i h . For this reason, we can then assign a portion of the trace τ S , B i k \tau_{S,B_{i_k}} τ S , B i k or τ S , B i h \tau_{S,B_{i_h}} τ S , B i h . We show that we can always assign the progressing trace τ S , B i k \tau_{S,B_{i_k}} τ S , B i k . Note that B i n ≤ P B i k B_{i_n} \le_P B_{i_k} B i n ≤ P B i k and B i n ≤ P B i h B_{i_n} \le_P B_{i_h} B i n ≤ P B i h , therefore τ S , B i k ( R ( B i n ) ) = τ S , B i n ( R ( B i n ) ) = τ S , B i h ( R ( B i n ) ) \tau_{S,B_{i_k}}(\mathcal{R}(B_{i_n})) = \tau_{S,B_{i_n}}(\mathcal{R}(B_{i_n})) = \tau_{S,B_{i_h}}(\mathcal{R}(B_{i_n})) τ S , B i k ( R ( B i n )) = τ S , B i n ( R ( B i n )) = τ S , B i h ( R ( B i n )) . The same applies for B i n + 1 B_{i_{n+1}} B i n + 1 , so τ S , B i k ( R ( B i n + 1 ) ) = τ S , B i n + 1 ( R ( B i n + 1 ) ) = τ S , B i h ( R ( B i n + 1 ) ) \tau_{S,B_{i_k}}(\mathcal{R}(B_{i_{n+1}})) = \tau_{S,B_{i_{n+1}}}(\mathcal{R}(B_{i_{n+1}})) = \tau_{S,B_{i_h}}(\mathcal{R}(B_{i_{n+1}})) τ S , B i k ( R ( B i n + 1 )) = τ S , B i n + 1 ( R ( B i n + 1 )) = τ S , B i h ( R ( B i n + 1 )) . Thus, both τ S , B i k \tau_{S,B_{i_k}} τ S , B i k or τ S , B i h \tau_{S,B_{i_h}} τ S , B i h follow the subpath and agree at R ( B i n ) \mathcal{R}(B_{i_n}) R ( B i n ) and R ( B i n + 1 ) \mathcal{R}(B_{i_{n+1}}) R ( B i n + 1 ) . We then always choose the trace τ S , B i k \tau_{S,B_{i_k}} τ S , B i k with the progress point, ensuring it infinitely appears in π ′ \pi' π ′ .
Let τ S , B i 1 , τ S , B i 2 , … τ S , B i n \tau_{S,B_{i_1}}, \tau_{S,B_{i_2}}, \ldots\tau_{S,B_{i_n}} τ S , B i 1 , τ S , B i 2 , … τ S , B i n be the traces associated to S S S in the manifold. It is easy to see that since τ \tau τ is constructed using the previously mentioned traces, then it must be the case that Inf ( τ ) ⊆ ⋃ j = 1 n Form ( τ S , B i j ) \operatorname{Inf}(\tau) \subseteq\bigcup_{j=1}^{n} \operatorname{Form}(\tau_{S,B_{i_j}}) Inf ( τ ) ⊆ ⋃ j = 1 n Form ( τ S , B i j ) . Hence, by property 3, F F F is minimal in ⋃ j = 1 n Form ( τ S , B i j ) \bigcup_{j=1}^{n} \operatorname{Form}(\tau_{S,B_{i_j}}) ⋃ j = 1 n Form ( τ S , B i j ) , which would also imply that it is minimal in Inf ( τ ) \operatorname{Inf}(\tau) Inf ( τ ) . Finally, we now that F ∈ Inf ( τ ) F \in\operatorname{Inf}(\tau) F ∈ Inf ( τ ) , since τ \tau τ infinitely visits the F F F -progress point, thus min ( Inf ( τ ) ) = F \min(\operatorname{Inf}(\tau)) = F min ( Inf ( τ )) = F .
Proof of Lemma 30. We use the same proof as Brotherston’s [5 ] , Lemma 7.2.8) to prove both directions, with slight modifications to accommodate for the modified properties of the validity condition:
⇒ \Rightarrow ⇒ ) Suppose that P = ( D , R ) \mathcal{P} = (\mathcal{D}, \mathcal{R}) P = ( D , R ) has a trace manifold. We will show how to construct an induction order ⊲ \lhd ⊲ for P \mathcal{P} P and ordered trace manifold with respect to ⊲ \lhd ⊲ . Take any maximal set B ⊆ Bud ( D ) \mathcal{B} \subseteq\operatorname{Bud}(\mathcal{D}) B ⊆ Bud ( D ) of weakly ≤ P \le_P ≤ P -connected buds, i.e. there is no other weakly ≤ P \le_P ≤ P -connected set of buds B ′ ⊆ Bud ( D ) \mathcal{B}' \subseteq\operatorname{Bud}(\mathcal{D}) B ′ ⊆ Bud ( D ) such that B ⊂ B ′ \mathcal{B} \subset\mathcal{B}' B ⊂ B ′ , and by Lemma 25 let S = ⋃ B ∈ B C B S = \bigcup_{B \in\mathcal{B}} C_B S = ⋃ B ∈ B C B . By definition of a trace manifold, there is some B k ∈ B B_k \in\mathcal{B} B k ∈ B such that the trace τ S , B k \tau_{S,B_k} τ S , B k progresses. It could be the case that that there are more than one bud that satisfy this property, in which case we pick any of the possible candidates. We choose B k B_k B k to be the ⊲ \lhd ⊲ -greatest element in B \mathcal{B} B , i.e. for all B i ∈ B B_i \in\mathcal{B} B i ∈ B then B i ⊲ B k B_i \lhd B_k B i ⊲ B k . Moreover, we define τ B k , B i = τ S , B i \tau_{B_k,B_i} = \tau_{S,B_i} τ B k , B i = τ S , B i for each B i ∈ B B_i \in\mathcal{B} B i ∈ B . Note that B k B_k B k also becomes a ⊲ \lhd ⊲ -greatest element in every subset of B \mathcal{B} B that contains B k B_k B k . We then consider all the maximal weakly ≤ P \le_P ≤ P -connected subsets of Bud ( D ) ∖ { B k } \operatorname{Bud}(\mathcal{D}) \setminus\{B_k\} Bud ( D ) ∖ { B k } and repeat this process until every weakly ≤ P \le_P ≤ P -connected subset of Bud ( D ) \operatorname{Bud}(\mathcal{D}) Bud ( D ) has a ⊲ \lhd ⊲ -greatest element. We check then that the constructed ⊲ \lhd ⊲ order indeed is a partial order and satisfies the two properties to be an induction order:
▪ ⊲ \lhd ⊲ is reflexive by construction. To prove transitivity, take some B i ⊲ B j B_i \lhd B_j B i ⊲ B j and B j ⊲ B k B_j \lhd B_k B j ⊲ B k . This implies that there are weakly ≤ P \le_P ≤ P -connected subsets B j , B k ⊆ Bud ( D ) \mathcal{B}_j, \mathcal{B}_k \subseteq\operatorname{Bud}(\mathcal{D}) B j , B k ⊆ Bud ( D ) such that B j B_j B j and B k B_k B k are ⊲ \lhd ⊲ -maximal elements in B j \mathcal{B}_j B j and B k \mathcal{B}_k B k respectively. Moreover, since B j ⊲ B k B_j \lhd B_k B j ⊲ B k , this implies that B j ⊆ B k \mathcal{B}_j \subseteq\mathcal{B}_k B j ⊆ B k , and since B i ◃ B j B_i \triangleleft B_j B i ◃ B j , it is the case that B i ∈ B j B_i \in\mathcal{B}_j B i ∈ B j . Hence, B i ∈ B k B_i \in\mathcal{B}_k B i ∈ B k and B i ◃ B k B_i \triangleleft B_k B i ◃ B k . For anti-symmetry, suppose B j ◃ B k B_j \triangleleft B_k B j ◃ B k and B k ◃ B j B_k \triangleleft B_j B k ◃ B j . This implies that there are weakly ≤ P \leq_{\mathcal{P}} ≤ P -connected subsets B j , B k ⊆ B u d ( D ) \mathcal{B}_j, \mathcal{B}_k \subseteq Bud(\mathcal{D}) B j , B k ⊆ B u d ( D ) such that B j B_j B j and B k B_k B k are ◃ \triangleleft ◃ -maximal elements in B j \mathcal{B}_j B j and B k \mathcal{B}_k B k respectively, and B k = B j \mathcal{B}_k = \mathcal{B}_j B k = B j since B j ⊆ B k \mathcal{B}_j \subseteq\mathcal{B}_k B j ⊆ B k and B k ⊆ B j \mathcal{B}_k \subseteq\mathcal{B}_j B k ⊆ B j . Therefore B j = B k B_j = B_k B j = B k , since they are the unique maximum of both subsets.
To check that it is forest-like, let B i ◃ B j B_i \triangleleft B_j B i ◃ B j and B i ◃ B k B_i \triangleleft B_k B i ◃ B k . By construction, there exist weakly ≤ P \leq_{\mathcal{P}} ≤ P -connected subset B j , B k ⊆ B u d ( D ) \mathcal{B}_j, \mathcal{B}_k \subseteq Bud(\mathcal{D}) B j , B k ⊆ B u d ( D ) such that B j B_j B j and B k B_k B k are ◃ \triangleleft ◃ -maximal elements in B j \mathcal{B}_j B j and B k \mathcal{B}_k B k respectively, where B i ∈ B j B_i \in\mathcal{B}_j B i ∈ B j and B i ∈ B k B_i \in\mathcal{B}_k B i ∈ B k . By construction, it must be the case that either B j ⊆ B k \mathcal{B}_j \subseteq\mathcal{B}_k B j ⊆ B k , in which case B j ◃ B k B_j \triangleleft B_k B j ◃ B k , or B k ⊆ B j \mathcal{B}_k \subseteq\mathcal{B}_j B k ⊆ B j , in which case B k ◃ B j B_k \triangleleft B_j B k ◃ B j , or B j = B k \mathcal{B}_j = \mathcal{B}_k B j = B k , in which case B j = B k B_j = B_k B j = B k .
Finally, we check that the constructed traces correspond to a ordered trace manifold, with respect to the induction order ◃ \triangleleft ◃ :
Let B i ◃ B j B_i \triangleleft B_j B i ◃ B j . By construction, τ B j , B i = τ S , B i \tau_{B_j,B_i} = \tau_{S,B_i} τ B j , B i = τ S , B i exists and is a trace following the basic cycle C B i C_{B_i} C B i in G P \mathcal{G}_{\mathcal{P}} G P and the same value at both instances of R ( B i ) \mathcal{R}(B_i) R ( B i ) , by the property 1 of the trace manifold.
Let B j , B i ◃ B k B_j, B_i \triangleleft B_k B j , B i ◃ B k and B j ≤ P B i B_j \leq_{\mathcal{P}} B_i B j ≤ P B i . By construction, τ B k , B j = τ S , B j \tau_{B_k,B_j} = \tau_{S,B_j} τ B k , B j = τ S , B j and τ B k , B i = τ S , B i \tau_{B_k,B_i} = \tau_{S,B_i} τ B k , B i = τ S , B i for some S S S such that S = ⋃ B ∈ B C B S = \bigcup_{B \in\mathcal{B}} C_B S = ⋃ B ∈ B C B and B k , B j , B i ∈ B B_k, B_j, B_i \in\mathcal{B} B k , B j , B i ∈ B with B k B_k B k as ◃ \triangleleft ◃ -maximal. By the property 2 of the trace manifold, then τ S , B i ( R ( B j ) ) = τ S , B j ( R ( B j ) ) \tau_{S,B_i}(\mathcal{R}(B_j)) = \tau_{S,B_j}(\mathcal{R}(B_j)) τ S , B i ( R ( B j )) = τ S , B j ( R ( B j )) , hence τ B k , B i ( R ( B j ) ) = τ B k , B j ( R ( B j ) ) \tau_{B_k,B_i}(\mathcal{R}(B_j)) = \tau_{B_k,B_j}(\mathcal{R}(B_j)) τ B k , B i ( R ( B j )) = τ B k , B j ( R ( B j )) .
Note that by contruction, For every k k k and any j j j , τ B k , B i j = τ S , B i j \tau_{B_k,B_i^j} = \tau_{S,B_i^j} τ B k , B i j = τ S , B i j for some S S S , where τ B k , B k \tau_{B_k,B_k} τ B k , B k has an F F F -progress point. Moreover, by property 3 of the trace manifold it is the case that F F F is minimal in ⋃ j = 1 n Form ( τ S , B i j ) \bigcup_{j=1}^{n}\operatorname{Form}(\tau_{S,B_i^j}) ⋃ j = 1 n Form ( τ S , B i j ) , hence F F F is minimal in ⋃ j = 1 n Form ( τ B k , B i j ) \bigcup_{j=1}^{n}\operatorname{Form}(\tau_{B_k,B_i^j}) ⋃ j = 1 n Form ( τ B k , B i j ) .
⇐ \Leftarrow ⇐ Let ◃ \triangleleft ◃ be an induction order for P = ( D , R ) \mathcal{P} = (\mathcal{D}, \mathcal{R}) P = ( D , R ) and suppose that P \mathcal{P} P has an ordered trace manifold with respect to ◃ \triangleleft ◃ . Now, consider any strongly connected subgraph S S S such that S = ⋃ B ∈ B S C B S = \bigcup_{B \in\mathcal{B}_S} C_B S = ⋃ B ∈ B S C B with B S ⊆ B u d ( D ) \mathcal{B}_S \subseteq Bud(\mathcal{D}) B S ⊆ B u d ( D ) and weakly ≤ P \leq_{\mathcal{P}} ≤ P -connected, by Lemma 25. By definition of induction order, B S \mathcal{B}_S B S has a ◃ \triangleleft ◃ -maximal element B k B_k B k . We then make τ S , B i = τ B k , B i \tau_{S,B_i} = \tau_{B_k,B_i} τ S , B i = τ B k , B i , for each i i i , and check that this construction satisfies the properties of a trace manifold:
By the property 1 of the ordered trace manifold, τ S , B i = τ B k , B i \tau_{S,B_i} = \tau_{B_k,B_i} τ S , B i = τ B k , B i is a trace following the basic cycle C B i C_{B_i} C B i , and takes the same value at both occurrences of R ( B i ) \mathcal{R}(B_i) R ( B i ) .
Let B i , B j ∈ B S B_i, B_j \in\mathcal{B}_S B i , B j ∈ B S , where B j ≤ P B i B_j \leq_{\mathcal{P}} B_i B j ≤ P B i . Therefore, since B k B_k B k is ◃ \triangleleft ◃ -maximal, then B j , B i ◃ B k B_j, B_i \triangleleft B_k B j , B i ◃ B k and by property 2 of the ordered trace manifold, we get that τ B k , B j ( R ( B j ) ) = τ B k , B i ( R ( B j ) ) \tau_{B_k,B_j}(\mathcal{R}(B_j)) = \tau_{B_k,B_i}(\mathcal{R}(B_j)) τ B k , B j ( R ( B j )) = τ B k , B i ( R ( B j )) , τ S , B j ( R ( B j ) ) = τ S , B i ( R ( B j ) ) \tau_{S,B_j}(\mathcal{R}(B_j)) = \tau_{S,B_i}(\mathcal{R}(B_j)) τ S , B j ( R ( B j )) = τ S , B i ( R ( B j )) .
Note that for any j j j , τ S , B i j = τ B k , B i j \tau_{S,B_i^j} = \tau_{B_k,B_i^j} τ S , B i j = τ B k , B i j where τ S , B k = τ B k , B k \tau_{S,B_k} = \tau_{B_k,B_k} τ S , B k = τ B k , B k has an F F F -progress point. Moreover, by property 3 of the ordered trace manifold, it is the case that F F F is minimal in ⋃ j = 1 n Form ( τ B k , B i j ) \bigcup_{j=1}^{n}\operatorname{Form}(\tau_{B_k,B_i^j}) ⋃ j = 1 n Form ( τ B k , B i j ) , hence F F F is minimal in ⋃ j = 1 n Form ( τ S , B i j ) \bigcup_{j=1}^{n}\operatorname{Form}(\tau_{S,B_i^j}) ⋃ j = 1 n Form ( τ S , B i j ) .
Injective bud-companion relation
Without loss of generality, for all pre-proofs P = ( D , R ) \mathcal{P} = (\mathcal{D}, \mathcal{R}) P = ( D , R ) we can consider that R \mathcal{R} R is an injective function. This can be done by defining the derivable rule (Id) which consists of using the mix rule (X) without making any change in the order of the formulas in the sequent. Now, let C C C be any companion node where C C C is labelled with ⊢ Γ \vdash\Gamma ⊢ Γ , and { B 1 , … , B n } \{B_1,\ldots,B_n\} { B 1 , … , B n } is the set of bud nodes such that for all i i i , R ( B i ) = C \mathcal{R}(B_i) = C R ( B i ) = C . Then C C C can be transformed as follows:
Γ ′ ( † ) B 1 , … , B n ⊢ Γ ( r ) ⇒ Γ ′ ( † ) B p ( n ) ⊢ Γ ⋮ ( I d ) ⋮ ( I d ) ( † ) B p ( 1 ) ⊢ Γ ( r ) \frac{\Gamma'}{(\dagger)B_1,\ldots,B_n\vdash\Gamma}\ (r)
\quad\Rightarrow\quad
\frac{\Gamma'}{\frac{(\dagger)B_{p(n)}\vdash\Gamma}{\vdots\ (Id)\qquad\vdots\ (Id)\qquad(\dagger)B_{p(1)}\vdash\Gamma}}\ (r) ( † ) B 1 , … , B n ⊢ Γ Γ ′ ( r ) ⇒ ⋮ ( I d ) ⋮ ( I d ) ( † ) B p ( 1 ) ⊢ Γ ( † ) B p ( n ) ⊢ Γ Γ ′ ( r )
Where p p p defines a permutation over { 1 , … , n } \{1,\ldots,n\} { 1 , … , n } . Note that this change in P \mathcal{P} P does not alter its validity, in case P \mathcal{P} P is a valid μ L L ω \mu\mathrm{LL}^{\omega} μ LL ω proof. Moreover, in case P \mathcal{P} P can be coherently decorated, this transformation preserves this property, since the rule (Id) maintains the same split during the ascending phase.
This also implies that for any companion node with multiple bud nodes B 1 , … , B n B_1,\ldots,B_n B 1 , … , B n , we can choose the order of the back-edges and therefore implicitly choose the relation ≤ P \leq_{\mathcal{P}} ≤ P over B 1 , … , B n B_1,\ldots,B_n B 1 , … , B n . For this reason, given a tree-compatible pre-proof P \mathcal{P} P , with non-injective R \mathcal{R} R , we can apply this same transformation to obtain a tree-compatible pre-proof R \mathcal{R} R , with injective R \mathcal{R} R .
The procedure chooses the adequate permutation p p p as follows: Consider any companion node C C C , with a set of bud nodes B = { B 1 , … , B n } \mathcal{B}=\{B_1,\ldots,B_n\} B = { B 1 , … , B n } such that for all i i i , R ( B i ) = C \mathcal{R}(B_i)=C R ( B i ) = C . Since every bud node points to the same companion then the set B \mathcal{B} B is weakly ≤ P \leq_{\mathcal{P}} ≤ P -connected, and by Definition 27, there is a B i ∈ B B_i\in\mathcal{B} B i ∈ B that is a ◃ \triangleleft ◃ -maximum for B \mathcal{B} B . Then we assign p ( 1 ) = i p(1)=i p ( 1 ) = i , in order to make B i B_i B i a ≤ P \leq_{\mathcal{P}} ≤ P -maximum for B \mathcal{B} B . We then consider the set B ∖ { B i } \mathcal{B}\setminus\{B_i\} B ∖ { B i } and repeat the process, until assigning an index in the permutation to every bud node in B \mathcal{B} B . The previously described assignment, clearly preserves tree-compatibility.
Example of non-coherently decorated pre-proof
Let G = ν X . μ Y . X G=\nu X.\mu Y.X G = ν X . μ Y . X and H = μ Y . G H=\mu Y.G H = μ Y . G . The following pre-proof, cannot be coherently decorated, since the split between its only companion and bud node are different:
( † ) ⊢ G , H ⊢ H , H ⊢ H , H ( X ) ( μ ) ( ν ) \frac{(\dagger)}
{\frac{\vdash G,H}{\frac{\vdash H,H}{\vdash H,H}\ (X)}\ (\mu)}\ (\nu) ⊢ H , H ⊢ H , H ( X ) ⊢ G , H ( μ ) ( † ) ( ν )
However, after the execution of the Algorithm 1, the resulting pre-proof is:
( † ) ⊢ G , H ⊢ H , H ⊢ H , H ⊢ G , H ⊢ H , H ⊢ H , H ⊢ H , H ( X ) ( μ ) ( ν ) ( μ ) ( ν ) ( μ ) ( ν ) \frac{(\dagger)}
{\frac{\vdash G,H}
{\frac{\vdash H,H}
{\frac{\vdash H,H}
{\frac{\vdash G,H}
{\frac{\vdash H,H}
{\frac{\vdash H,H}
{\vdash H,H}\ (X)}\ (\mu)}\ (\nu)}\ (\mu)}\ (\nu)}\ (\mu)}\ (\nu) ⊢ H , H ⊢ H , H ( X ) ⊢ H , H ( μ ) ⊢ G , H ( ν ) ⊢ H , H ( μ ) ⊢ H , H ( ν ) ⊢ G , H ( μ ) ( † ) ( ν )
Which can be coherently decorated.
Algorithms
Algorithms 1 and 2, can be found in the Figure 7 . In the following subsections, we present the proofs for their correctness.
Algorithm 1 Split Invariance
1: input 2: P 0 = ( D 0 , R 0 ) \mathcal{P}_0=(\mathcal{D}_0,\mathcal{R}_0) P 0 = ( D 0 , R 0 ) where D 0 = ( V 0 , s 0 , r 0 , p 0 ) \mathcal{D}_0=(V_0,s_0,r_0,p_0) D 0 = ( V 0 , s 0 , r 0 , p 0 ) with root v r ∈ V v_r\in V v r ∈ V , R 0 \mathcal{R}_0 R 0 injective and cut formulas C = { ( C 1 , C 1 ⊥ ) , … , ( C n , C n ⊥ ) } \mathcal{C}=\{(C_1,C_1^\perp),\ldots,(C_n,C_n^\perp)\} C = {( C 1 , C 1 ⊥ ) , … , ( C n , C n ⊥ )} 3: Partition L , R L,R L , R of { 1 , … , n } \{1,\ldots,n\} { 1 , … , n } where C L = Voc ( C i , i ∈ L ) \mathcal{C}^{L}=\operatorname{Voc}(C_i,i\in L) C L = Voc ( C i , i ∈ L ) and C R = Voc ( C i , i ∈ R ) \mathcal{C}^{R}=\operatorname{Voc}(C_i,i\in R) C R = Voc ( C i , i ∈ R ) . 4: ( Γ 0 , Δ 0 ) ∈ S (\Gamma_0,\Delta_0)\in S ( Γ 0 , Δ 0 ) ∈ S 5: output 6: P = ( D , R ) \mathcal{P}=(\mathcal{D},\mathcal{R}) P = ( D , R ) where D = ( V , s , r , p ) \mathcal{D}=(V,s,r,p) D = ( V , s , r , p ) where V ⊆ V 0 × N V\subseteq V_0\times\mathbb{N} V ⊆ V 0 × N 7: global variables 8: n ∈ N n\in\mathbb{N} n ∈ N 9: begin 10: n : = 0 n:=0 n := 0 ; p : = ∅ p:=\varnothing p := ∅ ; R : = ∅ \mathcal{R}:=\varnothing R := ∅ ; 11: V : = { ( v r , n ) } V:=\{(v_r,n)\} V := {( v r , n )} ; s : = { ( ( v r , n ) , s 0 ( v r ) ) } s:=\{((v_r,n),s_0(v_r))\} s := {(( v r , n ) , s 0 ( v r ))} ; r : = { ( ( v r , n ) , r 0 ( v r ) ) } r:=\{((v_r,n),r_0(v_r))\} r := {(( v r , n ) , r 0 ( v r ))} ; 12: splitInvariance ( v r , n ) ( Γ 0 , Δ 0 ) ∅ \operatorname{splitInvariance}(v_r,n)(\Gamma_0,\Delta_0)\ \varnothing splitInvariance ( v r , n ) ( Γ 0 , Δ 0 ) ∅ 13: end 14: procedure splitInvariance ( v , k ) ( Γ , Δ ) A \operatorname{splitInvariance}(v,k)(\Gamma,\Delta)\ \mathcal{A} splitInvariance ( v , k ) ( Γ , Δ ) A 15: parameters 16: ( v , k ) ∈ V 0 × N (v,k)\in V_0\times\mathbb{N} ( v , k ) ∈ V 0 × N 17: ( Γ , Δ ) ∈ S (\Gamma,\Delta)\in S ( Γ , Δ ) ∈ S 18: A : img ( R 0 ) × S → N \mathcal{A}:\operatorname{img}(\mathcal{R}_0)\times S\to\mathbb{N} A : img ( R 0 ) × S → N 19: begin 20: if v ∈ Bud ( D 0 ) v\in\operatorname{Bud}(\mathcal{D}_0) v ∈ Bud ( D 0 ) then 21: if ( ( R 0 ( v ) , ( Γ , Δ ) ) , i ) ∈ A ((\mathcal{R}_0(v),(\Gamma,\Delta)),i)\in\mathcal{A} (( R 0 ( v ) , ( Γ , Δ )) , i ) ∈ A for some i i i then 22: R : = R ∪ { ( ( v , k ) , ( R 0 ( v ) , i ) ) } \mathcal{R}:=\mathcal{R}\cup\{((v,k),(\mathcal{R}_0(v),i))\} R := R ∪ {(( v , k ) , ( R 0 ( v ) , i ))} 23: else 24: n : = n + 1 n:=n+1 n := n + 1 25: replace ( v , k ) (v,k) ( v , k ) with ( R 0 ( v ) , n ) (\mathcal{R}_0(v),n) ( R 0 ( v ) , n ) labelled s 0 ( R 0 ( v ) ) s_0(\mathcal{R}_0(v)) s 0 ( R 0 ( v )) with rule r 0 ( R 0 ( v ) ) r_0(\mathcal{R}_0(v)) r 0 ( R 0 ( v )) in D \mathcal{D} D 26: splitInvariance ( R 0 ( v ) , n ) ( Γ , Δ ) A \operatorname{splitInvariance}(\mathcal{R}_0(v),n)(\Gamma,\Delta)\ \mathcal{A} splitInvariance ( R 0 ( v ) , n ) ( Γ , Δ ) A 27: end if 28: else 29: if v ∈ img ( R 0 ) v\in\operatorname{img}(\mathcal{R}_0) v ∈ img ( R 0 ) and ( v , ( Γ , Δ ) ) ∉ dom ( A ) (v,(\Gamma,\Delta))\notin\operatorname{dom}(\mathcal{A}) ( v , ( Γ , Δ )) ∈ / dom ( A ) then 30: A : = A ∪ { ( v , ( Γ , Δ ) ) } \mathcal{A}:=\mathcal{A}\cup\{(v,(\Gamma,\Delta))\} A := A ∪ {( v , ( Γ , Δ ))} 31: end if 32: for each child v ′ v' v ′ of v v v in D 0 \mathcal{D}_0 D 0 do 33: add node ( v ′ , k ) (v',k) ( v ′ , k ) labelled s 0 ( v ′ ) s_0(v') s 0 ( v ′ ) as child of ( v , k ) (v,k) ( v , k ) to D \mathcal{D} D 34: if v ′ ∉ Bud ( D 0 ) v'\notin\operatorname{Bud}(\mathcal{D}_0) v ′ ∈ / Bud ( D 0 ) then define r ( v ′ , k ) = r 0 ( v ′ ) r(v',k)=r_0(v') r ( v ′ , k ) = r 0 ( v ′ ) else r ( v ′ , k ) r(v',k) r ( v ′ , k ) is undefined end if 35: calculate new split ( Γ ′ , Δ ′ ) (\Gamma',\Delta') ( Γ ′ , Δ ′ ) of s 0 ( v ′ ) s_0(v') s 0 ( v ′ ) based on ancestor relation of rule r 0 ( v ′ ) r_0(v') r 0 ( v ′ ) 36: if r 0 ( v ′ ) = ( Cut ) r_0(v')=(\operatorname{Cut}) r 0 ( v ′ ) = ( Cut ) with cut-formula C i C_i C i then 37: add C i C_i C i (resp. C i ⊥ C_i^\perp C i ⊥ ) to Γ ′ \Gamma' Γ ′ or Δ ′ \Delta' Δ ′ , according to the partitions C L \mathcal{C}^{L} C L and C R \mathcal{C}^{R} C R 38: end if 39: splitInvariance ( v ′ , k ) ( Γ ′ , Δ ′ ) A \operatorname{splitInvariance}(v',k)(\Gamma',\Delta')\ \mathcal{A} splitInvariance ( v ′ , k ) ( Γ ′ , Δ ′ ) A 40: end for 41: end if 42: end
Algorithm 2 Tree-compatibility [37 ]
1: input 2: P 0 = ( D 0 , R 0 ) \mathcal{P}_0=(\mathcal{D}_0,\mathcal{R}_0) P 0 = ( D 0 , R 0 ) where D 0 = ( V 0 , s 0 , r 0 , p 0 ) \mathcal{D}_0=(V_0,s_0,r_0,p_0) D 0 = ( V 0 , s 0 , r 0 , p 0 ) with root v r ∈ V v_r\in V v r ∈ V , and R 0 \mathcal{R}_0 R 0 injective 3: Ordered trace manifold for P 0 \mathcal{P}_0 P 0 , with induction order ( Bud ( D 0 ) , ◃ 0 ) (\operatorname{Bud}(\mathcal{D}_0),\triangleleft_0) ( Bud ( D 0 ) , ◃ 0 ) 4: output 5: P = ( D , R ) \mathcal{P}=(\mathcal{D},\mathcal{R}) P = ( D , R ) where D = ( V , s , r , p ) \mathcal{D}=(V,s,r,p) D = ( V , s , r , p ) where V ⊆ V 0 × N V\subseteq V_0\times\mathbb{N} V ⊆ V 0 × N 6: global variables 7: n ∈ N n\in\mathbb{N} n ∈ N 8: begin 9: n : = 0 n:=0 n := 0 ; p : = ∅ p:=\varnothing p := ∅ ; R : = ∅ \mathcal{R}:=\varnothing R := ∅ ; 10: V : = { ( v r , n ) } V:=\{(v_r,n)\} V := {( v r , n )} ; s : = { ( ( v r , n ) , s 0 ( v r ) ) } s:=\{((v_r,n),s_0(v_r))\} s := {(( v r , n ) , s 0 ( v r ))} ; r : = { ( ( v r , n ) , r 0 ( v r ) ) } r:=\{((v_r,n),r_0(v_r))\} r := {(( v r , n ) , r 0 ( v r ))} ; 11: treeCompatibility ( v r , n ) ∅ \operatorname{treeCompatibility}(v_r,n)\ \varnothing treeCompatibility ( v r , n ) ∅ 12: end 13: procedure treeCompatibility ( v , k ) A \operatorname{treeCompatibility}(v,k)\ \mathcal{A} treeCompatibility ( v , k ) A 14: parameters 15: ( v , k ) ∈ V 0 × N (v,k)\in V_0\times\mathbb{N} ( v , k ) ∈ V 0 × N 16: A : img ( R 0 ) → N \mathcal{A}:\operatorname{img}(\mathcal{R}_0)\to\mathbb{N} A : img ( R 0 ) → N 17: begin 18: if v ∈ Bud ( D 0 ) v\in\operatorname{Bud}(\mathcal{D}_0) v ∈ Bud ( D 0 ) then 19: if ( R 0 ( v ) , i ) ∈ A (\mathcal{R}_0(v),i)\in\mathcal{A} ( R 0 ( v ) , i ) ∈ A for some i i i then 20: R : = R ∪ { ( ( v , k ) , ( R 0 ( v ) , i ) ) } \mathcal{R}:=\mathcal{R}\cup\{((v,k),(\mathcal{R}_0(v),i))\} R := R ∪ {(( v , k ) , ( R 0 ( v ) , i ))} 21: else 22: n : = n + 1 n:=n+1 n := n + 1 23: replace ( v , k ) (v,k) ( v , k ) with ( R 0 ( v ) , n ) (\mathcal{R}_0(v),n) ( R 0 ( v ) , n ) labelled s 0 ( R 0 ( v ) ) s_0(\mathcal{R}_0(v)) s 0 ( R 0 ( v )) with rule r 0 ( R 0 ( v ) ) r_0(\mathcal{R}_0(v)) r 0 ( R 0 ( v )) in D \mathcal{D} D 24: treeCompatibility ( R 0 ( v ) , n ) A \operatorname{treeCompatibility}(\mathcal{R}_0(v),n)\ \mathcal{A} treeCompatibility ( R 0 ( v ) , n ) A 25: end if 26: else 27: if v ∈ img ( R 0 ) v\in\operatorname{img}(\mathcal{R}_0) v ∈ img ( R 0 ) and v ∉ dom ( A ) v\notin\operatorname{dom}(\mathcal{A}) v ∈ / dom ( A ) then 28: A : = { ( v , k ) } ∪ { ( v ′ , k ′ ) ∈ A ∣ B ◃ 0 B ′ , where R 0 ( v ) = B and R 0 ( v ′ ) = B ′ } \mathcal{A}:=\{(v,k)\}\cup\{(v',k')\in\mathcal{A}\mid B\triangleleft_0 B',\text{ where }\mathcal{R}_0(v)=B\text{ and }\mathcal{R}_0(v')=B'\} A := {( v , k )} ∪ {( v ′ , k ′ ) ∈ A ∣ B ◃ 0 B ′ , where R 0 ( v ) = B and R 0 ( v ′ ) = B ′ } 29: end if 30: for each child v ′ v' v ′ of v v v in D 0 \mathcal{D}_0 D 0 do 31: add node ( v ′ , k ) (v',k) ( v ′ , k ) labelled s 0 ( v ′ ) s_0(v') s 0 ( v ′ ) as child of ( v , k ) (v,k) ( v , k ) to D \mathcal{D} D 32: … \ldots …
Figure 7. Split Invariance and Tree-compatibility algorithms. Algorithm 2 was originally introduced by Sprenger and Dalim in [37 ] . Algorithm 1 can be seen as an extension of Algorithm 2, that uses the same unfolding procedure.
Correctness of Algorithm 1
Proof of Lemma 32. To prove the total correctness of the algorithm we show that: 1) the algorithm 1 terminates, 2) P = ( D , R ) \mathcal{P}=(\mathcal{D},\mathcal{R}) P = ( D , R ) is a pre-proof, 3) P \mathcal{P} P has a trace manifold, and 4) P \mathcal{P} P can be coherently decorated.
(1): After the initial call (line 12), for all bud node B ∈ Bud ( D 0 ) B\in\operatorname{Bud}(\mathcal{D}_0) B ∈ Bud ( D 0 ) , then the algorithm creates the unique path from the root ( v r , 0 ) (v_r,0) ( v r , 0 ) in D \mathcal{D} D to the bud node ( B , 0 ) (B,0) ( B , 0 ) in D \mathcal{D} D , while also propagating the split of the conclusion (lines 32–39). Moreover, it stores the information of the companion node R 0 ( B ) \mathcal{R}_0(B) R 0 ( B ) and its split ( Γ , Δ ) (\Gamma,\Delta) ( Γ , Δ ) in A \mathcal{A} A (lines 29–30). In the call $\operatorname{splitInvariance ( B , 0 ) (B,0) ( B , 0 ) ( Γ ′ , Δ ′ ) (\Gamma',\Delta') ( Γ ′ , Δ ′ ) A \mathcal{A} A , if ( Γ ′ , Δ ′ ) = ( Γ , Δ ) (\Gamma',\Delta')=(\Gamma,\Delta) ( Γ ′ , Δ ′ ) = ( Γ , Δ ) then the pair ( ( B , 0 ) ) , ( R 0 ( B ) , 0 ) ((B,0)),(\mathcal{R}_0(B),0) (( B , 0 )) , ( R 0 ( B ) , 0 ) is included in R \mathcal{R} R , and the unfolding is finished for that branch in the pre-proof D \mathcal{D} D (lines 21-22). On the contrary, if ( Γ ′ , Δ ′ ) ≠ ( Γ , Δ ) (\Gamma',\Delta')\ne(\Gamma,\Delta) ( Γ ′ , Δ ′ ) = ( Γ , Δ ) then the node ( B , 0 ) (B,0) ( B , 0 ) will be replaced by ( R 0 ( B ) , 1 ) (\mathcal{R}_0(B),1) ( R 0 ( B ) , 1 ) , and it will serve as the root of the new derivation tree that will be unfolded in the branch (lines 23-26) and the previously explained process is repeated. Therefore, to prove that there is no infinite unfolding of the pre-proof D \mathcal{D} D , we must show that eventually all bud nodes ( B , i ) (B,i) ( B , i ) in D \mathcal{D} D are matched with a companion node in A \mathcal{A} A with the same split. However, note that the number of possible splits of s 0 ( B ) s_0(B) s 0 ( B ) is finite, since s 0 ( B ) s_0(B) s 0 ( B ) is a finite list of formulas. Since the amount of splits is finite and the times we can unfold the proof is infinite, we can ensure that eventually, for some l l l , ( R 0 ( B ) , l ) (\mathcal{R}_0(B),l) ( R 0 ( B ) , l ) will appear in A \mathcal{A} A , with the same split as the bud node ( B , i ) (B,i) ( B , i ) .
