A proof of the focalization property of Linear Logic

Olivier LAURENT
Preuves Programmes Syst`mes e CNRS Universit Paris VII e UMR 7126 Case 7014 2, place Jussieu 75251 Paris Cedex 05 FRANCE

Olivier.Laurent@pps.jussieu.fr May 5, 2004

The focalization property of linear logic has been discovered by Jean-Marc Andreoli [1] in the beginning of the 90s. It is one of the main properties of Linear Logic that appeared after the original paper of Jean-Yves Girard [3]. This property is proved in various papers [1, 4, 2] but, as far as we know never for the usual LL sequent calculus (or with an intricate induction in Andreolis thesis, which is replaced here by a cut elimination property). The proof we give is a compilation of the previous proofs and proof techniques, our goal is just to give a complete presentation of a proof of this property which appears to be more and more important in the current research in Linear Logic [4, 2, 5, 6]. We decompose the usual focalization property into two technically dierent steps: weak focalization (the decompositions of two positive formulas are not interleaved, see LLfoc ) and reversing (negative rules are applied as late as possible from a top-down point of view, see LLFoc ).

Linear Logic and the focalization property

A ::= X | A A | A A | 1 | 0 | !A | X | A ` A | A & A | | | ?A

Formulas of Linear Logic are given by the usual grammar:

We split this grammar into two sub-classes: positive formulas and negative formulas. P N In the sequel: A, B, ... denote arbitrary formulas; P , Q, R denote positive formulas; N , M , L denote negative formulas; X, X , ... are called atoms or atomic formulas. 1 ::= X | A A | A A | 1 | 0 | !A ::= X | A ` A | A & A | | | ?A

We consider the usual rules of Linear Logic [3]: A , A ax , A , A cut ,

, A, B ` , A ` B , A , B & , A & B 1 ?, A ! ?, !A 1 , A ?d , ?A

, A , B , , A B , A 1 , A B , , , B 2 , A B , ?A, ?A ?c , ?A

?w , ?A

The main connectives of positive (resp. negative) formulas are called positive connectives (resp. negative connectives), that is X, , , 1, 0 and ! (resp. X , `, &, , and ?). The rules introducing positive (resp. negative) connectives are called positive rules (resp. negative rules). A positive (resp. negative) formula is strictly positive (resp. strictly negative) if its main connective is not ! (resp. ?). A positive (resp. negative) connective is strictly positive (resp. strictly negative) if it is not ! (resp. ?). A positive (resp. negative) rule is strictly positive (resp. strictly negative) if it is not introducing a !A formula (resp. ?A formula). Denition 1 (Main positive tree) If A is a formula, its main positive tree T + (A) is dened by: T + (N ) = T + (X) = X T + (A B) = T + (A) T + (B) T + (A B) = T + (A) T + (B) T + (1) = 1 T + (0) = 0 T + (!A) = ! T + (A) is a (possibly empty) tree whose nodes are , (with arity at most 2) and !, 1, 0 or X (with arity 0). Denition 2 (Weakly +-focalized proof ) A proof in LL is weakly +-focalized if it is cut-free and, for any subproof of with conclusion , A, the only positive rules of between two rules introducing connectives of T + (A) are rules introducing connectives of T + (A). Denition 3 (Strongly +-focalized proof ) A proof in LL is strongly +-focalized if it is cut-free and, for any subproof of with conclusion , A, the only rules of between two rules introducing connectives of T + (A) are rules introducing connectives of T + (A). 2

Denition 4 (Reversed proof ) A proof in LL is reversed if, for any subproof of with conclusion , N with N a strictly negative non-atomic formula, the last rule of is a strictly negative rule. Denition 5 (Focalized proof ) A proof in LL is focalized if it is both strongly +-focalized and reversed. Remark: Various systems for classical logic related with focalization constraints have been introduced. In particular Girards LC [4] corresponds to weak +-focalization, Danos-Joinet-Schellinxs LK [2] corresponds to strong +-focalization and Quatrini-Tortora de Falcos LK, [7] corresponds pol to focalization since their -constraint is a reversing constraint.

