Survey
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
Decidability of Linear AÆne Logic A. P. Kopylov Department of Mathematics and Mechanics Moscow State University 119899, Moscow, Russia Abstract Propositional linear logic is known to be undecidable. In the current paper we prove that full propositional linear aÆne logic containing all the multiplicatives, additives, exponentials, and constants is decidable. The proof is based on a reduction of linear aÆne logic to sequents of specic \normal forms", and on a generalization of Kanovich computational interpretation of linear logic adapted to these \normal forms". 1 Introduction and Summary Linear logic has been introduced by Girard [1]. The undecidability of linear logic is established in [5]. Linear aÆne logic is linear logic with the weakening rule [6]. We will abbreviate linear logic and linear aÆne logic as LL and LLW correspondingly. The decidability problem for linear aÆne logic remained open. In the current paper the decidability proof for linear aÆne logic is based on the three results: rstly, the entire LLW is reduced (according to Turing) to its certain fragment (normal fragment). Secondly, the derivability of a normal form sequent is characterized in terms of vector games. Thirdly, by means of this computational interpretation, we prove that the normal fragment of LLW is decidable. At the same time we prove that the entire linear logic is also reduced to its normal fragment and the derivability in this fragment is characterized in terms of analogous games. The inference rules of linear logic are shown in the Table 1a, and the weakening rule is in the Table 1b. Theorem 1 (Girard) The cut rule is eliminated both in LL and in LLW. 2 Normal fragment and its computational interpretation Now we give the denition of the normal fragment. The normal fragment is an expansion of the -Horn The research described in this publication was made possible in part by Grant No. NFQ000 from the International Science Foundation fragment [2, 3]. Let us recall some denitions from [2]. Denitions A simple product is a tensor product of literals and the constant 1. (For example: 1, p, p q). A Horn implication is an implication of the form: X Æ Y , where X and Y are simple products. A -Horn implication is an implication of the form: X Æ (Y Z), where X, Y and Z are simple products. Denitions A simple disjunction is a disjunction of the form: X Y , where X and Y are simple products. Horn implications, -Horn implications, and simple disjunctions are called normal formulas . A normal sequent is a sequent of the form .... ...... ............. .. W; ! ) ?; where W is a simple product, is a multiset of normal formulas, and is a multiset of simple products. The normal fragment is a fragment containing only normal sequents. Let us give the denition of the interpretation of normal sequents. We begin with the remark: there is a simple correspondence between simple products and vectors with natural coordinates (cf. [3]). Namely, we can associate a n-dimensional vector (x1 ; x2 ; : : : ; x ) with a simple product p1 p2 : : : p n . Here p = p p : : : p (x times) and p0 = 1. If X is ~ denotes the corresponding a simple product, then X vector. Let = (W; ! ) ?) be a normal sequent. Let us consider the following two single-person games associated with this sequent. x1 x x2 n x n Game A ~ are written on the 1. Initially, all vectors from blackboard. 2. We may write new vectors with natural coordinates by the following rules: (I) (L? ) ) ) ? A A A (Cut) (R? ) A; ) ; ) (L ) ) ) ) (L ) ) ) ) (L Æ) Æ ) (L?) ?) ) (L1) )1 .............. ........ .. 1 ............. 1; A; 1 A B; 1 A ............ B; 1; A; (R&) (R) (R>) B; A B; B; A 0; 2 B; 2 2 B; 1; A;B; ............. A ............ B; B; B; ; ) ) ) & ) ) ) ) ) A; &B; A ) 2 2 2 A A 2 ; (L&) (L) (L0) 1 2 A; 2 1; 1; ; A; ... ....... . ........... ... 2 2 2 A 1; 2 1; B; 1 B; 2 2 A; 2 A; 1 B; A; 1 ;A 1; ) ) (R ) ) (R ) )) (R Æ) ) )Æ (R?) ? )) (R1) )1 A;B; A ) ) )? ) 1 B; ) ) ) & ) ) ) ) )> A; B; A B; A; A B; B; A B; U; 2 ; ) ) ? (R!) !! ) ) ! ? ! ! ) ) (C!) ! ) ! ) ) ! ) ? (R?) (L?) ? ! )? )? )? ? (W ?) ))? (C?) )? Table 1a: The inference rules of LL. (L!) (W !) A; A; !A; A; A; A; A; A; A; A; A; A; A; A; A; (W ) (I) (L ) (H) (H ) (H ) ............... ....... .. (L!) (W !) (C!) (W ) A; ) ) ; where 0 ; 0 Table 1b: The weakening rule for LLW. 0 0 (Cut) (M) ) ) ~ = U: ~ ) ; where W W W W; U; 1 ;U 1 W; W; W V; 1; 2 Z; Z V; ) Æ ) ) ) Æ( ) ) ) ) ) Y X ) ) ) ) W; U; U;X Y1 Y; U; X Y1 Y2 U;X U1 ; U1 1 Y1 1 ............. U; Y2 ; U2 ;Y1 ............. Y2 ; Y2 1; U2 ; 2 2 1; 2 2 ) (R?) ) ) (W ?) ! ) ! ! ) (C?) ! ) Table 2a: The inference rules of W;A; W; ! W; A; W; W; ;Z ; Z W; W; A; W; W; A; A; W; W; A; W; ; Z ; Z; Z ; Z ) ) ; where 0 ; 0 ) ; ) Table 2b: The weakening rules for NLLW. W; W ) ) ? ) ) ? ) ? ? ) ? NLL. U; W; W; 0 0 1; 2 ) 2 (a) If X Æ Y 2 and a vector a + Y~ has been already written, for some vector a with nat~ ural coordinates, then we may write a + X. (b) If X Æ (Y1 Y2 ) 2 and vectors a + Y~1 and a + Y~2 have been already written for some ~ a 2 ! , then we may write a + X. ~ (c) If Y1 Y2 2 and vectors a1 + Y1 and a2 + Y~2 have been already written for some a1 ; a2 2 ! , then we may write a1 + a2 . ~. 