(2): We prove by induction on the recursion tree of the algorithm, that after each call of splitInvariance ( v , k ) ( Γ , Δ ) \operatorname{splitInvariance}(v,k)\;(\Gamma,\Delta) splitInvariance ( v , k ) ( Γ , Δ ) , then P \mathcal{P} P is a pre-proof. In the initial call (line 11), D \mathcal{D} D contains only the root node ( v r , 0 ) (v_r,0) ( v r , 0 ) . If r 0 ( v r ) r_0(v_r) r 0 ( v r ) is a rule with no premises, then no node is added and P \mathcal{P} P is a pre-proof with a single node. On the contrary, if r 0 ( v r ) r_0(v_r) r 0 ( v r ) has premises, then each of the nodes corresponding to these premises are added to D \mathcal{D} D according to the p 0 p_0 p 0 function of D 0 \mathcal{D}_0 D 0 (line 32), which by hypothesis is a well-defined edge and therefore, P \mathcal{P} P is a pre-proof. The case where r 0 ( v r ) r_0(v_r) r 0 ( v r ) is undefined, i.e. v r v_r v r is a bud node, is impossible since it contradicts the fact that P 0 \mathcal{P}_0 P 0 is pre-proof.
Now for the inductive step, suppose a call splitInvariance ( v , k ) ( Γ , Δ ) A \operatorname{splitInvariance}(v,k)\;(\Gamma,\Delta)\;\mathcal{A} splitInvariance ( v , k ) ( Γ , Δ ) A . We know by construction of the algorithm, that ( v , k ) (v,k) ( v , k ) was already added to D \mathcal{D} D in the previous step (line 24 or line 32). Moreover, by induction hypothesis we know that before the execution of the recursive call, P \mathcal{P} P is a pre-proof. It remains to show that after its execution, P \mathcal{P} P is still a pre-proof. We proceed by cases over r 0 ( v ) r_0(v) r 0 ( v ) . If r 0 ( v ) r_0(v) r 0 ( v ) is a rule with or without premises, we use the same arguments as in the base case. If r 0 ( v ) r_0(v) r 0 ( v ) is undefined, there are two cases: if ( ( R 0 ( v ) , ( Γ ′ , Δ ′ ) ) , i ) ((\mathcal{R}_0(v),(\Gamma',\Delta')),i) (( R 0 ( v ) , ( Γ ′ , Δ ′ )) , i ) is in A \mathcal{A} A for some i i i , and ( Γ , Δ ) = ( Γ ′ , Δ ′ ) (\Gamma,\Delta)=(\Gamma',\Delta') ( Γ , Δ ) = ( Γ ′ , Δ ′ ) , then the back edge ( ( v , k ) , ( R 0 ( v ) , i ) ) ((v,k),(\mathcal{R}_0(v),i)) (( v , k ) , ( R 0 ( v ) , i )) is added to R \mathcal{R} R (line 21). Since s 0 ( v ) = s 0 ( R 0 ( v ) ) s_0(v)=s_0(\mathcal{R}_0(v)) s 0 ( v ) = s 0 ( R 0 ( v )) , then s ( ( v , k ) ) = s ( ( R 0 ( v ) , i ) ) s((v,k))=s((\mathcal{R}_0(v),i)) s (( v , k )) = s (( R 0 ( v ) , i )) , and therefore P \mathcal{P} P is a pre-proof. Otherwise, if ( Γ , Δ ) ≠ ( Γ ′ , Δ ′ ) (\Gamma,\Delta)\ne(\Gamma',\Delta') ( Γ , Δ ) = ( Γ ′ , Δ ′ ) then ( v , k ) (v,k) ( v , k ) is replaced with the node ( R 0 ( v ) , n ) (\mathcal{R}_0(v),n) ( R 0 ( v ) , n ) with the rule r 0 ( R 0 ( v ) ) r_0(\mathcal{R}_0(v)) r 0 ( R 0 ( v )) and labelling s 0 ( R 0 ( v ) ) s_0(\mathcal{R}_0(v)) s 0 ( R 0 ( v )) (line 24), which constructs a well-defined edge, and therefore P \mathcal{P} P is a pre-proof.
(3): Consider any ( B , k ) ∈ Bud ( D ) (B,k)\in\operatorname{Bud}(\mathcal{D}) ( B , k ) ∈ Bud ( D ) and π B \pi_B π B as the unique path from R ( ( B , k ) ) \mathcal{R}((B,k)) R (( B , k )) to ( B , k ) (B,k) ( B , k ) in D \mathcal{D} D . By inspection on the Algorithm 1, π B \pi_B π B is obtained by unfolding a set of bud nodes U = { B 1 , … , B m } ⊆ Bud ( D 0 ) U=\{B_1,\ldots,B_m\}\subseteq\operatorname{Bud}(\mathcal{D}_0) U = { B 1 , … , B m } ⊆ Bud ( D 0 ) , represented by the following sequence of recursive calls:
c = ( ( R ( B , k ) ) , A 0 ) … ( ( B 1 , k 1 ) , A 1 ) … ( ( B m , k m ) , A m ) … ( ( B , k ) , A m + 1 ) c=((\mathcal{R}(B,k)),\mathcal{A}_0)\ldots((B_1,k_1),\mathcal{A}_1)\ldots((B_m,k_m),\mathcal{A}_m)\ldots((B,k),\mathcal{A}_{m+1}) c = (( R ( B , k )) , A 0 ) … (( B 1 , k 1 ) , A 1 ) … (( B m , k m ) , A m ) … (( B , k ) , A m + 1 )
Where each B j ∈ Bud ( D 0 ) B_j\in\operatorname{Bud}(\mathcal{D}_0) B j ∈ Bud ( D 0 ) , for 1 ≤ j ≤ m 1\le j\le m 1 ≤ j ≤ m , is an unpaired bud node that will be replaced by its companion in the line 25. Therefore, the path π B \pi_B π B will has the form:
π B = R ( B , k ) … ( R 0 ( B 1 ) , k 1 ′ ) … ( R 0 ( B m ) , k m ′ ) … ( B , k ) \pi_B=\mathcal{R}(B,k)\ldots(\mathcal{R}_0(B_1),k'_1)\ldots(\mathcal{R}_0(B_m),k'_m)\ldots(B,k) π B = R ( B , k ) … ( R 0 ( B 1 ) , k 1 ′ ) … ( R 0 ( B m ) , k m ′ ) … ( B , k )
Let B π B = U ∪ { B } ⊆ Bud ( D 0 ) \mathcal{B}_{\pi_B}=U\cup\{B\}\subseteq\operatorname{Bud}(\mathcal{D}_0) B π B = U ∪ { B } ⊆ Bud ( D 0 ) , and consider the path π ^ B = R 0 ( B ) … R 0 ( B 1 ) … R 0 ( B m ) … B \widehat{\pi}_B=\mathcal{R}_0(B)\ldots\mathcal{R}_0(B_1)\ldots\mathcal{R}_0(B_m)\ldots B π B = R 0 ( B ) … R 0 ( B 1 ) … R 0 ( B m ) … B , obtained after eliminating the numbers of the pairs describing the nodes in the path π B \pi_B π B . From the definition of the Algorithm 1, one deduces that π ^ B \widehat{\pi}_B π B describes a cycle contained in the strongly connected subgraph S π B = ⋃ B ′ ∈ B π B C B ′ S_{\pi_B}=\bigcup_{B'\in\mathcal{B}_{\pi_B}}C_{B'} S π B = ⋃ B ′ ∈ B π B C B ′ , in the graph G P 0 \mathcal{G}_{\mathcal{P}_0} G P 0 (here we implicitly use Lemma 25). Therefore, by Lemma 44, we get that:
The unique path π B \pi_B π B , represented as the basic cycle C B , k C_{B,k} C B , k in G P \mathcal{G}_{\mathcal{P}} G P , is formed by the composition of paths (of length > 0 >0 > 0 ) each of the form ( R 0 ( B i ) , k i ) … ( R 0 ( B j ) , k j ) (\mathcal{R}_0(B_i),k_i)\ldots(\mathcal{R}_0(B_j),k_j) ( R 0 ( B i ) , k i ) … ( R 0 ( B j ) , k j ) , where B i , B j ∈ B π B B_i,B_j\in\mathcal{B}_{\pi_B} B i , B j ∈ B π B and R 0 ( B i ) … R 0 ( B j ) \mathcal{R}_0(B_i)\ldots\mathcal{R}_0(B_j) R 0 ( B i ) … R 0 ( B j ) is a subpath of some cycle C B ′ ′ ⊆ S π B = ⋃ B ′ ∈ B π B C B ′ C_{B''}\subseteq S_{\pi_B}=\bigcup_{B'\in\mathcal{B}_{\pi_B}}C_{B'} C B ′′ ⊆ S π B = ⋃ B ′ ∈ B π B C B ′ in G P 0 \mathcal{G}_{\mathcal{P}_0} G P 0 . (A)
With this in mind, we show how to construct the traces for the trace manifold for P \mathcal{P} P . Take any strongly connected subgraph S S S of G P \mathcal{G}_{\mathcal{P}} G P . By Lemma 25 there exists a set B ⊆ B u d ( D ) \mathcal{B} \subseteq Bud(\mathcal{D}) B ⊆ B u d ( D ) such that S = ⋃ B ∈ B C B S = \bigcup_{B\in\mathcal{B}} C_B S = ⋃ B ∈ B C B . If B ∈ B B \in\mathcal{B} B ∈ B , we show how to construct the trace τ S , B \tau_{S,B} τ S , B up the basic cycle obtained from the unique path π B \pi_B π B from R ( B ) \mathcal{R}(B) R ( B ) to B B B in D \mathcal{D} D .
As mentioned before, π B \pi_B π B is defined by a set of unfolded bud nodes B π B ⊆ B u d ( D 0 ) \mathcal{B}_{\pi_B} \subseteq Bud(\mathcal{D}_0) B π B ⊆ B u d ( D 0 ) such that π B \pi_B π B corresponds to a strongly connected subgraph S π B = ⋃ B ′ ∈ B π B C B ′ S_{\pi_B} = \bigcup_{B'\in\mathcal{B}_{\pi_B}} C_{B'} S π B = ⋃ B ′ ∈ B π B C B ′ in G P 0 \mathcal{G}_{\mathcal{P}_0} G P 0 . Therefore, if we forget of the indexes in the strongly connected subgraph S S S and we call it S ^ \widehat{S} S , then S ^ = ⋃ B ∈ B S π B = ⋃ B ∈ B ( ⋃ B ′ ∈ B π B C B ′ ) \widehat{S} = \bigcup_{B\in\mathcal{B}} S_{\pi_B} = \bigcup_{B\in\mathcal{B}}(\bigcup_{B'\in\mathcal{B}_{\pi_B}} C_{B'}) S = ⋃ B ∈ B S π B = ⋃ B ∈ B ( ⋃ B ′ ∈ B π B C B ′ ) . Note that this is the union of intersecting strongly connected subgraphs. Indeed, S ^ \widehat{S} S describes a cycle in the original graph G P 0 \mathcal{G}_{\mathcal{P}_0} G P 0 . For this reason:
For all bud node B ∈ B ⊆ B u d ( D ) B \in\mathcal{B} \subseteq Bud(\mathcal{D}) B ∈ B ⊆ B u d ( D ) , and B ′ ∈ B π B ⊆ B u d ( D 0 ) B' \in\mathcal{B}_{\pi_B} \subseteq Bud(\mathcal{D}_0) B ′ ∈ B π B ⊆ B u d ( D 0 ) , then τ S ^ , B ′ \tau_{\widehat{S},B'} τ S , B ′ exists in the trace manifold for P 0 \mathcal{P}_0 P 0 , and is a trace following the basic cycle C B ′ C_{B'} C B ′ in G P 0 \mathcal{G}_{\mathcal{P}_0} G P 0 . (B)
The trace τ S , B \tau_{S,B} τ S , B for the basic cycle C B C_B C B obtained from the path π B \pi_B π B is constructed in the following way: take every segment ( R 0 ( B i ) , k i ) … ( R 0 ( B j ) , k j ) (\mathcal{R}_0(B_i),k_i)\ldots(\mathcal{R}_0(B_j),k_j) ( R 0 ( B i ) , k i ) … ( R 0 ( B j ) , k j ) of C B C_B C B , as described in (A). The segment R 0 ( B i ) … R 0 ( B j ) \mathcal{R}_0(B_i)\ldots\mathcal{R}_0(B_j) R 0 ( B i ) … R 0 ( B j ) is a subpath of some cycle C B ′ ⊆ ⋃ B ′ ′ ∈ B π B C B ′ ′ ⊆ S ^ C_{B'} \subseteq\bigcup_{B''\in\mathcal{B}_{\pi_B}} C_{B''} \subseteq\widehat{S} C B ′ ⊆ ⋃ B ′′ ∈ B π B C B ′′ ⊆ S where B ′ ∈ B π B B' \in\mathcal{B}_{\pi_B} B ′ ∈ B π B . Therefore, by (B) we can use the trace τ S ^ , B ′ \tau_{\widehat{S},B'} τ S , B ′ to cover the segment R 0 ( B i ) … R 0 ( B j ) \mathcal{R}_0(B_i)\ldots\mathcal{R}_0(B_j) R 0 ( B i ) … R 0 ( B j ) , and consequently the segment ( R 0 ( B i ) , k i ) … ( R 0 ( B j ) , k j ) (\mathcal{R}_0(B_i),k_i)\ldots(\mathcal{R}_0(B_j),k_j) ( R 0 ( B i ) , k i ) … ( R 0 ( B j ) , k j ) . In case the segment R 0 ( B i ) … R 0 ( B j ) \mathcal{R}_0(B_i)\ldots\mathcal{R}_0(B_j) R 0 ( B i ) … R 0 ( B j ) has a progress point, and it is contained in multiple basic cycles, we always choose the trace containing the formula that progresses, as explained in the proof of Lemma 29 .
Note that these traces can be composed, by the property 2 of the original trace manifold for P 0 \mathcal{P}_0 P 0 : consider any segment ( R 0 ( B i ) , k i ) … ( R 0 ( B j ) , k j ) … ( R 0 ( B h ) , k h ) (\mathcal{R}_0(B_i),k_i)\ldots(\mathcal{R}_0(B_j),k_j)\ldots(\mathcal{R}_0(B_h),k_h) ( R 0 ( B i ) , k i ) … ( R 0 ( B j ) , k j ) … ( R 0 ( B h ) , k h ) . Such that the segments R 0 ( B i ) … R 0 ( B j ) \mathcal{R}_0(B_i)\ldots\mathcal{R}_0(B_j) R 0 ( B i ) … R 0 ( B j ) and R 0 ( B j ) … R 0 ( B h ) \mathcal{R}_0(B_j)\ldots\mathcal{R}_0(B_h) R 0 ( B j ) … R 0 ( B h ) are included in the cycles C B 1 , C B 2 ⊆ ⋃ B ′ ′ ∈ B π B C B ′ ′ C_{B_1}, C_{B_2} \subseteq\bigcup_{B''\in\mathcal{B}_{\pi_B}} C_{B''} C B 1 , C B 2 ⊆ ⋃ B ′′ ∈ B π B C B ′′ respectively, in G P 0 \mathcal{G}_{\mathcal{P}_0} G P 0 . This implies that B j ≤ P 0 B 1 B_j \leq_{\mathcal{P}_0} B_1 B j ≤ P 0 B 1 and B j ≤ P 0 B 2 B_j \leq_{\mathcal{P}_0} B_2 B j ≤ P 0 B 2 , and therefore τ S ^ , B 1 ( R 0 ( B j ) ) = τ S ^ , B j ( R 0 ( B j ) ) = τ S ^ , B 2 ( R 0 ( B j ) ) \tau_{\widehat{S},B_1}(\mathcal{R}_0(B_j)) = \tau_{\widehat{S},B_j}(\mathcal{R}_0(B_j)) = \tau_{\widehat{S},B_2}(\mathcal{R}_0(B_j)) τ S , B 1 ( R 0 ( B j )) = τ S , B j ( R 0 ( B j )) = τ S , B 2 ( R 0 ( B j )) .
We show that this construction satisfies the three properties of trace manifolds. Remember that for all B = ( B i , k i ) ∈ B u d ( D ) B = (B_i,k_i) \in Bud(\mathcal{D}) B = ( B i , k i ) ∈ B u d ( D ) then R ( B ) = ( R 0 ( B i ) , k j ) \mathcal{R}(B) = (\mathcal{R}_0(B_i),k_j) R ( B ) = ( R 0 ( B i ) , k j ) .
By construction, the trace τ S , B \tau_{S,B} τ S , B is a trace over the basic cycle C B C_B C B in G P \mathcal{G}_{\mathcal{P}} G P . We check that τ S , B \tau_{S,B} τ S , B has a unique value at R ( B ) \mathcal{R}(B) R ( B ) . By construction τ S , B ( R ( B ) ) = τ S ^ , B ′ ( R ( B ) ‾ ) \tau_{S,B}(\mathcal{R}(B)) = \tau_{\widehat{S},B'}(\overline{\mathcal{R}(B)}) τ S , B ( R ( B )) = τ S , B ′ ( R ( B ) ) and τ S , B ( R ( B ) ) = τ S ^ , B ′ ′ ( R ( B ) ‾ ) \tau_{S,B}(\mathcal{R}(B)) = \tau_{\widehat{S},B''}(\overline{\mathcal{R}(B)}) τ S , B ( R ( B )) = τ S , B ′′ ( R ( B ) ) for some B ′ , B ′ ′ ∈ B π B B', B'' \in\mathcal{B}_{\pi_B} B ′ , B ′′ ∈ B π B . Note that R ( B ) ‾ \overline{\mathcal{R}(B)} R ( B ) is in C B ′ C_{B'} C B ′ and C B ′ ′ C_{B''} C B ′′ , implying B ^ ≤ P 0 B ′ \widehat{B} \leq_{\mathcal{P}_0} B' B ≤ P 0 B ′ and B ^ ≤ P 0 B ′ ′ \widehat{B} \leq_{\mathcal{P}_0} B'' B ≤ P 0 B ′′ . Then by property 2 of the trace manifold, τ S ^ , B ′ ( R ( B ) ‾ ) = τ S ^ , B ^ ( R ( B ) ‾ ) = τ S ^ , B ′ ′ ( R ( B ) ‾ ) \tau_{\widehat{S},B'}(\overline{\mathcal{R}(B)}) = \tau_{\widehat{S},\widehat{B}}(\overline{\mathcal{R}(B)}) = \tau_{\widehat{S},B''}(\overline{\mathcal{R}(B)}) τ S , B ′ ( R ( B ) ) = τ S , B ( R ( B ) ) = τ S , B ′′ ( R ( B ) ) . Thus, τ S , B \tau_{S,B} τ S , B takes the same value at R ( B ) \mathcal{R}(B) R ( B ) .
Let B 1 , B 2 ∈ B B_1,B_2 \in\mathcal{B} B 1 , B 2 ∈ B such that B 1 ≤ P B 2 B_1 \leq_{\mathcal{P}} B_2 B 1 ≤ P B 2 . By construction of τ S , B 1 ( R ( B 1 ) ) = τ S ^ , B ′ ( R ( B 1 ) ‾ ) \tau_{S,B_1}(\mathcal{R}(B_1)) = \tau_{\widehat{S},B'}(\overline{\mathcal{R}(B_1)}) τ S , B 1 ( R ( B 1 )) = τ S , B ′ ( R ( B 1 ) ) and τ S , B 2 ( R ( B 1 ) ) = τ S ^ , B ′ ′ ( R ( B 1 ) ‾ ) \tau_{S,B_2}(\mathcal{R}(B_1)) = \tau_{\widehat{S},B''}(\overline{\mathcal{R}(B_1)}) τ S , B 2 ( R ( B 1 )) = τ S , B ′′ ( R ( B 1 ) ) for some B ′ ∈ B π B 1 B' \in\mathcal{B}_{\pi_{B_1}} B ′ ∈ B π B 1 and B ′ ′ ∈ B π B 2 B'' \in\mathcal{B}_{\pi_{B_2}} B ′′ ∈ B π B 2 . Hence, R ( B 1 ) ‾ \overline{\mathcal{R}(B_1)} R ( B 1 ) is in C B ′ C_{B'} C B ′ and C B ′ ′ C_{B''} C B ′′ , which implies B ^ 1 ≤ P 0 B ′ \widehat{B}_1 \leq_{\mathcal{P}_0} B' B 1 ≤ P 0 B ′ and B ^ 1 ≤ P 0 B ′ ′ \widehat{B}_1 \leq_{\mathcal{P}_0} B'' B 1 ≤ P 0 B ′′ . Therefore, by property 2 of the trace manifold, τ S ^ , B ′ ( R ( B 1 ) ‾ ) = τ S ^ , B ^ 1 ( R ( B 1 ) ‾ ) = τ S ^ , B ′ ′ ( R ( B 1 ) ‾ ) \tau_{\widehat{S},B'}(\overline{\mathcal{R}(B_1)}) = \tau_{\widehat{S},\widehat{B}_1}(\overline{\mathcal{R}(B_1)}) = \tau_{\widehat{S},B''}(\overline{\mathcal{R}(B_1)}) τ S , B ′ ( R ( B 1 ) ) = τ S , B 1 ( R ( B 1 ) ) = τ S , B ′′ ( R ( B 1 ) ) . Thus τ S , B 1 ( R ( B 1 ) ) = τ S , B 2 ( R ( B 1 ) ) \tau_{S,B_1}(\mathcal{R}(B_1)) = \tau_{S,B_2}(\mathcal{R}(B_1)) τ S , B 1 ( R ( B 1 )) = τ S , B 2 ( R ( B 1 )) .
Let τ S , B 1 , … , τ S , B n \tau_{S,B_1},\ldots,\tau_{S,B_n} τ S , B 1 , … , τ S , B n be the traces associated to S S S in the trace manifold, and consider their respective unique paths π B 1 , … , π B n \pi_{B_1},\ldots,\pi_{B_n} π B 1 , … , π B n from companion to bud node. Assume that B \mathcal{B} B is the set of weakly ≤ P \leq_{\mathcal{P}} ≤ P -connected bud nodes in S S S . As explained before, we can recover the original strongly connected subgraph in G P 0 \mathcal{G}_{\mathcal{P}_0} G P 0 by removing the indexes of the vertices, which gives us S ^ = ⋃ B ∈ B S π B = ⋃ B ∈ B ( ⋃ B ′ ∈ B π B C B ′ ) \widehat{S} = \bigcup_{B\in\mathcal{B}} S_{\pi_B} = \bigcup_{B\in\mathcal{B}}(\bigcup_{B'\in\mathcal{B}_{\pi_B}} C_{B'}) S = ⋃ B ∈ B S π B = ⋃ B ∈ B ( ⋃ B ′ ∈ B π B C B ′ ) . By property 3 of the original trace manifold, then:
There exists a bud node B i ∈ B B_i \in\mathcal{B} B i ∈ B and a bud node B ′ ∈ B π B i B' \in\mathcal{B}_{\pi_{B_i}} B ′ ∈ B π B i , such that τ S , B ′ \tau_{S,B'} τ S , B ′ has an F F F -progress point, such that F F F is minimal in the set ⋃ B ′ ′ ∈ B π B 1 Form ( τ S , B ′ ′ ) ∪ ⋯ ∪ ⋃ B ′ ′ ∈ B π B n Form ( τ S , B ′ ′ ) \bigcup_{B'' \in\mathcal{B}_{\pi_{B_1}}} \operatorname{Form}(\tau_{S,B''}) \cup\cdots\cup\bigcup_{B'' \in\mathcal{B}_{\pi_{B_n}}} \operatorname{Form}(\tau_{S,B''}) ⋃ B ′′ ∈ B π B 1 Form ( τ S , B ′′ ) ∪ ⋯ ∪ ⋃ B ′′ ∈ B π B n Form ( τ S , B ′′ ) . (C)
Statement (C) also implies that this F F F -progress point is located in the basic cycle C B ′ C_{B'} C B ′ of G P 0 \mathcal{G}_{P_0} G P 0 , say in the vertex v v v . By construction, one can see that this vertex is also included in S S S , with the form ( v , k ) (v,k) ( v , k ) for some k k k . Since ( v , k ) (v,k) ( v , k ) is in S S S , it must be the case that there exists some B j ∈ B B_j \in\mathcal{B} B j ∈ B , such that ( v , k ) (v,k) ( v , k ) is included in π B j \pi_{B_j} π B j . By construction of the trace τ S , B j \tau_{S,B_j} τ S , B j , we also know that every time we pass through the segment ( R 0 ( B h ) , k h ) … ( R 0 ( B r ) , k r ) (R_0(B_h),k_h)\ldots(R_0(B_r),k_r) ( R 0 ( B h ) , k h ) … ( R 0 ( B r ) , k r ) , containing the vertex ( v , k ) (v,k) ( v , k ) , we used the segment of the trace τ S , B ′ \tau_{S,B'} τ S , B ′ to cover it. For this reason, τ S , B j \tau_{S,B_j} τ S , B j has an F F F -progress point. We now show that F F F is minimal in all the traces τ S , B 1 , … , τ S , B n \tau_{S,B_1},\ldots,\tau_{S,B_n} τ S , B 1 , … , τ S , B n .
Every trace τ S , B h \tau_{S,B_h} τ S , B h is obtained by the composition of segments of the traces τ S , B ′ ′ \tau_{S,B''} τ S , B ′′ , such that B ′ ′ ∈ B π B h B'' \in\mathcal{B}_{\pi_{B_h}} B ′′ ∈ B π B h . Thus, for all h h h we have that Form ( τ S , B h ) ⊆ ⋃ B ′ ′ ∈ B π B h Form ( τ S , B ′ ′ ) \operatorname{Form}(\tau_{S,B_h}) \subseteq\bigcup_{B'' \in\mathcal{B}_{\pi_{B_h}}} \operatorname{Form}(\tau_{S,B''}) Form ( τ S , B h ) ⊆ ⋃ B ′′ ∈ B π B h Form ( τ S , B ′′ ) . Therefore:
⋃ 1 ≤ h ≤ n Form ( τ S , B h ) ⊆ ⋃ 1 ≤ h ≤ n ( ⋃ B ′ ′ ∈ B π B h Form ( τ S , B ′ ′ ) ) \bigcup_{1 \le h \le n} \operatorname{Form}(\tau_{S,B_h}) \subseteq\bigcup_{1 \le h \le n} \left( \bigcup_{B'' \in\mathcal{B}_{\pi_{B_h}}} \operatorname{Form}(\tau_{S,B''}) \right) 1 ≤ h ≤ n ⋃ Form ( τ S , B h ) ⊆ 1 ≤ h ≤ n ⋃ B ′′ ∈ B π B h ⋃ Form ( τ S , B ′′ )
Therefore, by (C) and the fact that τ S , B j \tau_{S,B_j} τ S , B j has the F F F -progress point, one deduces that F F F is also minimal in ⋃ 1 ≤ h ≤ n Form ( τ S , B i ) \bigcup_{1 \le h \le n} \operatorname{Form}(\tau_{S,B_i}) ⋃ 1 ≤ h ≤ n Form ( τ S , B i ) .
(4): After the ascending phase, we obtain a coherent decoration of the derivation tree D \mathcal{D} D . By construction of the algorithm, it is clear that all pairs ( B , R ( B ) ) ∈ R (B,R(B)) \in\mathcal{R} ( B , R ( B )) ∈ R added in the line 22, are ensured to have the same split, by the conditional in line 21. Therefore, P \mathcal{P} P can be coherently decorated.
Correctness of Algorithm 2
For the proof of correctness, we briefly introduce the following lemma:
Definition 45 (Derivation graph homomorphism). Let G = ( V , s , r , p ) G=(V,s,r,p) G = ( V , s , r , p ) and G ′ = ( V ′ , s ′ , r ′ , p ′ ) G'=(V',s',r',p') G ′ = ( V ′ , s ′ , r ′ , p ′ ) be two derivation graphs. A derivation graph homomorphism from G G G to G ′ G' G ′ is a total function f : V → V ′ f: V \to V' f : V → V ′ satisfying that for any v ∈ V v \in V v ∈ V :
s ( v ) = s ′ ( f ( v ) ) r ( v ) ≃ r ′ ( f ( v ) ) f ( p j ( v ) ) ≃ p j ′ ( f ( v ) ) for each j ∈ N \begin{aligned}
s(v) &= s'(f(v)) \\
r(v) &\simeq r'(f(v)) \\
f(p_j(v)) &\simeq p'_j(f(v)) \quad\text{for each } j \in\mathbb{N}
\end{aligned} s ( v ) r ( v ) f ( p j ( v )) = s ′ ( f ( v )) ≃ r ′ ( f ( v )) ≃ p j ′ ( f ( v )) for each j ∈ N Where f ( x ) ≃ g ( y ) f(x) \simeq g(y) f ( x ) ≃ g ( y ) means that f ( x ) f(x) f ( x ) is defined iff g ( y ) g(y) g ( y ) is defined and if f ( x ) f(x) f ( x ) is defined then f ( x ) = g ( y ) f(x)=g(y) f ( x ) = g ( y ) . If G , G ′ G,G' G , G ′ are derivations graphs, we write f : G → G ′ f:G \to G' f : G → G ′ to mean that f f f is a derivation graph homomorphism from G G G to G ′ G' G ′ .
Lemma 46 (Brotherston [5 ] , Lemma 6.1.3). Let G = ( V , s , r , p ) G=(V,s,r,p) G = ( V , s , r , p ) and G ′ = ( V ′ , s ′ , r ′ , p ′ ) G'=(V',s',r',p') G ′ = ( V ′ , s ′ , r ′ , p ′ ) be derivation graphs such that there exists a derivation graph homomorphism f : G → G ′ f:G \to G' f : G → G ′ . Then for every infinite path v 0 v 1 v 2 … v_0v_1v_2\ldots v 0 v 1 v 2 … in G G G , there exists an infinite path v 0 ′ v 1 ′ v 2 ′ … v'_0v'_1v'_2\ldots v 0 ′ v 1 ′ v 2 ′ … in G ′ G' G ′ with the same rule sequent labelling, i.e. s ( v i ) = s ′ ( v i ′ ) s(v_i)=s'(v'_i) s ( v i ) = s ′ ( v i ′ ) and r ( v i ) = r ′ ( v i ′ ) r(v_i)=r'(v'_i) r ( v i ) = r ′ ( v i ′ ) for all i ≥ 0 i \ge0 i ≥ 0 . Furthermore, if f f f is surjective the converse holds.
The induction order ( B u d ( D ) , ⊲ ) (\mathit{Bud}(\mathcal{D}),\lhd) ( Bud ( D ) , ⊲ ) for the obtained circular proof P \mathcal{P} P after the execution of Algorithm 2, is given by lifting the original order ( B u d ( D 0 ) , ⊲ 0 ) (\mathit{Bud}(\mathcal{D}_0),\lhd_0) ( Bud ( D 0 ) , ⊲ 0 ) of P 0 \mathcal{P}_0 P 0 to P \mathcal{P} P . We define B ^ = v \widehat{B}=v B = v if B = ( v , k ) B=(v,k) B = ( v , k ) , where B ∈ B u d ( D ) B \in\mathit{Bud}(\mathcal{D}) B ∈ Bud ( D ) and B ^ ∈ B u d ( D 0 ) \widehat{B} \in\mathit{Bud}(\mathcal{D}_0) B ∈ Bud ( D 0 ) . Then the induction order ( B u d ( D ) , ⊲ ) (\mathit{Bud}(\mathcal{D}),\lhd) ( Bud ( D ) , ⊲ ) is defined as B ⊲ B ′ B \lhd B' B ⊲ B ′ iff B ^ ⊲ 0 B ^ ′ \widehat{B} \lhd_0 \widehat{B}' B ⊲ 0 B ′ or ( B ^ = B ^ ′ (\widehat{B}=\widehat{B}' ( B = B ′ and k ≤ k ′ ) k \le k') k ≤ k ′ ) , where B = ( v , k ) B=(v,k) B = ( v , k ) and B ′ = ( v ′ , k ′ ) B'=(v',k') B ′ = ( v ′ , k ′ ) .