2
2.1

Weakly Focalized Linear Logic

Sequent calculus LLfoc

A sequent of LLfoc has the shape ; where is a multi-set of formulas and is either empty or contains a unique positive formula. The rules are a linear logic version of the rules of Girards LC [4]: P ; P ; P ax ; P foc , P ; , P ; , P ; n-cut , ; ; P ; Q , ; P Q , N ; , M ; , ; N M

, P ; p-cut , ; , A, B ; ` , A ` B ;

; P , M ; , ; P M

, N ; ; Q , ; N Q , A ; , B ; & , A & B ;

; P 1 ; P B ; 1

, N ; 1 ; N B 1 ; , ;

; Q 2 ; A Q , ;

, M ; 2 ; A M

?, A ; ! ? ; !A

; P ?d , ?P ;

, N ; ?d , ?N ;

; ?w , ?A ;

, ?A, ?A ; ?c , ?A ;

2.2

Expansion of axioms

Given a positive formula P , the sequent P ; P is provable in LLfoc by means of the ax rule. Moreover, if P is not atomic, it is also possible to give a proof based only on ax rules applied to the sub-formulas of P (and thus, by an easy induction, to the atomic sub-formulas of P ): P ; P Q ; Q P , Q ; P Q ` P ` Q ; P Q M ; M foc P ; P M, M ; P , M ; P M ` P ` M ; P M Q ; Q P ; P 1 2 P ; P Q Q ; P Q & P & Q ; P Q M ; M foc M, M ; P ; P 1 2 P ; P M M ; P M & P & M ; P M 1 ; 1 ; 1 P ; P foc P , P ; ?d ?P , P ; ! ?P ; !P N ; N foc N, N ; Q ; Q N , Q ; N Q ` N ` Q ; N Q M ; M N ; N foc foc N, N ; M, M ; N , M ; N M ` N ` M ; N M N ; N foc Q ; Q N, N ; 1 2 N ; N Q Q ; N Q & N & Q ; N Q N ; N M ; M foc foc N, N ; M, M ; 1 2 N ; N M M ; N M & N & M ; N M ; 0

N ; N ?d ?N , N ; ! ?N ; !N

2.3

Embedding of LL (weak +-focalization)

The translation (.) of LL into LLfoc does not modify formulas, translates the sequent as ; and acts on proofs by adding a lot of cut rules: ax ax P , P P ; P , P , P cut , , A, B ` , A ` B 4 , P ; , P ; n-cut , ; , A, B ; ` , A ` B ;

 , P , Q , , P Q , P ;

ax ax P ; P Q ; Q P , Q ; P Q foc , Q ; P , Q , P Q ; n-cut , P , P Q ; n-cut , , P Q ; ax N ; N ax foc Q ; Q N, N ; N , Q ; N Q foc , Q ; N , Q , N Q ; n-cut , N , N Q ; n-cut , , N Q ; ax ax N ; N M ; M foc foc N, N ; M, M ; N , M ; N M foc , M ; N , M , N M ; n-cut , N , N M ; n-cut , , N M ; , A ; , B ; & , A & B ; ax P ; P 1 P ; P B foc P , P B ; , P ; n-cut , P B ; ax N ; N foc N, N ; 1 N ; N B foc , N ; N , N B ; n-cut , N B ;

, N , Q , , N Q , N ;

, N , M , , N M , N ; , A , B & , A & B

, P 1 , P B

, N 1 , N B

1 ; 1 foc 1; ; , ; , ;

, ,

 ?, A ! ?, !A

?, A ; ! ? ; !A foc ?, !A ; ax P ; P ?d P , ?P ; , P ; n-cut , ?P ; , N ; ?d , ?N ; ; ?w , ?A ; , ?A, ?A ; ?c , ?A ;

, P ?d , ?P , N ?d , ?N ?w , ?A , ?A, ?A ?c , ?A

2.4

Focalization in LLfoc

If is a proof in LLfoc then is the LL proof obtained by erasing all the ; in the sequents. Proposition 1 (Cut-free weak +-focalization) If is a cut-free proof of ; in LLfoc then is a weakly +-focalized proof of , in LL. Corollary 1.1 (Weak +-focalization) If is provable in LL, is provable with a weakly +-focalized proof. Proof: Starting from a proof of , we translate it into the proof of ; in LLfoc . Using the cut elimination property given in appendix A (corollary 5.2), this leads to a cut-free proof of ; in LLfoc . By proposition 1, ( ) is a weakly +-focalized proof of in LL. 2

3
3.1

Focalized Linear Logic

Sequent calculus LLFoc

In the spirit of [5], we dene a sub-system LLFoc of LLfoc in which the strictly negative formulas in the context of positive rules must be atomic. There are two kinds of sequents: P, N ; and P, ?, X ; P where X contains only negative atoms. In order to simplify the notations we will write P, N ; for either a sequent with empty or a sequent with = P and N = ?, X . X ; X ax P, ?, X ; P foc P, ?, X , P ;

