hh_meta.thm

% Meta-theory of HH in G (Abella's reasoning logic) % Reasoning on multisets -- this is needed in the statement % of the cut theorem. % The definition (remove G A D) asserts that G is D extended with A. Define remove : olist -> o -> olist -> prop by remove (A :: G) A G ; remove (B :: G) A (B :: D) := remove G A D. % If G is an extension of D, then all members of D are also members of G. Theorem remove_incl : forall G D A B, remove G A D -> member B D -> member B G. induction on 1. intros. case H1. search. case H2. search. apply IH to H3 H4. search. % If G is D extended with A, then any member of G is either A or a % member of D. Theorem remove_charac : forall G A D B, remove G A D -> member B G -> A = B \/ member B D. induction on 1. intros. case H1 (keep). case H2. search. search. case H2. search. apply IH to H3 H4. case H5. search. search. % Types of formulas and terms Kind fm, tm type. % Terms are left abstract, while formulas have the following constructors Type atm tm -> fm. Type and fm -> fm -> fm. Type top fm. Type imp fm -> fm -> fm. Type all (tm -> fm) -> fm. % Contexts are lists of formulas, but instead of defining a type of % formula lists, we just reuse the olist type and keep things of % the form (\$fm A) in them. In a polymorphically typed extension % of Abella, we can avoid this hack. Type \$fm fm -> o. % We will need to induct on the structure of formulas, so we write % an inductive definition of all formulas. Define is_fm : fm -> prop by is_fm (atm A) ; is_fm (and A B) := is_fm A /\ is_fm B ; is_fm top ; is_fm (imp A B) := is_fm A /\ is_fm B ; is_fm (all A) := forall x, is_fm (A x). % We keep the prog definition abstract. Type prog fm -> prop. % We have to assume that any prog clause is independent of nabla in % order for instantiation to be sound. Since there is no way to write % axioms in Abella, we state this as a theorem with an empty proof. Theorem prog_has_no_nablas : forall F, nabla (n:tm), prog (F n) -> exists E, F = x\ E. skip. % The backchaining calculus consists of two phases: seq and bch % % bch L F A stands for L ; [F] |- A (F backchainied) % seq L G stands for L |- G Define seq : olist -> fm -> prop, bch : olist -> fm -> tm -> prop by seq L (atm A) := exists F, member (\$fm F) L /\ bch L F A ; seq L (atm A) := exists F, prog F /\ bch L F A ; seq L (and G1 G2) := seq L G1 /\ seq L G2 ; seq L top ; seq L (imp F G) := seq (\$fm F :: L) G ; seq L (all A) := nabla x, seq L (A x) ; bch L (atm A) A ; bch L (and F1 F2) A := bch L F1 A \/ bch L F2 A ; bch L (imp G F) A := seq L G /\ bch L F A ; bch L (all F) A := exists t, bch L (F t) A. % Note: the third argument to bch always represents an atom. % This is because the atomic formula P(t1, ..., tn) is represented as % (atm (P t1 ... tn)), where P has type (tm -> ... -> tm -> tm). Theorem \$monotone : (forall L L' C, (forall E, member E L -> member E L') -> seq L C -> seq L' C) /\ (forall L L' F A, (forall E, member E L -> member E L') -> bch L F A -> bch L' F A). induction on 2 2. split. intros. case H2 (keep). apply IH1 to H1 H4. apply H1 to H3. search. apply IH1 to H1 H4. search. apply IH to H1 H3. apply IH to H1 H4. search. search. assert forall E, member E (\$fm F :: L) -> member E (\$fm F :: L'). intros. case H4. search. apply H1 to H5. search. apply IH to H4 H3. search. apply IH to H1 H3. search. intros. case H2 (keep). search. case H3. apply IH1 to H1 H4. search. apply IH1 to H1 H4. search. apply IH to H1 H3. apply IH1 to H1 H4. search. apply IH1 to H1 H3. search. Split \$monotone as monotone_seq, monotone_bch. Theorem \$weakening : (forall L L' B G, remove L' B L -> seq L G -> seq L' G) /\ (forall L L' B F A, remove L' B L -> bch L F A -> bch L' F A). split. intros. assert forall E, member E L -> member E L'. intros. backchain remove_incl. backchain monotone_seq. intros. assert forall E, member E L -> member E L'. intros. backchain remove_incl. backchain monotone_bch. Split \$weakening as weakening_seq, weakening_bch. Theorem member_inst : forall F (L:tm -> olist), nabla (n:tm), member (F n) (L n) -> forall t, member (F t) (L t). induction on 1. intros. case H1. search. apply IH to H2. apply H3 with t = t. search. % The instantiation lemma Theorem \$inst : (forall L G, nabla (n:tm), seq (L n) (G n) -> forall t, seq (L t) (G t)) /\ (forall L F A, nabla (n:tm), bch (L n) (F n) (A n) -> forall t, bch (L t) (F t) (A t)). induction on 1 1. split. intros. case H1. apply IH1 to H3. apply member_inst to H2. apply H4 with t = t. apply H5 with t = t. search. apply IH1 to H3. apply prog_has_no_nablas to H2. apply H4 with t = t. search. apply IH to H2. apply IH to H3. apply H4 with t = t. apply H5 with t = t. search. search. apply IH to H2. apply H3 with t = t. search. apply IH to H2. apply H3 with t = t. search. intros. case H1. search. case H2. apply IH1 to H3. apply H4 with t = t. search. apply IH1 to H3. apply H4 with t = t. search. apply IH to H2. apply IH1 to H3. apply H4 with t = t. apply H5 with t = t. search. apply IH1 to H2. apply H3 with t = t. search. Split \$inst as inst_seq, inst_bch. % The main cut-admissibility theorem. Theorem \$cut : (forall L K F G, is_fm F -> seq K F -> remove L (\$fm F) K -> seq L G -> seq K G) /\ (forall L K F F' A, is_fm F -> seq K F -> remove L (\$fm F) K -> bch L F' A -> bch K F' A) /\ (forall L F A, is_fm F -> seq L F -> bch L F A -> seq L (atm A)). % We proceed by nested induction on: % % - the structure of the cut-formula, then % - the structure of the container derivation (i.e., the derivation % that contains the cut-formula as a hypothesis). induction on 1 1 1. induction on 4 4 3. split. intros. case H4 (keep). apply remove_charac to H3 H5. case H7. case H1 (keep). case H6. search. case H6 (keep). case H10. apply IH4 to H1 H2 H3 H11. case H2. apply IH2 to H8 H13 H12. search. apply IH4 to H1 H2 H3 H11. case H2. apply IH2 to H9 H14 H12. search. case H6. case H6. apply IH3 to H1 H2 H3 H10. apply IH4 to H1 H2 H3 H11. case H2 (keep). apply IH to H8 H12 _ H14. apply IH2 to H9 H15 H13. search. case H6. apply IH4 to H1 H2 H3 H9. case H2. apply inst_seq to H11. apply H12 with t = t. apply H8 with x = t. apply IH2 to H14 H13 H10. search. apply IH4 to H1 H2 H3 H6. search. apply IH4 to H1 H2 H3 H6. search. apply IH3 to H1 H2 H3 H5. apply IH3 to H1 H2 H3 H6. search. search. assert remove (\$fm F1 :: L) (\$fm F) (\$fm F1 :: K). assert seq (\$fm F1 :: K) F. backchain weakening_seq. apply IH3 to H1 H7 H6 H5. search. apply IH3 to H1 H2 H3 H5. search. intros. case H4 (keep). search. case H5. apply IH4 to H1 H2 H3 H6. search. apply IH4 to H1 H2 H3 H6. search. apply IH3 to H1 H2 H3 H5. apply IH4 to H1 H2 H3 H6. search. apply IH4 to H1 H2 H3 H5. search. intros. case H3 (keep). search. case H4. case H1 (keep). case H2 (keep). apply IH2 to H6 H8 H5. search. case H1 (keep). case H2 (keep). apply IH2 to H7 H9 H5. search. case H1 (keep). case H2 (keep). apply IH to H6 H4 _ H8. apply IH2 to H7 H9 H5. search. case H1 (keep). case H2 (keep). apply H5 with x = t. apply inst_seq to H6. apply H8 with t = t. apply IH2 to H7 H9 H4. search. Split \$cut as cut, cut_commutative, cut_principal.