3. The aim of the game is to obtain the vector W n ............... ......... .. n Game B This game is the game A with the additional rule: 4. If a vector a has been written and a c then1 we may write c. In the current section we prove the following Theorem 2 (The computational interpretation) It is possible to reach the aim in the game A (B ), if and only if the sequent is derivable in LL (LLW). Proof To prove the theorem, we introduce auxiliary logics NLL and NLLW (normal linear logic and normal linear logic with the weakening rule). These logics are the development of calculus HLL, which was used in [2] for justication of the computational interpretation of the -Horn fragment of linear logic. The inference rules of NLL are represented in the Table 2a. And NLLW is NLL with the weakening rules (Table 2b). In the tables 2a, 2b U; V; W; X; Y; Z are simple products; ; 1 ; 2 are multisets of normal formulas and normal formulas with \!"; ; 1 ; 2 are multisets of simple products and simple products with \?"; and A is a normal formula. We dene the NLL and NLLW so that the set of normal sequents which are derivable in NLL (NLLW) is equal to the set of normal sequents which are derivable in LL (LLW). In order to verify it we use the following four lemmas. Lemma 2.1 If = P1 : : : P is a multiset of simple products and the sequent n ; ) is derivable in LL (LLW), then the sequent ; ) 1 We say that (x1 : : : xn ) (y1 : : : yn ) i 8 i (xi yi ). Lemma 2.3 Let the sequent ~ W; ; ! ~ ) ; ? (2) be derivable in NLL (NLLW) without the rule (M). ~ Here ; ~ are multisets of normal formulas and ; ~ are multisets of simple products. If = (W; ! ; ! ) ~ , then it is possible to reach the aim in the ?; ?) game A (B ). Lemma 2.4 Let be a normal sequent. Assume ~ by the rules of the that one can write the vector W game A (B ). Then the sequent is derivable in LL (LLW). It follows from the lemmas 2.1{2.4 that the following three assertions are equivalent: LL ` , NLL ` , It is possible to reach the aim in the game A . And by analogy for linear aÆne logic the following three assertions are equivalent: ` , NLLW ` , LLW It is possible to reach the aim in the game B . Hence, the theorem 2 is following from the lemmas 2.1{2.4. Now let us prove these lemmas by induction. For short, we pove these lemmas only for linear aÆne logic. In order to obtain the proofs for linear logic we need only except cases of weakening. Proof of Lemma 2.1 We prove the lemma by induction on the cut-free derivation of the sequent ; ) in LLW. Case 1. The last rule in the derivation is (L ). In this case = P1 ; P2 ; 0 and the derivation has the form: P1 ; P2 ; 0 ; ) ( ) P1 P2 ; 0 ; ) Hence, since (P1 ; P2 ; 0 ) = (P1 P2 ; 0 ) , then, by the induction hypothesis, the sequent (1) is derivable in NLLW. Case 2. The last rule in the derivation is (R ). L (1) is derivable in NLL (NLLW), where = P1 : : : P . n Lemma 2.2 We can eliminate the rule (M) both in NLL and NLLW. 1; ; 1 2 ) Z 1 ; 1 2 ; 2 ) Z2 ; 2 ( ) ( Z1 1 ) Z1 2 ; 1 2 ; 2 ) Z1 Z2 ; 2 ( 1 2 ; 1 ; 2 ) Z1 Z2 ; 1 ; 2 ( ) (1 ; 2 ) ; 1 ; 2 ) Z1 Z2 ; 1 ; 2 1 M M ) ) C ut L Table 3a: Case 2. 1; ; 1 2 ) X; 1 ( ) 1 ) X 2 ; 1 1 2 ; X Æ Y; (1 ; 2 ) ; X Æ Y; 1 2 ; 2 ) 2 ( X 2 ; X Æ Y; 2 ) 2 ( 2 ) 1 ; 2 ( ) 2 ) 1 ; 2 Y M 1; 1; H ) C ut ) L Table 3b: Case 3.1. 1; ; 1 2 ) X; 1 Y1 Y2 2 ; 2 ) 2 2 ; 2 ) 2 ( ) ( ) X ; X Æ (Y Y ); ) 1 ) X 2 ; 1 2 2 2 2 ( ) 1 2 ; X Æ (Y1 Y2 ); 1 ; 2 ) 1 ; 2 ( ) (1 ; 2 ) ; X Æ (Y1 Y2 ); 1 ; 2 ) 1 ; 2 1 M H C ut L Table 3c: Case 3.2. Table 3: The proof of the lemma 2.1 Case 5. The last rule in the derivation is (L1): In this case the derivation has the form: 1 ; 1 ) Z1 ; 1 2 ; 2 ) Z2 ; 2 1 ; 2 ; 1 ; 2 (R ) ) Z1 Z2 ; 1 ; 2 Hence, the sequent (1) is derivable in NLLW (see the Table 3a). Case 3. The last rule in the derivation is (L Æ). There are two subcases: Case 3.1. The derivation has the form: ) X; 1 1 ; 2 ; X Æ Y; 1 ; Y; 2 ; 1 1; 2 ) 2 2 ) 1 ; 2 (L Æ) Where Y is a simple product. In this case, the sequent (1) is derivable in NLLW (see the Table 3b). Case 3.2. The derivation has the form: ) X; 1 2 ; Y1 Y2 ; 2 ) 2 ( Æ) 1 ; 2 ; X Æ (Y1 Y2 ); 1 ; 2 ) 1 ; 2 1 ; 1 L ) 2 2 2 ) 2 ; Y2 ; 2 ; 2 ) 2 : Moreover, the derivation length of each of these sequents is shorter than the derivation length of sequent Y1 Y2 ; 2 ; 2 ) 2 . Hence, we can apply the induction hypothesis to these sequents. Therefore, the following sequents are derivable in NLLW: Y1 2 ; 2 ) 2 ; Y2 2 ; 2 ) 2 ; ; ) X; 1 : 1 1 The sequent (1) is derivable from them in NLLW (see the Table 3c). Case 4. The last rule in the derivation is (L ): ........ .... ..... .. .. ... Y1 ; 1 ; 1 ) 1 Y2 ; 2 ; 1 ; 2 ; Y1 Y2 1 ; ................... ...... 2 2 ) 2 ) 1 ; 2 (L......................... ) Hence, the sequent (1) is derivable in NLLW: Y1 ; 1 ) 1 Y2 ; 2 ) 2 1 1 2 ; Y1 Y2 ; 1 ; (1 ; 2 ) ; Y1 Y2 ; 1 ; ................. ....... ............... ....... .. 0 (1; 0 ) ; (L ) ) Case 6. Our sequent is an axiom (R1) : ) 1. Then, = ;, = 1, and the sequent (1) is an axiom in NLLW: 1 ) 1. Case 7. The last rule in the derivation is (W ): ) 0 ( ) 0 ; 1 ; ) 0 and 0 . Hence, the sequent (1) is 0 ; 0 W where derivable in NLLW: 0 is derivable in LLW then the following sequents are also derivable in LLW: Y1 ; 2 ; L 0 ; 0 ) 0 ; ) One can prove that if the sequent Y1 Y2 ; 2 ; 0 ) ( 1) 1; 0 ) Hence, since 0 = (1; 0 ) , the sequent (1) is derivable in NLLW: ; ) 2 ) 1 ; 2 ( ) ) ; 2 1 2 2 L (0 ; 1 ) ; ) (W ) (L ) Case 8. The last rule in the derivation is one of the rules: (L!), (W!), (C!), (R?), (W?), or (C?), then the sequent (1) is derivable by the same rule in NLLW. Proof of Lemma 2.2 Let the sequent W; ) Z; ; (3) be derivable in NLLW without the rule (M). We shall prove, by induction on derivation length, that the sequent W V; ) Z V; is also derivable without the rule (M). Case 1. The sequent (3) is an axiom in NLLW: W ) W . Then the sequent W V ) W V is also an axiom. Case 2. The sequent (3) is derivable by the rule (L ) in NLLW: W; U; (H .......................... ) 1 ) Z; ( ) ) Z; L ~ = U. ~ By the induction hypothesis the followHere W ing sequent is derivable in NLLW: W V; ) Z V; : The sequent U V; ) Z V; is derivable from it by the rule (L ) in NLLW, since W ! V = U ! V. Case 3. The sequent (3) is derivable by the rule (H): U; 0 ) Z; ( X U; X Æ Y; 0 ) Z; Y H ) By the induction hypothesis the following sequent is derivable: Y U V; 0 ) Z V; : The sequent X U V; X Æ Y; 0 ) Z V; is derivable from it by the rule (H). Case 4. The sequent (3) is derivable by the rule (H ): Y1 U; ) Z; Y2 U; 0 ) Z; ( ) X U; X Æ (Y1 Y2 ); 0 ) Z; 0 H By the induction hypothesis the following sequents are derivable: Y1 U V; Y2 U V; ) Z V; ; 0 ) Z V; : 0 The sequent X U V; X Æ (Y1 Y2 ); 0 ) Z V; is derivable from them by the rule (H ). Case 5. The sequent (3) is derivable by the rule (H ): ............... .......... Y1 U1 ; 1 ) Z; 1 U1 U2 ; Y1 Y2 ; 1 ; Y2 U2 ; .................. ....... 2 2 ) 2 ) Z; 1 ; 2 (H ......................... ) By the induction hypothesis the following sequent is derivable: Y1 U1 V; 1 ) Z V; 1 : The sequent U1 U2 V; Y1 Y2 ; 1 ; ) Z V; 1 ; 2 is derivable from it and from the sequent Y2 U; 2 ) ............ ........ ... 2 by the rule (H ). .................. ....... 2 Case 6. The sequent (3) is derivable by the rule (Cut). Case 6.1. The derivation has the form: W; 1 ) Z; 1 ; U U; 2 ) 2 W; 1 ; 2 (C ut) ) Z; 1 ; 2 By the induction hypothesis the following sequent is derivable W V; 1 ) Z V; 1 ; U: The sequent W V; 1 ; 2 ) Z V; 1 ; 2 is derivable from it and from the sequent U; by the rule (Cut). Case 6.2. The derivation has the form: W; 1 ) 1 ; U U; 2 ) Z; 2 W; 1 ; 2 ) Z; 1 ; 2 2 ) 2 , (C ut) By the induction hypothesis the following sequents are derivable W V; 1 ) 1 ; U V; U V; 2 ) 2 ; Z V: The sequent W V; 1 ; 2 ) Z V; 1 ; 2 is derivable from them by the rule (Cut). Case 7. The sequent (3) is derivable by the rule (W). Case 7.1. The derivation has the form: W; ) ( ) W; ) Z; W In this case the sequent W V; able: W; ) ) Z V; is deriv(W ) W V; ) ( ) W V; ) Z V; Case 7.2. If the derivation has some other form, then the rule (M) commutes with the rule (W). Hence, by the induction hypothesis, the sequent W V; ) Z V; is derivable without the rule (M). Case 8. The sequent (3) is derivable by the rule (L!), (W!), (C!), (R?), (W?), or (C?). Note, that the rule (M) commutes with all these rules. Hence, by the induction hypothesis, the sequent W V; ) Z V; is derivable without the rule (M). W Y1 U; 0; ! ~ ) ; ?~ Y2 U; 0; ! ~ ) ; ?~ ( ) ~ X U; X Æ (Y1 Y2 ); 0 ; ! ~ ) ; ? H Table 4a: Case 4. ~1 Y1 U1 ; 1 ; ! ~ 1 ) 1 ; ? U1 U2 ; Y1 .................. ....... ~2 Y2 U2 ; 2 ; ! ~ 2 ) 2 ; ? ~ 1 ; ? ~2 Y2 ; 1 ; 2 ; ! ~ 1 ; ! ~ 2 ) 1 ; 2 ; ? (H ........................ ) Table 4b: Case 5. ~ 1; U ~2 W; 1 ; ! ~ 1 ) 1 ; ? U; 2 ; ! ~ 2 ) 2 ; ? ~ 1 ; ? ~2 W; 1 ; 2 ; ! ~ 1 ; ! ~ 2 ) 1 ; 2 ; ? (C ut) Table 4c: Case 7. Table 4: The cases of the lemma 2.3 Proof of Lemma 2.3 We prove this lemma by the induction on the length of a derivation of the sequent (2). Case 1. The sequent (2) is an axiom W ) W . This case is trivial. Case 2. The sequent (2) is derivable by the rule (L ). This case is also trivial. Case 3. The sequent (2) is derivable by the rule (H): Y U; X U; X 0; !~ Æ Y; ) ; ?