P, N , A, B ; ` P, N , A ` B ; P, ?, X ; P P , ? , X ; Q 6 P, ?, X ; P P , ? , X , M ;

P, P , ?, ? , X , X ; P Q

P, P , ?, ? , X , X ; P M

 P, ?, X , N ;

P , ? , X ; Q

P, P , ?, ? , X , X ; N Q

P, ?, X , N ;

P , ? , X , M ;

P, P , ?, ? , X , X ; N M

P, N , A ; P, N , B ; & P, N , A & B ; P, ?, X ; P 1 P, ?, X ; P B P, ?, X ; Q 2 P, ?, X ; A Q ; 1 ?, A ; ! ? ; !A 1 P, N ; P, N , ; P, ?, X ; P ?d P, ?, X , ?P ; P, ?, X , N ; 1 P, ?, X ; N B P, ?, X , M ; 2 P, ?, X ; A M P, N , ;

P, ?, X , N ; ?d P, ?, X , ?N ;

P, ?, X ; ?w P, ?, X , ?A ;

P, ?, X , ?A, ?A ; ?c P, ?, X , ?A ;

3.2

Embedding of LLfoc

We embed proofs of LLfoc of sequents of the shape ; into LLFoc . We proceed in two steps by rst showing that LLFoc sequents (that is of the shape P, N ; or P, ?, X ; P ) provable in LLfoc are provable in LLFoc enriched with the two rules: P, ?, X ; P ?w P, ?, X , ?A ; P P, ?, X , ?A, ?A ; P ?c P, ?, X , ?A ; P

The size of a sequent ; is the sum of the sizes of the formulas of plus one. The size of a sequent ; P is the sum of the sizes of the formulas of plus the size of P . We consider a cut-free proof with expanded axioms (according to section 2.2). We proceed by induction on the size of the nal sequent. We look at the last rule of the proof. The non trivial cases are the foc rules and ?d rules on positive formulas (called a positive ?d rule), the other rules are immediately valid in LLFoc (in our restricted case where their conclusion is an LLFoc sequent). We only consider the foc rule (the case of the positive ?d rule is very similar). If the context is of the shape P, ?, X , the rule is valid in LLFoc . Otherwise this context contains a formula with main connective `, &, or : If we have a context of the shape , A ` B, using the required proof of section 2.2 without the dashed rule, we build: , A ` B ; P A, B ; A B p-cut , A, B ; P foc , A, B, P ; 7

By induction hypothesis, after cut elimination and expansion of axioms in LLfoc , this proof gives a proof of LLFoc . We conclude with: , A, B, P ; ` , A ` B, P ; If we have a context of the shape , A & B, in the same way we build: , A & B ; P A ; A B p-cut , A ; P foc , A, P ; and , A & B ; P B ; A B p-cut , B ; P foc , B, P ; After cut elimination and expansion of axioms, we apply the induction hypothesis and we conclude with: , A, P ; , B, P ; & , A & B, P ; If we have a context of the shape , , we build: , ; P ; 1 p-cut ; P foc , P ; After cut elimination and expansion of axioms, we apply the induction hypothesis and we conclude with: , P ; , , P ; If we have a context of the shape , , we can immediately use the LLFoc proof: , , P ;

We now have to eliminate the additional ?w and ?c rules. We show by induction on the size of the proof that if P, ?, X ; P (resp. P, N ; ) is provable in LLFoc with the two additional rules then P, ? , X ; P (resp. P, N ; ) is provable in LLFoc with ? ? (where ? ? means that for each formula ?A in ? there is at least an occurrence of ?A in ?, or equivalently that we have an inclusion of the underlying sets). If the last rule is not one of the two additional rules and neither a foc rule nor a positive ?d rule, we translate the rule by itself. Otherwise: If the last rule is a foc rule or a positive ?d rule, we apply the corresponding rule and we apply the required ?w and ?c rules to move from ? to ?. In the case of the added ?w rule, by induction hypothesis, P, ? , X ; P is provable in LLFoc with ? ?, thus we have the result since ? ?, ?A. In the case of the added ?c rule, by induction hypothesis, P, ? , X ; P is provable in LLFoc with ? ?, ?A, ?A, thus we have the result since ? ?, ?A. This shows that if ; is provable in LLfoc , then it is also provable in LLFoc .