Proof of lemma 35. To prove the total correctness of the algorithm we must show that: 1) P \mathcal{P} P is a pre-proof, 2) ( B u d ( D ) , ◃ ) (Bud(\mathcal{D}), \triangleleft) ( B u d ( D ) , ◃ ) is a tree-compatible induction order for P \mathcal{P} P , 3) There exists a ordered trace manifold with respect to ( B u d ( D ) , ◃ ) (Bud(\mathcal{D}), \triangleleft) ( B u d ( D ) , ◃ ) , 4) P \mathcal{P} P can be coherently decorated, 5) Algorithm 2 terminates. For the item 1), the proof follows the same approach as done for Algorithm 1, by doing induction on the execution tree for treeCompatibility. For the items 2) and 5) the reader can refer to the proof of Theorem 5 in [37 ] . We prove items 3) and 4):
(3): Consider any bud node ( B , k ) ∈ B u d ( D ) (B,k) \in Bud(\mathcal{D}) ( B , k ) ∈ B u d ( D ) with R ( B , k ) = ( R 0 ( B ) , i ) \mathcal{R}(B,k)=(\mathcal{R}_0(B),i) R ( B , k ) = ( R 0 ( B ) , i ) . Moreover, consider the sequence of calls:
c = ( ( R 0 ( B ) , i ) , A 0 ) … ( ( R 0 ( B 1 ) , k 1 ) , A 1 ) … ( ( R 0 ( B m ) , k m ) , A m ) … ( ( B , k ) , A m + 1 ) c = ((\mathcal{R}_0(B),i),\mathcal{A}_0)\ldots((\mathcal{R}_0(B_1),k_1),\mathcal{A}_1)\ldots((\mathcal{R}_0(B_m),k_m),\mathcal{A}_m)\ldots((B,k),\mathcal{A}_{m+1}) c = (( R 0 ( B ) , i ) , A 0 ) … (( R 0 ( B 1 ) , k 1 ) , A 1 ) … (( R 0 ( B m ) , k m ) , A m ) … (( B , k ) , A m + 1 )
Where each B j ∈ B u d ( D 0 ) B_j \in Bud(\mathcal{D}_0) B j ∈ B u d ( D 0 ) , for 1 ≤ j ≤ m 1 \leq j \leq m 1 ≤ j ≤ m . This sequence represents the execution calls of the Algorithm treeCompatibility, used to construct the unique path from ( R 0 ( B ) , i ) (\mathcal{R}_0(B),i) ( R 0 ( B ) , i ) to ( B , k ) (B,k) ( B , k ) in D \mathcal{D} D , where each call ( ( R 0 ( B j ) , k j ) , A j ) ((\mathcal{R}_0(B_j),k_j),\mathcal{A}_j) (( R 0 ( B j ) , k j ) , A j ) represents a recursive call over a companion node in the original proof P 0 \mathcal{P}_0 P 0 , with 1 ≤ j ≤ m 1 \leq j \leq m 1 ≤ j ≤ m . Since R ( B , k ) = ( R 0 ( B ) , i ) \mathcal{R}(B,k)=(\mathcal{R}_0(B),i) R ( B , k ) = ( R 0 ( B ) , i ) , it must be the case that the pair ( R 0 ( B ) , i ) (\mathcal{R}_0(B),i) ( R 0 ( B ) , i ) is added to A 0 \mathcal{A}_0 A 0 in the first call c 0 c_0 c 0 (line 28) and it appears in all subsequent A j \mathcal{A}_j A j with 1 ≤ j ≤ m + 1 1 \leq j \leq m+1 1 ≤ j ≤ m + 1 . Then, the pair ( ( B , k ) , ( R 0 ( B ) , i ) ) ((B,k),(\mathcal{R}_0(B),i)) (( B , k ) , ( R 0 ( B ) , i )) is added to R \mathcal{R} R in the call c m + 1 c_{m+1} c m + 1 (line 20). Now, consider any bud node B j B_j B j appearing in the call c j c_j c j with 1 ≤ j ≤ m 1 \leq j \leq m 1 ≤ j ≤ m . Since ( R 0 ( B ) , i ) (\mathcal{R}_0(B),i) ( R 0 ( B ) , i ) is in both A j \mathcal{A}_j A j and A j + 1 \mathcal{A}_{j+1} A j + 1 , and ( R 0 ( B j ) , k j ) (\mathcal{R}_0(B_j),k_j) ( R 0 ( B j ) , k j ) is added to A j \mathcal{A}_j A j in the call c j c_j c j (line 28), it must be the case that:
Now consider any ( B , k ) ∈ B u d ( D ) (B,k) \in Bud(\mathcal{D}) ( B , k ) ∈ B u d ( D ) and π B \pi_B π B as the unique path from R ( ( B , k ) ) \mathcal{R}((B,k)) R (( B , k )) to ( B , k ) (B,k) ( B , k ) in D \mathcal{D} D . By inspection on the Algorithm 2, π B \pi_B π B is obtained by unfolding a set of bud nodes U = { B 1 , … , B m } ⊆ B u d ( D 0 ) U=\{B_1,\ldots,B_m\} \subseteq Bud(\mathcal{D}_0) U = { B 1 , … , B m } ⊆ B u d ( D 0 ) , represented by the following sequence of recursive calls:
c = ( ( R ( B , k ) ) , A 0 ) … ( ( B 1 , k 1 ) , A 1 ) … ( ( B m , k m ) , A m ) … ( ( B , k ) , A m + 1 ) c = ((\mathcal{R}(B,k)),\mathcal{A}_0)\ldots((B_1,k_1),\mathcal{A}_1)\ldots((B_m,k_m),\mathcal{A}_m)\ldots((B,k),\mathcal{A}_{m+1}) c = (( R ( B , k )) , A 0 ) … (( B 1 , k 1 ) , A 1 ) … (( B m , k m ) , A m ) … (( B , k ) , A m + 1 )
Where each B j ∈ B u d ( D 0 ) B_j \in Bud(\mathcal{D}_0) B j ∈ B u d ( D 0 ) , for 1 ≤ j ≤ m 1 \leq j \leq m 1 ≤ j ≤ m , is an unpaired bud node that will be replaced by its companion in the line 23. Therefore, the path π B \pi_B π B will have the form:
π B = R ( B , k ) … ( R 0 ( B 1 ) , k 1 ′ ) … ( R 0 ( B m ) , k m ′ ) … ( B , k ) \pi_B = \mathcal{R}(B,k)\ldots(\mathcal{R}_0(B_1),k'_1)\ldots(\mathcal{R}_0(B_m),k'_m)\ldots(B,k) π B = R ( B , k ) … ( R 0 ( B 1 ) , k 1 ′ ) … ( R 0 ( B m ) , k m ′ ) … ( B , k )
Let B π B = U ∪ { B } ⊆ B u d ( D 0 ) \mathcal{B}_{\pi_B}=U\cup\{B\} \subseteq Bud(\mathcal{D}_0) B π B = U ∪ { B } ⊆ B u d ( D 0 ) , and consider the path π ^ B = R 0 ( B ) … R 0 ( B 1 ) … R 0 ( B m ) … B \widehat{\pi}_B=\mathcal{R}_0(B)\ldots\mathcal{R}_0(B_1)\ldots\mathcal{R}_0(B_m)\ldots B π B = R 0 ( B ) … R 0 ( B 1 ) … R 0 ( B m ) … B , obtained after eliminating the numbers of the pairs describing the nodes in the path π B \pi_B π B . From the definition of the Algorithm 2, one deduces that π ^ B \widehat{\pi}_B π B describes a cycle contained in the strongly connected subgraph S π B = ⋃ B ′ ∈ B π B C B ′ S_{\pi_B}=\bigcup_{B'\in\mathcal{B}_{\pi_B}}C_{B'} S π B = ⋃ B ′ ∈ B π B C B ′ , in the graph G P 0 \mathcal{G}_{\mathcal{P}_0} G P 0 (here we implicitly use Lemma 25 ). Therefore, by Lemma 44 , we get that:
The unique path π B \pi_B π B , represented as the basic cycle C ( B , k ) C_{(B,k)} C ( B , k ) in G P \mathcal{G}_{\mathcal{P}} G P , is formed by the composition of paths (of length > 0 >0 > 0 ) each of the form ( R 0 ( B i ) , k i ) … ( R 0 ( B j ) , k j ) (\mathcal{R}_0(B_i),k_i)\ldots(\mathcal{R}_0(B_j),k_j) ( R 0 ( B i ) , k i ) … ( R 0 ( B j ) , k j ) , where B i , B j ∈ B π B B_i,B_j \in\mathcal{B}_{\pi_B} B i , B j ∈ B π B and R 0 ( B i ) … R 0 ( B j ) \mathcal{R}_0(B_i)\ldots\mathcal{R}_0(B_j) R 0 ( B i ) … R 0 ( B j ) is a subpath of some cycle C B ′ ′ ⊆ S π B = ⋃ B ′ ∈ B π B C B ′ C_{B''} \subseteq S_{\pi_B}=\bigcup_{B'\in\mathcal{B}_{\pi_B}}C_{B'} C B ′′ ⊆ S π B = ⋃ B ′ ∈ B π B C B ′ in G P 0 \mathcal{G}_{\mathcal{P}_0} G P 0 . (B)
With this in mind, we show how to construct the traces for the ordered trace manifold for P \mathcal{P} P . Take any B , B ′ ∈ B u d ( D ) B,B' \in Bud(\mathcal{D}) B , B ′ ∈ B u d ( D ) such that B ◃ B ′ B \triangleleft B' B ◃ B ′ . We will construct the trace T B ′ , B \mathcal{T}_{B',B} T B ′ , B of the ordered trace manifold for P \mathcal{P} P , following the unique path π B \pi_B π B from R ( B ) \mathcal{R}(B) R ( B ) to B B B in D \mathcal{D} D . Take any B j B_j B j with 1 ≤ j ≤ m 1 \leq j \leq m 1 ≤ j ≤ m . After the call ( ( B j , k j ) , A j ) ((B_j,k_j),\mathcal{A}_j) (( B j , k j ) , A j ) in c c c , since ( B j , k j ) (B_j,k_j) ( B j , k j ) is not paired with any companion node, it implies that the conditional in line 19 was not satisfied, and therefore R 0 ( B j ) ∉ dom ( A j ) \mathcal{R}_0(B_j) \notin\operatorname{dom}(\mathcal{A}_j) R 0 ( B j ) ∈ / dom ( A j ) . This implies by (A), that B j ◃ 0 B B_j \triangleleft_0 B B j ◃ 0 B . Moreover, by definition of ◃ \triangleleft ◃ , we have that B ^ ◃ 0 B ′ ^ \widehat{B} \triangleleft_0 \widehat{B'} B ◃ 0 B ′ or B ^ = B ′ ^ \widehat{B} = \widehat{B'} B = B ′ . Thus, B j ◃ 0 B ^ ◃ 0 B ′ ^ B_j \triangleleft_0 \widehat{B} \triangleleft_0 \widehat{B'} B j ◃ 0 B ◃ 0 B ′ or B j ◃ 0 B ^ = B ′ ^ B_j \triangleleft_0 \widehat{B} = \widehat{B'} B j ◃ 0 B = B ′ , in both cases implying B j ◃ 0 B ′ B_j \triangleleft_0 B' B j ◃ 0 B ′ . Hence, we have that for all bud node B ′ ′ ∈ B π B B'' \in\mathcal{B}_{\pi_B} B ′′ ∈ B π B , then B ′ ′ ◃ 0 B ′ B'' \triangleleft_0 B' B ′′ ◃ 0 B ′ . Then, by definition of the original ordered trace manifold for P 0 P_0 P 0 :
For all bud node B ◃ B ′ B \triangleleft B' B ◃ B ′ and B ′ ′ ∈ B π B B'' \in\mathcal{B}_{\pi_B} B ′′ ∈ B π B , then τ B ′ , B ′ ′ \tau_{B',B''} τ B ′ , B ′′ exists in the ordered trace manifold for P 0 P_0 P 0 , and is a trace following the basic cycle C B ′ ′ C_{B''} C B ′′ in G P 0 \mathcal{G}_{P_0} G P 0 . (C)
The trace τ B ′ , B \tau_{B',B} τ B ′ , B for the path π B \pi_B π B is constructed in the following way: take every segment ( R 0 ( B i ) , k i ) … ( R 0 ( B j ) , k j ) (\mathcal{R}_0(B_i), k_i)\ldots(\mathcal{R}_0(B_j), k_j) ( R 0 ( B i ) , k i ) … ( R 0 ( B j ) , k j ) of π B \pi_B π B , as described in (B). The segment R 0 ( B i ) … R 0 ( B j ) \mathcal{R}_0(B_i)\ldots\mathcal{R}_0(B_j) R 0 ( B i ) … R 0 ( B j ) is a subpath of some cycle C B ′ ′ ⊆ S π B C_{B''} \subseteq S_{\pi_B} C B ′′ ⊆ S π B where B ′ ′ ∈ B π B B'' \in\mathcal{B}_{\pi_B} B ′′ ∈ B π B . Therefore, by (C) we can use the trace τ B ′ ^ , B ′ ′ \tau_{\widehat{B'},B''} τ B ′ , B ′′ to cover the segment R 0 ( B i ) … R 0 ( B j ) \mathcal{R}_0(B_i)\ldots\mathcal{R}_0(B_j) R 0 ( B i ) … R 0 ( B j ) , and consequently the segment ( R 0 ( B i ) , k i ) … ( R 0 ( B j ) , k j ) (\mathcal{R}_0(B_i), k_i)\ldots(\mathcal{R}_0(B_j), k_j) ( R 0 ( B i ) , k i ) … ( R 0 ( B j ) , k j ) . In case the segment R 0 ( B i ) … R 0 ( B j ) \mathcal{R}_0(B_i)\ldots\mathcal{R}_0(B_j) R 0 ( B i ) … R 0 ( B j ) has a progress point, and it is contained in multiple basic cycles, we always choose the trace containing the formula that progresses, as explained in the proof of Lemma 29 .
Note that these traces can be composed, by the property 2 of the original ordered trace manifold for P 0 P_0 P 0 : consider any segment ( R 0 ( B i ) , k i ) … ( R 0 ( B j ) , k j ) … ( R 0 ( B h ) , k h ) (\mathcal{R}_0(B_i), k_i)\ldots(\mathcal{R}_0(B_j), k_j)\ldots(\mathcal{R}_0(B_h), k_h) ( R 0 ( B i ) , k i ) … ( R 0 ( B j ) , k j ) … ( R 0 ( B h ) , k h ) . Such that the segments R 0 ( B i ) … R 0 ( B j ) \mathcal{R}_0(B_i)\ldots\mathcal{R}_0(B_j) R 0 ( B i ) … R 0 ( B j ) and R 0 ( B j ) … R 0 ( B h ) \mathcal{R}_0(B_j)\ldots\mathcal{R}_0(B_h) R 0 ( B j ) … R 0 ( B h ) are included in the cycles C B 1 , C B 2 ⊆ ⋃ B ′ ′ ∈ B π B C B ′ ′ C_{B_1}, C_{B_2} \subseteq\bigcup_{B''\in\mathcal{B}_{\pi_B}} C_{B''} C B 1 , C B 2 ⊆ ⋃ B ′′ ∈ B π B C B ′′ respectively, in G P 0 \mathcal{G}_{P_0} G P 0 . This implies that B j ≤ P 0 B 1 B_j \leq_{P_0} B_1 B j ≤ P 0 B 1 and B j ≤ P 0 B 2 B_j \leq_{P_0} B_2 B j ≤ P 0 B 2 . Moreover, since B 1 , B 2 , B j ∈ B π B B_1, B_2, B_j \in\mathcal{B}_{\pi_B} B 1 , B 2 , B j ∈ B π B , we know that B 1 , B 2 , B j ◃ 0 B ′ ^ B_1, B_2, B_j \triangleleft_0 \widehat{B'} B 1 , B 2 , B j ◃ 0 B ′ , therefore τ B ′ ^ , B 1 ( R 0 ( B j ) ) = τ B ′ ^ , B j ( R 0 ( B j ) ) = τ B ′ ^ , B 2 ( R 0 ( B j ) ) \tau_{\widehat{B'},B_1}(\mathcal{R}_0(B_j)) = \tau_{\widehat{B'},B_j}(\mathcal{R}_0(B_j)) = \tau_{\widehat{B'},B_2}(\mathcal{R}_0(B_j)) τ B ′ , B 1 ( R 0 ( B j )) = τ B ′ , B j ( R 0 ( B j )) = τ B ′ , B 2 ( R 0 ( B j )) .
We show that this construction satisfies the three properties of ordered trace manifolds. Remember that for all B = ( B i , k i ) ∈ B u d ( D ) B = (B_i, k_i) \in Bud(\mathcal{D}) B = ( B i , k i ) ∈ B u d ( D ) then R ( B ) = ( R 0 ( B i ) , k j ) \mathcal{R}(B) = (\mathcal{R}_0(B_i), k_j) R ( B ) = ( R 0 ( B i ) , k j ) .
By construction, the trace τ B ′ , B \tau_{B',B} τ B ′ , B is a trace over the basic cycle C B C_B C B in G P \mathcal{G}_P G P . We check that τ B ′ , B \tau_{B',B} τ B ′ , B has a unique value at R ( B ) \mathcal{R}(B) R ( B ) . By construction τ B ′ , B ( R ( B ) ) = τ B ′ ^ , B 1 ( R ( B ) ^ ) \tau_{B',B}(\mathcal{R}(B)) = \tau_{\widehat{B'},B_1}(\widehat{\mathcal{R}(B)}) τ B ′ , B ( R ( B )) = τ B ′ , B 1 ( R ( B ) ) and τ B ′ , B ( R ( B ) ) = τ B ′ ^ , B 2 ( R ( B ) ^ ) \tau_{B',B}(\mathcal{R}(B)) = \tau_{\widehat{B'},B_2}(\widehat{\mathcal{R}(B)}) τ B ′ , B ( R ( B )) = τ B ′ , B 2 ( R ( B ) ) for some B 1 , B 2 ∈ B π B B_1, B_2 \in\mathcal{B}_{\pi_B} B 1 , B 2 ∈ B π B . Note that R ( B ) ^ \widehat{\mathcal{R}(B)} R ( B ) is in C B 1 C_{B_1} C B 1 and C B 2 C_{B_2} C B 2 , implying B ^ ≤ P 0 B 1 \widehat{B} \leq_{P_0} B_1 B ≤ P 0 B 1 and B ^ ≤ P 0 B 2 \widehat{B} \leq_{P_0} B_2 B ≤ P 0 B 2 . Then by property 2 of the ordered trace manifold, τ B ′ ^ , B 1 ( R ( B ) ^ ) = τ B ′ ^ , B ^ ( R ( B ) ^ ) = τ B ′ ^ , B 2 ( R ( B ) ^ ) \tau_{\widehat{B'},B_1}(\widehat{\mathcal{R}(B)}) = \tau_{\widehat{B'},\widehat{B}}(\widehat{\mathcal{R}(B)}) = \tau_{\widehat{B'},B_2}(\widehat{\mathcal{R}(B)}) τ B ′ , B 1 ( R ( B ) ) = τ B ′ , B ( R ( B ) ) = τ B ′ , B 2 ( R ( B ) ) . Thus, τ S , B \tau_{S,B} τ S , B takes the same value at R ( B ) \mathcal{R}(B) R ( B ) .
Let B 1 , B 2 ◃ B ′ B_1, B_2 \triangleleft B' B 1 , B 2 ◃ B ′ such that B 1 ≤ P B 2 B_1 \leq_P B_2 B 1 ≤ P B 2 . By construction, τ B ′ , B 1 ( R ( B 1 ) ) = τ B ′ ^ , B ′ ′ ( R ( B 1 ) ^ ) \tau_{B',B_1}(\mathcal{R}(B_1)) = \tau_{\widehat{B'},B''}(\widehat{\mathcal{R}(B_1)}) τ B ′ , B 1 ( R ( B 1 )) = τ B ′ , B ′′ ( R ( B 1 ) ) and τ B ′ , B 2 ( R ( B 1 ) ) = τ B ′ ^ , B ′ ′ ′ ( R ( B 1 ) ^ ) \tau_{B',B_2}(\mathcal{R}(B_1)) = \tau_{\widehat{B'},B'''}(\widehat{\mathcal{R}(B_1)}) τ B ′ , B 2 ( R ( B 1 )) = τ B ′ , B ′′′ ( R ( B 1 ) ) for some B ′ ′ ∈ B π B 1 B'' \in\mathcal{B}_{\pi_{B_1}} B ′′ ∈ B π B 1 and B ′ ′ ′ ∈ B π B 2 B''' \in\mathcal{B}_{\pi_{B_2}} B ′′′ ∈ B π B 2 . Hence, R ( B 1 ) ^ \widehat{\mathcal{R}(B_1)} R ( B 1 ) is in C B ′ ′ C_{B''} C B ′′ and C B ′ ′ ′ C_{B'''} C B ′′′ , which implies B 1 ^ ≤ P 0 B ′ ′ \widehat{B_1} \leq_{P_0} B'' B 1 ≤ P 0 B ′′ and B 1 ^ ≤ P 0 B ′ ′ ′ \widehat{B_1} \leq_{P_0} B''' B 1 ≤ P 0 B ′′′ . Also note that B 1 ^ , B ′ ′ , B ′ ′ ′ ◃ 0 B ′ ^ \widehat{B_1}, B'', B''' \triangleleft_0 \widehat{B'} B 1 , B ′′ , B ′′′ ◃ 0 B ′ . Therefore, by property 2 of the ordered trace manifold, τ B ′ ^ , B ′ ′ ( R ( B 1 ) ^ ) = τ B ′ ^ , B 1 ^ ( R ( B 1 ) ^ ) = τ B ′ ^ , B ′ ′ ′ ( R ( B 1 ) ^ ) \tau_{\widehat{B'},B''}(\widehat{\mathcal{R}(B_1)}) = \tau_{\widehat{B'},\widehat{B_1}}(\widehat{\mathcal{R}(B_1)}) = \tau_{\widehat{B'},B'''}(\widehat{\mathcal{R}(B_1)}) τ B ′ , B ′′ ( R ( B 1 ) ) = τ B ′ , B 1 ( R ( B 1 ) ) = τ B ′ , B ′′′ ( R ( B 1 ) ) . Thus, τ B ′ , B 1 ( R ( B 1 ) ) = τ B ′ , B 2 ( R ( B 1 ) ) \tau_{B',B_1}(\mathcal{R}(B_1)) = \tau_{B',B_2}(\mathcal{R}(B_1)) τ B ′ , B 1 ( R ( B 1 )) = τ B ′ , B 2 ( R ( B 1 )) .
Let B 1 , … , B n , B ′ ∈ B u d ( D ) B_1,\ldots,B_n, B' \in Bud(\mathcal{D}) B 1 , … , B n , B ′ ∈ B u d ( D ) such that B 1 , … , B n ◃ B ′ B_1,\ldots,B_n \triangleleft B' B 1 , … , B n ◃ B ′ , i.e. all bud nodes that are below B ′ B' B ′ in the ◃ \triangleleft ◃ order (this includes B ′ B' B ′ itself). Moreover, consider the unique paths π B 1 , … , π B n \pi_{B_1},\ldots,\pi_{B_n} π B 1 , … , π B n from companion to bud node for B 1 , … , B 2 B_1,\ldots,B_2 B 1 , … , B 2 respectively. For any 1 ≤ i ≤ n 1 \leq i \leq n 1 ≤ i ≤ n , consider the trace τ B ′ , B i \tau_{B',B_i} τ B ′ , B i . By construction, τ B ′ , B i \tau_{B',B_i} τ B ′ , B i is obtained by the composition of segments of the traces τ B ′ ^ , B ′ ′ \tau_{\widehat{B'},B''} τ B ′ , B ′′ , such that B ′ ′ ∈ B π B i B'' \in\mathcal{B}_{\pi_{B_i}} B ′′ ∈ B π B i . Thus, for all i i i , Form ( τ B ′ , B i ) ⊆ ⋃ B ′ ′ ∈ B π B i Form ( τ B ′ ^ , B ′ ′ ) \operatorname{Form}(\tau_{B',B_i}) \subseteq\bigcup_{B''\in\mathcal{B}_{\pi_{B_i}}} \operatorname{Form}(\tau_{\widehat{B'},B''}) Form ( τ B ′ , B i ) ⊆ ⋃ B ′′ ∈ B π B i Form ( τ B ′ , B ′′ ) . Therefore, we have that:
⋃ 1 ≤ i ≤ n Form ( τ B ′ , B i ) ⊆ ⋃ 1 ≤ i ≤ n ( ⋃ B ′ ′ ∈ B π B i Form ( τ B ′ ^ , B ′ ′ ) ) \bigcup_{1\leq i\leq n}\operatorname{Form}(\tau_{B',B_i}) \subseteq\bigcup_{1\leq i\leq n}\left(\bigcup_{B''\in\mathcal{B}_{\pi_{B_i}}}\operatorname{Form}(\tau_{\widehat{B'},B''})\right) 1 ≤ i ≤ n ⋃ Form ( τ B ′ , B i ) ⊆ 1 ≤ i ≤ n ⋃ B ′′ ∈ B π B i ⋃ Form ( τ B ′ , B ′′ )
Then, the formula F F F that is minimal in the set ⋃ 1 ≤ i ≤ n ( ⋃ B ′ ′ ∈ B π B i Form ( τ B ′ ^ , B ′ ′ ) ) \bigcup_{1\leq i\leq n}\left(\bigcup_{B''\in\mathcal{B}_{\pi_{B_i}}}\operatorname{Form}(\tau_{\widehat{B'},B''})\right) ⋃ 1 ≤ i ≤ n ( ⋃ B ′′ ∈ B π B i Form ( τ B ′ , B ′′ ) ) by the property 3 in the ordered trace manifold for P 0 P_0 P 0 , is also minimal in the set ⋃ 1 ≤ i ≤ n Form ( τ B ′ , B i ) \bigcup_{1\leq i\leq n}\operatorname{Form}(\tau_{B',B_i}) ⋃ 1 ≤ i ≤ n Form ( τ B ′ , B i ) . It remains to be shown that τ B ′ , B ′ \tau_{B',B'} τ B ′ , B ′ contains an F F F -progress point. This can be done by inspection on the path π B ′ \pi_{B'} π B ′ from R ( B ′ ) R(B') R ( B ′ ) to B ′ B' B ′ in D \mathcal{D} D . For every vertex v v v in the basic cycle C B ′ ^ C_{\widehat{B'}} C B ′ , by the definition of the unfolding, one can always find a copy ( v , k ) (v,k) ( v , k ) of the original vertex in C B ′ C_{B'} C B ′ . For this reason, we can use the trace τ B ′ , B ′ \tau_{B',B'} τ B ′ , B ′ to cover the progress point, i.e. τ B ′ , B ′ ( v , k ) = τ B ′ ^ , B ′ ^ ( v ) \tau_{B',B'}(v,k)=\tau_{\widehat{B'},\widehat{B'}}(v) τ B ′ , B ′ ( v , k ) = τ B ′ , B ′ ( v ) , which contains the F F F -progress point.
(4): After the ascending phase, we obtain a coherent decoration of the derivation tree D \mathcal{D} D . It remains to be shown that the bud companion relation R \mathcal{R} R preserves the coherence of P \mathcal{P} P . Consider then any rooted path π \pi π in G P \mathcal{G}_{\mathcal{P}} G P . As explained before, π \pi π is obtained by unfolding a set of bud nodes U = { B 1 , … , B m } ⊆ B u d ( D 0 ) U=\{B_1,\ldots,B_m\}\subseteq Bud(\mathcal{D}_0) U = { B 1 , … , B m } ⊆ B u d ( D 0 ) . Therefore, π \pi π has the form:
π = ( v r , 0 ) … ( R 0 ( B 1 ) , k 1 ′ ) … ( R 0 ( B m ) , k m ′ ) … ( v f , k ) \pi=(v_r,0)\ldots(\mathcal{R}_0(B_1),k'_1)\ldots(\mathcal{R}_0(B_m),k'_m)\ldots(v_f,k) π = ( v r , 0 ) … ( R 0 ( B 1 ) , k 1 ′ ) … ( R 0 ( B m ) , k m ′ ) … ( v f , k )
where ( v r , 0 ) (v_r,0) ( v r , 0 ) is the root of D \mathcal{D} D and ( v f , k ) (v_f,k) ( v f , k ) is the final vertex of π \pi π . We can define a surjective derivation graph homomorphism from G P \mathcal{G}_{\mathcal{P}} G P to G P 0 \mathcal{G}_{\mathcal{P}_0} G P 0 , by mapping each copy ( v , k ) (v,k) ( v , k ) in G P \mathcal{G}_{\mathcal{P}} G P to its original vertex v v v in G P 0 \mathcal{G}_{\mathcal{P}_0} G P 0 . Therefore, by Lemma 46, there exists a path π ^ \widehat{\pi} π with the same rule sequent labelling as π \pi π .
Now consider the function ρ 0 : V 0 → S \rho_0:V_0\to S ρ 0 : V 0 → S , which stores the information of the decoration of the pre-proof P 0 \mathcal{P}_0 P 0 , i.e. it maps every node in D 0 \mathcal{D}_0 D 0 with their respective split ( Γ , Δ ) (\Gamma,\Delta) ( Γ , Δ ) after the ascending phase over the derivation tree D 0 \mathcal{D}_0 D 0 . Similarly, we consider the function ρ : V → S \rho:V\to S ρ : V → S , which stores the information of the decoration over the derivation tree D \mathcal{D} D of the pre-proof P \mathcal{P} P , that is obtained after executing the ascending phase. We would like to prove that:
If P 0 \mathcal{P}_0 P 0 can be coherently decorated, then ρ 0 ( v ) = ρ ( v , i ) \rho_0(v)=\rho(v,i) ρ 0 ( v ) = ρ ( v , i ) for every vertex ( v , i ) ∈ π (v,i)\in\pi ( v , i ) ∈ π and v ∈ π ^ v\in\widehat{\pi} v ∈ π . (D)
We proceed by induction on the length n n n of π \pi π . If n = 1 n=1 n = 1 , then π ^ = v r \widehat{\pi}=v_r π = v r and π = ( v r , 0 ) \pi=(v_r,0) π = ( v r , 0 ) . Since both P 0 \mathcal{P}_0 P 0 and P \mathcal{P} P have the same initial split, then ρ 0 ( v r ) = ρ ( v r , 0 ) \rho_0(v_r)=\rho(v_r,0) ρ 0 ( v r ) = ρ ( v r , 0 ) . In the case n + 1 n+1 n + 1 , then π ^ = π ^ n , v f \widehat{\pi}=\widehat{\pi}_n,v_f π = π n , v f and π = π n , ( v f , k ) \pi=\pi_n,(v_f,k) π = π n , ( v f , k ) . By induction hypothesis, for every ( v ′ , k ′ ) (v',k') ( v ′ , k ′ ) in π n \pi_n π n , then ρ 0 ( v ′ ) = ρ ( v ′ , k ′ ) \rho_0(v')=\rho(v',k') ρ 0 ( v ′ ) = ρ ( v ′ , k ′ ) . We will assume that ( v n , k ′ ) (v_n,k') ( v n , k ′ ) is the last vertex of the path π n \pi_n π n . We proceed by cases over the final node ( v f , k ) (v_f,k) ( v f , k ) . If ( v f , k ) (v_f,k) ( v f , k ) is one of the unfolded bud nodes in π \pi π , i.e. ( v f , k ) = ( R 0 ( B j ) , k j ′ ) (v_f,k)=(\mathcal{R}_0(B_j),k'_j) ( v f , k ) = ( R 0 ( B j ) , k j ′ ) for 1 ≤ j ≤ m 1\leq j\leq m 1 ≤ j ≤ m , it implies that ρ ( v f , k ) = ρ ( R 0 ( B j ) , k j ′ ) \rho(v_f,k)=\rho(\mathcal{R}_0(B_j),k'_j) ρ ( v f , k ) = ρ ( R 0 ( B j ) , k j ′ ) is obtained by propagating the split ρ ( v n , k ′ ) \rho(v_n,k') ρ ( v n , k ′ ) using the rule r ( v n , k ′ ) r(v_n,k') r ( v n , k ′ ) . Moreover, this implies that in D 0 \mathcal{D}_0 D 0 we have the subpath θ = v n , B j \theta=v_n,B_j θ = v n , B j . Therefore, ρ 0 ( B j ) \rho_0(B_j) ρ 0 ( B j ) is obtained by propagating the split ρ 0 ( v n ) \rho_0(v_n) ρ 0 ( v n ) using the rule r 0 ( v n ) r_0(v_n) r 0 ( v n ) . Since both ρ 0 ( v n ) = ρ ( v n , k ′ ) \rho_0(v_n)=\rho(v_n,k') ρ 0 ( v n ) = ρ ( v n , k ′ ) by induction hypothesis and r 0 ( v n ) = r ( v n , k ′ ) r_0(v_n)=r(v_n,k') r 0 ( v n ) = r ( v n , k ′ ) since π \pi π and π ^ \widehat{\pi} π have the same rule sequent labelling, we deduce that ρ ( v f , k ) = ρ ( R 0 ( B j ) , k j ′ ) = ρ 0 ( B j ) \rho(v_f,k)=\rho(\mathcal{R}_0(B_j),k'_j)=\rho_0(B_j) ρ ( v f , k ) = ρ ( R 0 ( B j ) , k j ′ ) = ρ 0 ( B j ) . Furthermore, since P 0 \mathcal{P}_0 P 0 can be coherently decorated, it implies that ρ 0 ( B j ) = ρ 0 ( R 0 ( B j ) ) \rho_0(B_j)=\rho_0(\mathcal{R}_0(B_j)) ρ 0 ( B j ) = ρ 0 ( R 0 ( B j )) , hence ρ ( v f , k ) = ρ ( R 0 ( B j ) , k j ′ ) = ρ 0 ( R 0 ( B j ) ) = ρ 0 ( v f ) \rho(v_f,k)=\rho(\mathcal{R}_0(B_j),k'_j)=\rho_0(\mathcal{R}_0(B_j))=\rho_0(v_f) ρ ( v f , k ) = ρ ( R 0 ( B j ) , k j ′ ) = ρ 0 ( R 0 ( B j )) = ρ 0 ( v f ) .
Otherwise, if ( v f , k ) ≠ ( R 0 ( B j ) , k j ′ ) (v_f,k)\neq(\mathcal{R}_0(B_j),k'_j) ( v f , k ) = ( R 0 ( B j ) , k j ′ ) for 1 ≤ j ≤ m 1\leq j\leq m 1 ≤ j ≤ m it implies that ρ ( v f , k ) \rho(v_f,k) ρ ( v f , k ) is obtained by propagating the split ρ ( v n , k ′ ) \rho(v_n,k') ρ ( v n , k ′ ) using the rule r ( v n , k ′ ) r(v_n,k') r ( v n , k ′ ) . Moreover, this implies that in D 0 \mathcal{D}_0 D 0 we have the subpath θ = v n , v f \theta=v_n,v_f θ = v n , v f . Therefore, ρ 0 ( v f ) \rho_0(v_f) ρ 0 ( v f ) is obtained by propagating the split ρ 0 ( v n ) \rho_0(v_n) ρ 0 ( v n ) using the rule r 0 ( v n ) r_0(v_n) r 0 ( v n ) . Since both ρ 0 ( v n ) = ρ ( v n , k ′ ) \rho_0(v_n)=\rho(v_n,k') ρ 0 ( v n ) = ρ ( v n , k ′ ) by induction hypothesis and r 0 ( v n ) = r ( v n , k ′ ) r_0(v_n)=r(v_n,k') r 0 ( v n ) = r ( v n , k ′ ) since π \pi π and π ^ \widehat{\pi} π have the same rule sequent labelling, we deduce that ρ ( v f , k ) = ρ 0 ( v f ) \rho(v_f,k)=\rho_0(v_f) ρ ( v f , k ) = ρ 0 ( v f ) .
Now consider any bud node ( B , k ) ∈ B u d ( D ) (B,k)\in Bud(\mathcal{D}) ( B , k ) ∈ B u d ( D ) and the rooted path π = ( v r , 0 ) , … , ( v p , k ) , R ( B , k ) \pi=(v_r,0),\ldots,(v_p,k),\mathcal{R}(B,k) π = ( v r , 0 ) , … , ( v p , k ) , R ( B , k ) in G P \mathcal{G}_{\mathcal{P}} G P , and π ^ = v r , … , v p , R ( B , k ) \widehat{\pi}=v_r,\ldots,v_p,\mathcal{R}(B,k) π = v r , … , v p , R ( B , k ) in G P 0 \mathcal{G}_{\mathcal{P}_0} G P 0 , where ( v p , k ) (v_p,k) ( v p , k ) is the predecessor of ( B , k ) (B,k) ( B , k ) in D \mathcal{D} D , and π \pi π and π ^ \widehat{\pi} π have the same rule sequent labelling. By (D), we know that ρ ( v p , k ) = ρ 0 ( v p ) \rho(v_p,k)=\rho_0(v_p) ρ ( v p , k ) = ρ 0 ( v p ) . Moreover, there is a subpath θ = ( v p , k ) , ( B , k ) \theta=(v_p,k),(B,k) θ = ( v p , k ) , ( B , k ) and θ ^ = v p , B \widehat{\theta}=v_p,B θ = v p , B in D \mathcal{D} D and D 0 \mathcal{D}_0 D 0 respectively. The split ρ ( B , k ) \rho(B,k) ρ ( B , k ) is obtained by propagating the split ρ ( v p , k ) \rho(v_p,k) ρ ( v p , k ) using the rule r ( v p , k ) r(v_p,k) r ( v p , k ) , while the split ρ 0 ( B ) \rho_0(B) ρ 0 ( B ) is obtained by propagating the split ρ 0 ( v p ) \rho_0(v_p) ρ 0 ( v p ) using the rule r 0 ( v p ) r_0(v_p) r 0 ( v p ) . Since ρ ( v p , k ) = ρ 0 ( v p ) \rho(v_p,k)=\rho_0(v_p) ρ ( v p , k ) = ρ 0 ( v p ) and r 0 ( v p ) = r ( v p , k ) r_0(v_p)=r(v_p,k) r 0 ( v p ) = r ( v p , k ) since π \pi π and π ^ \hat{\pi} π ^ have the same rule sequent labelling, we deduce that ρ ( B , k ) = ρ 0 ( B ) \rho(B,k)=\rho_0(B) ρ ( B , k ) = ρ 0 ( B ) .
Furthermore, by definition of the Algorithm 2 in the line 20, we have that R ( B , k ) = ( R 0 ( B ) , t ) \mathcal{R}(B,k)=(\mathcal{R}_0(B),t) R ( B , k ) = ( R 0 ( B ) , t ) for some t t t . Therefore, by (D) it implies that ρ ( R ( B , k ) ) = ρ ( R 0 ( B ) , t ) = ρ 0 ( R 0 ( B ) ) \rho(\mathcal{R}(B,k))=\rho(\mathcal{R}_0(B),t)=\rho_0(\mathcal{R}_0(B)) ρ ( R ( B , k )) = ρ ( R 0 ( B ) , t ) = ρ 0 ( R 0 ( B )) . Finally, since P 0 \mathcal{P}_0 P 0 can be coherently decorated, then ρ 0 ( R 0 ( B ) ) = ρ 0 ( B ) \rho_0(\mathcal{R}_0(B))=\rho_0(B) ρ 0 ( R 0 ( B )) = ρ 0 ( B ) , which implies ρ ( R ( B , k ) ) = ρ ( R 0 ( B ) , t ) = ρ 0 ( R 0 ( B ) ) = ρ 0 ( B ) = ρ ( B , k ) \rho(\mathcal{R}(B,k))=\rho(\mathcal{R}_0(B),t)=\rho_0(\mathcal{R}_0(B))=\rho_0(B)=\rho(B,k) ρ ( R ( B , k )) = ρ ( R 0 ( B ) , t ) = ρ 0 ( R 0 ( B )) = ρ 0 ( B ) = ρ ( B , k ) .
Proof of Theorem 37
Before the proof of the theorem, we explain the general strategy for proving this theorem. Our main objective is to adapt the proof-relevant interpolation method for circular proofs. However, our goal is not only to produce the interpolated circular pre-proof via cut-introduction, but also to prove that this construction is valid. Our strategy is to certify validity through the construction of ordered trace manifolds for the pre-proofs P l \mathcal{P}^l P l and P r \mathcal{P}^r P r .