~ ( ~ 0 ; ! ~ ) ; ? H ~0 W; 0 ; ! ~ 0 ) 0 ; ? ~ W U; ; ! ~ ) ; ? (W ) ~ by rules of the Similarly, one can write the vector W game B . Therefore, one can also write the vector ~ + U~ by the rules of this game (the rule (d)). W Case 7. The sequent (2) is derivable by the rule (Cut) (see the Table 4c). Consider the sequents ~ 1 ; ?U) 1 = (W; ! 1 ; ! ~ 1 ) ?1 ; ? ) ~ Consider the sequent = (Y U; ! 0 ; ! ~ ) ?; ?). By the induction hypothesis one can write the vector Y~ + U~ by the rules of the game B . Hence, one can write it by the rules of the game B too. Therefore, ~ + U~ by the rules of one can also write the vector X this game (the rule (a)). Case 4. The sequent (2) is derivable by the rule (H ) (see the Table 4a). Similarly to the previous case, one can write Y~1 + U~ and Y~2 + U~ by the rules of ~ + U~ by the game B . Therefore, one can also write X the rules of this game (the rule (b)). Case 5. The sequent (2) is derivable by the rule (H ) (see the Table 4b). Similarly, one can write the vectors Y~1 + U~ 1 and Y~2 + U~ 2 by rules of the game B . Therefore, one can also write the vector U~ 1 + U~ 2 by the rules of this game (the rule (c)). Case 6. The sequent (2) is derivable by the rule .... ......... ......... .. (W ): and ~ 2 ): 2 = (U; ! 2 ; ! ~ 2 ) ?2 ; ? By the induction hypothesis one can write the vector ~ by the rules of the game B and the vector U~ by W the rules of the game B . Hence, if one can write a vector by the rules of the game B , then one can write the same vector by the rules of the game B . In ~ by the rules of particular, one can write the vector W the game B . The other cases ((L!); (W !); (C!); (R?); (W ?); (C?)) are trivial. 1 2 1 Proof of Lemma 2.4 We prove this lemma by the induction on the number of moves, which are necessary ~ . If W ~ can be obtained at zero to obtain the vector W moves (i.e. W 2 ), then, apparently, the sequent W; ! ) ? is derivable. Otherwise, four cases may occur. U Y; ! ) ? U X; X Æ Y ) U Y U X; X Æ Y; ! ) ? ( ! !) U X; ! ) ? (C ut) L ;C Table 5a: Case 1. U; Y1 ; ! ) ? U; Y2 ; ! ) ? ( ) U; Y1 Y2 ; ! ) ? X; X Æ (Y1 Y2 ) ) Y1 Y2 U; X; X Æ (Y1 Y2 ); ! ) ? ( ) U X; X Æ (Y1 Y2 ); ! ) ? ( ! !) U X; ! ) ? L (C ut) L L ;C Table 5b: Case 2. Table 5: The proof of the lemma 2.4 ~ is obtained by the rule (a) from V~ . In Case 1. W this case W = U X, V =U Y, X ÆY 2 . By the induction hypothesis, the sequent V; ! ) ? is derivable in LLW. And the sequent W; ! ) ? is derivable from it in LLW (see the Table 5a). ~ is obtained by the rule (b) from V~1 and Case 2. W ~V2 . In this case W = U X, V 1 = U Y1 , V 2 = U Y2 , X Æ (Y1 Y2 ) 2 . By the induction hypothesis, the sequents V1 ; ! ) ? and V2 ; ! ) ? are derivable in LLW. Then, the following sequents are derivable: U; Y1; ! U; Y2; ! ) ?; ) ?; and the sequent W; ! ) ? is derivable from them in LLW (see the Table 5b). ~ is obtained by the rule (c) from V~1 and Case 3. W ~V2 . In this case W = U1 U2 , V1 = U1 Y1 , V2 = U2 Y2 , Y1 Y2 2 . By the induction hypothesis, the sequents V1 ; ! ) ? and V2 ; ! ) ? are derivable in LLW. Then the ................... . ...... following sequents are derivable: U1 ; Y 1 ; ! ) ?; ) ?; U2 ; Y 2 ; ! and the sequent W; ! ) ? is derivable from them in LLW: U1 ; Y1 ; ! ) ? U2 ; Y2 ; ! ) ? (L......................... ) ) ?; ? ( U1 U2 ; Y1 Y2 ; ! ) ? ( ! U1 U2 ; ! ) ? U1 ; U2 ; Y1 Y2 ; ! ; ! .... . ........ ......... ... L ............... ....... ... ! ;C ;C L ;C ?) !) ~ is obtained by the rule (d) from V~ . In Case 4. W this case W = V U. By the induction hypothesis, the sequent V; ! ) ? is derivable in LLW, and the sequent W; ! ) ? is derivable from it in LLW: V; ! ) ? (W ) ) ? ( ) V U; ! ) ? V; U; ! 3 L The decidability of the normal fragment of LLW Theorem 3 The normal fragment of linear aÆne logic is decidable. We begin with Denitions 1. The set of vectors implication X Æ Y if 8a 2 ! n a + Y~ A is closed under the Horn z 2 A ) a + X~ 2 A: (Here ! is the set of vectors with natural coordinates). 2. The set of vectors A is closed under the -Horn implication X Æ (Y1 Y2 ) if n 8a 2 ! a + Y~1 2 A; a + Y~2 2 A ) a + X~ 2 A: n 3. The set of vectors A is closed under the simple disjunction Y1 Y2 if ......... .... . .... . .. .... 8a1; a2 2 ! n Denition A cone with the vertex z is a set of vectors K = fy : y z g. Note, that any set of vectors, which is closed under the weakening rule is a union of cones. We shall prove that such set is a union of a nite set of cones. This fact follows from the central lemma: a1 + Y~1 2 A; a2 + Y~2 2 A ) a1 +a2 2 A Lemma 3.2 Any set of vectors from ! which are incomparable pairwise is nite. n Proof Let A ! and 8x; y 2 A (x 6< y). We shall prove that A is nite by induction on n. For n = 1 it is trivial. To verify the induction step, we consider any vector a 2 A. Let a = (a1 ; : : : ; a ). We dene the following sets: n n A = fx 2 A : x a g: 4. Let be a multiset of normal formulas. The set of vectors A is -closed if it is closed under all formulas from . 