3.3

Focalization in LLFoc

If is a proof in LLFoc then is the LL proof obtained by erasing all the ; in the sequents. Proposition 2 (Cut-free focalization) If is a cut-free proof of P, N ; in LLFoc then is a focalized proof of P, N , in LL. Corollary 2.1 (Focalization) If is provable in LL, is provable with a focalized proof. Proof: Starting from a proof of , we translate it into the proof of ; in LLfoc and then into a cut-free proof of ; in LLFoc . By proposition 2, is a focalized proof of in LL. 2 Remark: It is possible to dene the embedding of LL proofs into LLFoc directly (without using LLfoc as an intermediary step). We have chosen to decompose it into two steps in order to show that the key property is weak +-focalization. Strong +-focalization is then obtained through reversing which is in general easy to do.

4
4.1

Additional remarks
Decomposition of exponentials

The attentive reader has certainly remarked that an hidden decomposition of the exponential connectives underlies the whole text (as suggested by Girard [5]): !A = A ?A = A

with A negative and A positive, and and are used as the same connectives of LL (see pol appendix A.2.1). 9

The corresponding rules in LLfoc would be: ?, A ; () ?, A ; () , A ; ; A ; P ; P ; P , P ; , N ; ; N , N ; , N ;

However it is dicult to give a meaning to A (resp. A) under any other connective than (resp. ).

4.2

Quantiers

Our method can perfectly be extended to quantication (of any order) by adding A (resp. A) in positive (resp. negative) formulas and the following rules in LLfoc : ; P ; P , N ; ; N , A ; , A ;

with free neither in nor in in the rule.

References
[1] Jean-Marc Andreoli. Proposition pour une synth`se des paradigmes de la programmation logique e et de la programmation par objets. Th`se de doctorat, Universit Paris VI, June 1990. e e [2] Vincent Danos, Jean-Baptiste Joinet, and Harold Schellinx. A new deconstructive logic: linear logic. Journal of Symbolic Logic, 62(3):755807, September 1997. [3] Jean-Yves Girard. Linear logic. Theoretical Computer Science, 50:1102, 1987. [4] Jean-Yves Girard. A new constructive logic: classical logic. Mathematical Structures in Computer Science, 1(3):255296, 1991. [5] Jean-Yves Girard. Locus solum: From the rules of logic to the logic of rules. Mathematical Structures in Computer Science, 11(3):301506, June 2001. [6] Olivier Laurent. Etude de la polarisation en logique. Th`se de doctorat, Universit Aixe e Marseille II, March 2002. [7] Myriam Quatrini and Lorenzo Tortora de Falco. Polarisation des preuves classiques et renversement. Compte-Rendu de lAcadmie des Sciences de Paris, 323:113116, 1996. e

10

A
A.1

Cut elimination in LLfoc

Cut elimination steps
ax P ; P , P ; p-cut , P ; ; P ax P ; P p-cut ; P

Key steps , P ;

; P

; P foc , P ; , P ; n-cut , , P ; , P , Q ; ; P ; Q ` , ; P Q , P ` Q ; p-cut , , ;

; P , P ; p-cut , , P ;

; P , P , M ; ; P , M ; ` , P ` M ; , ; P M p-cut , , ; ; P , N , M ; , N ; , M ; ` , N ` M ; , ; N M p-cut , , ;

; Q , P , Q ; p-cut , , P ; p-cut , , ;

, M ; , P , M ; n-cut , , P ; p-cut , , ;

, N ; , P ; , B ; & , P & B ; p-cut , ;

, M ; , N , M ; n-cut , , N ; n-cut , , ; ; P , P ; p-cut , ;

; P 1 ; P B

, N ; 1 ; N B

, N ; , B ; & , N & B ; p-cut , ;

, N ; , N ; n-cut , ;

11

; 1

; , ; p-cut ;

; N ?, N ; ?d ! , ?N ; ? ; !N p-cut ?, ; , P ; ?, P ; ?d ! , ?P ; ? ; !P p-cut ?, ; ; ?, A ; ?w ! , ?A ; ? ; !A p-cut ?, ; , ?A , ?A ;

?, N ; ; N p-cut ?, ;

?, P ; , P ; n-cut ?, ; ; ?w ?, ;

?, A ; ?c ! ; , ?A ? ; !A p-cut ?, ;