To accomplish this, we follow the transformations induced by the descending phase, using induction on the subtrees of D 0 D_0 D 0 , and maintain as an invariant all the properties needed to construct ordered trace manifolds on both sides P l \mathcal{P}^l P l and P r \mathcal{P}^r P r . Specifically, with these invariants we want to capture the following properties of the interpolated pre-proof P \mathcal{P} P :
Both P 0 \mathcal{P}_0 P 0 and the right and left pre-proofs P l \mathcal{P}^l P l and P r \mathcal{P}^r P r have the same “skeleton”, i.e. their structural connectivity is exactly the same.
All traces in the original trace manifold for P 0 \mathcal{P}_0 P 0 live on formulas assigned to one side of the split Γ , Δ \Gamma,\Delta Γ , Δ . Therefore, each trace is transferred to exactly one of the two pre-proofs P l \mathcal{P}^l P l and P r \mathcal{P}^r P r .
The interpolant I I I and I ⊥ I^\perp I ⊥ can define new traces in the right and left pre-proofs P l \mathcal{P}^l P l and P r \mathcal{P}^r P r respectively, that can be included in the ordered trace manifold for P l \mathcal{P}^l P l and P r \mathcal{P}^r P r .
We will start by formally defining these properties: Consider a pre-proof P 0 = ( D 0 = ( V 0 , s 0 , r 0 , p 0 ) , R 0 ) \mathcal{P}_0=(D_0=(V_0,s_0,r_0,p_0),\mathcal{R}_0) P 0 = ( D 0 = ( V 0 , s 0 , r 0 , p 0 ) , R 0 ) that can be coherently decorated, with an ordered trace manifold with respect to a tree-compatible induction order ( Bud ( D 0 ) , ≺ 0 ) (\operatorname{Bud}(D_0),\prec_0) ( Bud ( D 0 ) , ≺ 0 ) , where R 0 \mathcal{R}_0 R 0 is injective. Assume Bud ( D 0 ) = { B 1 , … , B n } \operatorname{Bud}(D_0)=\{B_1,\ldots,B_n\} Bud ( D 0 ) = { B 1 , … , B n } . Since R 0 \mathcal{R}_0 R 0 is injective, we define the set of companion nodes of D 0 D_0 D 0 as Comp ( D 0 ) = { C B 1 , … , C B n } \operatorname{Comp}(D_0)=\{C_{B_1},\ldots,C_{B_n}\} Comp ( D 0 ) = { C B 1 , … , C B n } , where for all 1 ≤ i ≤ n 1\leq i\leq n 1 ≤ i ≤ n , R 0 ( B i ) = C B i \mathcal{R}_0(B_i)=C_{B_i} R 0 ( B i ) = C B i . Moreover, for any vertex v ∈ V 0 v\in V_0 v ∈ V 0 , we define T ( v ) T(v) T ( v ) as the derivation subtree in D 0 D_0 D 0 rooted in v v v . We equally define Bud ( T ( v ) ) ⊆ Bud ( D 0 ) \operatorname{Bud}(T(v))\subseteq\operatorname{Bud}(D_0) Bud ( T ( v )) ⊆ Bud ( D 0 ) as the set of bud nodes of D 0 D_0 D 0 appearing in T ( v ) T(v) T ( v ) , Comp ( T ( v ) ) ⊆ Comp ( D 0 ) \operatorname{Comp}(T(v))\subseteq\operatorname{Comp}(D_0) Comp ( T ( v )) ⊆ Comp ( D 0 ) as the set of companion nodes of D 0 D_0 D 0 appearing in T ( v ) T(v) T ( v ) .
Additionally, for any two vertex u , v u,v u , v in a derivation tree D ′ D' D ′ , we write u ⪯ D ′ v u\preceq_{D'}v u ⪯ D ′ v to say that there is a unique directed path from u u u to v v v of size m ≥ 0 m\geq0 m ≥ 0 in D ′ D' D ′ . Now assume there is a trace τ \tau τ over the path α \alpha α in a derivation tree D ′ D' D ′ . If α ′ \alpha' α ′ is a subpath of α \alpha α , we write τ ∣ α ′ \tau|_{\alpha'} τ ∣ α ′ for the trace obtained by restricting τ \tau τ to the subpath α ′ \alpha' α ′ .
The following proposition defines the necessary invariants that are used to describe the structure of the interpolated pre-proof during the execution of the descending phase:
Proposition 47. Consider the vertex v ∈ V 0 v\in V_0 v ∈ V 0 such that s ( v ) = ⊢ Γ , Δ s(v)=\vdash\Gamma,\Delta s ( v ) =⊢ Γ , Δ . Let C = { ( C 1 , C 1 ⊥ ) , … , ( C n , C n ⊥ ) } C=\{(C_1,C_1^\perp),\ldots,(C_n,C_n^\perp)\} C = {( C 1 , C 1 ⊥ ) , … , ( C n , C n ⊥ )} be the cut formulas of T ( v ) T(v) T ( v ) , and let L , R L,R L , R be any partitioning of { 1 , … , n } \{1,\ldots,n\} { 1 , … , n } and C L = { C i , i ∈ L } C^L=\{C_i,i\in L\} C L = { C i , i ∈ L } and C R = { C i , i ∈ R } C^R=\{C_i,i\in R\} C R = { C i , i ∈ R } . Then, the ascending and descending phase over T ( v ) T(v) T ( v ) produces two cut-free partial pre-proofs P l = ( D l , R l ) \mathcal{P}^l=(D^l,\mathcal{R}^l) P l = ( D l , R l ) and P r = ( D r , R r ) \mathcal{P}^r=(D^r,\mathcal{R}^r) P r = ( D r , R r ) with conclusions ⊢ Γ , I \vdash\Gamma,I ⊢ Γ , I and ⊢ Δ , I ⊥ \vdash\Delta,I^\perp ⊢ Δ , I ⊥ , such that P = ( D , R = R l ∪ R r ) \mathcal{P}=(D,\mathcal{R}=\mathcal{R}^l\cup\mathcal{R}^r) P = ( D , R = R l ∪ R r ) , with the form P = P l P r ⊢ Γ , Δ ( C u t ) \mathcal{P}=\frac{\mathcal{P}^l\quad\mathcal{P}^r}{\vdash\Gamma,\Delta}\;(\mathrm{Cut}) P = ⊢ Γ , Δ P l P r ( Cut ) and satisfying:
B u d ( T ( v ) ) = { B 1 , … , B n } Bud(T(v)) = \{B_1,\ldots,B_n\} B u d ( T ( v )) = { B 1 , … , B n } if and only if B u d ( D l ) = { B 1 l , … , B n l } Bud(\mathcal{D}^{l}) = \{B_1^{l},\ldots,B_n^{l}\} B u d ( D l ) = { B 1 l , … , B n l } and B u d ( D r ) = { B 1 r , … , B n r } Bud(\mathcal{D}^{r}) = \{B_1^{r},\ldots,B_n^{r}\} B u d ( D r ) = { B 1 r , … , B n r } , i.e. the set of bud nodes in B u d ( D l ) Bud(\mathcal{D}^{l}) B u d ( D l ) (resp. B u d ( D r ) Bud(\mathcal{D}^{r}) B u d ( D r ) ) is an exact copy of the set B u d ( T ( v ) ) = { B 1 , … , B n } Bud(T(v)) = \{B_1,\ldots,B_n\} B u d ( T ( v )) = { B 1 , … , B n } .
V o c ( I ) ⊆ ( V o c ( Γ ) ∪ C L ) ∩ ( V o c ( Δ ) ∪ C R ) Voc(I) \subseteq(Voc(\Gamma) \cup\mathcal{C}^{L}) \cap(Voc(\Delta) \cup\mathcal{C}^{R}) V oc ( I ) ⊆ ( V oc ( Γ ) ∪ C L ) ∩ ( V oc ( Δ ) ∪ C R ) . Moreover, the set of fixed-point variables of I I I is given by V a r ( I ) = { X B i ∣ B i ∈ B u d ( T ( v ) ) } Var(I) = \{X_{B_i} \mid B_i \in Bud(T(v))\} V a r ( I ) = { X B i ∣ B i ∈ B u d ( T ( v ))} while the set of bounded fixed-point variables is B V ( I ) = { X B i ∣ B i ∈ B u d ( T ( v ) ) ∧ C B i ∈ C o m p ( T ( v ) ) } BV(I) = \{X_{B_i} \mid B_i \in Bud(T(v)) \land C_{B_i} \in Comp(T(v))\} B V ( I ) = { X B i ∣ B i ∈ B u d ( T ( v )) ∧ C B i ∈ C o m p ( T ( v ))} .
R l \mathcal{R}^{l} R l and R r \mathcal{R}^{r} R r are injective, and for every B i ∈ B u d ( T ( v ) ) B_i \in Bud(T(v)) B i ∈ B u d ( T ( v )) such that C B i ∈ C o m p ( T ( v ) ) C_{B_i} \in Comp(T(v)) C B i ∈ C o m p ( T ( v )) , then R l \mathcal{R}^{l} R l (resp. R r \mathcal{R}^{r} R r ) is defined over B i l B_i^{l} B i l (resp. B i r B_i^{r} B i r ). We define C B i l = R l ( B i l ) C_{B_i}^{l} = \mathcal{R}^{l}(B_i^{l}) C B i l = R l ( B i l ) (resp. C B i r = R l ( B i r ) C_{B_i}^{r} = \mathcal{R}^{l}(B_i^{r}) C B i r = R l ( B i r ) ).
For all C B i , C B j ∈ C o m p ( T ( v ) ) C_{B_i}, C_{B_j} \in Comp(T(v)) C B i , C B j ∈ C o m p ( T ( v )) and B k ∈ B u d ( T ( v ) ) B_k \in Bud(T(v)) B k ∈ B u d ( T ( v )) , then:
a. C B i ⪯ T ( v ) C B j C_{B_i} \preceq_{T(v)} C_{B_j} C B i ⪯ T ( v ) C B j if and only if C B i l ⪯ D l C B j l C_{B_i}^{l} \preceq_{\mathcal{D}^{l}} C_{B_j}^{l} C B i l ⪯ D l C B j l and C B i r ⪯ D r C B j r C_{B_i}^{r} \preceq_{\mathcal{D}^{r}} C_{B_j}^{r} C B i r ⪯ D r C B j r
b. C B i ⪯ T ( v ) B k C_{B_i} \preceq_{T(v)} B_k C B i ⪯ T ( v ) B k if and only if C B i l ⪯ D l B k l C_{B_i}^{l} \preceq_{\mathcal{D}^{l}} B_k^{l} C B i l ⪯ D l B k l and C B i r ⪯ D r B k r C_{B_i}^{r} \preceq_{\mathcal{D}^{r}} B_k^{r} C B i r ⪯ D r B k r
Assume v ′ v' v ′ is the node labelled with ⊢ Γ , Δ \vdash\Gamma,\Delta ⊢ Γ , Δ at the root of P 0 \mathcal{P}_0 P 0 . Now, consider the original ordered trace manifold for P 0 \mathcal{P}_0 P 0 :
{ τ B j , B i ∣ B j , B i ∈ B u d ( D 0 ) , B i ⊲ 0 B j } \{\tau_{B_j,B_i} \mid B_j,B_i \in Bud(\mathcal{D}_0), B_i \lhd_0 B_j\} { τ B j , B i ∣ B j , B i ∈ B u d ( D 0 ) , B i ⊲ 0 B j }
with the induction order ( B u d ( D 0 ) , ⊲ 0 ) (Bud(\mathcal{D}_0), \lhd_0) ( B u d ( D 0 ) , ⊲ 0 ) . Let B ⊆ B u d ( D 0 ) \mathcal{B} \subseteq Bud(\mathcal{D}_0) B ⊆ B u d ( D 0 ) be the maximal set of elements such that for all B ′ ∈ B B' \in\mathcal{B} B ′ ∈ B , then B ′ ⊲ 0 B k B' \lhd_0 B_k B ′ ⊲ 0 B k , for some B k ∈ B u d ( D 0 ) B_k \in Bud(\mathcal{D}_0) B k ∈ B u d ( D 0 ) , with the associated trace τ B k , B ′ \tau_{B_k,B'} τ B k , B ′ , with B \mathcal{B} B being weakly ⪯ P 0 \preceq_{\mathcal{P}_0} ⪯ P 0 -connected.
a. Assume B i ∈ B u d ( T ( v ) ) B_i \in Bud(T(v)) B i ∈ B u d ( T ( v )) and B i ∈ B B_i \in\mathcal{B} B i ∈ B :
i. If C B i ∉ C o m p ( T ( v ) ) C_{B_i} \notin Comp(T(v)) C B i ∈ / C o m p ( T ( v )) , then there exists either a trace τ B k l , B i l \tau_{B_k^{l},B_i^{l}} τ B k l , B i l over the unique path from v ′ v' v ′ to B i l B_i^{l} B i l in D \mathcal{D} D , or a trace τ B k r , B i r \tau_{B_k^{r},B_i^{r}} τ B k r , B i r over the unique path from v ′ v' v ′ to B i r B_i^{r} B i r in D \mathcal{D} D , such that τ B k , B i ( v ) = τ B k l , B i l ( v ′ ) \tau_{B_k,B_i}(v) = \tau_{B_k^{l},B_i^{l}}(v') τ B k , B i ( v ) = τ B k l , B i l ( v ′ ) and τ B k , B i ( B i ) = τ B k l , B i l ( B i l ) \tau_{B_k,B_i}(B_i) = \tau_{B_k^{l},B_i^{l}}(B_i^{l}) τ B k , B i ( B i ) = τ B k l , B i l ( B i l ) or τ B k , B i ( v ) = τ B k r , B i r ( v ′ ) \tau_{B_k,B_i}(v) = \tau_{B_k^{r},B_i^{r}}(v') τ B k , B i ( v ) = τ B k r , B i r ( v ′ ) and τ B k , B i ( B i ) = τ B k r , B i r ( B i r ) \tau_{B_k,B_i}(B_i) = \tau_{B_k^{r},B_i^{r}}(B_i^{r}) τ B k , B i ( B i ) = τ B k r , B i r ( B i r ) depending on the case.
ii. If C B i ∈ C o m p ( T ( v ) ) C_{B_i} \in Comp(T(v)) C B i ∈ C o m p ( T ( v )) , then, there exists either a trace τ B k l , B i l \tau_{B_k^{l},B_i^{l}} τ B k l , B i l over the unique path from C B i l C_{B_i}^{l} C B i l to B i l B_i^{l} B i l in D l \mathcal{D}^{l} D l , or a trace τ B k r , B i r \tau_{B_k^{r},B_i^{r}} τ B k r , B i r over the unique path from C B i r C_{B_i}^{r} C B i r to B i r B_i^{r} B i r in D r \mathcal{D}^{r} D r , such that τ B k l , B i l ( C B i l ) = τ B k l , B i l ( B i l ) \tau_{B_k^{l},B_i^{l}}(C_{B_i}^{l}) = \tau_{B_k^{l},B_i^{l}}(B_i^{l}) τ B k l , B i l ( C B i l ) = τ B k l , B i l ( B i l ) or τ B k r , B i r ( C B i r ) = τ B k r , B i r ( B i r ) \tau_{B_k^{r},B_i^{r}}(C_{B_i}^{r}) = \tau_{B_k^{r},B_i^{r}}(B_i^{r}) τ B k r , B i r ( C B i r ) = τ B k r , B i r ( B i r ) depending on the case.
b. Assume B i , B j ∈ B B_i,B_j \in\mathcal{B} B i , B j ∈ B (i.e. B i , B j ⊲ 0 B k B_i,B_j \lhd_0 B_k B i , B j ⊲ 0 B k ) and B i , B j ∈ B u d ( T ( v ) ) B_i,B_j \in Bud(T(v)) B i , B j ∈ B u d ( T ( v )) , with B j ⪯ P 0 B i B_j \preceq_{\mathcal{P}_0} B_i B j ⪯ P 0 B i and C B j ∈ C o m p ( T ( v ) ) C_{B_j} \in Comp(T(v)) C B j ∈ C o m p ( T ( v )) . Then:
i. If C B i ∉ C o m p ( T ( v ) ) C_{B_i} \notin Comp(T(v)) C B i ∈ / C o m p ( T ( v )) then the companion node C B j l C_{B_j}^{l} C B j l (resp. C B j r C_{B_j}^{r} C B j r ) is in the unique path from v ′ v' v ′ to B i l B_i^{l} B i l (resp. B i r B_i^{r} B i r ) in D \mathcal{D} D , such that τ B k l , B i l ( C B j l ) = τ B k l , B j l ( C B j l ) \tau_{B_k^{l},B_i^{l}}(C_{B_j}^{l}) = \tau_{B_k^{l},B_j^{l}}(C_{B_j}^{l}) τ B k l , B i l ( C B j l ) = τ B k l , B j l ( C B j l ) or τ B k r , B i r ( C B j r ) = τ B k r , B j r ( C B j r ) \tau_{B_k^{r},B_i^{r}}(C_{B_j}^{r}) = \tau_{B_k^{r},B_j^{r}}(C_{B_j}^{r}) τ B k r , B i r ( C B j r ) = τ B k r , B j r ( C B j r ) .
ii. If C B i ∈ C o m p ( T ( v ) ) C_{B_i} \in Comp(T(v)) C B i ∈ C o m p ( T ( v )) then companion node C B j l C_{B_j}^{l} C B j l (resp. C B j r C_{B_j}^{r} C B j r ) is in the cycle C B i l C_{B_i}^{l} C B i l (resp. C B i r C_{B_i}^{r} C B i r ) in G P l \mathcal{G}_{\mathcal{P}^{l}} G P l (resp. G P r \mathcal{G}_{\mathcal{P}^{r}} G P r ) such that τ B k l , B i l ( C B j l ) = τ B k l , B j l ( C B j l ) \tau_{B_k^{l},B_i^{l}}(C_{B_j}^{l}) = \tau_{B_k^{l},B_j^{l}}(C_{B_j}^{l}) τ B k l , B i l ( C B j l ) = τ B k l , B j l ( C B j l ) or τ B k r , B i r ( C B j r ) = τ B k r , B j r ( C B j r ) \tau_{B_k^{r},B_i^{r}}(C_{B_j}^{r}) = \tau_{B_k^{r},B_j^{r}}(C_{B_j}^{r}) τ B k r , B i r ( C B j r ) = τ B k r , B j r ( C B j r ) .
c. Assume B i ∈ B u d ( T ( v ) ) B_i \in Bud(T(v)) B i ∈ B u d ( T ( v )) and B i ∈ B B_i \in\mathcal{B} B i ∈ B .
i. If C B i ∉ C o m p ( T ( v ) ) C_{B_i} \notin Comp(T(v)) C B i ∈ / C o m p ( T ( v )) , consider α \alpha α as the unique path from v v v to B i B_i B i in D 0 \mathcal{D}_0 D 0 . Then F o r m ( τ B k , B i ∣ α ) = F o r m ( τ B k l , B i l ) Form(\tau_{B_k,B_i} \mid\alpha) = Form(\tau_{B_k^{l},B_i^{l}}) F or m ( τ B k , B i ∣ α ) = F or m ( τ B k l , B i l ) (resp. F o r m ( τ B k , B i ∣ α ) = F o r m ( τ B k r , B i r ) Form(\tau_{B_k,B_i} \mid\alpha) = Form(\tau_{B_k^{r},B_i^{r}}) F or m ( τ B k , B i ∣ α ) = F or m ( τ B k r , B i r ) ). Moreover, if τ B k , B i ∣ α \tau_{B_k,B_i} \mid\alpha τ B k , B i ∣ α has an F F F -progress point, then τ B k l , B i l \tau_{B_k^{l},B_i^{l}} τ B k l , B i l (resp. τ B k r , B i r \tau_{B_k^{r},B_i^{r}} τ B k r , B i r ) also has an F F F -progress point.
ii. If C B i ∈ C o m p ( T ( v ) ) C_{B_i} \in Comp(T(v)) C B i ∈ C o m p ( T ( v )) then F o r m ( τ B k , B i ) = F o r m ( τ B k l , B i l ) Form(\tau_{B_k,B_i}) = Form(\tau_{B_k^{l},B_i^{l}}) F or m ( τ B k , B i ) = F or m ( τ B k l , B i l ) (resp. F o r m ( τ B k , B i ) = F o r m ( τ B k r , B i r ) Form(\tau_{B_k,B_i}) = Form(\tau_{B_k^{r},B_i^{r}}) F or m ( τ B k , B i ) = F or m ( τ B k r , B i r ) ). Moreover, if τ B k , B i \tau_{B_k,B_i} τ B k , B i has an F F F -progress point, then τ B k l , B i l \tau_{B_k^{l},B_i^{l}} τ B k l , B i l (resp. τ B k r , B i r \tau_{B_k^{r},B_i^{r}} τ B k r , B i r ) also has an F F F -progress point.
For every B i ∈ B u d ( T ( v ) ) B_i \in Bud(T(v)) B i ∈ B u d ( T ( v )) , there exist the traces ρ B i l \rho_{B_i^{l}} ρ B i l and ρ B i r \rho_{B_i^{r}} ρ B i r , called the interpolant traces of B i B_i B i , satisfying that for any choice of the operators σ j ∈ { μ , ν } \sigma_j \in\{\mu,\nu\} σ j ∈ { μ , ν } , with the dual
operator σ j ⊥ \sigma_j^\perp σ j ⊥ applied on the right side, performed as the rule of each companion node C B i l C_{B_i}^l C B i l (resp. C B i r C_{B_i}^r C B i r ) in C o m p ( T ( v ) ) Comp(T(v)) C o m p ( T ( v )) :
We have two cases:
If C B i ∉ C o m p ( T ( v ) ) C_{B_i}\notin Comp(T(v)) C B i ∈ / C o m p ( T ( v )) , then ρ B i l \rho_{B_i}^l ρ B i l (resp. ρ B i r \rho_{B_i}^r ρ B i r ) is a trace over the unique path from v ′ v' v ′ to B i l B_i^l B i l (resp. B i r B_i^r B i r ) in D D D , such that ρ B i l ( v ′ ) = I \rho_{B_i}^l(v')=I ρ B i l ( v ′ ) = I (resp. ρ B i r ( v ′ ) = I ⊥ \rho_{B_i}^r(v')=I^\perp ρ B i r ( v ′ ) = I ⊥ ) and ρ B i l ( B i l ) = X B i \rho_{B_i}^l(B_i^l)=X_{B_i} ρ B i l ( B i l ) = X B i (resp. ρ B i r ( B i r ) = X B i \rho_{B_i}^r(B_i^r)=X_{B_i} ρ B i r ( B i r ) = X B i ).
If C B i ∈ C o m p ( T ( v ) ) C_{B_i}\in Comp(T(v)) C B i ∈ C o m p ( T ( v )) , then ρ B i l \rho_{B_i}^l ρ B i l (resp. ρ B i r \rho_{B_i}^r ρ B i r ) is a trace over the unique path from C B i l C_{B_i}^l C B i l (resp. C B i r C_{B_i}^r C B i r ) to B i l B_i^l B i l (resp. B i r B_i^r B i r ) in D l D^l D l (resp. D r D^r D r ), such that ρ B i l ( B i l ) = ρ B i l ( C B i l ) \rho_{B_i}^l(B_i^l)=\rho_{B_i}^l(C_{B_i}^l) ρ B i l ( B i l ) = ρ B i l ( C B i l ) (resp. ρ B i r ( B i r ) = ρ B i r ( C B i r ) \rho_{B_i}^r(B_i^r)=\rho_{B_i}^r(C_{B_i}^r) ρ B i r ( B i r ) = ρ B i r ( C B i r ) ).
For every vertex u u u in the path followed by ρ B i l \rho_{B_i}^l ρ B i l (resp. ρ B i r \rho_{B_i}^r ρ B i r ), the formula ρ B i l ( u ) \rho_{B_i}^l(u) ρ B i l ( u ) (resp. ρ B i r ( u ) \rho_{B_i}^r(u) ρ B i r ( u ) ) is the unique occurrence of a subformula of I I I (resp. I ⊥ I^\perp I ⊥ ) in s ( u ) s(u) s ( u ) containing the occurrence of the fixed-point variable X B i X_{B_i} X B i or σ i X B i . I i \sigma_iX_{B_i}.I_i σ i X B i . I i (resp. σ i ⊥ X B i . I i ⊥ \sigma_i^\perp X_{B_i}.I_i^\perp σ i ⊥ X B i . I i ⊥ ). In particular, if B j ∈ B u d ( T ( v ) ) B_j\in Bud(T(v)) B j ∈ B u d ( T ( v )) with C B j ∈ C o m p ( T ( v ) ) C_{B_j}\in Comp(T(v)) C B j ∈ C o m p ( T ( v )) and B j ≤ P 0 B i B_j\leq_{P_0}B_i B j ≤ P 0 B i , then the companion node C B j l C_{B_j}^l C B j l (resp. C B j r C_{B_j}^r C B j r ) appears in the path followed by ρ B i l \rho_{B_i}^l ρ B i l (resp. ρ B i r \rho_{B_i}^r ρ B i r ), by the property (4), and ρ B i l ( C B j l ) = σ j X B j . I j \rho_{B_i}^l(C_{B_j}^l)=\sigma_jX_{B_j}.I_j ρ B i l ( C B j l ) = σ j X B j . I j (resp. ρ B i r ( C B j r ) = σ j ⊥ X B j . I j ⊥ \rho_{B_i}^r(C_{B_j}^r)=\sigma_j^\perp X_{B_j}.I_j^\perp ρ B i r ( C B j r ) = σ j ⊥ X B j . I j ⊥ ), where σ j X B j . I j \sigma_jX_{B_j}.I_j σ j X B j . I j (resp. σ j ⊥ X B j . I j ⊥ \sigma_j^\perp X_{B_j}.I_j^\perp σ j ⊥ X B j . I j ⊥ ) is the formula labelling the companion node C B j l C_{B_j}^l C B j l (resp. C B j r C_{B_j}^r C B j r ) in the current interpolated pre-proof.
If C B i ∈ C o m p ( T ( v ) ) C_{B_i}\in Comp(T(v)) C B i ∈ C o m p ( T ( v )) , then ρ B i l ( C B i l ) = ρ B i l ( B i l ) = σ i X B i . I i \rho_{B_i}^l(C_{B_i}^l)=\rho_{B_i}^l(B_i^l)=\sigma_iX_{B_i}.I_i ρ B i l ( C B i l ) = ρ B i l ( B i l ) = σ i X B i . I i (resp. ρ B i r ( C B i r ) = ρ B i r ( B i r ) = σ i ⊥ X B i . I i ⊥ \rho_{B_i}^r(C_{B_i}^r)=\rho_{B_i}^r(B_i^r)=\sigma_i^\perp X_{B_i}.I_i^\perp ρ B i r ( C B i r ) = ρ B i r ( B i r ) = σ i ⊥ X B i . I i ⊥ ), and the step of ρ B i l \rho_{B_i}^l ρ B i l (resp. ρ B i r \rho_{B_i}^r ρ B i r ) out of C B i l C_{B_i}^l C B i l (resp. C B i r C_{B_i}^r C B i r ) is the Fischer–Ladner unfolding of σ i X B i . I i \sigma_iX_{B_i}.I_i σ i X B i . I i (resp. σ i ⊥ X B i . I i ⊥ \sigma_i^\perp X_{B_i}.I_i^\perp σ i ⊥ X B i . I i ⊥ ).
For all C B j , C B m ∈ C o m p ( T ( v ) ) C_{B_j},C_{B_m}\in Comp(T(v)) C B j , C B m ∈ C o m p ( T ( v )) : C B m ⪯ T ( v ) C B j C_{B_m}\preceq_{T(v)}C_{B_j} C B m ⪯ T ( v ) C B j if and only if σ m X B m . I m \sigma_mX_{B_m}.I_m σ m X B m . I m is a subformula of σ j X B j . I j \sigma_jX_{B_j}.I_j σ j X B j . I j .
Proof. The proof of Proposition 47, is done by induction over the tree T ( v ) T(v) T ( v ) . The proof of property (6) can be done by reasoning on induction, and making simple observations of how the traces formed by the interpolant change during the descending phase, similar to the proof of the property (5).
Case Axiom:
Assume T ( v ) T(v) T ( v ) has the form π = ⊢ F , F ⊥ ⊢ ( A x ) \pi=\frac{\vdash F,F^\perp}{\phantom{\vdash}}^{(\mathrm{Ax})} π = ⊢ ⊢ F , F ⊥ ( Ax ) , and assume an arbitrary split ( Γ , Δ ) (\Gamma,\Delta) ( Γ , Δ ) of the sequent ⊢ F , F ⊥ \vdash F,F^\perp ⊢ F , F ⊥ . In this case B u d ( T ( v ) ) = C o m p ( T ( v ) ) = ∅ Bud(T(v))=Comp(T(v))=\varnothing B u d ( T ( v )) = C o m p ( T ( v )) = ∅ . After the ascending and descending phase we obtain the derivation trees D l = π 1 l D^l=\pi_1^l D l = π 1 l and D r = π 1 r D^r=\pi_1^r D r = π 1 r , where π 1 l , π 1 r \pi_1^l,\pi_1^r π 1 l , π 1 r are defined in the four transformations for the axiom case in the Appendix B.1, depending on the split ( Γ , Δ ) (\Gamma,\Delta) ( Γ , Δ ) .
We have that B u d ( T ( v ) ) = B u d ( D l ) = B u d ( D r ) = ∅ Bud(T(v))=Bud(D^l)=Bud(D^r)=\varnothing B u d ( T ( v )) = B u d ( D l ) = B u d ( D r ) = ∅ .
The constraint on the vocabulary of I I I is satisfied, by observation on all the cases in Appendix B.1. V a r ( I ) = B V ( I ) = ∅ Var(I)=BV(I)=\varnothing V a r ( I ) = B V ( I ) = ∅ since B u d ( T ( v ) ) = ∅ Bud(T(v))=\varnothing B u d ( T ( v )) = ∅ .
It is trivially true since R l = R r = ∅ R^l=R^r=\varnothing R l = R r = ∅ .
It is trivially true since C o m p ( T ( v ) ) = ∅ Comp(T(v))=\varnothing C o m p ( T ( v )) = ∅
It is trivially true since there is no trace passing through the node v v v .
The case for the leafs with the rules ( ⊤ ) (\top) ( ⊤ ) and ( 1 ) (1) ( 1 ) follow the same reasoning, as for the ( A x ) (\mathrm{Ax}) ( Ax ) case.
Case bud node:
Assume T ( v ) T(v) T ( v ) has the form π = ⊢ Γ , Δ \pi= \vdash\Gamma, \Delta π =⊢ Γ , Δ . In this case v v v is a bud node B i ∈ B u d ( D 0 ) B_i \in Bud(D_0) B i ∈ B u d ( D 0 ) , therefore B u d ( T ( v ) ) = { B i } Bud(T(v)) = \{B_i\} B u d ( T ( v )) = { B i } and C o m p ( T ( v ) ) = ∅ Comp(T(v)) = \emptyset C o m p ( T ( v )) = ∅ . After the ascending and descending phase, we obtain
⊢ Γ , X B i ⊢ Δ , X B i ⊢ Γ , Δ ( C u t ) \frac{\vdash\Gamma, X_{B_i} \qquad\vdash\Delta, X_{B_i}}{\vdash\Gamma, \Delta}\;(\mathrm{Cut}) ⊢ Γ , Δ ⊢ Γ , X B i ⊢ Δ , X B i ( Cut )
where we take D l = ⊢ Γ , X B i D^l = \vdash\Gamma, X_{B_i} D l =⊢ Γ , X B i and D r = ⊢ Δ , X B i D^r = \vdash\Delta, X_{B_i} D r =⊢ Δ , X B i as the derivation trees containing only one node.
The bud node B i B_i B i labelled with ⊢ Γ , Δ \vdash\Gamma, \Delta ⊢ Γ , Δ is divided into the bud nodes B i l , B i r B_i^l, B_i^r B i l , B i r labelled as ⊢ Γ , X B i \vdash\Gamma, X_{B_i} ⊢ Γ , X B i and ⊢ Δ , X B i \vdash\Delta, X_{B_i} ⊢ Δ , X B i respectively. Therefore B u d ( D l ) = { B i l } Bud(D^l) = \{B_i^l\} B u d ( D l ) = { B i l } and B u d ( D r ) = { B i r } Bud(D^r) = \{B_i^r\} B u d ( D r ) = { B i r } .
The constraint on the vocabulary of I I I is true since V o c ( I ) = ∅ Voc(I) = \emptyset V oc ( I ) = ∅ . Moreover, V a r ( I ) = { X B i } Var(I) = \{X_{B_i}\} V a r ( I ) = { X B i } and B V ( I ) = ∅ BV(I) = \emptyset B V ( I ) = ∅ .
It is trivially true since R l = R r = ∅ R^l = R^r = \emptyset R l = R r = ∅ .
It is trivially true since C o m p ( T ( v ) ) = ∅ Comp(T(v)) = \emptyset C o m p ( T ( v )) = ∅ .
In this case C B i ∉ C o m p ( T ( v ) ) C_{B_i} \notin Comp(T(v)) C B i ∈ / C o m p ( T ( v )) and we know that B i l , B i r B_i^l, B_i^r B i l , B i r exist by (1). Take any trace of the form τ B k , B i \tau_{B_k,B_i} τ B k , B i for a B k ∈ B u d ( D 0 ) B_k \in Bud(D_0) B k ∈ B u d ( D 0 ) , such that B i ◃ 0 B k B_i \triangleleft_0 B_k B i ◃ 0 B k in the original ordered trace manifold. Therefore, τ B k , B i ( v ) = G \tau_{B_k,B_i}(v) = G τ B k , B i ( v ) = G with either G ∈ Γ G \in\Gamma G ∈ Γ or G ∈ Δ G \in\Delta G ∈ Δ . Then:
a. If G ∈ Γ G \in\Gamma G ∈ Γ take τ B k , B i B i l = ( G , G ) \tau_{B_k,B_i}^{B_i^l} = (G,G) τ B k , B i B i l = ( G , G ) over the path from v ′ v' v ′ to B i l B_i^l B i l , where τ B k , B i ( v ) = τ B k , B i B i l ( v ′ ) = G \tau_{B_k,B_i}(v) = \tau_{B_k,B_i}^{B_i^l}(v') = G τ B k , B i ( v ) = τ B k , B i B i l ( v ′ ) = G and τ B k , B i ( B i ) = τ B k , B i B i l ( B i l ) = G \tau_{B_k,B_i}(B_i) = \tau_{B_k,B_i}^{B_i^l}(B_i^l) = G τ B k , B i ( B i ) = τ B k , B i B i l ( B i l ) = G . Otherwise, if G ∈ Δ G \in\Delta G ∈ Δ then take τ B k , B i B i r = ( G , G ) \tau_{B_k,B_i}^{B_i^r} = (G,G) τ B k , B i B i r = ( G , G ) over the path from v ′ v' v ′ to B i r B_i^r B i r , where τ B k , B i ( v ) = τ B k , B i B i r ( v ′ ) = G \tau_{B_k,B_i}(v) = \tau_{B_k,B_i}^{B_i^r}(v') = G τ B k , B i ( v ) = τ B k , B i B i r ( v ′ ) = G and τ B k , B i ( B i ) = τ B k , B i B i r ( B i r ) = G \tau_{B_k,B_i}(B_i) = \tau_{B_k,B_i}^{B_i^r}(B_i^r) = G τ B k , B i ( B i ) = τ B k , B i B i r ( B i r ) = G .
b. Trivially true since we have no other bud node.
c. F o r m ( τ B k , B i ∣ α 1 ) = F o r m ( τ B k , B i B i l ∣ α 2 ) = { G } Form(\tau_{B_k,B_i}|_{\alpha_1}) = Form(\tau_{B_k,B_i}^{B_i^l}|_{\alpha_2}) = \{G\} F or m ( τ B k , B i ∣ α 1 ) = F or m ( τ B k , B i B i l ∣ α 2 ) = { G } (resp. F o r m ( τ B k , B i ∣ α 1 ) = F o r m ( τ B k , B i B i r ∣ α 2 ) = { G } Form(\tau_{B_k,B_i}|_{\alpha_1}) = Form(\tau_{B_k,B_i}^{B_i^r}|_{\alpha_2}) = \{G\} F or m ( τ B k , B i ∣ α 1 ) = F or m ( τ B k , B i B i r ∣ α 2 ) = { G } ). There are no progress points.