5. The set of vectors A is closed under the weakening rule if 8a 2 A 8c a c 2 A: One can easily verify that dene the sets Lemma 3.1 Let = (W; ! ` ?) be a normal sequent. Then the following two conditions are equivalent: . By the induction hypothesis, the sets A are nite. Hence, A is nite too. (i) It is possible to reach the aim in the game B . (ii) For any A ! if A is -closed and closed under ~ A then W ~ 2 A. the weakening rule and n Proof (ii)=)(i) Let us consider the set A = fx : x can be obtained by the rules of B g: By the denition of the game B and closed sets, the set A is -closed and closed under the weakening rule. ~ Hence, according to Moreover, it includes the set . ~ 2 A. So, W ~ can be obtained by our hypothesis, W the rules of the game B . (i)=)(ii) Let A be any -closed set which is closed ~ A. Then, by the under the weakening rule, and denition of the game B and closed sets, A includes all vectors, which can be written by the rules of the ~ 2 A. game B . In particular, W Remark It is possible to formulate the analogous lemma for the game A . A i ij i A i = S A. =1 n i i Now we = fx 2 A : x = j g: i i S It is evident, that A = =1 A ai i ij j ij Corollary 3.3 Let A be a set of vectors with natural coordinates. In this case if A is closed under the weakening rule, then A is a union of nite set of cones. Proof Let B be a set of the minimal elements of the set A, i.e. B = fz 2 A : :9y 2 A Then [ A= 2B According to 3.2, B is nite. y < z g: K: z z Denition Let x; y 2 ! , then n max(x; y) := (max(x1 ; y1 ); : : : ; max(x ; y )): n n Denition Let A; B Z , x 2 Z (here Z is a set of integers), then n n A + x := fa + x : a 2 Ag; A + B := fa + b : a 2 A; b 2 Bg: S Lemma 3.4 Let A = 2B K , then z z 1. 2. A is closed under X Æ Y if and only if 8z 2 B max(z Y~ ; 0) + X~ 2 A: A is closed under X Æ (Y1 Y2 ) if and only if 8z1; z2 2 B 3. i.e. A is closed under Y1 8z1; z2 2 B max(z1 ................ ........ .. Y2 if and only if n Y~1 ; 0)+max(z2 Y~2 ; 0) 2 A ; a + Y~ ÆY n z z n ~ Y max z z ~; Y ~; Y ~ X [ z i.e. max z z K ( max z 2B 8z 2 B max(z 0)+X~ ~; Y A; ~ 2 A: Y~ ; 0) + X 2. The situation is similar to the previous item. is closed under X Æ (Y1 Y2 ) if and only if fa + X~ : a 2 w n A ; a + Y~1 2 A; a + Y~2 2 Ag A: But fa + X~ : a 2 w ; a + Y~1 2 A; a + Y~2 2 Ag = (A Y~1 ) \ (A Y~2 ) \ ! + X~ = S S ( K Y~1 ) \ ( K Y~2 ) \ ! + X~ = 2B 2B S S ~= ( K )\( K )\! +X 2B 2B S K \K \ ! + X~ = 2B S K ( 0)+ : 2B So, A is closed under X Æ (Y1 Y2 ) if and only if n n z z z z z1 ;z2 z1 ;z2 n z z ~ Y 1 z1 ~ Y 1 max z1 [ z1 ;z2 2B z2 ~ ;z Y 1 2 K n ~ Y 2 z z ~ Y 2 ~ ; Y 2 ( max z1 n ~ X ~ ;z Y 1 2 ~ ; Y 2 0)+X~ A; n z n z z ~ ; Y 1 max z z1 ;z2 2B max z z ~ ; Y 1 max z1 [ n z n z n z z n z1 ;z2 But fa + X~ : a 2 w ; a + Y~ 2 Ag = (A Y~ ) \ ! + X~ = S ~ = S [(K Y~ ) \ ! + X] ~ = ( K Y~ ) \ ! + X 2B 2B S [K \ ! + X] ~ = S [K ( ~ 0) + X] = 2B 2B SK ( 0)+ : 2B So, A is closed under X Æ Y if and only if z But fa1 + a2 : a1 ; a2 2 w ; a1 + Y~1 2 A; a2 + Y~2 2 Ag = (A Y~1 ) \ ! + (A Y~2 ) \ ! = S S ( K Y~1 ) \ ! + ( K Y~2 ) \ ! = S2BK S K2B + ( 0) ( 0) = 2B 2B S K ( 0)+ ( 0) : 2B So, A is closed under Y1 Y2 if and only if z 2 Ag A: n n n Proof 1. By the denition, A is closed under X if and only if fa + X~ : a 2 w .... ........ ........... .. ~ 2 A: Y~2 ; 0) + X Y~1 ; z2 max(z1 8z1; z2 2 B max(z1 Y~1 ; z2 Y~2 ; 0) + X~ 2 A: 3. A is closed under Y1 Y2 if and only if fa1 + a2 : a1 ; a2 2 w ; a1 + Y~1 2 A; a2 + Y~2 2 Ag A: ~ ; Y 2 ~ ; Y 2 max z2 ............... ......... .. K ( max z1 ~ ; Y 1 0)+max(z2 ~ ; Y 2 0) A; i.e. 8z1; z2 2 B max(z1 Y~1 ; 0)+max(z2 Y~2 ; 0) 2 A: Corollary 3.5 For any nite S set of vectors B the property of the set A = 2B K to be -closed is decidable. z z Proof Note that all conditions in the last lemma are decidable. Hence, the property to be -closed is also decidable. Now we can prove the theorem 3. Let us look at the condition (ii) from the lemma 3.1. By the corollary 3.3 this condition is equivalent to the following condition: S (iii) For any nite set B ! if 2B K is -closed ~ S 2B K then W ~ 2 S 2B K . and n z z z z z z By the corollary 3.5 the condition standing after the words \For any nite set B ! . . . " is decidable. Hence, the condition (iii) is recursively co-enumerable. By the lemma 3.1 the condition (iii) is equivalent to the condition (i). And by the theorem 2 the condition (i), in its turn, is equivalent to the condition, that LLW ` . Hence, the set of derivable normal sequents in LLW is co-enumerable. On the other hand it is clear that this set is recursively enumerable. So, by Post's theorem, it is decidable. n 4 Reduction to the normal fragment Theorem 4 Linear logic and linear aÆne logic are reduced to their normal fragments. ! ;A ) B ! ; !A ) B ! ; !A ) !B (L!) (R!) Proof of ! ' ! A ! ;A ) B ! ; A ) ?B ! ; ?A ) ?B (R?) (L?) A ! ; A? ) B ? Proof of A ? R Proof of ' B Proof of ? ' ? ! ;B ) A ! ; A? ; B ) ! ;A ) B C)C ( ) ! ;A ) B C ! ;C ) B C ! ;A C ) B C A (L? ) (R? ) B B ? (L) L (L&; W !) (L&) Proof of & ' & A C B ! ;A ) B C ) C ( ) ! ; A; C ) B C ( ) ! ;A C ) B C B C Æ ' A C Æ L R A Æ ' C ................... ...... Proof of C B ! ;B ) A C ) C ( Æ) ! ; B; A Æ C ) C ( Æ) ! ;A ÆC ) B ÆC B Æ ...... .... ........... .... L C R ! ;A ) B C ) C ( ! ; A C ) B; C ( ! ;A C ) B C R A L Proof of C Proof of ' ! ;A ) B C ) C ( Æ) ! ; C; C Æ A ) B ( Æ) ! ;C ÆA ) C ÆB Proof of C ! ;A ) B C)C ( &) ! ; A&C ) B ! ; A&C ) C ! ; A&C ) B&C B ' C (R; W !) ........... ....... ........ .... ...... ..... A ............ C ' C ............... L ............ ................. ........ R ) ) .... ...... .... B ............. C Table 6: The proof of lemma 4.1.1 Denition A \good" sequent is a sequent of the form: ; ! ` , where , and contain neither \!" nor \?". The proof of the theorem 4 consists of two steps. Firstly, we prove that the \good" fragment is reduced to the normal fragment (Lemma 4.1). And secondly, we prove that the whole LL and LLW are reduced to their \good" fragment (Lemma 4.2). Lemma 4.1 For any \good" sequent one can eectively construct a normal sequent such that LLW ` i LLW ` ; LL ` i LL ` : In order to prove the lemma 4.1 we introduce Denition Let A and B be formulas, and be a multiset of formulas. We say that A ' B in any logic (A is equal to B according to ), if the following sequents are derivable in the logic: ! ; A ) B; ! ; B ) A: If is an empty multiset, then we write A = B. By analogous to the traditional equality the following lemma is holds for LL as well as for LLW: Lemma 4.1.1 If A ' B then !A ' !B; A C ' B C; C Æ A ' C Æ B; ?A ' ?B; A&C ' B&C; A Æ C ' B Æ C; ? ? A ' B ; A C ' B C; A C ' B C: .................. ....... ..... ...... ........... ... Proof Let the sequents ! ; A ) B; ! ;B ) A be derivable. We must prove the derivability of the sequents of the form ! ; D(A) ) D(B) and of the form ! ; D(B) ) D(A), where D(p) = !p, p C, C Æ p, and so on. The derivations of the rst sequents are represented in the Table 6. The derivations of the second sequents are obtained by replacement A by B and B by A in the derivations in the Table 6. Lemma 4.1.2 If A ' B then the sequent ! ; ) is derivable if and only if the sequent ! ; 0 ) 0 is derivable, where 0 stands for substitution of the formula B for some occurrences of the formula A. Proof We may assume without loss of generality that 0 stands for substitution of the formula B for only one occurrence of the formula A. Let us suppose that the formula A occurs in the external formula C = C(A) (i.e., C 2 or C 2 ). Then C 0 = C(B) and according to the last lemma C ' C 0 . Case 1. C 2 , i.e. = C; 0 ; 0 = C 0 ; 0 ; 0 = . Then the sequent ! ; 0 ) 0 is derivable from the sequent ! ; ) : ! ; C; 0 ) ! ; C 0 ) C ! ; ! ; C 0 ; 0 ) ! ; C 0 ; 0 ) (C !) (C ut) (4) Case 2. C 2 , i.e. = C; 0 ; 0 = C 0 ; 0 ; 0 = . Then the sequent ! ; 0 ) 0 is derivable from the sequent ! ; ) : ! ; C ) C0 ! ; ! ; ) C 0 ; 0 ( !) ! ; ) C 0 ; 0 ! ; ) C; 0 (C ut) C (5) Replace the formula C by the formula C 0 and the formula C 0 by the formula C in the derivations (4) and (5). We obtain the derivation of the sequent ! ; ) from the sequent ! ; 0 ) 0 . Lemma 4.1.3 The sequent !(A Æ p); !(p Æ A); ) is derivable, where 0 stands for substitution of the formula A for all occurrences of the literal p. =A Æ p; p Æ A: Then A ' p. Hence, if the sequent 0 ) 0 is derivable, then the sequent ! ; ) is derivable too (by the rule (W !) and the last lemma). Conversely, suppose that the sequent !(A Æp); !(p Æ A); ) is derivable. If we substitute the formula A for the literal p in this sequent, then we obtain the derivable sequent: !(A Æ A); !(A Æ A); 0 ) 0 : The sequent 0 ) 0 is derivable from this sequent: !(A Æ A); !(A Æ A); 0 ) 0 ( !(A Æ A); 0 ) 0 0 ) 0 C !) .... ........... . ..... ... !(r !(r !(r !(r Æ (p q)); Æ (p q)); Æ p?); Æ 1); !((p q) Æ r); !((p q) Æ r); !(p? Æ r); !(1 Æ r): Table 7. After all these replacements we obtain a \good" sequent of the form is derivable if and only if the sequent 0 ) 0 Proof Let Namely, we replace the formulas A&B by (A? B ? )? , A B by (A? B ? )? , A Æ B by (A B ? )? , ? by 1?, and 0 by >?. Secondly, using that 1 = > in LLW, we replace > by 1. After that we obtain a \good" sequent 1 which contains only the following connectives: !, , , ? , and 1. Then, using the lemma 4.1.3, we may replace all formulas of the forms p q, p q, p? and 1 by new literals and add formulas of the form from the Table 7 to the antecedent. ) !(A Æ A) Using this lemma, let us prove the lemma 4.1 for LLW. Let we have a \good" sequent . We shall transform this sequent, watching that the new sequent will be always derivable in LLW if and only if the old one is derivable too. Firstly, by De Morgan equalities, we eliminate from our sequent the following connectives: &, , Æ, ?, 0. .................. ...... 