?, A ; ! ? ; !A , ?A , ?A ; ?, A ; p-cut ! ? ; !A ?, , ?A ; p-cut ?, ?, ; ?c ?, ;

Left commutative p-steps , A, B ; P ` , A ` B ; P , P ; p-cut , , A ` B ; , A, B ; P , P ; p-cut , , A, B ; ` , , A ` B ;

, A ; P , B ; P & , A & B ; P , P ; p-cut , , A & B ; , A ; P , P ; , B ; P , P ; p-cut p-cut , , A ; , , B ; & , , A & B ; ; P , ; P , P ; p-cut , , ; , ; P , P ; p-cut , , ; ; P ?w , ?A ; P , P ; p-cut , , ?A ; ; P , P ; p-cut , ; , , ; , , ; ; P

, P ; p-cut , ; ?w , , ?A ;

12

 , ?A, ?A ; P ?c , ?A ; P , P ; p-cut , , ?A ; Right commutative p-steps , P ; P foc ; P , P , P ; p-cut , , P ; , A, B, P ; ` ; P , A ` B, P ; p-cut , , A ` B ; , P ; P ; Q ; P Q , , P ; P p-cut , , ; P Q , P , N ; ; Q ; N Q ; P , , P p-cut , , ; N Q , N ; , P ; Q , , P ; N Q ; P p-cut , , ; N Q , P , N ; , M ; ; N M ; P , , P p-cut , , ; N M

, ?A, ?A ; P , P ; p-cut , , ?A, ?A ; ?c , , ?A ;

; P

, P ; P p-cut , ; P foc , , P ;

; P

, A, B, P ; p-cut , , A, B ; ` , , A ` B ;

; P

, P ; P p-cut , ; P ; Q Q , , ; P , P , N ; p-cut , , N ; ; Q , , ; N Q ; P

; P

, P ; Q p-cut , N ; , ; Q , , ; N Q ; P , P , N ; p-cut , , N ; , M ; , , ; N M

; P

, P , A ; , P , B ; & , P , A & B ; p-cut , , A & B ; ; P , P , B ; ; P , P , A ; p-cut p-cut , , A ; , , B ; & , , A & B ; , P ; P 1 ; P , P ; P B p-cut , ; P B , P , N ; 1 , P ; N B ; P p-cut , ; N B ; P , P ; P p-cut , ; P 1 , ; P B

; P

, P , N ; p-cut , , N ; 1 , ; N B

13

 , P ; , P , ; ; P p-cut , , ; ; P , , P ; p-cut , , ;

; P

, P ; p-cut , ; , , ; , , ;

?, ?A , B ; ! ?, ?A ; !B ? ; !A p-cut ?, ? ; !B , P ; P ?d ; P , P , ?P ; p-cut , , ?P ; , P , N ; ?d ; P , P , ?N ; p-cut , , ?N ; , P ; ?w , P , ?A ; ; P p-cut , , ?A ; , P , ?A, ?A ; ?c , P , ?A ; ; P p-cut , , ?A ; Commutative n-steps , P ; P foc , P, P ; , P ; n-cut , , P ; , A, B, P ; ` , A ` B, P ; , P ; n-cut , , A ` B ; , P ; P ; Q , , P ; P Q , P ; n-cut , , ; P Q , P, N ; ; Q , , P ; N Q , P ; n-cut , , ; N Q

? ; !A ?, ?A , B ; p-cut ?, ?, B ; ! ?, ? ; !B ; P , P ; P p-cut , ; P ?d , , ?P ; , P , N ; p-cut , , N ; ?d , , ?N ;

; P

; P

, P ; p-cut , ; ?w , , ?A ;

; P

, P , ?A, ?A ; p-cut , , ?A, ?A ; ?c , , ?A ;

, P ; P , P ; n-cut , ; P foc , , P ; , A, B, P ; , P ; n-cut , , A, B ; ` , , A ` B ; , P ; P , P ; n-cut , ; P ; Q Q , , ; P , P, N ; , P ; n-cut , , N ; ; Q , , ; N Q

14

 , N ; , P ; Q , , P ; N Q , P ; n-cut , , ; N Q , P, N ; , M ; , , P ; N M , P ; n-cut , , ; N M

