@transitive_le [3: @IHL |2: /2 width=2/ | skip ]
qed.
+lemma cw_tpair_sn: ∀I,L,V,T. #[L, V] < #[L, ②{I}V.T].
+#I #L #V #T normalize in ⊢ (? % %); //
+qed.
+
+lemma cw_tpair_dx: ∀I,L,V,T. #[L, T] < #[L, ②{I}V.T].
+#I #L #V #T normalize in ⊢ (? % %); //
+qed.
+
+lemma cw_tpair_dx_sn: ∀I1,I2,L,V1,V2,T. #[L, V2] < #[L, ②{I1}V1.②{I2}V2.T].
+#I1 #I2 #L #V1 #V2 #T normalize in ⊢ (? % %); /2 width=1/
+qed.
+
+lemma cw_tpair_dx_sn_shift: ∀I1,I2,L,V1,V2,T. #[L.ⓑ{I2}V2, T] < #[L, ②{I1}V1.②{I2}V2.T].
+#I1 #I2 #L #V1 #V2 #T normalize in ⊢ (? % %); /2 width=1/
+qed.
+
(* Basic_1: removed theorems 6:
flt_thead_sx flt_thead_dx flt_arith0 flt_arith1 flt_arith2 flt_trans
*)
--- /dev/null
+(**************************************************************************)
+(* ___ *)
+(* ||M|| *)
+(* ||A|| A project by Andrea Asperti *)
+(* ||T|| *)
+(* ||I|| Developers: *)
+(* ||T|| The HELM team. *)
+(* ||A|| http://helm.cs.unibo.it *)
+(* \ / *)
+(* \ / This file is distributed under the terms of the *)
+(* v GNU General Public License Version 2 *)
+(* *)
+(**************************************************************************)
+
+include "basic_2/reducibility/ltpr_ldrop.ma".
+include "basic_2/static/aaa_ltps.ma".
+include "basic_2/static/lsuba_aaa.ma".
+
+(* CONTEXT-FREE PARALLEL REDUCTION ON LOCAL ENVIRONMENTS ********************)
+
+(* Properties about atomic arity assignment on terms ************************)
+
+fact aaa_ltpr_tpr_conf_aux: ∀L,T,L1,T1,A. L1 ⊢ T1 ÷ A → L = L1 → T = T1 →
+ ∀L2. L1 ➡ L2 → ∀T2. T1 ➡ T2 → L2 ⊢ T2 ÷ A.
+#L #T @(cw_wf_ind … L T) -L -T #L #T #IH
+#L1 #T1 #A * -L1 -T1 -A
+[ -IH #L1 #k #H1 #H2 #L2 #_ #T2 #H destruct
+ >(tpr_inv_atom1 … H) -H //
+| #I #L1 #K1 #V1 #B #i #HLK1 #HK1 #H1 #H2 #L2 #HL12 #T2 #H destruct
+ >(tpr_inv_atom1 … H) -T2
+ lapply (ldrop_pair2_fwd_fw … HLK1 (#i)) #HKV1
+ elim (ltpr_ldrop_conf … HLK1 … HL12) -HLK1 -HL12 #X #HLK2 #H
+ elim (ltpr_inv_pair1 … H) -H #K2 #V2 #HK12 #HV12 #H destruct
+ lapply (IH … HKV1 … HK1 … HK12 … HV12) // -L1 -K1 -V1 /2 width=5/
+| #L1 #V1 #T1 #B #A #HB #HA #H1 #H2 #L2 #HL12 #X #H destruct
+ elim (tpr_inv_abbr1 … H) -H *
+ [ #V2 #T0 #T2 #HV12 #HT10 #HT02 #H destruct
+ lapply (tps_lsubs_conf … HT02 (L2.ⓓV2) ?) -HT02 /2 width=1/ #HT02
+ lapply (IH … HB … HL12 … HV12) -HB /width=5/ #HB
+ lapply (IH … HA … (L2.ⓓV2) … HT10) -IH -HA -HT10 /width=5/ -T1 /2 width=1/ -L1 -V1 /3 width=5/
+ | -B #T #HT1 #HTX
+ elim (lift_total X 0 1) #X1 #HX1
+ lapply (tpr_lift … HTX … HT1 … HX1) -T #HTX1
+ lapply (IH … HA … (L2.ⓓV1) … HTX1) /width=5/ -T1 /2 width=1/ -L1 #HA
+ @(aaa_inv_lift … HA … HX1) /2 width=1/
+ ]
+| #L1 #V1 #T1 #B #A #HB #HA #H1 #H2 #L2 #HL12 #X #H destruct
+ elim (tpr_inv_abst1 … H) -H #V2 #T2 #HV12 #HT12 #H destruct
+ lapply (IH … HB … HL12 … HV12) -HB /width=5/ #HB