Case n-ary rule:
Assume the derivation tree T ( v ) T(v) T ( v ) has the form:
π 1 ′ ⋯ π n ′ ⊢ Γ , Δ ( r ) \frac{\pi'_1 \quad\cdots\quad\pi'_n}{\vdash\Gamma, \Delta}\;(r) ⊢ Γ , Δ π 1 ′ ⋯ π n ′ ( r )
After the ascending and descending phase on every subtree π i ′ \pi'_i π i ′ , we obtain the elementary CPRIS:
π 1 L ⊢ Γ 1 , I 1 π 1 R ⊢ I 1 ⊥ , Δ 1 ⊢ Γ 1 , Δ 1 ( C u t ) ⋯ π n L ⊢ Γ n , I n π n R ⊢ I n ⊥ , Δ n ⊢ Γ n , Δ n ⊢ Γ 1 , Δ 1 ⋯ ⊢ Γ n , Δ n ⊢ Γ , Δ ( r ) \frac{
\frac{\pi_1^L}{\vdash\Gamma_1, I_1}
\qquad
\frac{\pi_1^R}{\vdash I_1^\perp, \Delta_1}
}{
\vdash\Gamma_1, \Delta_1
}\;(\mathrm{Cut})
\quad\cdots\quad
\frac{
\frac{\pi_n^L}{\vdash\Gamma_n, I_n}
\qquad
\frac{\pi_n^R}{\vdash I_n^\perp, \Delta_n}
}{
\vdash\Gamma_n, \Delta_n
}
\quad
\frac{\vdash\Gamma_1, \Delta_1 \quad\cdots\quad\vdash\Gamma_n, \Delta_n}{\vdash\Gamma, \Delta}\;(r) ⊢ Γ 1 , Δ 1 ⊢ Γ 1 , I 1 π 1 L ⊢ I 1 ⊥ , Δ 1 π 1 R ( Cut ) ⋯ ⊢ Γ n , Δ n ⊢ Γ n , I n π n L ⊢ I n ⊥ , Δ n π n R ⊢ Γ , Δ ⊢ Γ 1 , Δ 1 ⋯ ⊢ Γ n , Δ n ( r )
Where π 1 , … , π n \pi_1,\ldots,\pi_n π 1 , … , π n are the subtrees, rooted at the nodes labelled with ( ⊢ Γ 1 , Δ 1 ) , … , ( ⊢ Γ n , Δ n ) (\vdash\Gamma_1,\Delta_1),\ldots,(\vdash\Gamma_n,\Delta_n) ( ⊢ Γ 1 , Δ 1 ) , … , ( ⊢ Γ n , Δ n ) respectively. The cut-introduction step transforms the previous derivation tree into a solved CPRIS, of the form:
π L ⊢ Γ , I π R ⊢ I ⊥ , Δ ⊢ Γ , Δ ( C u t ) \frac{\pi^L}{\vdash\Gamma, I}
\qquad
\frac{\pi^R}{\vdash I^\perp, \Delta}
\over
\vdash\Gamma, \Delta\;(\mathrm{Cut}) ⊢ Γ , Δ ( Cut ) ⊢ Γ , I π L ⊢ I ⊥ , Δ π R
For each rule ( r ) (r) ( r ) , the transformation and the derivation trees π L \pi^L π L and π R \pi^R π R are defined in the Appendix § 7 .§ 1 and Figure 4 (b). Then, we have two cases: either the root v v v is a companion node or not. If it is not a companion node, then the descending phase concludes with the cut-introduction step specified above. In this case, we define D l = π L D^l = \pi^L D l = π L and D r = π R D^r = \pi^R D r = π R .
Otherwise, if the root v v v is a companion, say v = C B k v=C_{B_k} v = C B k , then the previous derivation tree is therefore transformed into:
π L [ σ X B k . I / X B k ] ⊢ Γ , I [ σ X B k . I / X B k ] ⊢ Γ , σ X B k . I ( σ ) π R [ σ ⊥ X B k . I ⊥ / X B k ] ⊢ Δ , I ⊥ [ σ ⊥ X B k . I ⊥ / X B k ] ⊢ Δ , σ ⊥ X B k . I ⊥ ( σ ⊥ ) ⊢ Γ , σ X B k . I ⊢ Δ , σ ⊥ X B k . I ⊥ ⊢ Γ , Δ ( C u t ) \begin{aligned}
&\frac{\pi^{L}[\sigma X_{B_k}.I/X_{B_k}]}{\frac{\vdash\Gamma,I[\sigma X_{B_k}.I/X_{B_k}]}{\vdash\Gamma,\sigma X_{B_k}.I}\;(\sigma)}
\qquad
\frac{\pi^{R}[\sigma^{\perp}X_{B_k}.I^{\perp}/X_{B_k}]}{\frac{\vdash\Delta,I^{\perp}[\sigma^{\perp}X_{B_k}.I^{\perp}/X_{B_k}]}{\vdash\Delta,\sigma^{\perp}X_{B_k}.I^{\perp}}\;(\sigma^{\perp})}
\\
&\hspace{4cm}\frac{\vdash\Gamma,\sigma X_{B_k}.I\qquad\vdash\Delta,\sigma^{\perp}X_{B_k}.I^{\perp}}{\vdash\Gamma,\Delta}\;(\mathrm{Cut})
\end{aligned} ⊢ Γ , σ X B k . I ⊢ Γ , I [ σ X B k . I / X B k ] ( σ ) π L [ σ X B k . I / X B k ] ⊢ Δ , σ ⊥ X B k . I ⊥ ⊢ Δ , I ⊥ [ σ ⊥ X B k . I ⊥ / X B k ] ( σ ⊥ ) π R [ σ ⊥ X B k . I ⊥ / X B k ] ⊢ Γ , Δ ⊢ Γ , σ X B k . I ⊢ Δ , σ ⊥ X B k . I ⊥ ( Cut )
In this case, in R l \mathcal{R}^{l} R l (resp. R r \mathcal{R}^{r} R r ) we pair the bud node B k l B^{l}_{k} B k l (resp. B k r B^{r}_{k} B k r ) with the node corresponding to the left (resp. right) premise of the cut. We define D l = π L [ σ X B k . I / X B k ] ⊢ Γ , I [ σ X B k . I / X B k ] ⊢ Γ , σ X B k . I ( σ ) \mathcal{D}^{l}=\frac{\pi^{L}[\sigma X_{B_k}.I/X_{B_k}]}{\frac{\vdash\Gamma,I[\sigma X_{B_k}.I/X_{B_k}]}{\vdash\Gamma,\sigma X_{B_k}.I}\;(\sigma)} D l = ⊢ Γ , σ X B k . I ⊢ Γ , I [ σ X B k . I / X B k ] ( σ ) π L [ σ X B k . I / X B k ]
and D r = π R [ σ ⊥ X B k . I ⊥ / X B k ] ⊢ Δ , I ⊥ [ σ ⊥ X B k . I ⊥ / X B k ] ⊢ Δ , σ ⊥ X B k . I ⊥ ( σ ⊥ ) \mathcal{D}^{r}=\frac{\pi^{R}[\sigma^{\perp}X_{B_k}.I^{\perp}/X_{B_k}]}{\frac{\vdash\Delta,I^{\perp}[\sigma^{\perp}X_{B_k}.I^{\perp}/X_{B_k}]}{\vdash\Delta,\sigma^{\perp}X_{B_k}.I^{\perp}}\;(\sigma^{\perp})} D r = ⊢ Δ , σ ⊥ X B k . I ⊥ ⊢ Δ , I ⊥ [ σ ⊥ X B k . I ⊥ / X B k ] ( σ ⊥ ) π R [ σ ⊥ X B k . I ⊥ / X B k ] .
In both cases, whether v v v is a companion or not, one observes that:
For the unary rules { ℘ , ⊥ , ⊕ i , ? d , ? w , ? c , ! p , ν , μ } \{\mathord{\wp},\bot,\oplus_i,?d,?w,?c,!p,\nu,\mu\} { ℘ , ⊥ , ⊕ i , ? d , ? w , ? c , ! p , ν , μ } , we have that B u d ( T ( v ) ) = B u d ( π 1 ) = { B 1 , … , B m } Bud(T(v))=Bud(\pi_1)=\{B_1,\ldots,B_m\} B u d ( T ( v )) = B u d ( π 1 ) = { B 1 , … , B m } if and only if B u d ( π 1 l ) = { B 1 l , … , B m l } Bud(\pi_1^l)=\{B_1^l,\ldots,B_m^l\} B u d ( π 1 l ) = { B 1 l , … , B m l } and B u d ( π 1 r ) = { B 1 r , … , B m r } Bud(\pi_1^r)=\{B_1^r,\ldots,B_m^r\} B u d ( π 1 r ) = { B 1 r , … , B m r } by the induction hypothesis. Then, by the structure of the derivation trees π L \pi^L π L and π R \pi^R π R , one derives that B u d ( D l ) = B u d ( π L ) = B u d ( π 1 L ) Bud(\mathcal{D}^l)=Bud(\pi^L)=Bud(\pi_1^L) B u d ( D l ) = B u d ( π L ) = B u d ( π 1 L ) and B u d ( D r ) = B u d ( π R ) = B u d ( π 1 R ) Bud(\mathcal{D}^r)=Bud(\pi^R)=Bud(\pi_1^R) B u d ( D r ) = B u d ( π R ) = B u d ( π 1 R ) .
For the binary rules { ⊗ , & , c u t } \{\otimes,\&,cut\} { ⊗ , & , c u t } , we have that B u d ( T ( v ) ) = B u d ( π 1 ) ∪ B u d ( π 2 ) = { B 1 , … , B m } ∪ { B 1 ′ , … , B k ′ } Bud(T(v))=Bud(\pi_1)\cup Bud(\pi_2)=\{B_1,\ldots,B_m\}\cup\{B'_1,\ldots,B'_k\} B u d ( T ( v )) = B u d ( π 1 ) ∪ B u d ( π 2 ) = { B 1 , … , B m } ∪ { B 1 ′ , … , B k ′ } if and only if B u d ( π 1 L ) = { B 1 l , … , B m l } Bud(\pi_1^L)=\{B_1^l,\ldots,B_m^l\} B u d ( π 1 L ) = { B 1 l , … , B m l } , B u d ( π 1 R ) = { B 1 r , … , B m r } Bud(\pi_1^R)=\{B_1^r,\ldots,B_m^r\} B u d ( π 1 R ) = { B 1 r , … , B m r } , B u d ( π 2 L ) = { B 1 ′ l , … , B k ′ l } Bud(\pi_2^L)=\{B_1^{\prime l},\ldots,B_k^{\prime l}\} B u d ( π 2 L ) = { B 1 ′ l , … , B k ′ l } , and B u d ( π 2 R ) = { B 1 ′ r , … , B k ′ r } Bud(\pi_2^R)=\{B_1^{\prime r},\ldots,B_k^{\prime r}\} B u d ( π 2 R ) = { B 1 ′ r , … , B k ′ r } by the induction hypothesis. Then, by the structure of the derivation trees π L \pi^L π L and π R \pi^R π R , one derives that B u d ( D l ) = B u d ( π L ) = B u d ( π 1 L ) ∪ B u d ( π 2 L ) Bud(\mathcal{D}^l)=Bud(\pi^L)=Bud(\pi_1^L)\cup Bud(\pi_2^L) B u d ( D l ) = B u d ( π L ) = B u d ( π 1 L ) ∪ B u d ( π 2 L ) and B u d ( D r ) = B u d ( π R ) = B u d ( π 1 R ) ∪ B u d ( π 2 R ) Bud(\mathcal{D}^r)=Bud(\pi^R)=Bud(\pi_1^R)\cup Bud(\pi_2^R) B u d ( D r ) = B u d ( π R ) = B u d ( π 1 R ) ∪ B u d ( π 2 R ) .
By the induction hypothesis we have that V o c ( I i ) ⊆ ( V o c ( Γ i ) ∪ C i L ) ∩ ( V o c ( Δ i ) ∪ C i R ) Voc(I_i)\subseteq(Voc(\Gamma_i)\cup C_i^L)\cap(Voc(\Delta_i)\cup C_i^R) V oc ( I i ) ⊆ ( V oc ( Γ i ) ∪ C i L ) ∩ ( V oc ( Δ i ) ∪ C i R ) , V a r ( I i ) = { X B j ∣ B j ∈ B u d ( π i ) } Var(I_i)=\{X_{B_j}\mid B_j\in Bud(\pi_i)\} V a r ( I i ) = { X B j ∣ B j ∈ B u d ( π i )} , and B V ( I i ) = { X B j ∣ B j ∈ B u d ( π i ) ∧ C B j ∈ C o m p ( π i ) } BV(I_i)=\{X_{B_j}\mid B_j\in Bud(\pi_i)\land C_{B_j}\in Comp(\pi_i)\} B V ( I i ) = { X B j ∣ B j ∈ B u d ( π i ) ∧ C B j ∈ C o m p ( π i )} , for all i i i . We have two cases:
If v v v is not a companion node, then by the induction hypothesis on every I i I_i I i and observing the transformations for every rule ( r ) (r) ( r ) , used to construct the interpolant I I I in the Appendix B.1 and Figure 4 (b), one can observe that V o c ( I ) ⊆ ( V o c ( Γ ) ∪ C L ) ∩ ( V o c ( Δ ) ∪ C R ) Voc(I)\subseteq(Voc(\Gamma)\cup C^L)\cap(Voc(\Delta)\cup C^R) V oc ( I ) ⊆ ( V oc ( Γ ) ∪ C L ) ∩ ( V oc ( Δ ) ∪ C R ) . Moreover, none of these transformations include or bound fixed-point variables in I I I , therefore V a r ( I ) = ⋃ 1 ≤ i ≤ n V a r ( I i ) = ⋃ 1 ≤ i ≤ n { X B j ∣ B j ∈ B u d ( π i ) } = { X B j ∣ B j ∈ B u d ( T ( v ) ) } Var(I)=\bigcup_{1\leq i\leq n}Var(I_i)=\bigcup_{1\leq i\leq n}\{X_{B_j}\mid B_j\in Bud(\pi_i)\}=\{X_{B_j}\mid B_j\in Bud(T(v))\} V a r ( I ) = ⋃ 1 ≤ i ≤ n V a r ( I i ) = ⋃ 1 ≤ i ≤ n { X B j ∣ B j ∈ B u d ( π i )} = { X B j ∣ B j ∈ B u d ( T ( v ))} , and B V ( I ) = ⋃ 1 ≤ i ≤ n B V ( I i ) = ⋃ 1 ≤ i ≤ n { X B j ∣ B j ∈ B u d ( π i ) ∧ C B j ∈ C o m p ( π i ) } = { X B j ∣ B j ∈ B u d ( T ( v ) ) ∧ C B j ∈ C o m p ( T ( v ) ) } BV(I)=\bigcup_{1\leq i\leq n}BV(I_i)=\bigcup_{1\leq i\leq n}\{X_{B_j}\mid B_j\in Bud(\pi_i)\land C_{B_j}\in Comp(\pi_i)\}=\{X_{B_j}\mid B_j\in Bud(T(v))\land C_{B_j}\in Comp(T(v))\} B V ( I ) = ⋃ 1 ≤ i ≤ n B V ( I i ) = ⋃ 1 ≤ i ≤ n { X B j ∣ B j ∈ B u d ( π i ) ∧ C B j ∈ C o m p ( π i )} = { X B j ∣ B j ∈ B u d ( T ( v )) ∧ C B j ∈ C o m p ( T ( v ))} .
If v v v is a companion node, say C B k C_{B_k} C B k , then the interpolant is σ X B k . I \sigma X_{B_k}.I σ X B k . I , where I I I satisfies that V o c ( I ) ⊆ ( V o c ( Γ ) ∪ C L ) ∩ ( V o c ( Δ ) ∪ C R ) Voc(I)\subseteq(Voc(\Gamma)\cup C^L)\cap(Voc(\Delta)\cup C^R) V oc ( I ) ⊆ ( V oc ( Γ ) ∪ C L ) ∩ ( V oc ( Δ ) ∪ C R ) , V a r ( I ) = { X B j ∣ B j ∈ B u d ( T ( v ) ) } Var(I)=\{X_{B_j}\mid B_j\in Bud(T(v))\} V a r ( I ) = { X B j ∣ B j ∈ B u d ( T ( v ))} and B V ( I ) = { X B j ∣ B j ∈ B u d ( T ( v ) ) ∧ C B j ∈ C o m p ( T ( v ) ) } BV(I)=\{X_{B_j}\mid B_j\in Bud(T(v))\land C_{B_j}\in Comp(T(v))\} B V ( I ) = { X B j ∣ B j ∈ B u d ( T ( v )) ∧ C B j ∈ C o m p ( T ( v ))} , since we already showed it in the case where v v v is not a companion. It is then clear that V o c ( σ X B k . I ) = V o c ( I ) Voc(\sigma X_{B_k}.I)=Voc(I) V oc ( σ X B k . I ) = V oc ( I ) and V a r ( σ X B k . I ) = V a r ( I ) Var(\sigma X_{B_k}.I)=Var(I) V a r ( σ X B k . I ) = V a r ( I ) , therefore satisfying both constraints. Finally, since the variable X B k X_{B_k} X B k is bounded in σ X B k . I \sigma X_{B_k}.I σ X B k . I , we get that B V ( σ X B k . I ) = B V ( I ) ∪ { X B k } BV(\sigma X_{B_k}.I)=BV(I)\cup\{X_{B_k}\} B V ( σ X B k . I ) = B V ( I ) ∪ { X B k } , therefore satisfying the property, since v = C B k ∈ C o m p ( T ( v ) ) v=C_{B_k}\in Comp(T(v)) v = C B k ∈ C o m p ( T ( v )) .
Let R i l \mathcal{R}_i^l R i l and R i r \mathcal{R}_i^r R i r be the partial bud-companion relation for the derivation trees π i L \pi_i^L π i L and π i R \pi_i^R π i R respectively, for all i i i . By the induction hypothesis, we have that R i l \mathcal{R}_i^l R i l and R i r \mathcal{R}_i^r R i r are injective, and for every B j ∈ B u d ( π i ) B_j\in Bud(\pi_i) B j ∈ B u d ( π i ) such that C B j ∈ C o m p ( π i ) C_{B_j}\in Comp(\pi_i) C B j ∈ C o m p ( π i ) , then R i l \mathcal{R}_i^l R i l (resp. R i r \mathcal{R}_i^r R i r ) is defined over B j l B_j^l B j l (resp. B j r B_j^r B j r ). We know B j l B_j^l B j l and B j r B_j^r B j r exist, by induction hypothesis on (1). We have two cases:
Assume v v v is not a companion node. As explained for (1), for the unary rules { ⊗ , ⊥ , ⊕ i , ? d , ? w , ? c , ! p , ν , μ } \{\otimes,\bot,\oplus_i,?d,?w,?c,!p,\nu,\mu\} { ⊗ , ⊥ , ⊕ i , ? d , ? w , ? c , ! p , ν , μ } , B u d ( D l ) = B u d ( π 1 l ) Bud(D^l)=Bud(\pi^l_1) B u d ( D l ) = B u d ( π 1 l ) and B u d ( D r ) = B u d ( π 1 r ) Bud(D^r)=Bud(\pi^r_1) B u d ( D r ) = B u d ( π 1 r ) . Since v v v is not a companion, one can also deduce that C o m p ( D l ) = C o m p ( π 1 l ) Comp(D^l)=Comp(\pi^l_1) C o m p ( D l ) = C o m p ( π 1 l ) and C o m p ( D r ) = C o m p ( π 1 r ) Comp(D^r)=Comp(\pi^r_1) C o m p ( D r ) = C o m p ( π 1 r ) . In this case the relations R 1 l \mathcal{R}^l_1 R 1 l and R 1 r \mathcal{R}^r_1 R 1 r are not altered, therefore R l = R 1 l \mathcal{R}^l=\mathcal{R}^l_1 R l = R 1 l and R r = R 1 r \mathcal{R}^r=\mathcal{R}^r_1 R r = R 1 r , where both relations satisfy (3) by the induction hypothesis. For the binary rules { ⊗ , & , c u t } \{\otimes,\mathbin{\&},cut\} { ⊗ , & , c u t } , the argument is the same: one derives that B u d ( D l ) = B u d ( π 1 l ) ∪ B u d ( π 2 l ) Bud(D^l)=Bud(\pi^l_1)\cup Bud(\pi^l_2) B u d ( D l ) = B u d ( π 1 l ) ∪ B u d ( π 2 l ) , B u d ( D r ) = B u d ( π 1 r ) ∪ B u d ( π 2 r ) Bud(D^r)=Bud(\pi^r_1)\cup Bud(\pi^r_2) B u d ( D r ) = B u d ( π 1 r ) ∪ B u d ( π 2 r ) , C o m p ( D l ) = C o m p ( π 1 l ) ∪ C o m p ( π 2 l ) Comp(D^l)=Comp(\pi^l_1)\cup Comp(\pi^l_2) C o m p ( D l ) = C o m p ( π 1 l ) ∪ C o m p ( π 2 l ) , and C o m p ( D r ) = C o m p ( π 1 r ) ∪ C o m p ( π 2 r ) Comp(D^r)=Comp(\pi^r_1)\cup Comp(\pi^r_2) C o m p ( D r ) = C o m p ( π 1 r ) ∪ C o m p ( π 2 r ) . Thus, both R l = R 1 l ∪ R 2 l \mathcal{R}^l=\mathcal{R}^l_1\cup\mathcal{R}^l_2 R l = R 1 l ∪ R 2 l and R r = R 1 r ∪ R 2 r \mathcal{R}^r=\mathcal{R}^r_1\cup\mathcal{R}^r_2 R r = R 1 r ∪ R 2 r satisfy property (3).
In both cases, where (r) is either a unary or binary rule, we have defined R l \mathcal{R}^l R l and R r \mathcal{R}^r R r , both satisfying (3), by the previous proof given where v v v is not a companion. Now assume v v v is a companion node, say C B k C_{B_k} C B k . We will name C B k l C^l_{B_k} C B k l (resp. C B k r C^r_{B_k} C B k r ) the root of the derivation tree D l D^l D l (resp. D r D^r D r ), labelled with the sequent ⊢ Γ , σ X B k . I \vdash\Gamma,\sigma\mathsf{X}_{B_k}.I ⊢ Γ , σ X B k . I (resp. ⊢ Δ , σ ⊥ X B k . I ⊥ \vdash\Delta,\sigma^\bot\mathsf{X}_{B_k}.I^\bot ⊢ Δ , σ ⊥ X B k . I ⊥ ). Then, we update the relations R l \mathcal{R}^l R l and R r \mathcal{R}^r R r with R l = R l ∪ { ( B k l , C B k l ) } \mathcal{R}^l=\mathcal{R}^l\cup\{(B^l_k,C^l_{B_k})\} R l = R l ∪ {( B k l , C B k l )} and R r = R r ∪ { ( B k r , C B k r ) } \mathcal{R}^r=\mathcal{R}^r\cup\{(B^r_k,C^r_{B_k})\} R r = R r ∪ {( B k r , C B k r )} . This update maintains injectivity since B k B_k B k has no other companion node assigned, and ensures that B k l ∈ dom ( R l ) B^l_k\in\operatorname{dom}(\mathcal{R}^l) B k l ∈ dom ( R l ) (resp. B k r ∈ dom ( R r ) B^r_k\in\operatorname{dom}(\mathcal{R}^r) B k r ∈ dom ( R r ) ), therefore satisfying (3).
By the induction hypothesis, we have that for all C B j , C B l ∈ C o m p ( π i ′ ) C_{B_j},C_{B_l}\in Comp(\pi'_i) C B j , C B l ∈ C o m p ( π i ′ ) and B h ∈ B u d ( π i ′ ) B_h\in Bud(\pi'_i) B h ∈ B u d ( π i ′ ) , then:
a. C B j ⪯ π i ′ C B l C_{B_j}\preceq_{\pi'_i}C_{B_l} C B j ⪯ π i ′ C B l if and only if C j B l ⪯ π i l C l B l C^{B_l}_j\preceq_{\pi^l_i}C^{B_l}_l C j B l ⪯ π i l C l B l and C j B r ⪯ π i r C l B r C^{B_r}_j\preceq_{\pi^r_i}C^{B_r}_l C j B r ⪯ π i r C l B r
b. C B j ⪯ π i ′ B h C_{B_j}\preceq_{\pi'_i}B_h C B j ⪯ π i ′ B h if and only if C j B l ⪯ π i l B h l C^{B_l}_j\preceq_{\pi^l_i}B^l_h C j B l ⪯ π i l B h l and C j B r ⪯ π i r B h r C^{B_r}_j\preceq_{\pi^r_i}B^r_h C j B r ⪯ π i r B h r .
for every i i i . We can ensure that B h l , C j B l , C l B l B^l_h,C^{B_l}_j,C^{B_l}_l B h l , C j B l , C l B l (resp. B h r , C j B r , C l B r B^r_h,C^{B_r}_j,C^{B_r}_l B h r , C j B r , C l B r ) exist by the induction hypothesis on (1) and (3). Now let C B j , C B l ∈ C o m p ( T ( v ) ) C_{B_j},C_{B_l}\in Comp(T(v)) C B j , C B l ∈ C o m p ( T ( v )) and B h ∈ B u d ( T ( v ) ) B_h\in Bud(T(v)) B h ∈ B u d ( T ( v )) . We have two cases:
Assume v v v is not a companion node. Therefore, every path starting from a companion in T ( v ) T(v) T ( v ) must be strictly inside some derivation tree π i ′ \pi'_i π i ′ , for some i i i . In other words, C B j ⪯ T ( v ) C B l C_{B_j}\preceq_{T(v)}C_{B_l} C B j ⪯ T ( v ) C B l if and only if C B j ⪯ π i ′ C B l C_{B_j}\preceq_{\pi'_i}C_{B_l} C B j ⪯ π i ′ C B l , and C B j ⪯ T ( v ) B h C_{B_j}\preceq_{T(v)}B_h C B j ⪯ T ( v ) B h if and only if C B j ⪯ π i ′ B h C_{B_j}\preceq_{\pi'_i}B_h C B j ⪯ π i ′ B h , for some i i i . Then, by the induction hypothesis:
a. C B j ⪯ T ( v ) C B l C_{B_j}\preceq_{T(v)}C_{B_l} C B j ⪯ T ( v ) C B l if and only if C j B l ⪯ π i l C l B l C^{B_l}_j\preceq_{\pi^l_i}C^{B_l}_l C j B l ⪯ π i l C l B l and C j B r ⪯ π i r C l B r C^{B_r}_j\preceq_{\pi^r_i}C^{B_r}_l C j B r ⪯ π i r C l B r
b. C B j ⪯ T ( v ) B h C_{B_j}\preceq_{T(v)}B_h C B j ⪯ T ( v ) B h if and only if C j B l ⪯ π i l B h l C^{B_l}_j\preceq_{\pi^l_i}B^l_h C j B l ⪯ π i l B h l and C j B r ⪯ π i r B h r C^{B_r}_j\preceq_{\pi^r_i}B^r_h C j B r ⪯ π i r B h r .
Either if (r) is unary or binary, in both cases we have that the transformations in the Appendix B.1 do not introduce new companion nodes, and preserve intact the derivations trees π i l \pi^l_i π i l and π i r \pi^r_i π i r . Therefore:
a. C j B l ⪯ D l C l B l C^{B_l}_j\preceq_{D^l}C^{B_l}_l C j B l ⪯ D l C l B l (resp. C j B r ⪯ D r C l B r C^{B_r}_j\preceq_{D^r}C^{B_r}_l C j B r ⪯ D r C l B r ) if and only if C j B l ⪯ π i l C l B l C^{B_l}_j\preceq_{\pi^l_i}C^{B_l}_l C j B l ⪯ π i l C l B l (resp. C j B r ⪯ π i r C l B r C^{B_r}_j\preceq_{\pi^r_i}C^{B_r}_l C j B r ⪯ π i r C l B r ).
b. C j B l ⪯ D l B h l C^{B_l}_j\preceq_{D^l}B^l_h C j B l ⪯ D l B h l (resp. C j B r ⪯ D r B h r C^{B_r}_j\preceq_{D^r}B^r_h C j B r ⪯ D r B h r ) if and only if C j B l ⪯ π i l B h l C^{B_l}_j\preceq_{\pi^l_i}B^l_h C j B l ⪯ π i l B h l (resp. C j B r ⪯ π i r B h r C^{B_r}_j\preceq_{\pi^r_i}B^r_h C j B r ⪯ π i r B h r ).
Which implies:
a. C B j ⪯ T ( v ) C B l C_{B_j}\preceq_{T(v)}C_{B_l} C B j ⪯ T ( v ) C B l if and only if C j B l ⪯ D l C l B l C^{B_l}_j\preceq_{D^l}C^{B_l}_l C j B l ⪯ D l C l B l and C j B r ⪯ D r C l B r C^{B_r}_j\preceq_{D^r}C^{B_r}_l C j B r ⪯ D r C l B r
b. C B j ⪯ T ( v ) B h C_{B_j}\preceq_{T(v)}B_h C B j ⪯ T ( v ) B h if and only if C j B l ⪯ D l B h l C^{B_l}_j\preceq_{D^l}B^l_h C j B l ⪯ D l B h l and C j B r ⪯ D r B h r C^{B_r}_j\preceq_{D^r}B^r_h C j B r ⪯ D r B h r .
Assume v v v is a companion node, say C B k C_{B_k} C B k . In the previous case, we have shown that (4) is satisfied by all companion nodes inside some derivation tree π i ′ \pi'_i π i ′ . It remains to be shown that it is also satisfied by C B k C_{B_k} C B k , i.e.
a. C B k ⪯ T ( v ) C B l C_{B_k}\preceq_{T(v)}C_{B_l} C B k ⪯ T ( v ) C B l if and only if C k B l ⪯ D l C l B l C^{B_l}_k\preceq_{D^l}C^{B_l}_l C k B l ⪯ D l C l B l and C k B r ⪯ D r C l B r C^{B_r}_k\preceq_{D^r}C^{B_r}_l C k B r ⪯ D r C l B r
b. C B k ⪯ T ( v ) B h C_{B_k}\preceq_{T(v)}B_h C B k ⪯ T ( v ) B h if and only if C k B l ⪯ D l B h l C^{B_l}_k\preceq_{D^l}B^l_h C k B l ⪯ D l B h l and C k B r ⪯ D r B h r C^{B_r}_k\preceq_{D^r}B^r_h C k B r ⪯ D r B h r .
This is true since C B k C_{B_k} C B k is the root of T ( v ) T(v) T ( v ) , C k B l C^{B_l}_k C k B l (resp. C k B r C^{B_r}_k C k B r ) is the root of D l D^l D l (resp. D r D^r D r ), and there is always a path from the root to any vertex in a tree.
Let B ⊆ B u d ( D 0 ) B \subseteq Bud(D_0) B ⊆ B u d ( D 0 ) be the set of all bud nodes under B h ∈ D 0 B_h \in D_0 B h ∈ D 0 in the ◃ 0 \triangleleft_0 ◃ 0 relation, i.e. for all B i ∈ B B_i \in B B i ∈ B then B i ◃ 0 B h B_i \triangleleft_0 B_h B i ◃ 0 B h . We will implicitly use the properties (1) and (3) to refer to the copies of the bud and companion nodes inside every subtree π i \pi_i π i , with their respective injective relations R i l R_i^l R i l and R i r R_i^r R i r . Moreover, we name α \alpha α the unique path from v v v to B i B_i B i in D 0 D_0 D 0 , and if B i B_i B i is in π k ′ \pi_k' π k ′ with k ∈ { 1 , 2 } k \in\{1,2\} k ∈ { 1 , 2 } , then we name α k ′ \alpha_k' α k ′ the unique path from v k v_k v k to B i B_i B i in π k ′ \pi_k' π k ′ , where v k v_k v k is the root of π k ′ \pi_k' π k ′ . We then define τ B h , B i ∣ α k ′ = τ B h , B i k \tau_{B_h,B_i}|_{\alpha_k'} = \tau_{B_h,B_i}^k τ B h , B i ∣ α k ′ = τ B h , B i k .
a. Assume B i ∈ B u d ( T ( v ) ) B_i \in Bud(T(v)) B i ∈ B u d ( T ( v )) and B i ∈ B B_i \in B B i ∈ B . We refer to the traces τ B h , B i 1 \tau_{B_h,B_i}^{1} τ B h , B i 1 , τ B h , B i 1 \tau_{B_h,B_i}^{1} τ B h , B i 1 , τ B h , B i 2 \tau_{B_h,B_i}^{2} τ B h , B i 2 , τ B h , B i 2 \tau_{B_h,B_i}^{2} τ B h , B i 2 as the traces obtained by induction hypothesis over the subtrees π 1 1 \pi_1^1 π 1 1 , π 1 2 \pi_1^2 π 1 2 , π 2 1 \pi_2^1 π 2 1 , π 2 2 \pi_2^2 π 2 2 respectively, where v 1 ′ v_1' v 1 ′ , v 2 ′ v_2' v 2 ′ are the roots of the subtrees π 1 \pi_1 π 1 and π 2 \pi_2 π 2 respectively. We name α \alpha α the unique path from v v v to B i B_i B i in D 0 D_0 D 0 .
i. Assume C B i ∉ C o m p ( T ( v ) ) CB_i \notin Comp(T(v)) C B i ∈ / C o m p ( T ( v )) . We proceed by cases on the rule ( r ) (r) ( r ) :
Case (Cut): Without loss of generality we assume that the cut formulas are in the right split. Note that no trace through v v v follows the cut formulas C , C ⊥ C,C^\perp C , C ⊥ : they do not occur in the conclusion of the rule (Cut), and the ancestor relation of the cut relates only the context formulas. Traces following the cut formulas are internal to a premise subtree and are given by the induction hypothesis.
Let τ B h , B i ∣ α = ( G , G ) ∘ τ B h , B i k \tau_{B_h,B_i}|_\alpha= (G,G) \circ\tau_{B_h,B_i}^k τ B h , B i ∣ α = ( G , G ) ∘ τ B h , B i k and τ B h , B i ( v ) = G \tau_{B_h,B_i}(v) = G τ B h , B i ( v ) = G with G ∈ Γ k G \in\Gamma_k G ∈ Γ k , for k ∈ { 1 , 2 } k \in\{1,2\} k ∈ { 1 , 2 } . Then τ B h , B i l \tau_{B_h,B_i}^{l} τ B h , B i l exists and τ B h , B i l ( v 1 ′ ) = G \tau_{B_h,B_i}^{l}(v_1') = G τ B h , B i l ( v 1 ′ ) = G with G ∈ Γ k G \in\Gamma_k G ∈ Γ k . We take τ B h ′ , B i l = ( G , G , G ) ∘ τ B h , B i k \tau_{B_h',B_i}^{l} = (G,G,G) \circ\tau_{B_h,B_i}^k τ B h ′ , B i l = ( G , G , G ) ∘ τ B h , B i k .
Let τ B h , B i ∣ α = ( G , G ) ∘ τ B h , B i k \tau_{B_h,B_i}|_\alpha= (G,G) \circ\tau_{B_h,B_i}^k τ B h , B i ∣ α = ( G , G ) ∘ τ B h , B i k and τ B h , B i ( v ) = G \tau_{B_h,B_i}(v) = G τ B h , B i ( v ) = G with G ∈ Δ k G \in\Delta_k G ∈ Δ k , for k ∈ { 1 , 2 } k \in\{1,2\} k ∈ { 1 , 2 } . Then τ B h , B i r \tau_{B_h,B_i}^{r} τ B h , B i r exists and τ B h , B i r ( v 1 ′ ) = G \tau_{B_h,B_i}^{r}(v_1') = G τ B h , B i r ( v 1 ′ ) = G with G ∈ Δ k G \in\Delta_k G ∈ Δ k . We take τ B h ′ , B i r = ( G , G , G , G ) ∘ τ B h , B i k \tau_{B_h',B_i}^{r} = (G,G,G,G) \circ\tau_{B_h,B_i}^k τ B h ′ , B i r = ( G , G , G , G ) ∘ τ B h , B i k .
Case ( ⊗ ) (\otimes) ( ⊗ ) :
Let τ B h , B i ∣ α = ( G , G ) ∘ τ B h , B i k \tau_{B_h,B_i}|_\alpha= (G,G) \circ\tau_{B_h,B_i}^k τ B h , B i ∣ α = ( G , G ) ∘ τ B h , B i k and τ B h , B i ( v ) = G \tau_{B_h,B_i}(v) = G τ B h , B i ( v ) = G with G ∈ Γ G \in\Gamma G ∈ Γ , for k ∈ { 1 , 2 } k \in\{1,2\} k ∈ { 1 , 2 } . Then τ B h , B i 1 \tau_{B_h,B_i}^{1} τ B h , B i 1 exists and τ B h , B i 1 ( v 1 ′ ) = G \tau_{B_h,B_i}^{1}(v_1') = G τ B h , B i 1 ( v 1 ′ ) = G with G ∈ Γ G \in\Gamma G ∈ Γ . We take τ B h ′ , B i l = ( G , G ) ∘ τ B h ′ , B i l \tau_{B_h',B_i}^{l} = (G,G) \circ\tau_{B_h',B_i}^{l} τ B h ′ , B i l = ( G , G ) ∘ τ B h ′ , B i l .