, P ; Q , P ; n-cut , ; Q , N ; , , ; N Q

, P, N ; , P ; n-cut , , N ; , M ; , , ; N M , P, A ; , P, B ; & , P, A & B ; , P ; n-cut , , A & B ; , P, A ; , P ; , P, B ; , P ; n-cut n-cut , , A ; , , B ; & , , A & B ; , P ; P 1 , P ; P B , P ; n-cut , ; P B , P, N ; 1 , P ; N B , P ; n-cut , ; N B , P ; , P, ; , P ; n-cut , , ; , P, ; , P ; n-cut , , ; , P ; P ?d , P, ?P ; , P ; n-cut , , ?P ; , P, N ; ?d , P, ?N ; , P ; n-cut , , ?N ; , P ; ?w , P, ?A ; , P ; n-cut , , ?A ; , P ; P , P ; n-cut , ; P 1 , ; P B , P, N ; , P ; n-cut , , N ; 1 , ; N B , P ; , P ; n-cut , ; , , ; , , ;

, P ; P , P ; n-cut , ; P ?d , , ?P ; , P, N ; , P ; n-cut , , N ; ?d , , ?N ; , P ; , P ; n-cut , ; ?w , , ?A ;

15

 , P, ?A, ?A ; ?c , P, ?A ; , P ; n-cut , , ?A ;

, P, ?A, ?A ; , P ; n-cut , , ?A, ?A ; ?c , , ?A ;

A.2
A.2.1

Cut elimination property

LL pol

The system LL is obtained from LL by restricting formulas to the following grammar: pol P N ::= X | P P ::= X | N ` N | P P | N &N | 1 | 0 | !N | | | ?P | N | P

with the following rules for the and connectives: N,N N , N where N is a multi-set of negative formulas. A.2.2 LLP , P , P

The system LLP is obtained from LL by restricting formulas to the following grammar: P N ::= X | P P ::= X | N ` N | P P | N &N | 1 | 0 | !N | | | ?P

with the following generalizations of the exponential rules: N,N ! N , !N , P ?d , ?P ?w , N , N, N ?c , N

where N is a multi-set of negative formulas. Proposition 3 (Strong normalization) There is no innite sequence of reductions in LLP if we forbid commutations of cuts. Proof: Such a sequence of reductions can only contain nitely many steps between two steps that correspond to a reduction step in proof-nets. Thus by strong normalization for proof-nets [6] we can conclude. 2 Corollary 3.1 (Cut elimination) If is provable in LLP, then is provable without the cut rule. A.2.3 Simulations

The translation (.)! of LL into LLP is obtained by replacing N by !N and P by ?P and the two pol lifting rules by promotion and dereliction. Lemma 1 (Polarized formulas) If A is a positive (resp. negative) formula in LL then A! is a positive (resp. negative) formula in pol LLP. 16

Proposition 4 (One-to-one simulation) If reduces to in LL by one step of reduction then ! reduces to ! in LLP by one step of pol reduction. Corollary 4.1 (Strong normalization) There is no innite sequence of reductions in LL if we forbid commutations of cuts. pol Corollary 4.2 (Cut elimination) If is provable in LL , then is provable without the cut rule. pol The translation (.) of LLfoc into LL is obtained by adding to formulas exactly the required pol liftings to get a polarized formula, and by translating the sequent P, N ; by P , N , . In particular (!P ` 1) = !P ` 1. Proposition 5 (Strict simulation) If reduces to in LLfoc by one step of reduction then reduces to in LL by at least one pol step of reduction. Corollary 5.1 (Strong normalization) There is no innite sequence of reductions in LLfoc if we forbid commutations of cuts. Corollary 5.2 (Cut elimination) If ; is provable in LLfoc , then ; is provable without the cut rules. Proof: We consider a cut rule without any cut above it. We look at the two dierent cases: If it is a n-cut, we look at the rule above the premise , P ; . If the rule above it introduces P , it is either a foc rule or a rule and we apply the corresponding key step (and the n-cut becomes a p-cut) or commutative n-step. Otherwise this rule cannot be an ax rule, a 1 rule or a ! rule and we can apply the corresponding commutative n-step. If it is a p-cut, we rst look at the premise ; P . If P is not a main formula, we can apply a left commutative p-step. If P is a main formula and P is not, we can apply the corresponding right commutative p-step (notice that the rule above P cannot be a 1 rule). We just have to verify that we can apply the right commutative p-step in the case of a ! rule above P : since P is main, the rule above it is either an ax rule or a ! rule and we can apply the reduction step. If both P and P are main, we apply the corresponding key step. So that, either the proof is cut-free or a reduction step can be applied, and we conclude by strong normalization. 2

17