1 ; 2 ; 3 ) where 1 is a multiset of the formulas of the kind from the Table 7. 2 is a multiset of new literals, and 3 is a multiset of the formulas of the kind !p, where p is a new literal. Then we transfer formulas from 1 and 3 to equivalent normal formulas. Namely, since p = (1 Æ p), we may replace formulas !p from 3 by the Horn implication !(1 Æ p). Since p? Æ q = p q, we may replace !(p? Æ q) by the simple disjunction !(p q). And using that !((p q) Æ r) = !(p Æ r) !(q Æ r), we may replace the formulas of the form !((p q) Æ r) by the pair of the Horn implications !(p Æ r) and !(q Æ r). Finally, since !(r Æ p? ) = (?(r p))? , we may remove the formulas !(r Æ p? ) from the antecedent, and add the formula ?(r p) to the consequent. After all, we obtain a sequent of the form: .............. ........ .. ............... ......... .. 2 ; ! ) ?; where is a list of normal formulas and is a list of simple products. Then we replace the multiset of literals 2 by simple product W = 2 and obtain a normal sequent = (W; ! ) ?). This sequent is derivable in LLW if and only if the sequent is. For LL this lemma is proved similarly except the point, where we use the fact that > = 1 in LLW. For LL it is proved that LL is reduced to the fragment without constants 0 and >. Lemma 4.2 Linear logic and linear aÆne logic are reduced (according to Turing) to the problems of the derivability of \good" sequents in the corresponding logic. Proof Note, that since ?A = (!A? )? we can conne ourselves without loss of generality to the sequents without \?". Let we have any sequent without \?". Consider the set X = fA : !A 2 Sub g. With any formula A from X we associate a new literal p . For any formula B 2 Sub we dene the formula B 0 as a formula obtained from B by replacement of subformulas of the form !A by the literal p . (We replace only formulas !A which is not a subformula of any other formula of the kind !C). For example, if B = (!q Æ !(r !t)) then B 0 = (p Æ p( ! ) ). We dene also the inverse operation ` as follows: A A q r t `B = B[!A1 =p ; : : : ; !A =p n ]; A1 n A where X = fA1 ; : : : ; A g. It is clear that `(B 0 ) = B. Consider the multiset X consisting of the following formulas: (p Æ p p ); (p Æ 1); (p Æ A0 ); n A A A A n i A2 An is derivable, then (p A1 ::: p An Æ p ) 2 X : (7) C (b) X is the least set that meets the property (a), ~ i.e. if any set ~ meets the property (a) then . This set encodes the rule (L!). It is easy to verify that the desired least set exists and nite. Moreover, if we have decidability algorithm for the \good" sequents of LL (LLW) then we can eectively construct this set. Really, one can construct this set by the following way. Initially, let X = ;. Then, while there is a derivable sequent of the form (6), such that the formula (7) is not in X , we add this formula in X . (We can verify whether sequents of the kind (6) are derivable, because they are \good" sequents). Since the set X is nite, then the number of the formulas of the form (7) is also nite. Hence, this algorithm is always terminal. X Lemma 4.2.1 If B !`B is derivable. X X B = (p 2 X ; X , then the sequent ) A1 ::: p Æp An C ) (where A ; C 2 X , and all A 's are dierent), such that the sequent ) !`B is derivable. Let us verify that the set ~ X meets the property (a). Indeed, suppose that the sequent !X ; !~ X ; p ; : : : ; p n ) C 0 i i A1 A is derivable, then we verify that (p : : : p n ÆC) 2 ~ X . If we substitute the formulas !A for all literals p in the last sequent, then we get the derivable sequent !`X ; !`~ X ; !A1 ; : : : ; !A ) C: (8) A1 A A n For any formula B 2 X ; ~ X the sequent ) !`B is derivable (item 1 of this lemma and the denition of ~ X ). The sequent A for any A 2 X . These formulas encode the rules (C!), (W !) and (R!) correspondingly. We shall also consider the multiset X with the following properties: (a) For any set of the formulas A1 ; : : : ; A ; C 2 X (all A 's are dierent) if the sequent !X ; !X ; p ; p ; : : : ; p ) C 0 (6) A1 Proof 1. If B 2 , then `B = (!A Æ !A !A), or (!A Æ 1), or (!A Æ A). In all these cases the sequent ) `B is easily derivable, and so the sequent ) !`B is. 2. In order to prove this lemma for B 2 , we consider the set ~ X , which consists of the formulas of the form !A1 ; : : : ; !A n )C is derivable from these sequents and the sequent (8) by the rule (Cut). Thus, the sequent ) !(!A1 : : : !A Æ !C) n is derivable. Hence, by the denition of the set ~ X , (p : : : p n Æ p ) 2 ~ X : A1 A C So, the set ~ X meets the property (a). And therefore, by the property (b), X ~ X . Hence, if B 2 X then the sequent ) !`B is derivable. Lemma 4.2.2 If we have the sequent = ( ) ) (which does not contain any \?"), then the sequent is derivable if and only if the sequent !X ; !X ; 0 ) 0 is derivable. Proof Firstly, we prove that if the sequent !X ; !X ; 0 ) 0 is derivable then the sequent is derivable too. If we substitute the formulas !A for the literals p in this sequent, then we obtain the derivable sequent A !