Let τ B h , B i ∣ α = ( G , G ) ∘ τ B h , B i k \tau_{B_h,B_i}|_\alpha= (G,G) \circ\tau_{B_h,B_i}^k τ B h , B i ∣ α = ( G , G ) ∘ τ B h , B i k and τ B h , B i ( v ) = G \tau_{B_h,B_i}(v) = G τ B h , B i ( v ) = G with G ∈ Δ G \in\Delta G ∈ Δ . Then τ B h , B i 1 \tau_{B_h,B_i}^{1} τ B h , B i 1 exists and τ B h , B i 1 ( v 1 ′ ) = G \tau_{B_h,B_i}^{1}(v_1') = G τ B h , B i 1 ( v 1 ′ ) = G with G ∈ Δ G \in\Delta G ∈ Δ . We take τ B h ′ , B i r = ( G , G , G ) ∘ τ B h ′ , B i r \tau_{B_h',B_i}^{r} = (G,G,G) \circ\tau_{B_h',B_i}^{r} τ B h ′ , B i r = ( G , G , G ) ∘ τ B h ′ , B i r .
Let τ B h , B i ∣ α = ( A ⊗ B , A ) ∘ τ B h , B i 1 \tau_{B_h,B_i}|_\alpha= (A \otimes B,A) \circ\tau_{B_h,B_i}^{1} τ B h , B i ∣ α = ( A ⊗ B , A ) ∘ τ B h , B i 1 and τ B h , B i ( v ) = A ⊗ B \tau_{B_h,B_i}(v) = A \otimes B τ B h , B i ( v ) = A ⊗ B with ancestor A A A after the rule ( ⊗ ) (\otimes) ( ⊗ ) . Then τ B h , B i 1 \tau_{B_h,B_i}^{1} τ B h , B i 1 exists and τ B h , B i 1 ( v 1 ′ ) = A \tau_{B_h,B_i}^{1}(v_1') = A τ B h , B i 1 ( v 1 ′ ) = A . We take τ B h ′ , B i r = ( A ⊗ B , A ⊗ B , A ) ∘ τ B h ′ , B i r \tau_{B_h',B_i}^{r} = (A \otimes B,A \otimes B,A) \circ\tau_{B_h',B_i}^{r} τ B h ′ , B i r = ( A ⊗ B , A ⊗ B , A ) ∘ τ B h ′ , B i r .
Let τ B h , B i ∣ α = ( A ⊗ B , B ) ∘ τ B h , B i 2 \tau_{B_h,B_i}|_\alpha= (A \otimes B,B) \circ\tau_{B_h,B_i}^{2} τ B h , B i ∣ α = ( A ⊗ B , B ) ∘ τ B h , B i 2 and τ B h , B i ( v ) = A ⊗ B \tau_{B_h,B_i}(v) = A \otimes B τ B h , B i ( v ) = A ⊗ B with ancestor B B B after the rule ( ⊗ ) (\otimes) ( ⊗ ) . Then τ B h , B i 2 \tau_{B_h,B_i}^{2} τ B h , B i 2 exists and τ B h , B i 2 ( v 2 ′ ) = B \tau_{B_h,B_i}^{2}(v_2') = B τ B h , B i 2 ( v 2 ′ ) = B . We take τ B h ′ , B i r = ( A ⊗ B , A ⊗ B , B ) ∘ τ B h ′ , B i r \tau_{B_h',B_i}^{r} = (A \otimes B,A \otimes B,B) \circ\tau_{B_h',B_i}^{r} τ B h ′ , B i r = ( A ⊗ B , A ⊗ B , B ) ∘ τ B h ′ , B i r .
τ B h , B i B r , B r = ( A ⊗ B , A ⊗ B , A ⊗ B , B ) ∘ τ B h , B i 2 \tau_{B_h,B_i}^{B_r,B_r} = (A \otimes B,A \otimes B,A \otimes B,B) \circ\tau_{B_h,B_i}^{2} τ B h , B i B r , B r = ( A ⊗ B , A ⊗ B , A ⊗ B , B ) ∘ τ B h , B i 2
Case ( ⊥ ) (\perp) ( ⊥ ) :
Let τ B h , B i ∣ α = ( G , G ) ∘ τ B h , B i 1 \tau_{B_h,B_i}|_{\alpha} = (G,G) \circ\tau_{B_h,B_i}^{1} τ B h , B i ∣ α = ( G , G ) ∘ τ B h , B i 1 and τ B h , B i ( v ) = G \tau_{B_h,B_i}(v) = G τ B h , B i ( v ) = G with G ∈ Γ G \in\Gamma G ∈ Γ . Then τ B h , B i 1 \tau_{B_h,B_i}^{1} τ B h , B i 1 exists and τ B h , B i 1 ( v 1 ′ ) = G \tau_{B_h,B_i}^{1}(v_1') = G τ B h , B i 1 ( v 1 ′ ) = G with G ∈ Γ G \in\Gamma G ∈ Γ . We take τ B h , B i 1 = ( G , G , G ) ∘ τ B h , B i 1 \tau_{B_h,B_i}^{1} = (G,G,G) \circ\tau_{B_h,B_i}^{1} τ B h , B i 1 = ( G , G , G ) ∘ τ B h , B i 1 .
Let τ B h , B i ∣ α = ( G , G ) ∘ τ B h , B i 1 \tau_{B_h,B_i}|_{\alpha} = (G,G) \circ\tau_{B_h,B_i}^{1} τ B h , B i ∣ α = ( G , G ) ∘ τ B h , B i 1 and τ B h , B i ( v ) = G \tau_{B_h,B_i}(v) = G τ B h , B i ( v ) = G with G ∈ Δ G \in\Delta G ∈ Δ . Then τ B h , B i 1 \tau_{B_h,B_i}^{1} τ B h , B i 1 exists and τ B h , B i 1 ( v 1 ′ ) = G \tau_{B_h,B_i}^{1}(v_1') = G τ B h , B i 1 ( v 1 ′ ) = G with G ∈ Δ G \in\Delta G ∈ Δ . We take τ B h , B i 1 = ( G , G ) ∘ τ B h , B i 1 \tau_{B_h,B_i}^{1} = (G,G) \circ\tau_{B_h,B_i}^{1} τ B h , B i 1 = ( G , G ) ∘ τ B h , B i 1 .
Case ( & ) (\&) ( & ) :
Let τ B h , B i ∣ α = ( G , G ) ∘ τ B h , B i k \tau_{B_h,B_i}|_{\alpha} = (G,G) \circ\tau_{B_h,B_i}^{k} τ B h , B i ∣ α = ( G , G ) ∘ τ B h , B i k and τ B h , B i ( v ) = G \tau_{B_h,B_i}(v) = G τ B h , B i ( v ) = G with G ∈ Γ G \in\Gamma G ∈ Γ . Then τ B h , B i k \tau_{B_h,B_i}^{k} τ B h , B i k exists and τ B h , B i k ( v k ′ ) = G \tau_{B_h,B_i}^{k}(v_k') = G τ B h , B i k ( v k ′ ) = G with G ∈ Γ G \in\Gamma G ∈ Γ . We take τ B h , B i k = ( G , G , G , G ) ∘ τ B h , B i k \tau_{B_h,B_i}^{k} = (G,G,G,G) \circ\tau_{B_h,B_i}^{k} τ B h , B i k = ( G , G , G , G ) ∘ τ B h , B i k .
Let τ B h , B i ∣ α = ( G , G ) ∘ τ B h , B i k \tau_{B_h,B_i}|_{\alpha} = (G,G) \circ\tau_{B_h,B_i}^{k} τ B h , B i ∣ α = ( G , G ) ∘ τ B h , B i k and τ B h , B i ( v ) = G \tau_{B_h,B_i}(v) = G τ B h , B i ( v ) = G with G ∈ Δ G \in\Delta G ∈ Δ . Then τ B h , B i k \tau_{B_h,B_i}^{k} τ B h , B i k exists and τ B h , B i k ( v k ′ ) = G \tau_{B_h,B_i}^{k}(v_k') = G τ B h , B i k ( v k ′ ) = G with G ∈ Δ G \in\Delta G ∈ Δ . We take τ B h , B i k = ( G , G , G ) ∘ τ B h , B i k \tau_{B_h,B_i}^{k} = (G,G,G) \circ\tau_{B_h,B_i}^{k} τ B h , B i k = ( G , G , G ) ∘ τ B h , B i k .
Let τ B h , B i ∣ α = ( F 1 & F 2 , F k ) ∘ τ B h , B i k \tau_{B_h,B_i}|_{\alpha} = (F_1 \& F_2,F_k) \circ\tau_{B_h,B_i}^{k} τ B h , B i ∣ α = ( F 1 & F 2 , F k ) ∘ τ B h , B i k and τ B h , B i ( v ) = F 1 & F 2 \tau_{B_h,B_i}(v) = F_1 \& F_2 τ B h , B i ( v ) = F 1 & F 2 with ancestor F k F_k F k after the rule ( & ) (\&) ( & ) . Then τ B h , B i k \tau_{B_h,B_i}^{k} τ B h , B i k exists and τ B h , B i k ( v k ′ ) = F k \tau_{B_h,B_i}^{k}(v_k') = F_k τ B h , B i k ( v k ′ ) = F k . We take τ B h , B i k = ( F 1 & F 2 , F 1 & F 2 , F k , F k ) ∘ τ B h , B i k \tau_{B_h,B_i}^{k} = (F_1 \& F_2,F_1 \& F_2,F_k,F_k) \circ\tau_{B_h,B_i}^{k} τ B h , B i k = ( F 1 & F 2 , F 1 & F 2 , F k , F k ) ∘ τ B h , B i k .
Case ( ? d ) (?d) ( ? d ) :
Let τ B h , B i ∣ α = ( G , G ) ∘ τ B h , B i 1 \tau_{B_h,B_i}|_{\alpha} = (G,G) \circ\tau_{B_h,B_i}^{1} τ B h , B i ∣ α = ( G , G ) ∘ τ B h , B i 1 and τ B h , B i ( v ) = G \tau_{B_h,B_i}(v) = G τ B h , B i ( v ) = G with G ∈ Γ G \in\Gamma G ∈ Γ . Then τ B h , B i 1 \tau_{B_h,B_i}^{1} τ B h , B i 1 exists and τ B h , B i 1 ( v 1 ′ ) = G \tau_{B_h,B_i}^{1}(v_1') = G τ B h , B i 1 ( v 1 ′ ) = G with G ∈ Γ G \in\Gamma G ∈ Γ . We take τ B h , B i 1 = ( G , G , G ) ∘ τ B h , B i 1 \tau_{B_h,B_i}^{1} = (G,G,G) \circ\tau_{B_h,B_i}^{1} τ B h , B i 1 = ( G , G , G ) ∘ τ B h , B i 1 .
Let τ B h , B i ∣ α = ( G , G ) ∘ τ B h , B i 1 \tau_{B_h,B_i}|_{\alpha} = (G,G) \circ\tau_{B_h,B_i}^{1} τ B h , B i ∣ α = ( G , G ) ∘ τ B h , B i 1 and τ B h , B i ( v ) = G \tau_{B_h,B_i}(v) = G τ B h , B i ( v ) = G with G ∈ Δ G \in\Delta G ∈ Δ . Then τ B h , B i 1 \tau_{B_h,B_i}^{1} τ B h , B i 1 exists and τ B h , B i 1 ( v 1 ′ ) = G \tau_{B_h,B_i}^{1}(v_1') = G τ B h , B i 1 ( v 1 ′ ) = G with G ∈ Δ G \in\Delta G ∈ Δ . We take τ B h , B i 1 = ( G , G ) ∘ τ B h , B i 1 \tau_{B_h,B_i}^{1} = (G,G) \circ\tau_{B_h,B_i}^{1} τ B h , B i 1 = ( G , G ) ∘ τ B h , B i 1 .
Let τ B h , B i ∣ α = ( ? , F ) ∘ τ B h , B i 1 \tau_{B_h,B_i}|_{\alpha} = (?,F) \circ\tau_{B_h,B_i}^{1} τ B h , B i ∣ α = ( ? , F ) ∘ τ B h , B i 1 and τ B h , B i ( v ) = ? F \tau_{B_h,B_i}(v) = ?F τ B h , B i ( v ) = ? F with ancestor F F F after the rule ( ? d ) (?d) ( ? d ) . Then τ B h , B i 1 \tau_{B_h,B_i}^{1} τ B h , B i 1 exists and τ B h , B i 1 ( v 1 ′ ) = F \tau_{B_h,B_i}^{1}(v_1') = F τ B h , B i 1 ( v 1 ′ ) = F . We take τ B h , B i 1 = ( ? , F , F , F ) ∘ τ B h , B i 1 \tau_{B_h,B_i}^{1} = (?,F,F,F) \circ\tau_{B_h,B_i}^{1} τ B h , B i 1 = ( ? , F , F , F ) ∘ τ B h , B i 1 .
Case ( ! p ) (!p) ( ! p ) :
Let τ B h , B i ∣ α = ( G , G ) ∘ τ B h , B i 1 \tau_{B_h,B_i}|_{\alpha} = (G,G) \circ\tau_{B_h,B_i}^{1} τ B h , B i ∣ α = ( G , G ) ∘ τ B h , B i 1 and τ B h , B i ( v ) = G \tau_{B_h,B_i}(v) = G τ B h , B i ( v ) = G with G ∈ ? Γ G \in?\Gamma G ∈ ? Γ . Then τ B h , B i 1 \tau_{B_h,B_i}^{1} τ B h , B i 1 exists and τ B h , B i 1 ( v 1 ′ ) = G \tau_{B_h,B_i}^{1}(v_1') = G τ B h , B i 1 ( v 1 ′ ) = G with G ∈ ? Γ G \in?\Gamma G ∈ ? Γ . We take τ B h , B i 1 = ( G , G , G , G ) ∘ τ B h , B i 1 \tau_{B_h,B_i}^{1} = (G,G,G,G) \circ\tau_{B_h,B_i}^{1} τ B h , B i 1 = ( G , G , G , G ) ∘ τ B h , B i 1 .
Let τ B h , B i ∣ α = ( G , G ) ∘ τ B h , B i 1 \tau_{B_h,B_i}|_{\alpha} = (G,G) \circ\tau_{B_h,B_i}^{1} τ B h , B i ∣ α = ( G , G ) ∘ τ B h , B i 1 and τ B h , B i ( v ) = G \tau_{B_h,B_i}(v) = G τ B h , B i ( v ) = G with G ∈ ? Δ G \in?\Delta G ∈ ? Δ . Then τ B h , B i 1 \tau_{B_h,B_i}^{1} τ B h , B i 1 exists and τ B h , B i 1 ( v 1 ′ ) = G \tau_{B_h,B_i}^{1}(v_1') = G τ B h , B i 1 ( v 1 ′ ) = G with G ∈ ? Δ G \in?\Delta G ∈ ? Δ . We take τ B h , B i 1 = ( G , G , G ) ∘ τ B h , B i 1 \tau_{B_h,B_i}^{1} = (G,G,G) \circ\tau_{B_h,B_i}^{1} τ B h , B i 1 = ( G , G , G ) ∘ τ B h , B i 1 .
Let τ B h , B i ∣ α = ( ! F , F ) ∘ τ B h , B i 1 \tau_{B_h,B_i}|_{\alpha} = (!F,F) \circ\tau_{B_h,B_i}^{1} τ B h , B i ∣ α = ( ! F , F ) ∘ τ B h , B i 1 and τ B h , B i ( v ) = ! F \tau_{B_h,B_i}(v) = !F τ B h , B i ( v ) = ! F with ancestor F F F after the rule ( ! p ) (!p) ( ! p ) . Then τ B h , B i 1 \tau_{B_h,B_i}^{1} τ B h , B i 1 exists and τ B h , B i 1 ( v 1 ′ ) = F \tau_{B_h,B_i}^{1}(v_1') = F τ B h , B i 1 ( v 1 ′ ) = F . We take τ B h , B i 1 = ( ! F , ! F , F , F ) ∘ τ B h , B i 1 \tau_{B_h,B_i}^{1} = (!F,!F,F,F) \circ\tau_{B_h,B_i}^{1} τ B h , B i 1 = ( ! F , ! F , F , F ) ∘ τ B h , B i 1 .
Case ( ? w ) (?w) ( ? w ) :
Let τ B h , B i ∣ α = ( G , G ) ∘ τ B h , B i 1 \tau_{B_h,B_i}|_{\alpha} = (G,G) \circ\tau_{B_h,B_i}^{1} τ B h , B i ∣ α = ( G , G ) ∘ τ B h , B i 1 and τ B h , B i ( v ) = G \tau_{B_h,B_i}(v) = G τ B h , B i ( v ) = G with G ∈ Γ G \in\Gamma G ∈ Γ . Then τ B h , B i 1 \tau_{B_h,B_i}^{1} τ B h , B i 1 exists and τ B h , B i 1 ( v 1 ′ ) = G \tau_{B_h,B_i}^{1}(v_1') = G τ B h , B i 1 ( v 1 ′ ) = G with G ∈ Γ G \in\Gamma G ∈ Γ . We take τ B h , B i 1 = ( G , G , G ) ∘ τ B h , B i 1 \tau_{B_h,B_i}^{1} = (G,G,G) \circ\tau_{B_h,B_i}^{1} τ B h , B i 1 = ( G , G , G ) ∘ τ B h , B i 1 .
Let τ B h , B i ∣ α = ( G , G ) ∘ τ B h , B i 1 \tau_{B_h,B_i}|_{\alpha} = (G,G) \circ\tau_{B_h,B_i}^{1} τ B h , B i ∣ α = ( G , G ) ∘ τ B h , B i 1 and τ B h , B i ( v ) = G \tau_{B_h,B_i}(v) = G τ B h , B i ( v ) = G with G ∈ Δ G \in\Delta G ∈ Δ . Then τ B h , B i 1 \tau_{B_h,B_i}^{1} τ B h , B i 1 exists and τ B h , B i 1 ( v 1 ′ ) = G \tau_{B_h,B_i}^{1}(v_1') = G τ B h , B i 1 ( v 1 ′ ) = G with G ∈ Δ G \in\Delta G ∈ Δ . We take τ B h , B i 1 = ( G , G ) ∘ τ B h , B i 1 \tau_{B_h,B_i}^{1} = (G,G) \circ\tau_{B_h,B_i}^{1} τ B h , B i 1 = ( G , G ) ∘ τ B h , B i 1 .
Case (?c): The principal formula ? F ?F ? F has two ancestor occurrences ? F , ? F ?F, ?F ? F , ? F in the premise of the rule (?c). Without loss of generality, we assume the trace follows the occurrence of ? F ?F ? F in the first position of the premise sequent, i.e. the leftmost one; the case of the rightmost occurrence is identical.
Let τ B h , B i ∣ α = ( G , G ) ∘ τ B h , B i 1 \tau_{B_h,B_i}|_{\alpha} = (G,G) \circ\tau^1_{B_h,B_i} τ B h , B i ∣ α = ( G , G ) ∘ τ B h , B i 1 and τ B h , B i ( v ) = G \tau_{B_h,B_i}(v) = G τ B h , B i ( v ) = G with G ∈ Γ G \in\Gamma G ∈ Γ . Then τ B h ′ , B i ′ 1 \tau^{1}_{B_h',B_i'} τ B h ′ , B i ′ 1 exists and τ B h ′ , B i ′ 1 ( v 1 ′ ) = G \tau^{1}_{B_h',B_i'}(v'_1) = G τ B h ′ , B i ′ 1 ( v 1 ′ ) = G with G ∈ Γ G \in\Gamma G ∈ Γ . We take τ B h ′ , B i ′ B l , B l = ( G , G , G ) ∘ τ B h ′ , B i ′ 1 \tau^{B_l,B_l}_{B_h',B_i'} = (G,G,G) \circ\tau^{1}_{B_h',B_i'} τ B h ′ , B i ′ B l , B l = ( G , G , G ) ∘ τ B h ′ , B i ′ 1 .
Let τ B h , B i ∣ α = ( G , G ) ∘ τ B h , B i 1 \tau_{B_h,B_i}|_{\alpha} = (G,G) \circ\tau^1_{B_h,B_i} τ B h , B i ∣ α = ( G , G ) ∘ τ B h , B i 1 and τ B h , B i ( v ) = G \tau_{B_h,B_i}(v) = G τ B h , B i ( v ) = G with G ∈ Δ G \in\Delta G ∈ Δ . Then τ B r ′ , B i ′ 1 \tau^{1}_{B_r',B_i'} τ B r ′ , B i ′ 1 exists and τ B r ′ , B i ′ 1 ( v 1 ′ ) = G \tau^{1}_{B_r',B_i'}(v'_1) = G τ B r ′ , B i ′ 1 ( v 1 ′ ) = G with G ∈ Δ G \in\Delta G ∈ Δ . We take τ B r ′ , B r ′ B r , B r = ( G , G ) ∘ τ B r ′ , B r ′ 1 \tau^{B_r,B_r}_{B_r',B_r'} = (G,G) \circ\tau^{1}_{B_r',B_r'} τ B r ′ , B r ′ B r , B r = ( G , G ) ∘ τ B r ′ , B r ′ 1 .
Let τ B h , B i ∣ α = ( ? F , ? F ) ∘ τ B h , B i 1 \tau_{B_h,B_i}|_{\alpha} = (?F,?F) \circ\tau^1_{B_h,B_i} τ B h , B i ∣ α = ( ? F , ? F ) ∘ τ B h , B i 1 and τ B h , B i ( v ) = ? F \tau_{B_h,B_i}(v) = ?F τ B h , B i ( v ) = ? F with ancestor the leftmost occurrence of ? F ?F ? F after the rule (?c). Then τ B l ′ , B l ′ 1 \tau^{1}_{B_l',B_l'} τ B l ′ , B l ′ 1 exists and τ B l ′ , B l ′ 1 ( v 1 ′ ) = ? F \tau^{1}_{B_l',B_l'}(v'_1) = ?F τ B l ′ , B l ′ 1 ( v 1 ′ ) = ? F . We take τ B h ′ , B i ′ B l , B l = ( ? F , ? F , ? F ) ∘ τ B h ′ , B i ′ 1 \tau^{B_l,B_l}_{B_h',B_i'} = (?F,?F,?F) \circ\tau^{1}_{B_h',B_i'} τ B h ′ , B i ′ B l , B l = ( ? F , ? F , ? F ) ∘ τ B h ′ , B i ′ 1 .
Case ( σ ) (\sigma) ( σ ) , with σ ∈ { ( ν ) , ( μ ) } \sigma\in\{(\nu),(\mu)\} σ ∈ {( ν ) , ( μ )} :
Let τ B h , B i ∣ α = ( G , G ) ∘ τ B h , B i 1 \tau_{B_h,B_i}|_{\alpha} = (G,G) \circ\tau^1_{B_h,B_i} τ B h , B i ∣ α = ( G , G ) ∘ τ B h , B i 1 and τ B h , B i ( v ) = G \tau_{B_h,B_i}(v) = G τ B h , B i ( v ) = G with G ∈ Γ G \in\Gamma G ∈ Γ . Then τ B l ′ , B l ′ 1 \tau^{1}_{B_l',B_l'} τ B l ′ , B l ′ 1 exists and τ B l ′ , B l ′ 1 ( v 1 ′ ) = G \tau^{1}_{B_l',B_l'}(v'_1) = G τ B l ′ , B l ′ 1 ( v 1 ′ ) = G with G ∈ Γ G \in\Gamma G ∈ Γ . We take τ B l ′ , B l ′ B l , B l = ( G , G , G ) ∘ τ B l ′ , B l ′ 1 \tau^{B_l,B_l}_{B_l',B_l'} = (G,G,G) \circ\tau^{1}_{B_l',B_l'} τ B l ′ , B l ′ B l , B l = ( G , G , G ) ∘ τ B l ′ , B l ′ 1 .
Let τ B h , B i ∣ α = ( G , G ) ∘ τ B h , B i 1 \tau_{B_h,B_i}|_{\alpha} = (G,G) \circ\tau^1_{B_h,B_i} τ B h , B i ∣ α = ( G , G ) ∘ τ B h , B i 1 and τ B h , B i ( v ) = G \tau_{B_h,B_i}(v) = G τ B h , B i ( v ) = G with G ∈ Δ G \in\Delta G ∈ Δ . Then τ B r ′ , B r ′ 1 \tau^{1}_{B_r',B_r'} τ B r ′ , B r ′ 1 exists and τ B r ′ , B r ′ 1 ( v 1 ′ ) = G \tau^{1}_{B_r',B_r'}(v'_1) = G τ B r ′ , B r ′ 1 ( v 1 ′ ) = G with G ∈ Δ G \in\Delta G ∈ Δ . We take τ B r ′ , B r ′ B r , B r = ( G , G ) ∘ τ B r ′ , B r ′ 1 \tau^{B_r,B_r}_{B_r',B_r'} = (G,G) \circ\tau^{1}_{B_r',B_r'} τ B r ′ , B r ′ B r , B r = ( G , G ) ∘ τ B r ′ , B r ′ 1 .
Let τ B h , B i ∣ α = ( σ X . F , F [ σ X . F / X ] ) ∘ τ B h , B i 1 \tau_{B_h,B_i}|_{\alpha} = (\sigma X.F,F[\sigma X.F/X]) \circ\tau^1_{B_h,B_i} τ B h , B i ∣ α = ( σ X . F , F [ σ X . F / X ]) ∘ τ B h , B i 1 and τ B h , B i ( v ) = σ X . F \tau_{B_h,B_i}(v) = \sigma X.F τ B h , B i ( v ) = σ X . F with ancestor F [ σ X . F / X ] F[\sigma X.F/X] F [ σ X . F / X ] after the rule ( σ ) (\sigma) ( σ ) . Then τ B l ′ , B l ′ 1 \tau^{1}_{B_l',B_l'} τ B l ′ , B l ′ 1 exists and τ B l ′ , B l ′ 1 ( v 1 ′ ) = F [ σ X . F / X ] \tau^{1}_{B_l',B_l'}(v'_1) = F[\sigma X.F/X] τ B l ′ , B l ′ 1 ( v 1 ′ ) = F [ σ X . F / X ] . We take τ B l ′ , B l ′ B l , B l = ( σ X . F , σ X . F , F [ σ X . F / X ] ) ∘ τ B l ′ , B l ′ 1 \tau^{B_l,B_l}_{B_l',B_l'} = (\sigma X.F,\sigma X.F,F[\sigma X.F/X]) \circ\tau^{1}_{B_l',B_l'} τ B l ′ , B l ′ B l , B l = ( σ X . F , σ X . F , F [ σ X . F / X ]) ∘ τ B l ′ , B l ′ 1 .
In case v v v is a companion node and τ B l ′ , B l ′ B l , B l ( v 1 ′ ) = G \tau^{B_l,B_l}_{B_l',B_l'}(v'_1) = G τ B l ′ , B l ′ B l , B l ( v 1 ′ ) = G (resp. τ B r ′ , B r ′ B r , B r ( v 1 ′ ) = G \tau^{B_r,B_r}_{B_r',B_r'}(v'_1) = G τ B r ′ , B r ′ B r , B r ( v 1 ′ ) = G ), then we obtain the new trace by concatenating a new instance of G G G to the trace, due to the newly introduced rule σ \sigma σ (resp. σ ⊥ \sigma^\bot σ ⊥ ), where σ ∈ { ν , μ } \sigma\in\{\nu,\mu\} σ ∈ { ν , μ } . In all cases, τ B h , B i ( v ) = τ B l ′ , B l ′ B l , B l ( v 1 ′ ) \tau_{B_h,B_i}(v) = \tau^{B_l,B_l}_{B_l',B_l'}(v'_1) τ B h , B i ( v ) = τ B l ′ , B l ′ B l , B l ( v 1 ′ ) (resp. τ B h , B i ( v ) = τ B r ′ , B r ′ B r , B r ( v 1 ′ ) ) \tau_{B_h,B_i}(v) = \tau^{B_r,B_r}_{B_r',B_r'}(v'_1)) τ B h , B i ( v ) = τ B r ′ , B r ′ B r , B r ( v 1 ′ )) and τ B h , B i ( B i ) = τ B l ′ , B l ′ B l , B l ( B l ′ ) \tau_{B_h,B_i}(B_i) = \tau^{B_l,B_l}_{B_l',B_l'}(B_l') τ B h , B i ( B i ) = τ B l ′ , B l ′ B l , B l ( B l ′ ) (resp. τ B h , B i ( B i ) = τ B r ′ , B r ′ B r , B r ( B r ′ ) \tau_{B_h,B_i}(B_i) = \tau^{B_r,B_r}_{B_r',B_r'}(B_r') τ B h , B i ( B i ) = τ B r ′ , B r ′ B r , B r ( B r ′ ) ).
ii. Assume C B i ∈ Comp ( T ( v ) ) C_{B_i} \in\operatorname{Comp}(T(v)) C B i ∈ Comp ( T ( v )) . There are two cases: If v ≠ C B i v \ne C_{B_i} v = C B i , it means that C B i C_{B_i} C B i appears in one of the subtrees π 1 L \pi^L_1 π 1 L or π 2 R \pi^R_2 π 2 R . If C B i C_{B_i} C B i appears in π k \pi^k π k with k ∈ { 1 , 2 } k \in\{1,2\} k ∈ { 1 , 2 } , by induction hypothesis there exists either a trace τ B l ′ , B l ′ B l , B l \tau^{B_l,B_l}_{B_l',B_l'} τ B l ′ , B l ′ B l , B l over the unique path from C B l ′ C_{B_l'} C B l ′ to B l ′ B_l' B l ′ in π k L \pi^L_k π k L or a trace τ B r ′ , B r ′ B r , B r \tau^{B_r,B_r}_{B_r',B_r'} τ B r ′ , B r ′ B r , B r over the unique path from C B r ′ C_{B_r'} C B r ′ to B r ′ B_r' B r ′ in π k R \pi^R_k π k R , satisfying that τ B l ′ , B l ′ B l , B l ( B l ′ ) = τ B l ′ , B l ′ B l , B l ( C B l ′ ) \tau^{B_l,B_l}_{B_l',B_l'}(B_l') = \tau^{B_l,B_l}_{B_l',B_l'}(C_{B_l'}) τ B l ′ , B l ′ B l , B l ( B l ′ ) = τ B l ′ , B l ′ B l , B l ( C B l ′ ) and τ B r ′ , B r ′ B r , B r ( B r ′ ) = τ B r ′ , B r ′ B r , B r ( C B r ′ ) \tau^{B_r,B_r}_{B_r',B_r'}(B_r') = \tau^{B_r,B_r}_{B_r',B_r'}(C_{B_r'}) τ B r ′ , B r ′ B r , B r ( B r ′ ) = τ B r ′ , B r ′ B r , B r ( C B r ′ ) . In all rules (r) (could be unary or binary), we have that τ B l ′ , B l ′ B l , B l \tau^{B_l,B_l}_{B_l',B_l'} τ B l ′ , B l ′ B l , B l will appear in D l D^l D l and τ B r ′ , B r ′ B r , B r \tau^{B_r,B_r}_{B_r',B_r'} τ B r ′ , B r ′ B r , B r will appear in D r D^r D r , since π k L \pi^L_k π k L and π k R \pi^R_k π k R are not modified in the cut-introduction step that defines π L \pi^L π L and π R \pi^R π R .
We have proven in the previous point (5.a.i), that there exists a trace τ B l ′ , B l ′ B l , B l \tau^{B_l,B_l}_{B_l',B_l'} τ B l ′ , B l ′ B l , B l over the unique path v ′ v' v ′ to B l ′ B_l' B l ′ in D D D , or a trace τ B r ′ , B r ′ B r , B r \tau^{B_r,B_r}_{B_r',B_r'} τ B r ′ , B r ′ B r , B r over the unique path v ′ v' v ′ to B r ′ B_r' B r ′ in D D D , such that τ B h , B i ( v ) = τ B l ′ , B l ′ B l , B l ( v ′ ) \tau_{B_h,B_i}(v) = \tau^{B_l,B_l}_{B_l',B_l'}(v') τ B h , B i ( v ) = τ B l ′ , B l ′ B l , B l ( v ′ ) and τ B h , B i ( B i ) = τ B l ′ , B l ′ B l , B l ( B l ′ ) \tau_{B_h,B_i}(B_i) = \tau^{B_l,B_l}_{B_l',B_l'}(B_l') τ B h , B i ( B i ) = τ B l ′ , B l ′ B l , B l ( B l ′ ) , or τ B h , B i ( v ) = τ B r ′ , B r ′ B r , B r ( v ′ ) \tau_{B_h,B_i}(v) = \tau^{B_r,B_r}_{B_r',B_r'}(v') τ B h , B i ( v ) = τ B r ′ , B r ′ B r , B r ( v ′ ) and τ B h , B i ( B i ) = τ B r ′ , B r ′ B r , B r ( B r ′ ) \tau_{B_h,B_i}(B_i) = \tau^{B_r,B_r}_{B_r',B_r'}(B_r') τ B h , B i ( B i ) = τ B r ′ , B r ′ B r , B r ( B r ′ ) . If v = C B i v = C_{B_i} v = C B i then D D D is obtained by cutting the D l = π L [ σ X B k . I / X B k ] ⊢ Γ , I [ σ X B k . I / X B k ] D^l = \frac{\pi^L[\sigma X_{B_k}.I/X_{B_k}]}{\vdash\Gamma,I[\sigma X_{B_k}.I/X_{B_k}]} D l = ⊢ Γ , I [ σ X B k . I / X B k ] π L [ σ X B k . I / X B k ] and D r = π R [ σ ⊥ X B k . I ⊥ / X B k ] ⊢ Δ , I ⊥ [ σ ⊥ X B k . I ⊥ / X B k ] D^r = \frac{\pi^R[\sigma^\bot X_{B_k}.I^\bot/X_{B_k}]}{\vdash\Delta,I^\bot[\sigma^\bot X_{B_k}.I^\bot/X_{B_k}]} D r = ⊢ Δ , I ⊥ [ σ ⊥ X B k . I ⊥ / X B k ] π R [ σ ⊥ X B k . I ⊥ / X B k ] . In this case, s ( C B l ′ ) = ( ⊢ Γ , σ X B k . I ) s(C_{B_l'}) = (\vdash\Gamma,\sigma X_{B_k}.I) s ( C B l ′ ) = ( ⊢ Γ , σ X B k . I ) and s ( C B r ′ ) = ( ⊢ Δ , σ ⊥ X B k . I ⊥ ) s(C_{B_r'}) = (\vdash\Delta,\sigma^\bot X_{B_k}.I^\bot) s ( C B r ′ ) = ( ⊢ Δ , σ ⊥ X B k . I ⊥ ) . It is clear that the rules (Cut), ( σ ) (\sigma) ( σ ) , ( σ ⊥ ) (\sigma^\bot) ( σ ⊥ ) preserve the context, which implies that τ B l ′ , B l ′ B l , B l ( v ′ ) = τ B l ′ , B l ′ B l , B l ( C B l ′ ) \tau^{B_l,B_l}_{B_l',B_l'}(v') = \tau^{B_l,B_l}_{B_l',B_l'}(C_{B_l'}) τ B l ′ , B l ′ B l , B l ( v ′ ) = τ B l ′ , B l ′ B l , B l ( C B l ′ ) and τ B r ′ , B r ′ B r , B r ( v ′ ) = τ B r ′ , B r ′ B r , B r ( C B r ′ ) \tau^{B_r,B_r}_{B_r',B_r'}(v') = \tau^{B_r,B_r}_{B_r',B_r'}(C_{B_r'}) τ B r ′ , B r ′ B r , B r ( v ′ ) = τ B r ′ , B r ′ B r , B r ( C B r ′ ) . Moreover, by the property 1 of the ordered trace manifold for P 0 P_0 P 0 , we know that τ B h , B i ( B i ) = τ B h , B i ( C B i ) \tau_{B_h,B_i}(B_i) = \tau_{B_h,B_i}(C_{B_i}) τ B h , B i ( B i ) = τ B h , B i ( C B i ) . Thus:
τ B l ′ , B l ′ B l , B l ( B l ′ ) = τ B h , B i ( B i ) = τ B h , B i ( C B i ) = τ B l ′ , B l ′ B l , B l ( v ′ ) = τ B l ′ , B l ′ B l , B l ( C B l ′ ) \tau^{B_l,B_l}_{B_l',B_l'}(B_l') = \tau_{B_h,B_i}(B_i) = \tau_{B_h,B_i}(C_{B_i}) = \tau^{B_l,B_l}_{B_l',B_l'}(v') = \tau^{B_l,B_l}_{B_l',B_l'}(C_{B_l'}) τ B l ′ , B l ′ B l , B l ( B l ′ ) = τ B h , B i ( B i ) = τ B h , B i ( C B i ) = τ B l ′ , B l ′ B l , B l ( v ′ ) = τ B l ′ , B l ′ B l , B l ( C B l ′ )
τ B r ′ , B r ′ B r , B r ( B r ′ ) = τ B h , B i ( B i ) = τ B h , B i ( C B i ) = τ B r ′ , B r ′ B r , B r ( v ′ ) = τ B r ′ , B r ′ B r , B r ( C B r ′ ) \tau^{B_r,B_r}_{B_r',B_r'}(B_r') = \tau_{B_h,B_i}(B_i) = \tau_{B_h,B_i}(C_{B_i}) = \tau^{B_r,B_r}_{B_r',B_r'}(v') = \tau^{B_r,B_r}_{B_r',B_r'}(C_{B_r'}) τ B r ′ , B r ′ B r , B r ( B r ′ ) = τ B h , B i ( B i ) = τ B h , B i ( C B i ) = τ B r ′ , B r ′ B r , B r ( v ′ ) = τ B r ′ , B r ′ B r , B r ( C B r ′ )
Therefore τ B h , B i B l \tau_{B_h,B_i}^{B_l} τ B h , B i B l or τ B h , B i B r \tau_{B_h,B_i}^{B_r} τ B h , B i B r satisfy (5.a.ii).
b. We implicitly use the property (5.a), to refer to the traces τ B h , B i B l \tau_{B_h,B_i}^{B_l} τ B h , B i B l or τ B h , B i B r \tau_{B_h,B_i}^{B_r} τ B h , B i B r , for all B i ∈ B B_i \in\mathcal{B} B i ∈ B . First we proof the following auxiliary propositions:
Proposition 48. Let τ \tau τ be any trace in the ordered trace manifold for P 0 P_0 P 0 . Then either for every vertex u ∈ V 0 u \in V_0 u ∈ V 0 with a split ( Γ , Δ ) (\Gamma,\Delta) ( Γ , Δ ) , then τ ( u ) ∈ Γ \tau(u) \in\Gamma τ ( u ) ∈ Γ or for every vertex u ∈ V 0 u \in V_0 u ∈ V 0 with a split ( Γ , Δ ) (\Gamma,\Delta) ( Γ , Δ ) , then τ ( u ) ∈ Δ \tau(u) \in\Delta τ ( u ) ∈ Δ .