+ lapply (IH … HA … (L2.ⓛV2) … HT12) -IH -HA -HT12 /width=5/ -T1 /2 width=1/
+| #L1 #V1 #T1 #B #A #HV1 #HT1 #H1 #H2 #L2 #HL12 #X #H destruct
+ elim (tpr_inv_appl1 … H) -H *
+ [ #V2 #T2 #HV12 #HT12 #H destruct
+ lapply (IH … HV1 … HL12 … HV12) -HV1 -HV12 /width=5/ #HB
+ lapply (IH … HT1 … HL12 … HT12) -IH -HT1 -HL12 -HT12 /width=5/ /2 width=3/
+ | #V2 #W2 #T0 #T2 #HV12 #HT02 #H1 #H2 destruct
+ elim (aaa_inv_abst … HT1) -HT1 #B0 #A0 #HB0 #HA0 #H destruct
+ lapply (IH … HV1 … HL12 … HV12) -HV1 -HV12 /width=5/ #HB
+ lapply (IH … HB0 … HL12 W2 ?) -HB0 /width=5/ #HB0
+ lapply (IH … HA0 … (L2.ⓛW2) … HT02) -IH -HA0 -HT02 /width=5/ -T0 /2 width=1/ -L1 -V1 /4 width=7/
+ | #V0 #V2 #W0 #W2 #T0 #T2 #HV10 #HW02 #HT02 #HV02 #H1 #H2 destruct
+ elim (aaa_inv_abbr … HT1) -HT1 #B0 #HW0 #HT0
+ lapply (IH … HW0 … HL12 … HW02) -HW0 /width=5/ #HW2
+ lapply (IH … HV1 … HL12 … HV10) -HV1 -HV10 /width=5/ #HV0
+ lapply (IH … HT0 … (L2.ⓓW2) … HT02) -IH -HT0 -HT02 /width=5/ -V1 -T0 /2 width=1/ -L1 -W0 #HT2
+ @(aaa_abbr … HW2) -HW2
+ @(aaa_appl … HT2) -HT2 /3 width=7/ (**) (* explict constructors, /5 width=7/ is too slow *)
+ ]
+| #L1 #V1 #T1 #A #HV1 #HT1 #H1 #H2 #L2 #HL12 #X #H destruct
+ elim (tpr_inv_cast1 … H) -H
+ [ * #V2 #T2 #HV12 #HT12 #H destruct
+ lapply (IH … HV1 … HL12 … HV12) -HV1 -HV12 /width=5/ #HV2
+ lapply (IH … HT1 … HL12 … HT12) -IH -HT1 -HL12 -HT12 /width=5/ -L1 -V1 -T1 /2 width=1/
+ | -HV1 #HT1X
+ lapply (IH … HT1 … HL12 … HT1X) -IH -HT1 -HL12 -HT1X /width=5/
+ ]
+]
+qed.
+
+lemma aaa_ltpr_tpr_conf: ∀L1,T1,A. L1 ⊢ T1 ÷ A → ∀L2. L1 ➡ L2 →
+ ∀T2. T1 ➡ T2 → L2 ⊢ T2 ÷ A.
+/2 width=9/ qed.
+
+lemma aaa_ltpr_conf: ∀L1,T,A. L1 ⊢ T ÷ A → ∀L2. L1 ➡ L2 → L2 ⊢ T ÷ A.
+/2 width=5/ qed.
+
+lemma aaa_tpr_conf: ∀L,T1,A. L ⊢ T1 ÷ A → ∀T2. T1 ➡ T2 → L ⊢ T2 ÷ A.
+/2 width=5/ qed.
/3 width=4/
]
qed.
+
+lemma aaa_inv_lift: ∀L2,T2,A. L2 ⊢ T2 ÷ A → ∀L1,d,e. ⇩[d, e] L2 ≡ L1 →
+ ∀T1. ⇧[d, e] T1 ≡ T2 → L1 ⊢ T1 ÷ A.
+#L2 #T2 #A #H elim H -L2 -T2 -A
+[ #L2 #k #L1 #d #e #_ #T1 #H
+ >(lift_inv_sort2 … H) -H //
+| #I #L2 #K2 #V2 #B #i #HLK2 #_ #IHB #L1 #d #e #HL21 #T1 #H
+ elim (lift_inv_lref2 … H) -H * #Hid #H destruct
+ [ elim (ldrop_conf_lt … HL21 … HLK2 ?) -L2 // -Hid /3 width=8/
+ | lapply (ldrop_conf_ge … HL21 … HLK2 ?) -L2 // -Hid /3 width=8/
+ ]
+| #L2 #V2 #T2 #B #A #_ #_ #IHB #IHA #L1 #d #e #HL21 #X #H
+ elim (lift_inv_bind2 … H) -H #V1 #T1 #HV12 #HT12 #H destruct
+ /4 width=4/
+| #L2 #V2 #T2 #B #A #_ #_ #IHB #IHA #L1 #d #e #HL21 #X #H
+ elim (lift_inv_bind2 … H) -H #V1 #T1 #HV12 #HT12 #H destruct
+ /4 width=4/
+| #L2 #V2 #T2 #B #A #_ #_ #IHB #IHA #L1 #d #e #HL21 #X #H
+ elim (lift_inv_flat2 … H) -H #V1 #T1 #HV12 #HT12 #H destruct
+ /3 width=4/
+| #L2 #V2 #T2 #A #_ #_ #IH1 #IH2 #L1 #d #e #HL21 #X #H
+ elim (lift_inv_flat2 … H) -H #V1 #T1 #HV12 #HT12 #H destruct
+ /3 width=4/
+]
+qed-.