`X ; !`X ; ) : (9) !X ; !X ; p ; p ; 00 ) 0 !X ; !X ; p p ; 00 ) 0 !X ; !X ; p ; p A A A A Æp p ) p p Æ p p ; 00 ) 0 ( ! !) !X ; !X ; p ; 00 ) 0 A p ;p A A A A A A A A (C ut) A L ;C A Table 8. For any formula B 2 X ; X the sequent ) !`B is derivable (Lemma 4.2.1). The sequent = ( ) ) is derivable from these sequents and the sequent (9) by the rule (Cut). Now we prove by induction on the derivation of the sequent that if this sequent is derivable, then the sequent !X ; !X ; 0 ) 0 is derivable too. Case 1. is derived by the rule (L!): ) ( !) !A; 0 ) Hence, (p Æ A0 ) 2 A; 0 Then A 2 X . X . According to the induction hypothesis the sequent !X ; !X ; A0 ; 00 ) 0 is derivable. Hence, the sequent !X ; !X ; p ; 00 ) 0 is also derivable: !X ; !X ; A0 ; 00 ) 0 p ; p Æ A0 ) A0 ( ) !X ; !X ; p ; p Æ A0 ; 00 ) 0 A A A C ut A A (L!; C !) !X ; !X ; p ; 00 ) 0 ) ( !) !A; 0 ) Then A 2 X . Hence, (p Æ 1) 2 X . By the induction hypothesis the sequent !X ; !X ; 00 ) 0 is derivable. Hence, the sequent !X ; !X ; p ; 00 ) 0 is also derivable: !X ; !X ; 00 ) 0 A A A C ut !X ; !X ; p ; p A p ;p Æ1 ) 1 ( Æ 1; 00 ) 0 A A (L!; C !) !X ; !X ; p ; 00 ) 0 A Case 3. is derived by the rule (C!): !A; !A; !A; ) ( 0) 0 C !) 0 A is derivable. The derivation of the sequent !X ; !X ; p ; 00 ) 0 from this sequent is represented in the Table 8. Case 4. is derived by the rule (R!): A !A1 ; : : : ; !A !A1 ; : : : ; !A n )C ( ) !C R !) By the induction hypothesis the sequent !X ; !X ; p ; : : : ; p ) C 0 A1 An is derivable. Consider formulas B1 ; : : : ; B , such that all B 's are dierent and m i fB1 ; : : : ; B g = fA1 ; : : : ; A g: m n The sequent !X ; !X ; p ; : : : ; p (p W (L1) A Bm ) C0 is derivable (similarly to the case 3). Hence, by denition of the set X , Case 2. is derived by the rule (W !): !X ; !X ; 1; 00 ) 0 A A B1 A 0 A n L A Then A 2 X . Hence, (p Æ p p ) 2 X . By the induction hypothesis the sequent !X ; !X ; p ; p ; 0 ) 0 ) B1 ::: p Bm Æ p ) 2 X : C Therefore, the sequent !X ; !X ; p ; : : : ; p B1 Bm )p C is derivable. Similarly to the case 2, the sequent !X ; !X ; p ; : : : ; p A1 An )p C : is derivable from the last sequent. The other cases not associated with \!", are trivial. It is easy to see that the lemma 4.2 follows from the last lemma. And the theorem 4 follows from the lemmas 4.1 and 4.2. The theorems 1{3 provide the main Theorem Linear aÆne logic is decidable. 5 Conclusion remarks Here we consider one more corollary from the computational interpretation of the normal fragments. In [4] it was considered the certain fragment H2 LL(!; ; Æ; ). The sequents from this fragment were interpreted in terms of so-called non-determinate Petri Nets with conditional transitions. In fact, it is possible to express the rules of the game A in the terms of these nets and vice versa. As a corollary of this fact we have that the whole linear logic is reduced to the fragment H2 LL(!; ; Æ; ). By analogous the multiplicative-exponential fragment of linear logic (MELL = LL(!; ?;? ; Æ; ; ; 1; ?)) is reduced to H2 LL(!; ; Æ). Hence, the following fragments are decidable or undecidable simultaneously: (i) MELL = LL(!; ?; ; ; Æ; ? ; 1; ?) (ii) LL(!; ; Æ) (iii) The normal fragment of LL(!; ?; ; ; Æ) The decidability problem of these fragments remains open. ............... ........ .. ............... ........ .. ............... ......... .. Acknowledgments I am very thankful to S.N. Artemov for orientation of my research and for support of my work. I owe M.I. Kanovich a debt of gratitude for the fruitful discussion of the results and his helpful comments. This work would be impossible without their participation. References [1] J.-Y.Girard. Linear Logic. Theoretical Computer Science, 50, 1{102, 1987. [2] M.I.Kanovich. Horn Programming in Linear Logic is NP-complete. In Proc. 7-th Annual IEEE Symposium on Logic in Computer Science, Santa Cruz, June 1992, pp.200-210 [3] M.I.Kanovich. Petri Nets, Horn Programs, Linear Logic, and Vector Games. Proceedings of the International Symposium Theoretical Aspects of Computer Software, TACS'94, Sendai, Japan, April 1994. In Lecture Notes in Computer Science, (ed. M.Hagiya and J.Mitchell), 1994, 789, p.642{666 [4] M.I.Kanovich. Simulating Linear Logic in 1-Only Linear Logic. CNRS, Laboratoire de Mathematiques Discretes, Pretirage nÆ 94-02, January 1994, 81 p. Available by anonymous ftp from host lmd.univ-mrs.fr and the le pub/kanovich/unit-only.dvi. [5] P.Lincoln, J.Mitchell, A.Scedrov, and N.Shankar. Decision Problems for Propositional Linear Logic. In Proc. 31st IEEE Symp. on Foundations of Computer Science, 662{671, 1990. [6] A.S.Troelstra. Lectures on Linear Logic. CSLI lecture notes; no.29, 1992