Proof. Done by induction on the derivation tree D 0 \mathcal{D}_0 D 0 . It holds since every split and trace are propagated using the ancestor relation.
Proposition 49. For any set { B i 1 , … , B i n } ⊆ B u d ( T ( v ) ) \{B_{i_1}, \ldots, B_{i_n}\} \subseteq Bud(T(v)) { B i 1 , … , B i n } ⊆ B u d ( T ( v )) such that { B i 1 , … , B i m } ⊆ B \{B_{i_1}, \ldots, B_{i_m}\} \subseteq\mathcal{B} { B i 1 , … , B i m } ⊆ B . Then either τ B h , B i 1 B l , … , τ B h , B i m B l \tau_{B_h,B_{i_1}}^{B_l}, \ldots, \tau_{B_h,B_{i_m}}^{B_l} τ B h , B i 1 B l , … , τ B h , B i m B l exist, or τ B h , B i 1 B r , … , τ B h , B i m B r \tau_{B_h,B_{i_1}}^{B_r}, \ldots, \tau_{B_h,B_{i_m}}^{B_r} τ B h , B i 1 B r , … , τ B h , B i m B r exist (all traces generated from B \mathcal{B} B are on the same side of the interpolated pre-proof).
Proof. Let B i , B j ∈ B B_i, B_j \in\mathcal{B} B i , B j ∈ B (i.e. B i , B j ◃ 0 B k B_i, B_j \mathrel{\triangleleft_0} B_k B i , B j ◃ 0 B k ). Note that B \mathcal{B} B is a set of weakly ≤ P 0 \leq_{P_0} ≤ P 0 -connected, so either B i ≤ P 0 B j B_i \leq_{P_0} B_j B i ≤ P 0 B j , B j ≤ P 0 B i B_j \leq_{P_0} B_i B j ≤ P 0 B i or B i , B j B_i, B_j B i , B j are not directly related:
If B i ≤ P 0 B j B_i \leq_{P_0} B_j B i ≤ P 0 B j then by property 2 of the ordered trace manifold τ B h , B i ( C B i ) = τ B h , B j ( C B i ) \tau_{B_h,B_i}(C_{B_i}) = \tau_{B_h,B_j}(C_{B_i}) τ B h , B i ( C B i ) = τ B h , B j ( C B i ) . Therefore, if s ( C B i ) s(C_{B_i}) s ( C B i ) is split into ( Γ ′ , Δ ′ ) (\Gamma', \Delta') ( Γ ′ , Δ ′ ) , then either τ B h , B i ( C B i ) , τ B h , B j ( C B i ) ∈ Γ ′ \tau_{B_h,B_i}(C_{B_i}), \tau_{B_h,B_j}(C_{B_i}) \in\Gamma' τ B h , B i ( C B i ) , τ B h , B j ( C B i ) ∈ Γ ′ or τ B h , B i ( C B i ) , τ B h , B j ( C B i ) ∈ Δ ′ \tau_{B_h,B_i}(C_{B_i}), \tau_{B_h,B_j}(C_{B_i}) \in\Delta' τ B h , B i ( C B i ) , τ B h , B j ( C B i ) ∈ Δ ′ . Therefore, by Proposition 48 for all nodes u ∈ V 0 u \in V_0 u ∈ V 0 , τ B h , B i ( u ) \tau_{B_h,B_i}(u) τ B h , B i ( u ) and τ B h , B j ( u ) \tau_{B_h,B_j}(u) τ B h , B j ( u ) are on the same side of the split of u u u . This implies that for the root vertex v v v , then either τ B h , B i ( v ) , τ B h , B j ( v ) ∈ Γ \tau_{B_h,B_i}(v), \tau_{B_h,B_j}(v) \in\Gamma τ B h , B i ( v ) , τ B h , B j ( v ) ∈ Γ or τ B h , B i ( v ) , τ B h , B j ( v ) ∈ Δ \tau_{B_h,B_i}(v), \tau_{B_h,B_j}(v) \in\Delta τ B h , B i ( v ) , τ B h , B j ( v ) ∈ Δ . Finally, based on every case of the construction made in (5.a.i), one can derive that τ B h , B i ( v ) , τ B h , B j ( v ) ∈ Γ \tau_{B_h,B_i}(v), \tau_{B_h,B_j}(v) \in\Gamma τ B h , B i ( v ) , τ B h , B j ( v ) ∈ Γ if and only if τ B h , B i B l , τ B h , B j B l \tau_{B_h,B_i}^{B_l}, \tau_{B_h,B_j}^{B_l} τ B h , B i B l , τ B h , B j B l exist, or respectively τ B h , B i ( v ) , τ B h , B j ( v ) ∈ Δ \tau_{B_h,B_i}(v), \tau_{B_h,B_j}(v) \in\Delta τ B h , B i ( v ) , τ B h , B j ( v ) ∈ Δ if and only if τ B h , B i B r , τ B h , B j B r \tau_{B_h,B_i}^{B_r}, \tau_{B_h,B_j}^{B_r} τ B h , B i B r , τ B h , B j B r exist.
The case B j ≤ P 0 B i B_j \leq_{P_0} B_i B j ≤ P 0 B i is symmetrical to the previous one.
Assume B i B_i B i and B j B_j B j are not related. Since B \mathcal{B} B is weakly ≤ P 0 \leq_{P_0} ≤ P 0 -connected, then there is a chain of bud nodes B l 1 ′ , … , B l m ′ B'_{l_1}, \ldots, B'_{l_m} B l 1 ′ , … , B l m ′ in B \mathcal{B} B , where B l 1 ′ = B i B'_{l_1} = B_i B l 1 ′ = B i and B l m ′ = B j B'_{l_m} = B_j B l m ′ = B j satisfying that for all B l i ′ B'_{l_i} B l i ′ and B l i + 1 ′ B'_{l_{i+1}} B l i + 1 ′ then B l i ′ ≤ P 0 B l i + 1 ′ B'_{l_i} \leq_{P_0} B'_{l_{i+1}} B l i ′ ≤ P 0 B l i + 1 ′ or B l i + 1 ′ ≤ P 0 B l i ′ B'_{l_{i+1}} \leq_{P_0} B'_{l_i} B l i + 1 ′ ≤ P 0 B l i ′ , such that B l i ′ , B l i + 1 ′ ◃ 0 B h B'_{l_i}, B'_{l_{i+1}} \mathrel{\triangleleft_0} B_h B l i ′ , B l i + 1 ′ ◃ 0 B h . Therefore, by property 2 of the ordered trace manifold for P 0 P_0 P 0 , either τ B h , B l i ′ ( C B l i ′ ) = τ B h , B l i + 1 ′ ( C B l i ′ ) \tau_{B_h,B'_{l_i}}(C_{B'_{l_i}}) = \tau_{B_h,B'_{l_{i+1}}}(C_{B'_{l_i}}) τ B h , B l i ′ ( C B l i ′ ) = τ B h , B l i + 1 ′ ( C B l i ′ ) or τ B h , B l i ′ ( C B l i + 1 ′ ) = τ B h , B l i + 1 ′ ( C B l i + 1 ′ ) \tau_{B_h,B'_{l_i}}(C_{B'_{l_{i+1}}}) = \tau_{B_h,B'_{l_{i+1}}}(C_{B'_{l_{i+1}}}) τ B h , B l i ′ ( C B l i + 1 ′ ) = τ B h , B l i + 1 ′ ( C B l i + 1 ′ ) . In both cases, by Proposition 48 one can deduce that for all nodes u ∈ V 0 u \in V_0 u ∈ V 0 , τ B h , B l i ′ ( u ) \tau_{B_h,B'_{l_i}}(u) τ B h , B l i ′ ( u ) and τ B h , B l i + 1 ′ ( u ) \tau_{B_h,B'_{l_{i+1}}}(u) τ B h , B l i + 1 ′ ( u ) are on the same side of the split of u u u . Therefore, by transitivity on the chain B l 1 ′ , … , B l m ′ ∈ B B'_{l_1}, \ldots, B'_{l_m} \in\mathcal{B} B l 1 ′ , … , B l m ′ ∈ B we derive that for all nodes u ∈ V 0 u \in V_0 u ∈ V 0 , τ B h , B i ( u ) \tau_{B_h,B_i}(u) τ B h , B i ( u ) and τ B h , B j ( u ) \tau_{B_h,B_j}(u) τ B h , B j ( u ) are on the same side of the split of $u. We conclude by using the same argument used for the case B i ≤ P 0 B j B_i \leq_{P_0} B_j B i ≤ P 0 B j .
Now assume B i , B j ∈ B B_i, B_j \in\mathcal{B} B i , B j ∈ B and B i , B j ∈ B u d ( T ( v ) ) B_i, B_j \in Bud(T(v)) B i , B j ∈ B u d ( T ( v )) , where B j ≤ P 0 B i B_j \leq_{P_0} B_i B j ≤ P 0 B i and C B j ∈ C o m p ( T ( v ) ) C_{B_j} \in Comp(T(v)) C B j ∈ C o m p ( T ( v )) . Note that B j ≤ P 0 B i B_j \leq_{P_0} B_i B j ≤ P 0 B i implies by Proposition 49 that the traces can be the clauses (5.a) are both on the same side, i.e. either τ B k , B j B l , τ B k , B i B l \tau_{B_k,B_j}^{B_l}, \tau_{B_k,B_i}^{B_l} τ B k , B j B l , τ B k , B i B l exist, or τ B k , B j B r , τ B k , B i B r \tau_{B_k,B_j}^{B_r}, \tau_{B_k,B_i}^{B_r} τ B k , B j B r , τ B k , B i B r exist. We will assume that τ B k , B j B l , τ B k , B i B l \tau_{B_k,B_j}^{B_l}, \tau_{B_k,B_i}^{B_l} τ B k , B j B l , τ B k , B i B l exist. The case for the right side is symmetrical. For k ∈ { 1 , 2 } k \in\{1,2\} k ∈ { 1 , 2 } , we will refer to the root of π k \pi_k π k as v k ′ v'_k v k ′ , and v k L v_k^L v k L (resp. v k R v_k^R v k R ) to the root of π k L \pi_k^L π k L (resp. π k R \pi_k^R π k R ). ii. Assume C B i ∉ C o m p ( T ( v ) ) CB_i \notin Comp(T(v)) C B i ∈ / C o m p ( T ( v )) . We have two cases: if v ≠ C B j v \ne CB_j v = C B j then C B j CB_j C B j is in one of the subtrees π k ′ \pi'_k π k ′ . By the induction hypothesis, C B j L C^{L}_{B_j} C B j L (resp. C B j R C^{R}_{B_j} C B j R ) is in the unique path from v k ′ L v^{\prime L}_k v k ′ L to B i L B^L_i B i L (resp. B i R B^R_i B i R ) in π k \pi_k π k , such that τ R B k , B j L ( C B j L ) = τ R B k , B i L ( C B j R ) \tau^{R^L_{B_k,B_j}}(C^{L}_{B_j}) = \tau^{R^L_{B_k,B_i}}(C^{R}_{B_j}) τ R B k , B j L ( C B j L ) = τ R B k , B i L ( C B j R ) . This implies that the unique path α \alpha α from the root v k L v^L_k v k L (resp. v k R v^R_k v k R ) of π k L \pi^L_k π k L (resp. π k R \pi^R_k π k R ) to B i L B^L_i B i L (resp. B i R B^R_i B i R ) in π k L \pi^L_k π k L (resp. π k R \pi^R_k π k R ) includes the companion C B j B i L C^{B^L_i}_{B_j} C B j B i L (resp. C B j B i R C^{B^R_i}_{B_j} C B j B i R ). Furthermore, as shown before for all rule ( r ) (r) ( r ) , the cut-introduction preserve π k L \pi^L_k π k L (resp. π k R \pi^R_k π k R ). Therefore, the path from v ′ v' v ′ to v k L v^L_k v k L (resp. v k R v^R_k v k R ) composed with α \alpha α which ends in B i L B^L_i B i L (resp. B i R B^R_i B i R ), satisfies the property.
If v = C B j v = CB_j v = C B j , then we have a path from v ′ v' v ′ to C B j B i L C^{B^L_i}_{B_j} C B j B i L (resp. C B j B i R C^{B^R_i}_{B_j} C B j B i R ), and since C B j B i L C^{B^L_i}_{B_j} C B j B i L (resp. C B j B i R C^{B^R_i}_{B_j} C B j B i R ) is the root of D L D^L D L (resp. D R D^R D R ), it must be the case that there is a path from C B j B i L C^{B^L_i}_{B_j} C B j B i L (resp. C B j B i R C^{B^R_i}_{B_j} C B j B i R ) to B i L B^L_i B i L (resp. B i R B^R_i B i R ). Moreover, by property (5.a) we know that τ B h , B j ( C B j ) = τ B h B i L , B j L ( v ′ ) \tau_{B_h,B_j}(CB_j) = \tau^{B^L_i,B^L_j}_{B_h}(v') τ B h , B j ( C B j ) = τ B h B i L , B j L ( v ′ ) and τ B h , B i ( C B j ) = τ B h B i R , B j R ( v ′ ) \tau_{B_h,B_i}(CB_j) = \tau^{B^R_i,B^R_j}_{B_h}(v') τ B h , B i ( C B j ) = τ B h B i R , B j R ( v ′ ) . Since the (Cut) rule preserve the contexts, then τ B h , B j ( C B j ) = τ B h B i L , B j L ( C B j B i L ) \tau_{B_h,B_j}(CB_j) = \tau^{B^L_i,B^L_j}_{B_h}(C^{B^L_i}_{B_j}) τ B h , B j ( C B j ) = τ B h B i L , B j L ( C B j B i L ) and τ B h , B i ( C B j ) = τ B h B i R , B j R ( C B j B i R ) \tau_{B_h,B_i}(CB_j) = \tau^{B^R_i,B^R_j}_{B_h}(C^{B^R_i}_{B_j}) τ B h , B i ( C B j ) = τ B h B i R , B j R ( C B j B i R ) . Finally, by property (2) of the ordered trace manifold for P 0 P_0 P 0 , then τ B h , B j ( C B j ) = τ B h , B i ( C B j ) \tau_{B_h,B_j}(CB_j) = \tau_{B_h,B_i}(CB_j) τ B h , B j ( C B j ) = τ B h , B i ( C B j ) , which implies τ B h B i L , B j L ( C B j B i L ) = τ B h B i R , B j R ( C B j B i R ) \tau^{B^L_i,B^L_j}_{B_h}(C^{B^L_i}_{B_j}) = \tau^{B^R_i,B^R_j}_{B_h}(C^{B^R_i}_{B_j}) τ B h B i L , B j L ( C B j B i L ) = τ B h B i R , B j R ( C B j B i R ) .
ii. Assume C B i ∈ C o m p ( T ( v ) ) CB_i \in Comp(T(v)) C B i ∈ C o m p ( T ( v )) . We have two cases: if v ≠ C B i v \ne CB_i v = C B i then C B i CB_i C B i and C B j CB_j C B j are in one of the subtrees π k ′ \pi'_k π k ′ . By the induction hypothesis, C B i B j L C^{B^L_j}_{B_i} C B i B j L (resp. C B i B j R C^{B^R_j}_{B_i} C B i B j R ) is in the cycle C C B i B j L C_{C^{B^L_j}_{B_i}} C C B i B j L (resp. C C B i B j R C_{C^{B^R_j}_{B_i}} C C B i B j R ) in P π k L P^L_{\pi_k} P π k L (resp. P π k R P^R_{\pi_k} P π k R ), which is the left (resp. right) partial pre-proof obtained after the descending phase over π k ′ \pi'_k π k ′ , such that τ B k B j L , B i L ( C B i B j L ) = τ B k B j R , B i R ( C B i B j R ) \tau^{B^L_j,B^L_i}_{B_k}(C^{B^L_j}_{B_i}) = \tau^{B^R_j,B^R_i}_{B_k}(C^{B^R_j}_{B_i}) τ B k B j L , B i L ( C B i B j L ) = τ B k B j R , B i R ( C B i B j R ) . Note that the derivation tree for P π k L P^L_{\pi_k} P π k L (resp. P π k R P^R_{\pi_k} P π k R ) is π k L \pi^L_k π k L (resp. π k R \pi^R_k π k R ). As shown before, all the cut-introductions preserve π k L \pi^L_k π k L (resp. π k R \pi^R_k π k R ), and includes it in D L D^L D L (resp. D R D^R D R ). Thus, the property also holds for P π k L P^L_{\pi_k} P π k L (resp. P π k R P^R_{\pi_k} P π k R ).
If v = C B i v = CB_i v = C B i then C B j CB_j C B j is in one of the subtrees π k ′ \pi'_k π k ′ , since C B i ≠ C B j CB_i \ne CB_j C B i = C B j because R 0 R_0 R 0 is injective. By the induction hypothesis, C B i B j L C^{B^L_j}_{B_i} C B i B j L (resp. C B i B j R C^{B^R_j}_{B_i} C B i B j R ) is in the unique path from v k L v^L_k v k L to B i L B^L_i B i L (resp. B i R B^R_i B i R ) in π k \pi_k π k , such that τ B i B k L , B j L ( C B i B j L ) = τ B i B k R , B j R ( C B i B j R ) \tau^{B^L_k,B^L_j}_{B_i}(C^{B^L_j}_{B_i}) = \tau^{B^R_k,B^R_j}_{B_i}(C^{B^R_j}_{B_i}) τ B i B k L , B j L ( C B i B j L ) = τ B i B k R , B j R ( C B i B j R ) . This implies that the unique path α \alpha α from the root v k L v^L_k v k L (resp. v k R v^R_k v k R ) of π k L \pi^L_k π k L (resp. π k R \pi^R_k π k R ) to B i L B^L_i B i L (resp. B i R B^R_i B i R ) in π k L \pi^L_k π k L (resp. π k R \pi^R_k π k R ) includes the companion C B j B i L C^{B^L_i}_{B_j} C B j B i L (resp. C B j B i R C^{B^R_i}_{B_j} C B j B i R ). Furthermore, as shown before for all rule ( r ) (r) ( r ) , the cut-introductions preserve π k L \pi^L_k π k L (resp. π k R \pi^R_k π k R ). Note that in this case C B j B i L C^{B^L_i}_{B_j} C B j B i L (resp. C B j B i R C^{B^R_i}_{B_j} C B j B i R ) is the root of D L D^L D L (resp. D R D^R D R ). Therefore, the path from C B j B i L C^{B^L_i}_{B_j} C B j B i L to v k L v^L_k v k L (resp. v k R v^R_k v k R ) composed with α \alpha α which ends in B i L B^L_i B i L (resp. B i R B^R_i B i R ), which represent the cycle C C B j B i L C_{C^{B^L_i}_{B_j}} C C B j B i L (resp. C C B j B i R C_{C^{B^R_i}_{B_j}} C C B j B i R ), and satisfies the property.
c. Let B i ∈ B u d ( T ( v ) ) B_i \in Bud(T(v)) B i ∈ B u d ( T ( v )) and B i ∈ B B_i \in\mathcal{B} B i ∈ B . We directly refer to the traces τ B h B i L , B i L \tau^{B^L_i,B^L_i}_{B_h} τ B h B i L , B i L or τ B h B i R , B i R \tau^{B^R_i,B^R_i}_{B_h} τ B h B i R , B i R , constructed in the proof of the property (5.a). As before, we refer to the traces τ B h , B i L 1 \tau^1_{B_h,B^L_i} τ B h , B i L 1 , τ B h , B i R 1 \tau^1_{B_h,B^R_i} τ B h , B i R 1 , τ B h , B i L 2 \tau^2_{B_h,B^L_i} τ B h , B i L 2 , τ B h , B i R 2 \tau^2_{B_h,B^R_i} τ B h , B i R 2 as the traces obtained by induction hypothesis on (5.a) over the subtrees π 1 L \pi^L_1 π 1 L , π 1 R \pi^R_1 π 1 R , π 2 L \pi^L_2 π 2 L , π 2 R \pi^R_2 π 2 R respectively. Moreover, we name α \alpha α the unique path from v v v to B i B_i B i in D 0 D_0 D 0 , and if B i B_i B i is in π k ′ \pi'_k π k ′ with k ∈ { 1 , 2 } k \in\{1,2\} k ∈ { 1 , 2 } , then we name α k ′ \alpha'_k α k ′ the unique path from v k v_k v k to B i B_i B i in π k ′ \pi'_k π k ′ , where v k v_k v k is the root of π k ′ \pi'_k π k ′ . We then define τ B h , B i ∣ α k ′ = τ B h , B i k \tau_{B_h,B_i} \mid\alpha'_k = \tau^k_{B_h,B_i} τ B h , B i ∣ α k ′ = τ B h , B i k .
i. Let C B i ∉ C o m p ( T ( v ) ) CB_i \notin Comp(T(v)) C B i ∈ / C o m p ( T ( v )) and assume v v v is not a companion. By induction hypothesis, we know that F o r m ( τ B h , B i ) = F o r m ( τ B h L , B i L ) Form(\tau_{B_h,B_i}) = Form(\tau^{B^L_h,B^L_i}) F or m ( τ B h , B i ) = F or m ( τ B h L , B i L ) (resp. F o r m ( τ B h , B i ) = F o r m ( τ B h R , B i R ) Form(\tau_{B_h,B_i}) = Form(\tau^{B^R_h,B^R_i}) F or m ( τ B h , B i ) = F or m ( τ B h R , B i R ) ), where τ B h L , B i L k \tau^k_{B^L_h,B^L_i} τ B h L , B i L k (resp. τ B h R , B i R k \tau^k_{B^R_h,B^R_i} τ B h R , B i R k ) has an F F F -progress point if τ B h , B i k \tau^k_{B_h,B_i} τ B h , B i k has an F F F -progress point. For every rule ( r ) (r) ( r ) , we can use this induction hypothesis in every case of the proof for (5.a.i) to prove that F o r m ( τ B h , B i ∣ α ) = F o r m ( τ B h L , B i L ) Form(\tau_{B_h,B_i}\mid\alpha) = Form(\tau^{B^L_h,B^L_i}) F or m ( τ B h , B i ∣ α ) = F or m ( τ B h L , B i L ) or F o r m ( τ B h , B i ∣ α ) = F o r m ( τ B h R , B i R ) Form(\tau_{B_h,B_i}\mid\alpha) = Form(\tau^{B^R_h,B^R_i}) F or m ( τ B h , B i ∣ α ) = F or m ( τ B h R , B i R ) holds, depending on whether τ B h L , B i L \tau^{B^L_h,B^L_i} τ B h L , B i L or τ B h R , B i R \tau^{B^R_h,B^R_i} τ B h R , B i R exists. If τ B h , B i ∣ α \tau_{B_h,B_i}\mid\alpha τ B h , B i ∣ α has a progress point, there are two cases:
If the progress point is inside τ B h , B i k \tau^{k}_{B_h,B_i} τ B h , B i k , then τ B h ′ , B i ′ k \tau^{k}_{B'_h,B'_i} τ B h ′ , B i ′ k (resp. τ B r ′ , B i ′ k \tau^{k}_{B'_r,B'_i} τ B r ′ , B i ′ k ) also has a progress point since it contains the trace τ B h ′ , B i ′ k \tau^{k}_{B'_h,B'_i} τ B h ′ , B i ′ k (resp. τ B r ′ , B i ′ k \tau^{k}_{B'_r,B'_i} τ B r ′ , B i ′ k ) for all rule ( r ) (r) ( r ) .
If the progress point is in v v v , i.e. ( r ) = ν (r)=\nu ( r ) = ν and τ B h , B i ( v ) \tau_{B_h,B_i}(v) τ B h , B i ( v ) is the principal formula F = ν X . F ′ F=\nu X.F' F = ν X . F ′ , by observation on the case σ ∈ { ν , μ } \sigma\in\{\nu,\mu\} σ ∈ { ν , μ } in the proof for (5.a.i), then this progress point is also included in τ B h ′ , B i ′ \tau_{B'_h,B'_i} τ B h ′ , B i ′ (resp. τ B r ′ , B i ′ \tau_{B'_r,B'_i} τ B r ′ , B i ′ ).
In case v v v is a companion, the property still is preserved since the rules ( C u t ) , ( ν ) , ( μ ) (\mathrm{Cut}),(\nu),(\mu) ( Cut ) , ( ν ) , ( μ ) preserve the context, therefore Form ( τ B h ′ , B i ′ ) \operatorname{Form}(\tau_{B'_h,B'_i}) Form ( τ B h ′ , B i ′ ) (resp. Form ( τ B r ′ , B i ′ ) \operatorname{Form}(\tau_{B'_r,B'_i}) Form ( τ B r ′ , B i ′ ) ) remain the same.
ii. Let C B i ∈ Comp ( T ( v ) ) C_{B_i}\in\operatorname{Comp}(T(v)) C B i ∈ Comp ( T ( v )) then there are two cases: if v ≠ C B i v\ne C_{B_i} v = C B i , then C B i C_{B_i} C B i is in one of the subtrees π k ′ \pi'_k π k ′ . This case follows directly from the induction hypothesis.
If v = C B i v=C_{B_i} v = C B i , we have already proven in the previous item that Form ( τ B h , B i ∣ α ) = Form ( τ B h ′ , B i ′ ) \operatorname{Form}(\tau_{B_h,B_i|_\alpha})=\operatorname{Form}(\tau_{B'_h,B'_i}) Form ( τ B h , B i ∣ α ) = Form ( τ B h ′ , B i ′ ) (resp. Form ( τ B h , B i ∣ α ) = Form ( τ B r ′ , B i ′ ) \operatorname{Form}(\tau_{B_h,B_i|_\alpha})=\operatorname{Form}(\tau_{B'_r,B'_i}) Form ( τ B h , B i ∣ α ) = Form ( τ B r ′ , B i ′ ) ), and that F F F -progress points are preserved. However, note that since v = C B i v=C_{B_i} v = C B i , then τ B h , B i ∣ α = τ B h , B i \tau_{B_h,B_i|_\alpha}=\tau_{B_h,B_i} τ B h , B i ∣ α = τ B h , B i . Therefore, the property is satisfied.
With this we conclude the proof of Proposition 47. ◀
Given these invariants described in Proposition 47, it is possible to deduce the final structure of the pre-proofs P l P^l P l and P r P^r P r , which is stated in the Lemma 51. Before defining the lemma, we include the following proposition which generates maximal elements of weakly ≤ P \le_P ≤ P -connected set of bud nodes, based on tree-compatibility.
Proposition 50. Let P = ( D , R ) P=(D,R) P = ( D , R ) be a μ L L ω \mu\mathrm{LL}_{\omega} μ LL ω pre-proof with injective R R R , let ◃ \triangleleft ◃ be a tree-compatible induction order for P P P , and let B ⊆ Bud ( D ) \mathcal{B}\subseteq\operatorname{Bud}(D) B ⊆ Bud ( D ) be a weakly ≤ P \le_P ≤ P -connected set of bud nodes. Let B i ∈ B B_i\in\mathcal{B} B i ∈ B be such that the companion C B i C_{B_i} C B i is closest to the root among the companions in B \mathcal{B} B , i.e. for all B j ∈ B B_j\in\mathcal{B} B j ∈ B , then C B j C_{B_j} C B j does not appear in the path from the root of D D D to C B i C_{B_i} C B i . Then B i B_i B i is a ◃ \triangleleft ◃ maximum for B \mathcal{B} B .
Proof. We define the sets L 0 = { B i } \mathcal{L}_0=\{B_i\} L 0 = { B i } and L n + 1 = L n ∪ { B ∈ B ∣ B ≤ P B ′ or B ′ ≤ P B , for some B ′ ∈ L n } \mathcal{L}_{n+1}=\mathcal{L}_n\cup\{B\in\mathcal{B}\mid B\le_P B'\text{ or }B'\le_P B,\text{ for some }B'\in\mathcal{L}_n\} L n + 1 = L n ∪ { B ∈ B ∣ B ≤ P B ′ or B ′ ≤ P B , for some B ′ ∈ L n } . We prove by induction on n n n that for all B ∈ L n B\in\mathcal{L}_n B ∈ L n , then B ◃ B i B\triangleleft B_i B ◃ B i , and moreover either C B C_B C B appears on the basic cycle C B i C_{B_i} C B i , or B ≤ P B ′ B\le_P B' B ≤ P B ′ for some B ′ ∈ L n − 1 B'\in\mathcal{L}_{n-1} B ′ ∈ L n − 1 . If n = 0 n=0 n = 0 , then B = B i ◃ B i B=B_i\triangleleft B_i B = B i ◃ B i and C B i C_{B_i} C B i appears on C B i C_{B_i} C B i . If n > 0 n>0 n > 0 , take B ∈ L n ∖ L n − 1 B\in\mathcal{L}_n\setminus\mathcal{L}_{n-1} B ∈ L n ∖ L n − 1 , related to some B ′ ∈ L n − 1 B'\in\mathcal{L}_{n-1} B ′ ∈ L n − 1 , where B ′ ◃ B i B'\triangleleft B_i B ′ ◃ B i by the induction hypothesis. We have two cases:
If B ≤ P B ′ B\le_P B' B ≤ P B ′ , then by tree-compatibility B ◃ B ′ B\triangleleft B' B ◃ B ′ , and by transitivity B ◃ B i B\triangleleft B_i B ◃ B i . In this case, since B ′ ∈ L n − 1 B'\in\mathcal{L}_{n-1} B ′ ∈ L n − 1 the second part of the hypothesis is satisfied.
If B ′ ≤ P B B'\le_P B B ′ ≤ P B , by multiple applications of the second part of the induction hypothesis, we can form a chain B ′ ≤ P B 1 ≤ P B 2 ⋯ ≤ P B m B'\le_P B_1\le_P B_2\cdots\le_P B_m B ′ ≤ P B 1 ≤ P B 2 ⋯ ≤ P B m such that B m ≤ P B i B_m\le_P B_i B m ≤ P B i , and each element in the chain is from a deeper level. Then, B m − 1 ≤ P B m B_{m-1}\le_P B_m B m − 1 ≤ P B m imply that C B m − 1 C_{B_{m-1}} C B m − 1 appears in the path from C B m C_{B_m} C B m to B m B_m B m , and B m ≤ P B i B_m\le_P B_i B m ≤ P B i imply that C B m C_{B_m} C B m appears in the path from C B i C_{B_i} C B i to B i B_i B i . Therefore B m − 1 ≤ P B i B_{m-1}\le_P B_i B m − 1 ≤ P B i . We can repeat this argument, until reaching B ′ ≤ P B i B'\le_P B_i B ′ ≤ P B i . Again, this implies that C B ′ C_{B'} C B ′ is in the path from C B i C_{B_i} C B i to B i B_i B i . Moreover, since B ′ ≤ P B B'\le_P B B ′ ≤ P B then C B ′ C_{B'} C B ′ appears in the path from C B C_B C B to B B B . However, note that C B i C_{B_i} C B i is the closest companion to the root, for this reason, it must be the case that B ≤ P B i B\le_P B_i B ≤ P B i . Then, by tree-compatibility B ◃ B i B\triangleleft B_i B ◃ B i .
Finally note that for some n n n , B = L n \mathcal{B}=\mathcal{L}_n B = L n , hence B i B_i B i is a ◃ \triangleleft ◃ maximum for B \mathcal{B} B .
Lemma 51. The ascending and descending phase over P 0 \mathcal{P}_0 P 0 produces two pre-proofs P l = ( D l , R l ) P^l=(D^l,R^l) P l = ( D l , R l ) and P r = ( D r , R r ) P^r=(D^r,R^r) P r = ( D r , R r ) with conclusions ⊢ Γ , I \vdash\Gamma,I ⊢ Γ , I and ⊢ Δ , I ⊥ \vdash\Delta,I^\perp ⊢ Δ , I ⊥ , such that P = P l ⊢ Γ , I P r ⊢ Δ , I ⊥ ⊢ Γ , Δ P=\frac{P^l}{\vdash\Gamma,I}\frac{P^r}{\vdash\Delta,I^\perp}\vdash\Gamma,\Delta P = ⊢ Γ , I P l ⊢ Δ , I ⊥ P r ⊢ Γ , Δ (Cut) and satisfying:
Bud ( D 0 ) = { B 1 , … , B n } \operatorname{Bud}(\mathcal{D}_0)=\{B_1,\ldots,B_n\} Bud ( D 0 ) = { B 1 , … , B n } if and only if Bud ( D l ) = { B 1 l , … , B n l } \operatorname{Bud}(\mathcal{D}^l)=\{B^l_1,\ldots,B^l_n\} Bud ( D l ) = { B 1 l , … , B n l } and Bud ( D r ) = { B 1 r , … , B n r } \operatorname{Bud}(\mathcal{D}^r)=\{B^r_1,\ldots,B^r_n\} Bud ( D r ) = { B 1 r , … , B n r } .
I I I is a μ L L ω \mu\mathrm{LL}^{\omega} μ LL ω formula such that Voc ( I ) ⊆ ( Voc ( Γ ) ∪ C l ) ∩ ( Voc ( Δ ) ∪ C r ) \operatorname{Voc}(I)\subseteq(\operatorname{Voc}(\Gamma)\cup C^l)\cap(\operatorname{Voc}(\Delta)\cup C^r) Voc ( I ) ⊆ ( Voc ( Γ ) ∪ C l ) ∩ ( Voc ( Δ ) ∪ C r ) , and every fixed-point variable in I I I is bounded.
R l \mathcal{R}^l R l and R r \mathcal{R}^r R r are total and injective bud-companion relations. We define C B i l = R l ( B i l ) C_{B^l_i}=\mathcal{R}^l(B^l_i) C B i l = R l ( B i l ) (resp. C B i r = R l ( B i r ) C_{B^r_i}=\mathcal{R}^l(B^r_i) C B i r = R l ( B i r ) ).
For all B 1 , B 2 ∈ Bud ( D 0 ) B_1,B_2\in\operatorname{Bud}(\mathcal{D}_0) B 1 , B 2 ∈ Bud ( D 0 ) , then B 1 ≤ P 0 B 2 B_1\leq_{\mathcal{P}_0}B_2 B 1 ≤ P 0 B 2 if and only if B 1 l ≤ P l B 2 l B^l_1\leq_{\mathcal{P}_l}B^l_2 B 1 l ≤ P l B 2 l and B 1 r ≤ P r B 2 r B^r_1\leq_{\mathcal{P}_r}B^r_2 B 1 r ≤ P r B 2 r .
Let ( Bud ( D l ) , ◃ l ) (\operatorname{Bud}(\mathcal{D}^l),\triangleleft_l) ( Bud ( D l ) , ◃ l ) and ( Bud ( D r ) , ◃ r ) (\operatorname{Bud}(\mathcal{D}^r),\triangleleft_r) ( Bud ( D r ) , ◃ r ) be defined as:
B i l ◃ l B j l ⟺ B i ◃ 0 B j and B i r ◃ r B j r ⟺ B i ◃ 0 B j B^l_i\triangleleft_l B^l_j\Longleftrightarrow B_i\triangleleft_0 B_j\quad\text{and}\quad B^r_i\triangleleft_r B^r_j\Longleftrightarrow B_i\triangleleft_0 B_j B i l ◃ l B j l ⟺ B i ◃ 0 B j and B i r ◃ r B j r ⟺ B i ◃ 0 B j
If ( Bud ( D 0 ) , ◃ 0 ) (\operatorname{Bud}(\mathcal{D}_0),\triangleleft_0) ( Bud ( D 0 ) , ◃ 0 ) is a tree-compatible induction order for P 0 \mathcal{P}_0 P 0 , then ( Bud ( D l ) , ◃ l ) (\operatorname{Bud}(\mathcal{D}^l),\triangleleft_l) ( Bud ( D l ) , ◃ l ) and ( Bud ( D r ) , ◃ r ) (\operatorname{Bud}(\mathcal{D}^r),\triangleleft_r) ( Bud ( D r ) , ◃ r ) are tree-compatible induction orders for P l \mathcal{P}_l P l and P r \mathcal{P}_r P r respectively.
B = { B 1 , … , B n } ⊆ Bud ( D 0 ) \mathcal{B}=\{B_1,\ldots,B_n\}\subseteq\operatorname{Bud}(\mathcal{D}_0) B = { B 1 , … , B n } ⊆ Bud ( D 0 ) is the weakly ≤ P 0 \leq_{\mathcal{P}_0} ≤ P 0 -connected set of all bud nodes such that B k ∈ Bud ( D 0 ) B_k\in\operatorname{Bud}(\mathcal{D}_0) B k ∈ Bud ( D 0 ) is a ◃ 0 \triangleleft_0 ◃ 0 maximum for B \mathcal{B} B if and only if B l = { B 1 l , … , B n l } ⊆ Bud ( D l ) \mathcal{B}^l=\{B^l_1,\ldots,B^l_n\}\subseteq\operatorname{Bud}(\mathcal{D}^l) B l = { B 1 l , … , B n l } ⊆ Bud ( D l ) (resp. B r = { B 1 r , … , B n r } ⊆ Bud ( D r ) \mathcal{B}^r=\{B^r_1,\ldots,B^r_n\}\subseteq\operatorname{Bud}(\mathcal{D}^r) B r = { B 1 r , … , B n r } ⊆ Bud ( D r ) ) is the weakly ≤ P l \leq_{\mathcal{P}_l} ≤ P l -connected (resp. ≤ P r \leq_{\mathcal{P}_r} ≤ P r -connected) set of all bud nodes such that B k l ∈ Bud ( D l ) B^l_k\in\operatorname{Bud}(\mathcal{D}^l) B k l ∈ Bud ( D l ) (resp. B k r ∈ Bud ( D r ) B^r_k\in\operatorname{Bud}(\mathcal{D}^r) B k r ∈ Bud ( D r ) ) is a ◃ l \triangleleft_l ◃ l (resp. ◃ r \triangleleft_r ◃ r ) maximum for B l \mathcal{B}^l B l (resp. B r \mathcal{B}^r B r ).
Let B l = { B 1 l , … , B n l } \mathcal{B}^l=\{B^l_1,\ldots,B^l_n\} B l = { B 1 l , … , B n l } and B r = { B 1 r , … , B n r } \mathcal{B}^r=\{B^r_1,\ldots,B^r_n\} B r = { B 1 r , … , B n r } be the sets described in the item (6). Then either the traces τ B 1 l , B 1 l , … , τ B n l , B n l \tau_{B^l_1,B^l_1},\ldots,\tau_{B^l_n,B^l_n} τ B 1 l , B 1 l , … , τ B n l , B n l or τ B 1 r , B 1 r , … , τ B n r , B n r \tau_{B^r_1,B^r_1},\ldots,\tau_{B^r_n,B^r_n} τ B 1 r , B 1 r , … , τ B n r , B n r exist, satisfying:
a. For all B i l ∈ B l B^l_i\in\mathcal{B}^l B i l ∈ B l (resp. B i r ∈ B r B^r_i\in\mathcal{B}^r B i r ∈ B r ), then τ B j l , B i l \tau_{B^l_j,B^l_i} τ B j l , B i l (resp. τ B j r , B i r \tau_{B^r_j,B^r_i} τ B j r , B i r ) is a trace over the unique path from C B i l C_{B^l_i} C B i l (resp. C B i r C_{B^r_i} C B i r ) to B i l B^l_i B i l (resp. B i r B^r_i B i r ) in D l \mathcal{D}^l D l (resp. D r \mathcal{D}^r D r ), such that τ B j l , B i l ( C B i l ) = τ B j l , B i l ( B i l ) \tau_{B^l_j,B^l_i}(C_{B^l_i})=\tau_{B^l_j,B^l_i}(B^l_i) τ B j l , B i l ( C B i l ) = τ B j l , B i l ( B i l ) (resp. τ B j r , B i r ( C B i r ) = τ B j r , B i r ( B i r ) \tau_{B^r_j,B^r_i}(C_{B^r_i})=\tau_{B^r_j,B^r_i}(B^r_i) τ B j r , B i r ( C B i r ) = τ B j r , B i r ( B i r ) ).
b. Assume B i l , B j l ◃ l B k l B^l_i,B^l_j\triangleleft_l B^l_k B i l , B j l ◃ l B k l (resp. B i r , B j r ◃ r B k r B^r_i,B^r_j\triangleleft_r B^r_k B i r , B j r ◃ r B k r ) with B j l ≤ P l B i l B^l_j\leq_{\mathcal{P}_l}B^l_i B j l ≤ P l B i l (resp. B j r ≤ P r B i r B^r_j\leq_{\mathcal{P}_r}B^r_i B j r ≤ P r B i r ). Then τ B k l , B i l ( C B j l ) = τ B k l , B j l ( C B j l ) \tau_{B^l_k,B^l_i}(C_{B^l_j})=\tau_{B^l_k,B^l_j}(C_{B^l_j}) τ B k l , B i l ( C B j l ) = τ B k l , B j l ( C B j l ) (resp. τ B k r , B i r ( C B j r ) = τ B k r , B j r ( C B j r ) \tau_{B^r_k,B^r_i}(C_{B^r_j})=\tau_{B^r_k,B^r_j}(C_{B^r_j}) τ B k r , B i r ( C B j r ) = τ B k r , B j r ( C B j r ) ).
c. τ B k l , B k l \tau_{B^l_k,B^l_k} τ B k l , B k l (resp. τ B k r , B k r \tau_{B^r_k,B^r_k} τ B k r , B k r ) has at least one F F F -progress point and F F F is minimal in ⋃ i = 1 n Form ( τ B k l , B i l ) \bigcup_{i=1}^{n}\operatorname{Form}(\tau_{B^l_k,B^l_i}) ⋃ i = 1 n Form ( τ B k l , B i l ) (resp. ⋃ i = 1 n Form ( τ B k r , B i r ) \bigcup_{i=1}^{n}\operatorname{Form}(\tau_{B^r_k,B^r_i}) ⋃ i = 1 n Form ( τ B k r , B i r ) ) with respect to the subformula relation.
Let B l = { B 1 l , … , B n l } \mathcal{B}^l=\{B^l_1,\ldots,B^l_n\} B l = { B 1 l , … , B n l } and B r = { B 1 r , … , B n r } \mathcal{B}^r=\{B^r_1,\ldots,B^r_n\} B r = { B 1 r , … , B n r } be the sets described in the item (6). Then the interpolant traces ρ B 1 l , … , ρ B n l \rho_{B^l_1},\ldots,\rho_{B^l_n} ρ B 1 l , … , ρ B n l and ρ B 1 r , … , ρ B n r \rho_{B^r_1},\ldots,\rho_{B^r_n} ρ B 1 r , … , ρ B n r exist, satisfying, for any choice of the binders σ i ∈ { μ , ν } \sigma_i\in\{\mu,\nu\} σ i ∈ { μ , ν } performed at the closure steps of the descending phase:
a. If B k l ∈ B l B^l_k\in\mathcal{B}^l B k l ∈ B l (resp. B k r ∈ B r B^r_k\in\mathcal{B}^r B k r ∈ B r ) is closest to the root among the companions in B l \mathcal{B}^l B l (resp. B r \mathcal{B}^r B r ), then B i l B^l_i B i l (resp. B i r B^r_i B i r ) is a ◃ l \triangleleft_l ◃ l (resp. ◃ r \triangleleft_r ◃ r ) maximum for B l \mathcal{B}^l B l (resp. B r \mathcal{B}^r B r ).
b. For all B i l ∈ B l B^l_i\in\mathcal{B}^l B i l ∈ B l (resp. B i r ∈ B r B^r_i\in\mathcal{B}^r B i r ∈ B r ), then ρ B i l \rho_{B^l_i} ρ B i l (resp. ρ B i r \rho_{B^r_i} ρ B i r ) is a trace over the unique path from C B i l C_{B^l_i} C B i l (resp. C B i r C_{B^r_i} C B i r ) to B i l B^l_i B i l (resp. B i r B^r_i B i r ) in D l \mathcal{D}^l D l (resp. D r \mathcal{D}^r D r ), such that ρ B i l ( C B i l ) = ρ B i l ( B i l ) = σ i X B i . I i \rho_{B^l_i}(C_{B^l_i})=\rho_{B^l_i}(B^l_i)=\sigma_iX_{B_i}.I_i ρ B i l ( C B i l ) = ρ B i l ( B i l ) = σ i X B i . I i (resp. ρ B i r ( C B i r ) = ρ B i r ( B i r ) = σ i l X B i . I i l \rho_{B^r_i}(C_{B^r_i})=\rho_{B^r_i}(B^r_i)=\sigma^l_iX_{B_i}.I^l_i ρ B i r ( C B i r ) = ρ B i r ( B i r ) = σ i l X B i . I i l ), and the step of ρ B i l \rho_{B^l_i} ρ B i l (resp. ρ B i r \rho_{B^r_i} ρ B i r ) out of C B i l C_{B^l_i} C B i l (resp. C B i r C_{B^r_i} C B i r ) is the Fischer-Ladner unfolding of the σ i \sigma_i σ i (resp. σ i l \sigma^l_i σ i l ) formula. This step is an F F F -progress point with F = σ i X B i . I i F=\sigma_iX_{B_i}.I_i F = σ i X B i . I i (resp. F = σ i l X B i . I i l F=\sigma^l_iX_{B_i}.I^l_i F = σ i l X B i . I i l ) if and only if σ i = ν \sigma_i=\nu σ i = ν (resp. σ i = μ \sigma_i=\mu σ i = μ ).
c. Assume B i l , B j l ◃ l B k l B^l_i,B^l_j\triangleleft_l B^l_k B i l , B j l ◃ l B k l (resp. B i r , B j r ◃ r B k r B^r_i,B^r_j\triangleleft_r B^r_k B i r , B j r ◃ r B k r ) with B j l ≤ P l B i l B^l_j\leq_{\mathcal{P}_l}B^l_i B j l ≤ P l B i l (resp. B j r ≤ P r B i r B^r_j\leq_{\mathcal{P}_r}B^r_i B j r ≤ P r B i r ). Then ρ B i l ( C B j l ) = ρ B j l ( C B j l ) \rho_{B^l_i}(C_{B^l_j})=\rho_{B^l_j}(C_{B^l_j}) ρ B i l ( C B j l ) = ρ B j l ( C B j l ) (resp. ρ B i r ( C B j r ) = ρ B j r ( C B j r ) \rho_{B^r_i}(C_{B^r_j})=\rho_{B^r_j}(C_{B^r_j}) ρ B i r ( C B j r ) = ρ B j r ( C B j r ) ).
d. Consider B k l ∈ B l B^l_k\in\mathcal{B}^l B k l ∈ B l (resp. B k r ∈ B r B^r_k\in\mathcal{B}^r B k r ∈ B r ) as in the item (a). Then ρ B k l ( C B k l ) = σ k X B k . I k \rho_{B^l_k}(C_{B^l_k})=\sigma_kX_{B_k}.I_k ρ B k l ( C B k l ) = σ k X B k . I k (resp. ρ B k r ( C B k r ) = σ k l X B k . I k l \rho_{B^r_k}(C_{B^r_k})=\sigma^l_kX_{B_k}.I^l_k ρ B k r ( C B k r ) = σ k l X B k . I k l ) is minimal in ⋃ i = 1 n Form ( ρ B i l ) \bigcup_{i=1}^{n}\operatorname{Form}(\rho_{B^l_i}) ⋃ i = 1 n Form ( ρ B i l ) (resp. ⋃ i = 1 n Form ( ρ B i r ) \bigcup_{i=1}^{n}\operatorname{Form}(\rho_{B^r_i}) ⋃ i = 1 n Form ( ρ B i r ) ) with respect to the subformula relation.
Proof. 1. It follows from Proposition 47 (1), over the derivation tree D 0 \mathcal{D}_0 D 0 .
It follows from 47 (2) over the derivation tree D 0 \mathcal{D}_0 D 0 .
It follows from Proposition 47 (3) over the derivation tree D 0 \mathcal{D}_0 D 0 .
Let B 1 , B 2 ∈ Bud ( D 0 ) B_1, B_2 \in\operatorname{Bud}(D_0) B 1 , B 2 ∈ Bud ( D 0 ) . Then:
B 1 ≤ P 0 B 2 ⟺ D e f 24 C B 2 ⪯ D 0 C B 1 ⪯ D 0 B 2 ⟺ P r o p 47.4 C B 2 l ⪯ D l C B 1 l ⪯ D l B 2 l ⟺ D e f 24 B 1 l ≤ P l B 2 l . B_1 \le_{P_0} B_2 \mathrel{\mathop{\Longleftrightarrow}^{\mathrm{Def}\ 24}} C_{B_2} \preceq_{D_0} C_{B_1} \preceq_{D_0} B_2 \mathrel{\mathop{\Longleftrightarrow}^{\mathrm{Prop}\ 47.4}} C_{B_2^l} \preceq_{D_l} C_{B_1^l} \preceq_{D_l} B_2^l \mathrel{\mathop{\Longleftrightarrow}^{\mathrm{Def}\ 24}} B_1^l \le_{P_l} B_2^l. B 1 ≤ P 0 B 2 ⟺ Def 24 C B 2 ⪯ D 0 C B 1 ⪯ D 0 B 2 ⟺ Prop 47.4 C B 2 l ⪯ D l C B 1 l ⪯ D l B 2 l ⟺ Def 24 B 1 l ≤ P l B 2 l .
We can use the same argument to derive B 1 r ≤ P r B 2 r B_1^r \le_{P_r} B_2^r B 1 r ≤ P r B 2 r .
Assume ( Bud ( D 0 ) , ◃ 0 ) (\operatorname{Bud}(D_0), \triangleleft_0) ( Bud ( D 0 ) , ◃ 0 ) is a tree-compatible induction order. By definition, ◃ l \triangleleft_l ◃ l and ◃ r \triangleleft_r ◃ r are also induction orders since they are exact copies of ◃ 0 \triangleleft_0 ◃ 0 , which is an induction order. Moreover, since the relations ≤ P l \le_{P_l} ≤ P l and ≤ P r \le_{P_r} ≤ P r are exact copies of ≤ P 0 \le_{P_0} ≤ P 0 by the item (4), where R l R^l R l and R r R^r R r are injective by the item (3), then tree-compatibility is preserved.
Follows directly from the fact that ≤ P l \le_{P_l} ≤ P l and ≤ P r \le_{P_r} ≤ P r are exact copies of ≤ P 0 \le_{P_0} ≤ P 0 by the item (4), and that ◃ l \triangleleft_l ◃ l and ◃ r \triangleleft_r ◃ r are also copies of ◃ 0 \triangleleft_0 ◃ 0 .
The fact that either the traces τ B k l , B i l … τ B k l , B n l \tau_{B_k^l,B_i^l}\ldots\tau_{B_k^l,B_n^l} τ B k l , B i l … τ B k l , B n l or τ B k r , B i r … τ B k r , B n r \tau_{B_k^r,B_i^r}\ldots\tau_{B_k^r,B_n^r} τ B k r , B i r … τ B k r , B n r exist, follows from Proposition 49. The properties (a) and (b) are direct consequence of Proposition 47.5.a.ii and 47.5.b.ii respectively. Note that τ B k , B k \tau_{B_k,B_k} τ B k , B k has an F F F -progress point such that F F F is minimal in ⋃ j = 1 n Form ( τ B k , B i ) \bigcup_{j=1}^{n}\operatorname{Form}(\tau_{B_k,B_i}) ⋃ j = 1 n Form ( τ B k , B i ) with respect to the subformula relation in the original ordered trace manifold. Therefore, by Proposition 47.5.c.ii we derive property (c).
Property (a) is a direct application of Proposition 50. The property (b) follows directly from Proposition 47.6.a.ii and 47.6.c. For property (c) we have that τ B k l , B i l ( C B i l ) = σ j X B j . I j = τ B k l , B i l ( C B i l ) \tau_{B_k^l,B_i^l}(C_{B_i^l}) = \sigma_j X_{B_j}.I_j = \tau_{B_k^l,B_i^l}(C_{B_i^l}) τ B k l , B i l ( C B i l ) = σ j X B j . I j = τ B k l , B i l ( C B i l ) (resp. τ B k r , B i r ( C B j r ) = σ j X B j . I j r = τ B k r , B i r ( C B i r ) \tau_{B_k^r,B_i^r}(C_{B_j^r}) = \sigma_j X_{B_j}.I_j^r = \tau_{B_k^r,B_i^r}(C_{B_i^r}) τ B k r , B i r ( C B j r ) = σ j X B j . I j r = τ B k r , B i r ( C B i r ) ) by Proposition 47.6.b and 47.6.c. For the proof of property (d) we have:
We prove the left case since the right case is symmetrical. First, note that by Proposition 47.6.b, every formula in ⋃ i = 1 n Form ( ρ B i l ) \bigcup_{i=1}^{n}\operatorname{Form}(\rho_{B_i^l}) ⋃ i = 1 n Form ( ρ B i l ) is either a σ \sigma σ -formula σ j X B j . I j \sigma_jX_{B_j}.I_j σ j X B j . I j , taken at a companion node C B j l C_{B_j^l} C B j l on the path followed by some ρ B i l \rho_{B_i^l} ρ B i l , or a decomposition of a formula into its subformulas, containing σ \sigma σ -formula σ i X B i . I i \sigma_iX_{B_i}.I_i σ i X B i . I i as a strict subformula. In the first case B j ∈ B B_j \in\mathcal{B} B j ∈ B , since if C B j l C_{B_j^l} C B j l is on the path of ρ B i l \rho_{B_i^l} ρ B i l , it implies B j ≤ P 0 B i B_j \le_{P_0} B_i B j ≤ P 0 B i by the item (4), and since B i ◃ 0 B k B_i \triangleleft_0 B_k B i ◃ 0 B k , by tree-compatibility and transitivity then B j ◃ 0 B k B_j \triangleleft_0 B_k B j ◃ 0 B k . Therefore, it suffices to show that no σ \sigma σ formula σ j X B j . I j \sigma_jX_{B_j}.I_j σ j X B j . I j , where B j ∈ B B_j \in\mathcal{B} B j ∈ B is a strict subformula of σ k X B k . I k \sigma_kX_{B_k}.I_k σ k X B k . I k . Assume then that σ j X B j . I j \sigma_jX_{B_j}.I_j σ j X B j . I j is a strict subformula of σ k X B k . I k \sigma_kX_{B_k}.I_k σ k X B k . I k , for some B j ∈ B B_j \in\mathcal{B} B j ∈ B . By the property Proposition 47.6.d, this implies C B j l ⪯ D l C B k l C_{B_j^l} \preceq_{D_l} C_{B_k^l} C B j l ⪯ D l C B k l with C B j l ≠ C B k l C_{B_j^l} \ne C_{B_k^l} C B j l = C B k l , i.e. the companion C B j l C_{B_j^l} C B j l is strictly closer to the root than C B k l C_{B_k^l} C B k l . This contradicts the item (a), since C B k l C_{B_k^l} C B k l is closest to the root among the companions in B l \mathcal{B}^l B l . Therefore, σ k X B k . I k \sigma_kX_{B_k}.I_k σ k X B k . I k is minimal.
Intuitively, the construction of the ordered trace manifolds for P l \mathcal{P}^l P l and P r \mathcal{P}^r P r can be summarized in the following series of steps:
After the descending phase, the structure of P 0 \mathcal{P}_0 P 0 is duplicated on both sides. By Lemma 51.1, 51.3 and 51.5, every bud node, companion node and basic cycle of P 0 \mathcal{P}_0 P 0 has a copy in P l \mathcal{P}^l P l and a copy in P r \mathcal{P}^r P r , and by Lemma 51.4, the structural connectivity ≤ P 0 \le_{P_0} ≤ P 0 and the induction order ◃ 0 \triangleleft_0 ◃ 0 are transported to both sides. In particular, both P l \mathcal{P}^l P l and P r \mathcal{P}^r P r must validate a copy of every cycle of P 0 \mathcal{P}_0 P 0 .
However, the traces of the original ordered trace manifold are not duplicated. This happens since each trace lies entirely on one side of the split (Proposition 48). More precisely, by Lemma 51.6, for each maximal weakly ≤ P 0 \le_{P_0} ≤ P 0 -connected set B \mathcal{B} B with the associated sets of copies B l \mathcal{B}^l B l and B r \mathcal{B}^r B r , then the traces related to B \mathcal{B} B in the original manifold are transferred as traces over the cycles of B l \mathcal{B}^l B l in P l \mathcal{P}^l P l , or over the cycles of B r \mathcal{B}^r B r in P r \mathcal{P}^r P r , but not both. These traces satisfy the conditions of ordered trace manifolds on the side that includes them.
The cycles on the side that did not receive the transferred traces of B \mathcal{B} B must therefore be validated by other means: the interpolant traces ρ B i l , ρ B i r \rho_{B_i^l},\rho_{B_i^r} ρ B i l , ρ B i r of Lemma 51.8, which exist on both sides. Since the unfolding step at a companion node is a progress point exactly when the binder is ν \nu ν on that side, choosing each σ i \sigma_i σ i opposite to the side receiving the transferred traces of the original ordered trace manifold makes the interpolant traces progress precisely on the unsupported side. The minimality condition then follows from tree-compatibility, through Proposition 50 and Lemma 51.8.d
This procedure is formalized in the following theorem:
Theorem 52 (Weak interpolation for μ L L ω \mu\mathrm{LL}^{\omega} μ LL ω ). Let P 0 = ( D 0 , R 0 ) \mathcal{P}_0=(\mathcal{D}_0,\mathcal{R}_0) P 0 = ( D 0 , R 0 ) be a μ L L ω \mu\mathrm{LL}^{\omega} μ LL ω proof with a split conclusion ⊢ Γ , Δ \vdash\Gamma,\Delta ⊢ Γ , Δ and a trace manifold. Moreover, let C = { ( C 1 , C 1 ⊥ ) , … , ( C n , C n ⊥ ) } \mathcal{C}=\{(C_1,C_1^{\perp}),\ldots,(C_n,C_n^{\perp})\} C = {( C 1 , C 1 ⊥ ) , … , ( C n , C n ⊥ )} be the cut formulas of D 0 \mathcal{D}_0 D 0 , let L , R L,R L , R be any partitioning of { 1 , … , n } \{1,\ldots,n\} { 1 , … , n } and C L = Voc ( C i , i ∈ L ) \mathcal{C}^{L}=\operatorname{Voc}(C_i,i\in L) C L = Voc ( C i , i ∈ L ) and C R = Voc ( C i , i ∈ R ) \mathcal{C}^{R}=\operatorname{Voc}(C_i,i\in R) C R = Voc ( C i , i ∈ R ) . After the ascending and descending phase over P 0 \mathcal{P}_0 P 0 , we obtain an interpolated pre-proof P = ⊢ Γ , I ⊢ Δ , I ⊥ ⊢ Γ , Δ ( C u t ) \mathcal{P}=\frac{\vdash\Gamma,I\qquad\vdash\Delta,I^{\perp}}{\vdash\Gamma,\Delta}\;(\mathrm{Cut}) P = ⊢ Γ , Δ ⊢ Γ , I ⊢ Δ , I ⊥ ( Cut ) satisfying:
P l = ( D l , R l ) \mathcal{P}^{l}=(\mathcal{D}^{l},\mathcal{R}^{l}) P l = ( D l , R l ) and P r = ( D r , R r ) \mathcal{P}^{r}=(\mathcal{D}^{r},\mathcal{R}^{r}) P r = ( D r , R r ) are two pre-proofs with conclusions ⊢ Γ , I \vdash\Gamma,I ⊢ Γ , I and ⊢ Δ , I ⊥ \vdash\Delta,I^{\perp} ⊢ Δ , I ⊥ respectively, where R l \mathcal{R}^{l} R l (resp. R r \mathcal{R}^{r} R r ) is injective and Voc ( I ) ⊆ ( Voc ( Γ ) ∪ C L ) ∩ ( Voc ( Δ ) ∪ C R ) \operatorname{Voc}(I)\subseteq(\operatorname{Voc}(\Gamma)\cup\mathcal{C}^{L})\cap(\operatorname{Voc}(\Delta)\cup\mathcal{C}^{R}) Voc ( I ) ⊆ ( Voc ( Γ ) ∪ C L ) ∩ ( Voc ( Δ ) ∪ C R ) .
There exists an induction order, and an ordered trace manifold for P l \mathcal{P}^{l} P l and P r \mathcal{P}^{r} P r respectively, implying that P \mathcal{P} P is a valid proof by Lemmas 29 and 30.
P → ( C u t ) ω P 0 \mathcal{P}\mathrel{\to^{\omega}_{(\mathrm{Cut})}}\mathcal{P}_0 P → ( Cut ) ω P 0 .
Proof. Without loss of generality, we may assume that the bud-companion relations are always injective as explained in the Appendix C.2. Firstly, we transform the trace manifold for P 0 \mathcal{P}_0 P 0 into an ordered trace manifold using the Lemma 30. Then using Algorithm 1 and 2, we obtain a new proof P 0 ′ = ( D 0 ′ , R 0 ′ ) \mathcal{P}'_0=(\mathcal{D}'_0,\mathcal{R}'_0) P 0 ′ = ( D 0 ′ , R 0 ′ ) that can be coherently decorated, and has an ordered trace manifold with respect to an induction order ( Bud ( D 0 ′ ) , ◃ ′ ) (\operatorname{Bud}(\mathcal{D}'_0),\mathrel{\triangleleft'} ) ( Bud ( D 0 ′ ) , ◃ ′ ) that is tree-compatible. Therefore, after the ascending and descending phase over P 0 ′ \mathcal{P}'_0 P 0 ′ we obtain an interpolated pre-proof P = ⊢ Γ , I ⊢ Δ , I ⊥ ⊢ Γ , Δ ( C u t ) \mathcal{P}=\frac{\vdash\Gamma,I\qquad\vdash\Delta,I^{\perp}}{\vdash\Gamma,\Delta}\;(\mathrm{Cut}) P = ⊢ Γ , Δ ⊢ Γ , I ⊢ Δ , I ⊥ ( Cut ) such that:
By Lemma 51.2 and 51.3, the property (1) is satisfied.
Consider the tree-compatible induction orders ( Bud ( D l ) , ◃ l ) (\operatorname{Bud}(\mathcal{D}^{l}),\mathrel{\triangleleft^{l}}) ( Bud ( D l ) , ◃ l ) and ( Bud ( D r ) , ◃ r ) (\operatorname{Bud}(\mathcal{D}^{r}),\mathrel{\triangleleft^{r}}) ( Bud ( D r ) , ◃ r ) for the pre-proofs P l \mathcal{P}^{l} P l and P r \mathcal{P}^{r} P r respectively, given by the Lemma 51.5. We will build an ordered trace manifold for P l \mathcal{P}^{l} P l and P r \mathcal{P}^{r} P r as follows: Consider the bud nodes B k l ∈ Bud ( D l ) B_k^{l}\in\operatorname{Bud}(\mathcal{D}^{l}) B k l ∈ Bud ( D l ) and B k r ∈ Bud ( D r ) B_k^{r}\in\operatorname{Bud}(\mathcal{D}^{r}) B k r ∈ Bud ( D r ) , given by the Lemma 51.1. Now, consider the sets of bud nodes B l = { B l ∣ B l ∈ Bud ( D l ) and B l ◃ l B k l } \mathcal{B}^{l}=\{B^{l}\mid B^{l}\in\operatorname{Bud}(\mathcal{D}^{l})\text{ and }B^{l}\mathrel{\triangleleft^{l}} B_k^{l}\} B l = { B l ∣ B l ∈ Bud ( D l ) and B l ◃ l B k l } and B r = { B r ∣ B r ∈ Bud ( D r ) and B r ◃ r B k r } \mathcal{B}^{r}=\{B^{r}\mid B^{r}\in\operatorname{Bud}(\mathcal{D}^{r})\text{ and }B^{r}\mathrel{\triangleleft^{r}} B_k^{r}\} B r = { B r ∣ B r ∈ Bud ( D r ) and B r ◃ r B k r } , given by the Lemma 51.6. By Lemma 51.7, either for all B i l ∈ B l B_i^{l}\in\mathcal{B}^{l} B i l ∈ B l then τ B k l , B i l \tau_{B_k^{l},B_i^{l}} τ B k l , B i l exists, or for all B i r ∈ B r B_i^{r}\in\mathcal{B}^{r} B i r ∈ B r then τ B k r , B i r \tau_{B_k^{r},B_i^{r}} τ B k r , B i r exists. Assume the first case (the other scenario is symmetrical). Therefore, all the traces associated to B k l B_k^{l} B k l for the trace manifold in P l \mathcal{P}^{l} P l are covered, and satisfy the ordered trace manifold properties by Lemma 51.7. It remains to be shown that we can construct the traces for P r \mathcal{P}^{r} P r using the interpolant.
We set τ B k r , B i r = ρ B i r \tau_{B_k^{r},B_i^{r}}=\rho_{B_i^{r}} τ B k r , B i r = ρ B i r , for all B i r ∈ B r B_i^{r}\in\mathcal{B}^{r} B i r ∈ B r as specified in Lemma 51.8. We check that these traces satisfy the three properties of the ordered trace manifolds. Properties (1) and (2), they are derived directly from Lemma 51.8.b and Lemma 51.8.c. We now check property (3). First, assume B n r ∈ B r B_n^{r}\in\mathcal{B}^{r} B n r ∈ B r , has the closest companion to the root of D r \mathcal{D}^{r} D r in the set B r \mathcal{B}^{r} B r . By Proposition 50, then B n r B_n^{r} B n r is a ◃ r \triangleleft^{r} ◃ r maximum for B r \mathcal{B}^{r} B r . However, this implies that B k r ◃ r B n r B_k^{r}\mathrel{\triangleleft^{r}}B_n^{r} B k r ◃ r B n r and B k r ◃ r B n r B_k^{r}\mathrel{\triangleleft^{r}}B_n^{r} B k r ◃ r B n r . Thus, by antisymmetry B n r = B k r B_n^{r}=B_k^{r} B n r = B k r , which implies that B k r B_k^{r} B k r has the closest companion to the root of D r \mathcal{D}^{r} D r in the set B r \mathcal{B}^{r} B r . Therefore, ρ B i r ( C B k r ) = σ k l X B k r . I k ⊥ \rho_{B_i^{r}}(C_{B_k^{r}})=\sigma_k^{l}X_{B_k^{r}}.I_k^{\perp} ρ B i r ( C B k r ) = σ k l X B k r . I k ⊥ is minimal in ⋃ B i r ∈ B r Form ( ρ B i r ) \bigcup_{B_i^{r}\in\mathcal{B}^{r}}\operatorname{Form}(\rho_{B_i^{r}}) ⋃ B i r ∈ B r Form ( ρ B i r ) with respect to the subformula relation by Lemma 51.8.d. Finally, by Lemma 51.8.b if we pick σ k ⊥ = ν \sigma_k^{\perp} = \nu σ k ⊥ = ν , then ρ k B r \rho^{\mathrm{Br}}_k ρ k Br has a progress point, which completes the proof of property (3).
For the last part of the theorem one can follow the interpolation proof-sequence backward. Remark 36.(i) ensures that it consists in a cut-elimination reduction sequence while Remark 36.(ii) ensures that the cut-reduction can be performed by unfolding the circular proofs and preserving the synchronisation of the two circular proofs so that at the limit, the proof object is regular and is the infinite unfolding of the original circular proof P 0 P_0 P 0 .
As done in the Section 3, if we restrict the vocabulary of the cut formulas in the pre-proof P 0 P_0 P 0 , we obtain an interpolation theorem with cuts.
Theorem 53 (Interpolation theorem with cuts for μ L L ω \mu\mathrm{LL}^{\omega} μ LL ω ). Let P 0 = ( D 0 , R 0 ) P_0 = (D_0, R_0) P 0 = ( D 0 , R 0 ) be a μ L L ω \mu\mathrm{LL}^{\omega} μ LL ω proof with a split conclusion ⊢ Γ , Δ \vdash\Gamma, \Delta ⊢ Γ , Δ and a trace manifold. Moreover, let C = { ( C 1 , C 1 ⊥ ) , … , ( C n , C n ⊥ ) } \mathcal{C} = \{(C_1, C_1^{\perp}), \ldots, (C_n, C_n^{\perp})\} C = {( C 1 , C 1 ⊥ ) , … , ( C n , C n ⊥ )} be the cut formulas of D 0 D_0 D 0 , let L , R L, R L , R be any partitioning of { 1 , … , n } \{1, \ldots, n\} { 1 , … , n } and C L = Voc ( C i , i ∈ L ) \mathcal{C}^{L} = \operatorname{Voc}(C_i, i \in L) C L = Voc ( C i , i ∈ L ) and C R = Voc ( C i , i ∈ R ) \mathcal{C}^{R} = \operatorname{Voc}(C_i, i \in R) C R = Voc ( C i , i ∈ R ) , such that ( C L ∖ Voc ( Γ ) ) ∩ ( Voc ( Δ ) ∪ C R ) = ∅ (\mathcal{C}^{L} \setminus\operatorname{Voc}(\Gamma)) \cap(\operatorname{Voc}(\Delta) \cup\mathcal{C}^{R}) = \varnothing ( C L ∖ Voc ( Γ )) ∩ ( Voc ( Δ ) ∪ C R ) = ∅ and ( C R ∖ Voc ( Δ ) ) ∩ ( Voc ( Γ ) ∪ C L ) = ∅ (\mathcal{C}^{R} \setminus\operatorname{Voc}(\Delta)) \cap(\operatorname{Voc}(\Gamma) \cup\mathcal{C}^{L}) = \varnothing ( C R ∖ Voc ( Δ )) ∩ ( Voc ( Γ ) ∪ C L ) = ∅ . After the ascending and descending phase over P 0 P_0 P 0 , we obtain an interpolated pre-proof P = ⊢ Γ , I ⊢ Δ , I ⊥ ⊢ Γ , Δ ( C u t ) P = \frac{\vdash\Gamma, I \qquad\vdash\Delta, I^{\perp}}{\vdash\Gamma, \Delta}\,(\mathrm{Cut}) P = ⊢ Γ , Δ ⊢ Γ , I ⊢ Δ , I ⊥ ( Cut ) satisfying:
P l = ( D l , R l ) P^{l} = (D^{l}, R^{l}) P l = ( D l , R l ) and P r = ( D r , R r ) P^{r} = (D^{r}, R^{r}) P r = ( D r , R r ) are two pre-proofs with conclusions ⊢ Γ , I \vdash\Gamma, I ⊢ Γ , I and ⊢ Δ , I ⊥ \vdash\Delta, I^{\perp} ⊢ Δ , I ⊥ respectively, where R l R^{l} R l (resp. R r R^{r} R r ) is injective and Voc ( I ) ⊆ Voc ( Γ ) ∩ Voc ( Δ ) \operatorname{Voc}(I) \subseteq\operatorname{Voc}(\Gamma) \cap\operatorname{Voc}(\Delta) Voc ( I ) ⊆ Voc ( Γ ) ∩ Voc ( Δ ) .
There exists an induction order, and an ordered trace manifold for P l P^{l} P l and P r P^{r} P r respectively, implying that P P P is a valid proof by Lemmas 29 and 30.
P → ( C u t ) ω P 0 P \to^{\omega}_{(\mathrm{Cut})} P_0 P → ( Cut ) ω P 0 .
In the end, Theorem 37 is derived as a corollary of Theorem 53.