From b4f26632954d3d2e9ae6802ddf37380395115e1c Mon Sep 17 00:00:00 2001 From: affeldt-aist <33154536+affeldt-aist@users.noreply.github.com> Date: Fri, 26 Jan 2024 23:26:41 +0900 Subject: [PATCH] changelog for version 1.0.0 (#1165) * changelog for version 1.0.0 --- CHANGELOG.md | 647 +++++++++++++++++++++++++++++++++++- CHANGELOG_UNRELEASED.md | 637 ----------------------------------- INSTALL.md | 30 +- README.md | 4 +- coq-mathcomp-analysis.opam | 1 + coq-mathcomp-classical.opam | 1 + 6 files changed, 665 insertions(+), 655 deletions(-) diff --git a/CHANGELOG.md b/CHANGELOG.md index bf0a8172d..a56942f52 100644 --- a/CHANGELOG.md +++ b/CHANGELOG.md @@ -1,6 +1,651 @@ # Changelog -Latest releases: [[0.7.0] - 2024-01-19](#070---2024-01-19) and [[0.6.7] - 2024-01-09](#067---2024-01-09) +Latest releases: [[1.0.0] - 2024-01-26](#100---2024-01-26) and [[0.7.0] - 2024-01-19](#070---2024-01-19) + +## [1.0.0] - 2024-01-26 + +### Added + +- in `constructive_ereal.v`: + + definition `dEFin` + + notations `%:dE`, `%:E` (`ereal_dual_scope`) + + notation `\bar^d ...` (`type_scope`) for dual extended real numbers + + instance using `isNmodule.Build` for `\bar` + + instances using `Choice.on` and `isNmodule.Build` for `\bar^d` + + lemma `EFin_semi_additive` + + lemmas `dEFinE`, `dEFin_semi_additive` + + instance using `isSemiAdditive.Build` for `\bar^d` + + canonical `dEFin_snum` + +- in `reals.v`: + + definition `Rint_pred` + +- in `topology.v` + + definition `set_system`, identity coercion `set_system_to_set` + with instances using `Equality.on`, `Choice.on`, `Pointed.on`, + `isFiltered.Build` + + mixin `selfFiltered`, factory `hasNbhs`, structure `Nbhs`, + type `nbhsType` + + instance for matrices using `selfFiltered.Build` + + lemmas `cvg_in_ex`, `cvg_inP`, `cvg_in_toP`, `dvg_inP`, `cvg_inNpoint`, + `eq_is_cvg_in` + + notations `E @[ x \oo ]`, `limn`, `cvgn` + + definition `continuous_at` + + definitions `weak_topology`, `sup_topology`, `prod_topology` + + definition `prod_topo_apply` + + definition `discrete_topology` + + instead of `zmodType` using `isPointed.Build` + + definition `pointwise_cvgE`, + instance using `Uniform.copy` for `{ptws _ -> _}` + + definition `compact_open_of_nbhs`, lemmas `compact_openK_nbhsE_subproof`, + `compact_openK_openE_subproof` + +- in `cantor.v`: + + definition `pointed_principal_filter`, instances using `Pointed.on` and `hasNbhs.Build` + + definition `pointed_discrete_topology` + + lemma `discrete_pointed` + + lemma `discrete_bool_compact` + +- in `normedtype.v`: + + definition `urysohnType` with instances using `Pointed.on` and `isUniform.Build` + +- in `derive.v`: + + lemma `cvg_at_rightE`, `cvg_at_leftE` + +- in `convex.v`: + + definition `convex_lmodType` with instances using `Choice.on` and `isConvexSpace.Build` + + definition `convex_realDomainType` with instance using `isConvexSpace.Build` + +- in `lebesgue_stieltjes_measure.v`: + + instance on `ocitv_type` using `Pointed.on` + +- in `lebesgue_integral.v`: + + mixin `isNonNegFun`, notations `{nnfun _ >-> _}`, `[nnfun of _]` + + section `ring` + * lemmas `fimfun_mulr_closed`, instances using `GRing.isMulClosed.Build`, + `[SubZmodule_isSubRing of ... by <:]` + * lemmas `fimfunM`, `fimfun1`, `fimfun_prod`, `fimfunX`, + * lemma `indic_fimfun_subproof`, instance using `indic_fimfun_subproof` + * definition `indic_fimfun` + * instance using `FImFun.copy`, definition `scale_fimfun` + + section `comring` + * instance using `[SubRing_isSubComRing of ... by <:]` + * instance using `FImFun.copy` + + lemmas `fimfunE`, `fimfunEord`, `trivIset_preimage1`, `trivIset_preimage1_in` + + section `fimfun_bin` + * lemma `max_fimfun_subproof`, instance using `max_fimfun_subproof` + + factory `FiniteDecomp` + +- in `charge.v`: + + `cscale` instances using `SigmaFinite_isFinite.Build` and `isAdditiveCharge.Build` + +### Changed + +- in `boolp.v`: + + in lemma `gen_choiceMixin`: `Choice.mixin_of` -> `hasChoice` + + in definition `gen_eqMixin`: `EqMixin` -> `hasDecEq.Build` + + canonical `dep_arrow_eqType` -> instance using `gen_eqMixin` + + canonical `dep_arrow_choiceType` -> instance using `gen_choiceMixin` + + canonical `Prop_eqType` -> instance using `gen_eqMixin` + + canonical `Prop_choiceType` -> instance using `gen_choiceMixin` + + canonical `classicType_eqType` -> instance using `gen_eqMixin` + + canonical `classicType_choiceType` -> instance using `gen_choiceMixin` + + canonical `eclassicType_eqType` -> instance using `Equality.copy` + + canonical `eclassicType_choiceType` -> instance using `gen_choiceMixin` + + definition `porderMixin` and canonical `porderType` -> instance using `isPOrder.Build` + + definition `latticeMixin` and canonical `latticeType` -> instance using `POrder_isLattice.Build` + +- in `classical_sets.v`: + + canonicals + `setU_monoid`, `setU_comoid`, `setU_mul_monoid`, `setI_monoid`, + `setI_comoid`, `setI_mul_monoid`, `setU_add_monoid`, `setI_add_monoid` + -> instances using + `isComLaw.Build`, `isMulLaw.Build`, `isComLaw.Build`, `isMulLaw.Build`, `isAddLaw.Build`, `isAddLaw.Build` + + module `Pointed` (packed class) -> mixin `isPointed`, structure `Pointed` + + canonical `arrow_pointedType` and definition `dep_arrow_pointedType` -> instance using `isPointed.Build` + + canonicals + `unit_pointedType`, `bool_pointedType`, `Prop_pointedType`, `nat_pointedType`, + `prod_pointedType`, `matrix_pointedType`, `option_pointedType`, `pointed_fset` + -> instances using + `isPointed.Build` + + module `Empty` (packed class) -> mixin `isEmpty`, structure `Empty`, factories `Choice_isEmpty`, `Type_isEmpty` + + definition `False_emptyMixin` and canonicals `False_eqType`, + `False_choiceType`, `False_countType`, `False_finType`, `False_emptyType` -> + instance using `Type_isEmpty.Build` + + definition `void_emptyMixin` and canonical `void_emptyType` -> instance using `isEmpty.Build` + + definition `orderMixin` and canonicals `porderType`, `latticeType`, `distrLatticeType` -> + instances using `Choice.copy` and `isMeetJoinDistrLattice.Build` + + canonicals `bLatticeType`, `tbLatticeType`, `bDistrLatticeType`, `tbDistrLatticeType` -> + instances using `hasBottom.Build` and `hasTop.Build` + + canonical `cbDistrLatticeType` -> instance using `hasRelativeComplement.Build` + + canonical `ctbDistrLatticeType` -> instance using `hasComplement.Build` + +- in `functions.v`: + + notation `split` + + notation `\_` moved from `fun_scope` to `function_scope` + + notations `pinv`, `pPbij`, `pPinj`, `injpPfun`, `funpPinj` + + in definition `fct_zmodMixin`: `ZmodMixin` -> `isZmodule.Build` + + canonical `fct_zmodType` -> instance using `fct_zmodMixin` + + in definition `fct_ringMixin`: `RingMixin` -> `Zmodule_isRing.Build` + + canonical `fct_ringType` -> instance using `fct_ringMixin` + + canonical `fct_comRingType` -> + definition and instance using `Ring_hasCommutativeMul.Build` and `fct_comRingType` + + definition `fct_lmodMixin` and canonical `fct_lmodType` -> + definition `fct_lmodMixin` and instance using `fct_lmodMixin` + +- in `cardinality.v`: + + canonical `rat_pointedType` -> instance using `isPointed.Build` + + canonical `fimfun_subType` -> instance using `isSub.Build` + + definition `fimfuneqMixin` and canonical `fimfuneqType` -> instance using `[Equality of ... by <:]` + + definition `fimfunchoiceMixin` and canonical `fimfunchoiceType` -> instance using `[Choice of ... by <:]` + + canonicals `fimfun_add`, `fimfun_zmod`, `fimfun_zmodType`, and definition `fimfun_zmodMixin` -> + instances using `isZmodClosed.Build` and `[SubChoice_isSubZmodule of ... <:]` + +- in `signed.v`: + + definitions `signed_subType`, `signed_choiceMixin`, `signed_porderMixin`, + canonicals `signed_eqMixin`, `signed_eqType`, `signed_choiceType`, `signed_porderType` + -> instances using `[isSub for ...]` and `[POrder of ... by <:]` + + in lemma `signed_le_total`: `totalPOrderMixin` -> `total` + + canonicals `signed_latticeType`, `signed_distrLatticeType`, `signed_orderType` + -> instance using `Order.POrder_isTotal.Build` + +- in `constructive_ereal.v`: + + definition `ereal_eqMixin` and canonical `ereal_eqType` -> instance using `hasDecEq.Build` + + definition `ereal_choiceMixin` and canonical `ereal_choiceType` -> instance using `Choice.copy` + + definition `ereal_countMixin` and `ereal_countType` -> instance using `PCanIsCountable` + + definition `ereal_porderMixin` and canonical `Choice.copy` -> instance using `isPOrder.Build` + + in lemma `le_total_ereal` : `le_total_ereal` -> `total` + + canonicals `ereal_latticeType`, `ereal_distrLatticeType`, `ereal_orderType`, `ereal_blatticeType`, + `ereal_tblatticeType`, lemmas `ereal_blatticeMixin`, `ereal_blatticeMixin` -> + instances using `POrder_isTotal.Build`, `hasBottom.Build`, `hasTop.Build` + + canonicals `adde_monoid`, `adde_comoid`, `mule_mulmonoid` -> instance using `isMulLaw.Build` + + notations `maxe`, `mine`: `fun_scope` -> `function_scope` + + canonicals `mule_monoid`, `mule_comoid` -> instance using `isComLaw.Build` + + canonicals `maxe_monoid`, `maxe_comoid` -> instance using `isLaw.Build` + +- in `reals.v`: + + module `Real` (packed class) -> mixin `ArchimedeanField_isReal` + with fields `sup_upper_bound_subdef`, `sup_adherent_subdef`, + structure `Real` + + canonicals `Rint_keyed`, `Rint_opprPred`, `Rint_addrPred`, `Rint_mulrPred`, `Rint_zmodPred`, + `Rint_semiringPred`, `Rint_smulrPred`, `Rint_subringPred` + -> instance using `GRing.isSubringClosed.Build` + +- in `topology.v`: + + canonicals `linear_eqType`, `linear_choiceType` -> + instances using `gen_eqMixin`, `gen_choiceMixin` + + canonical `gen_choiceMixin` -> instance using `isPointed.Build` + + module `Filtered` (packed class) -> mixin `isFiltered` with field `nbhs`, + structure `Filtered` + + now use `set_system`: + * definitions `filter_from`, `filter_prod`, `cvg_to`, `type_of_filter`, `lim_in`, + `Build_ProperFilter`, `filter_ex`, `fmap`, `fmapi`, `globally`, `in_filter_prod`, + `within`, `subset_filter`, `powerset_filter_from`, `principal_filter`, `locally_of`, + `sup_subbase`, `cluster`, `compact`, `near_covering`, `near_covering_within`, + `compact_near`, `nbhs_`, `weak_ent`, `sup_ent`, `cauchy`, `cvg_cauchy`, `cauchy_ex.Build`, + `cauchy_ball` + * classes `Filter`, `ProperFilter'`, `UltraFilter` + * instances `fmap_proper_filter`, `fmapi_filter`, `fmapi_proper_filter`, + `filter_prod_filter`, `filter_prod1`, `filter_prod2` + * record `in_filter` + * structure `filter_on` + * variant `nbhs_subspace_spec` + * lemmas `nearE`, `eq_near`, `nbhs_filterE`, `cvg_refl`, `cvg_trans`, `near2_curry`, + `near_swap`, `filterP_strong`, `filter_nbhsT`, `nearT`, `filter_not_empty_ex`, + `filter_ex_subproof`, `filter_getP`, `near`, `nearW`, `filterE`, `filter_app`, + `filter_app2`, `filter_app3`, `filterS2`, `filterS3`, `nearP_dep`, `filter2P`, + `filter_ex2`, `filter_fromP`, `filter_fromTP`, `filter_bigIP`, `filter_forall`, + `filter_imply`, `fmapEP`, `fmapiE`, `cvg_id`, `appfilterP`, `cvg_app`, `cvgi_app`, + `cvg_comp`, `cvgi_comp`, `near_eq_cvg`, `eq_cvg`, `neari_eq_loc`, `cvg_near_const`, + `near_pair`, `near_map`, `near_map2`, `near_mapi`, `filter_pair_set`, + `filter_pair_near_of`, `cvg_pair`, `cvg_comp2`, `near_powerset_map`, + `near_powerset_map_monoE`, `cvg_fmap`, `continuous_cvg`, `continuous_is_cvg`, + `continuous2_cvg`, `cvg_near_cst`, `is_cvg_near_cst`, `cvg_cst`, `is_cvg_cst`, + `fmap_within_eq`, `cvg_image`, `cvg_fmap2`, `cvg_within_filter`, `cvg_app_within`, + `meets_openr`, `meets_openl`, `meetsxx`, `proper_meetsxx`, `ultra_cvg_clusterE`, + `ultraFilterLemma`, `compact_ultra`, `proper_image`, `in_ultra_setVsetC`, + `ultra_image`, `filter_finI`, `close_cvg`, `discrete_cvg`, `nbhs_E`, `cvg_closeP`, + `cvg_mx_entourageP`, `cvg_fct_entourageP`, `fcvg_ball2P`, `cvg_ball2P`, `cauchy_cvgP`, + `mx_complete`, `Uniform_isComplete.Build`, `cauchy_ballP`, `cauchy_exP`, `cauchyP`, + `compact_cauchy_cvg`, `pointwise_cvgE`, `pointwise_uniform_cvg`, `cvg_sigL`, `uniform_restrict_cvg`, + `cvg_uniformU`, `cvg_uniform_set0`, `fam_cvgP`, `family_cvg_subset`, + `family_cvg_finite_covers`, `fam_cvgE`, `Nbhs_isTopological`, `compact_open_fam_compactP`, + `compact_cvg_within_compact`, `nbhs_subspace`, `subspace_cvgP`, `uniform_limit_continuous`, + `uniform_limit_continuous_subspace`, `pointwise_compact_cvg` + * `t` in module type `PropInFilterSig` + + canonical `matrix_filtered` -> instance using `isFiltered.Build` + + now use `nbhs` instead of `[filter of ...]` + * notations `-->`, `E @[ x --> F ]`, `f @ F`, ```E `@[ x --> F ]```, ```f `@ G```, + `{ptws, F --> f }` + + notation `lim` is now a definition + + canonical `filtered_prod` -> instances using `isFiltered.Build`, `selfFiltered.Build` + + now use `set_system` and also `nbhsType` instead of `filteredType ...` + * lemmas `cvg_ex`, `cvgP`, `cvg_toP`, `dvgP`, `cvgNpoint`, `eq_is_cvg` + + canonicals `filter_on_eqType`, `filter_on_choiceType`, `filter_on_PointedType`, + `filter_on_FilteredType` -> instances using `gen_eqMixin`, `gen_choiceMixin`, + `isPointed.Build`, `isFiltered.Build` + + canonical `bool_discrete_filter` -> instance using `hasNbhs.Build` + + module `Topological` (packed class) -> mixin `Nbhs_isTopological`, + structure `Topological`, type `topologicalType` + * definition `open` now a field of the mixin + + notation `continuous` now uses definition `continuous_at` + + section `TopologyOfFilter` -> factory `Nbhs_isNbhsTopological` + + section `TopologyOfOpen` -> factory `Pointed_isOpenTopological` + + section `TopologyOfBase` -> factory `Pointed_isBaseTopological` + + section `TopologyOfSubbase` -> factory `Pointed_isSubBaseTopological` + + definition `Pointed_isSubBaseTopological`, + canonicals `nat_filteredType`, `nat_topologicalType` -> + instance using `Pointed_isBaseTopological.Build` + + filter now explicit in the notation `X --> Y` + * lemmas `cvg_addnr`, `cvg_subnr`, `cvg_mulnl`, `cvg_mulnr`, `cvg_divnr` + + definition `prod_topologicalTypeMixin`, canonical `prod_topologicalType` -> + instances using `hasNbhs.Build`, `Nbhs_isNbhsTopological.Build` + + definition `matrix_topologicalTypeMixin`, canonical `matrix_topologicalType` -> + instance using `Nbhs_isNbhsTopological.Build` + + definitions `weak_topologicalTypeMixin`, `weak_topologicalType` -> + instances using `Pointed.on`, `Pointed_isOpenTopological.Build` + + definitions `sup_topologicalTypeMixin`, `sup_topologicalType` -> + instances using `Pointed.on` and `Pointed_isSubBaseTopological.Build` + + definition `product_topologicalType` -> definition `product_topology_def` + and instance using `Topological.copy` + + in `lim_id`: `nbhs` now explicit + + canonical `bool_discrete_topology` -> instance using `bool_discrete_topology` + + module `Uniform` (packed class) -> mixin `Nbhs_isUniform_mixin`, + structure `Uniform`, type `uniformType`, + factories `Nbhs_isUniform`, `isUniform` + + definition `prod_uniformType_mixin`, canonical `prod_uniformType` -> + instance using `Nbhs_isUniform.Build` + + definition `matrix_uniformType_mixin`, canonical `matrix_uniformType` -> + instance using `Nbhs_isUniform.Build` + + definitions `weak_uniform_mixin`, `weak_uniformType` -> instance + using `Nbhs_isUniform.Build` + + definitions `fct_uniformType_mixin`, `fct_topologicalTypeMixin`, + `generic_source_filter`, `fct_topologicalType`, `fct_uniformType` -> + definition `arrow_uniform` and instance using `arrow_uniform` + + definitions `sup_uniform_mixin`, `sup_uniformType` -> instance using + `Nbhs_isUniform.Build` + + definition `product_uniformType` -> instance using `Uniform.copy` + + definition `discrete_uniformType` -> instance using `Choice.on`, + `Choice.on`, `discrete_uniform_mixin` + + module `PseudoMetric` (packed class) -> factory `Nbhs_isPseudoMetric` + + definition `ball` now a field of factory `Nbhs_isPseudoMetric` + + definition `matrix_pseudoMetricType_mixin`, canonical `matrix_pseudoMetricType` -> + instance using `Uniform_isPseudoMetric.Build` + + definition `prod_pseudoMetricType_mixin`, canonical `prod_pseudoMetricType` -> + instance using `Uniform_isPseudoMetric.Build` + + definition `fct_pseudoMetricType_mixin`, canonical `fct_pseudoMetricType` -> + instance using `Uniform_isPseudoMetric.Build` + + canonical `quotient_subtype` -> instance using `Quotient.copy` + + canonical `quotient_eq` -> instance using `[Sub ... of ... by %/]` + + canonical `quotient_choice` -> instance using `[Choice of ... by <:]` + + canonical `quotient_pointed` -> instance using `isPointed.Build` + + in definition `quotient_topologicalType_mixin`: + `topologyOfOpenMixin` -> `Pointed_isOpenTopological.Build` + + canonical `quotient_topologicalType` -> instance using `quotient_topologicalType_mixin` + + lemma `repr_comp_continuous` uses the notation `\pi_` instead of `... == ... %[mod ...]` + + definition `discrete_pseudoMetricType` -> instead using `discrete_pseudometric_mixin` + + module `Complete` (packed class) -> mixin `Uniform_isComplete`, + structure `Complete`, type `completeType` + * lemma `cauchy_cvg` now a mixin field + + canonical `matrix_completeType` -> instance using `Uniform_isComplete.Build` + + canonical `fun_completeType` -> instance using `Uniform_isComplete.Build` + + module `CompletePseudoMetric` (packed class) -> structure `CompletePseudoMetric` + + matrix instance using `Uniform_isComplete.Build` + + function instance using `Uniform_isComplete.Build` + + module `regular_topology` -> instances using ` Pointed.on`, + `hasNbhs.Build`, `Nbhs_isPseudoMetric.Build` + + in module `numFieldTopology`: + * `realType`, `rcfType`, `archiFieldType`, `realFieldType`, `numClosedFieldType`, + `numFieldType` instances using `PseudoMetric.copy` + + definition `fct_RestrictedUniform`, `fct_RestrictedUniformTopology`, canonical + `fct_RestrictUniformFilteredType`, `fct_RestrictUniformTopologicalType`, `fct_restrictedUniformType` + -> + definition `uniform_fun`, instance using `Unifom.copy` for ```{uniform` _ -> _}``` + + definitions `fct_Pointwise`, `fct_PointwiseTopology`, canonicals `fct_PointwiseFilteredType`, + `fct_PointwiseTopologicalType` -> definition `pointwise_fun`, instance using `Topological.copy` + + definition `compact_openK_topological_mixin`, canonical `compact_openK_filter`, + `compact_openK_topological` -> instances using `Pointed.on`, + `hasNbhs.Build`, `compact_openK_openE_subproof` for `compact_openK` + + canonical `compact_open_pointedType`, definition `compact_open_topologicalType`, + canonicals `compact_open_filtered`, `compact_open_filtered` -> definition `compact_open_def`, + instances using `Pointed.on`, `Nbhs.copy`, `Pointed.on`, `Nbhs_isTopological` + + definitions `weak_pseudoMetricType_mixin`, `weak_pseudoMetricType` -> lemmas + `weak_pseudo_metric_ball_center`, `weak_pseudo_metric_entourageE`, instance using `niform_isPseudoMetric.Build` + + definition `countable_uniform_pseudoMetricType_mixin` -> module `countable_uniform` + with definition `type`, instances using `Uniform.on`, `Uniform_isPseudoMetric.Build`, + lemma `countable_uniform_bounded`, notation `countable_uniform` + + definitions `sup_pseudoMetric_mixin`, `sup_pseudoMetricType`, `product_pseudoMetricType` -> + instances using `PseudoMetric.on`, `PseudoMetric.copy` + + definitions `subspace_pointedType`, `subspace_topologicalMixin`, canonicals `subspace_filteredType`, + `subspace_topologicalType` -> instance using `Choice.copy`, `isPointed.Build`, `hasNbhs.Build`, + lemmas `nbhs_subspaceP_subproof`, `nbhs_subspace_singleton`, `nbhs_subspace_nbhs`, instance using + `Nbhs_isNbhsTopological.Build` + + definition `subspace_uniformMixin`, canonical `subspace_uniformType` -> instance using + `Nbhs_isUniform_mixin.Build` + + definition `subspace_pseudoMetricType_mixin`, canonical `subspace_pseudoMetricType` -> lemmas `subspace_pm_ball_center`, + `subspace_pm_ball_sym`, `subspace_pm_ball_triangle`, `subspace_pm_entourageE`, instance using + `Uniform_isPseudoMetric.Build` + + section `gauges` -> module `gauge` + * `gauge_pseudoMetricType` -> `gauge.type` (instances using `Uniform.on`, `PseudoMetric.on`) + * `gauge_uniformType` -> `gauge.type` + +- in `cantor.v`: + + in definition `tree_of` and lemma `cantor_like_finite_prod`: + `pointed_discrete` -> `pointed_discrete_topology` + +- in `normedtype.v`: + + module `PseudoMetricNormedZmodule` (packed class) -> + mixin `NormedZmod_PseudoMetric_eq` (with field `pseudo_metric_ball_norm`), + structure `PseudoMetricNormedZmod` + + now use `set_system`: + * definitions `pinfty_nbhs`, `ninfty_nbhs`, `dominated_by`, `strictly_dominated_by`, + `bounded_near`, `sub_klipschitz`, `lipschitz_on`, `sub_lipschitz` + * lemmas `cvgrnyP`, `cvgenyP`, `fcvgrPdist_lt`, `cvgrPdist_lt`, `cvgrPdistC_lt`, + `cvgr_dist_lt`, `cvgr_distC_lt`, `cvgr_dist_le`, `cvgr_distC_le`, `cvgr0Pnorm_lt`, + `cvgr0_norm_lt`, `cvgr0_norm_le`, `cvgrPdist_le`, `cvgrPdist_ltp`, `cvgrPdist_lep`, + `cvgrPdistC_le`, `cvgrPdistC_ltp`, `cvgrPdistC_lep`, `cvgr0Pnorm_le`, `cvgr0Pnorm_ltp`, + `cvgr0Pnorm_lep`, `cvgr_norm_lt`, `cvgr_norm_le`, `cvgr_norm_gt`, `cvgr_norm_ge`, + `cvgr_neq0`, `real_cvgr_lt`, `real_cvgr_le`, `real_cvgr_gt`, `real_cvgr_ge`, `cvgr_lt`, + `cvgr_le`, `cvgr_gt`, `cvgr_ge`, `sub_dominatedl`, `sub_dominatedr`, `ex_dom_bound`, + `ex_strict_dom_bound`, `sub_boundedr`, `sub_boundedl`, `ex_bound`, `ex_strict_bound`, + `ex_strict_bound_gt0`, `klipschitzW`, `cvg_bounded`, `fcvgr2dist_ltP`, `cvgr2dist_ltP`, + `cvgr2dist_lt` + + module `NormedModule` (packed class) -> + mixin `PseudoMetricNormedZmod_Lmodule_isNormedModule`, + structure `NormedModule` + + module and section `regular_topology` -> section `regular_topology` + with instances using `Num.NormedZmodule.on`, `NormedZmod_PseudoMetric_eq.Build`, + `seudoMetricNormedZmod_Lmodule_isNormedModule.Build` + + in module `numFieldNormedType` + * `realType` instances using `GRing.ComAlgebra.copy`, `Vector.copy`, `NormedModule.copy` + * `rcfType` instances using `GRing.ComAlgebra.copy`, `Vector.copy`, `NormedModule.copy` + * `archiFieldType` instances using `GRing.ComAlgebra.copy`, `Vector.copy`, `NormedModule.copy` + * `realFieldType` instances using `GRing.ComAlgebra.copy`, `Vector.copy`, `NormedModule.copy`, `Num.RealField.on` + * `numClosedFieldType` instances using `GRing.ComAlgebra.copy`, `Vector.copy`, `NormedModule.copy`, `Num.ClosedField.on` + * `numFieldType` instances using `GRing.ComAlgebra.copy`, `Vector.copy`, `NormedModule.copy`, `Num.NumField.on` + + in lemma `norm_lim_id`: now explicit use of `nbhs` + + definition `matrix_PseudoMetricNormedZmodMixin` and canonical `matrix_normedModType` + -> instance using `PseudoMetricNormedZmod_Lmodule_isNormedModule.Build` + + definition `prod_pseudoMetricNormedZmodMixin` and canonical `prod_normedModType` + -> instance using `PseudoMetricNormedZmod_Lmodule_isNormedModule.Build` + + module `CompleteNormedModule` (packed class) -> structure `CompleteNormedModule` + + canonicals `R_regular_completeType`, `R_regular_CompleteNormedModule` -> + instance using `Uniform_isComplete.Build` + + canonicals `R_completeType` and `R_CompleteNormedModule` -> instance using `Complete.on` + + now use `cvgn` instead of `cvg`: + * lemma `cvg_seq_bounded` + +- in `Rstruct.v`: + + canonicals `R_eqMixin`, `R_eqType` -> instance using `hasDecEq.Build` + + definition `R_choiceMixin` and canonical `R_choiceType` -> instance using `hasChoice.Build` + + definition `R_zmodMixin` and canonical `R_zmodType` -> instance using `isZmodule.Build` + + definition `R_ringMixin` and canonicals `R_ringType`, `R_comRingType` -> + instances using `Zmodule_isRing.Build`, `Ring_hasCommutativeMul.Build` + + canonicals `Radd_monoid`, `Radd_comoid` -> instance using `isComLaw.Build` + + canonicals `Rmul_monoid`, `Rmul_comoid` -> instance using `isComLaw.Build` + + canonical `Rmul_mul_law` -> instance using `isMulLaw.Build` + + canonical `Radd_add_law` -> instance using `isAddLaw.Build` + + definition `R_unitRingMixin` and canonical `R_unitRing` -> instance using `Ring_hasMulInverse.Build` + + canonicals `R_comUnitRingType` and `R_idomainType` -> + instance using `ComUnitRing_isIntegral.Build` + + in lemma `R_fieldMixin`: `GRing.Field.mixin_of` -> `GRing.field_axiom` + + definition `Definition` and canonical `R_fieldType` -> instance using `UnitRing_isField.Build` + + definition `R_numMixin`, canonicals `R_porderType`, `R_numDomainType`, `R_normedZmodType`, `R_numFieldType` + -> instance using `IntegralDomain_isNumRing.Build` + + in lemma `R_total`: `totalPOrderMixin` -> `total` + + canonicals `R_latticeType`, `R_distrLatticeType`, `R_orderType`, `R_realDomainType`, `R_realFieldType` + -> instance using `POrder_isTotal.Build` + + in lemmas `Rarchimedean_axiom`, `Rreal_closed_axiom`: `R_numDomainType` -> `[the numDomainType of R : Type]` + + canonical `R_realArchiFieldType` -> instance using `RealField_isArchimedean.Build` + + canonical `R_rcfType` -> instance using `RealField_isClosed.Build` + + definition `real_realMixin` and canonical `real_realType` -> instance using `ArchimedeanField_isReal.Build` +- in `prodnormedzmodule.v`: + + definition `normedZmodMixin` and canonical `normedZmodType` -> + instance using `Num.Zmodule_isNormed.Build` + +- in `ereal.v`: + + canonical `ereal_pointed` -> instance using `isPointed.Build` + + definitions `ereal_dnbhs`, `ereal_nbhs` -> now use `set_system` + + canonical `ereal_ereal_filter` -> instance using `hasNbhs.Build` + + definition `ereal_topologicalMixin`, canonical `ereal_topologicalType`, + definitions `ereal_pseudoMetricType_mixin`, `ereal_uniformType_mixin`, + canonicals `ereal_uniformType`, `ereal_pseudoMetricType` -> + instance using `Nbhs_isPseudoMetric.Build` + +- "moved" from `normedtype.v` to `Rstruct.v`: + + canonicals `R_pointedType`, `R_filteredType`, `R_topologicalType`, `R_uniformType`, + `R_pseudoMetricType` -> instance using `PseudoMetric.copy` + +- in `realfun.v`: + + now explicitly display the filter in the notation `X --> Y`: + * lemma s`cvg_at_rightP`, `cvg_at_leftP`, `cvge_at_rightP`, `cvge_at_leftP` + +- in `sequences.v`: + + the lemmas and the notations (in particular, bigop notations) that were using + `cvg` or `cvg (... @ \oo)`/`lim` are now using `cvgn`/`limn` + and now explicitly mention the filter in the notation `X --> Y` + +- in `trigo.v`: + + now make explicit mention of the filter: + * definitions `sin`, `cos` + * lemmas `cvg_series_cvg_series_group`, `lt_sum_lim_series`, `is_cvg_series_sin_coeff`, + `sinE`, `cvg_sin_coeff'`, `is_cvg_series_cos_coeff`, `cosE`, `cvg_cos_coeff'` + +- in `itv.v`: + + canonical `itv_subType` -> instance using `[isSub for ... ]` + + definitions `itv_eqMixin`, `itv_choiceMixin` and canonicals `itv_eqType`, `itv_choiceType` + -> instance using `[Choice of ... by <:]` + + definition `itv_porderMixin` and canonical `itv_porderType` -> instance using + `[SubChoice_isSubPOrder of ... by <: with ...]` + +- in `landau.v`: + + now use `set_system` + * structures `littleo_type`, `bigO_type`, `bigOmega_type`, `bigTheta_type` + * lemmas `littleo_class`, `littleoE`, `littleo`, `bigO_exP`, `bigO_class`, + `bigO_clone`, `bigOP`, `bigOE`, `bigOmegaP`, `bigThetaP` + * definitions `littleo_clone`, `the_littleo`, `littleoP`, `the_bigO`, + `bigOmega_clone`, `the_bigOmega`, `is_bigOmega`, `bigTheta_clone`, + `is_bigTheta` + * variants `littleo_spec`, `bigOmega_spec`, `bigTheta_spec` + * notation `PhantomF` + * facts `is_bigOmega_key`, `is_bigTheta_key` + * canonicals `the_littleo_littleo`, `the_bigO_bigO`, `the_littleo_bigO`, + `is_bigOmega_keyed`, `the_bigOmega_bigOmega`, `is_bigTheta_keyed`, + `the_bigTheta_bigTheta` + + canonical `littleo_subtype` -> instance using `[isSub for ...]` + + canonical `bigO_subtype` -> instance using `[isSub for ...]` + + in `linear_for_continuous`: + * `GRing.Scale.op s_law` -> `GRing.Scale.Law.sort` + * argument `s_law` removed + + canonical `bigOmega_subtype` -> instance using `[isSub for ...]` + + canonical `bigTheta_subtype` -> instance using `[isSub for ...]` + +- in `forms.v`: + + module `Bilinear` (packed class) -> mixin `isBilinear`, structure `Bilinear`, + definition `bilinear_for`, factory `bilinear_isBilinear`, new module `Bilinear` + containing the definition `map` + + canonical `mulmx_bilinear` -> lemma `mulmx_is_bilinear` and instance using + `bilinear_isBilinear.Build` + +- in `derive.v` + + in notation `'d`, `differentiable`, `is_diff`: `[filter of ...]` -> `nbhs F` + + canonical `mulr_linear` -> instance using `isLinear.Build` + + canonical `mulr_rev_linear` -> instance using `isLinear.Build` + + canonical `mulr_bilinear` -> lemma `mulr_is_bilinear` and instance using `bilinear_isBilinear.Build` + + `set (set ...)` -> `set_system ...` +- in `esum.v`: + + several occurrences of `cvg`/`lim` changed to `cvgn`/`limn` and + usages of the notation `X --> Y` changed to `X @ F --> Y` (with an explicit filter) + * `is_cvg_pseries_inside_norm` + * `is_cvg_pseries_inside` + * `pseries_diffs_equiv` + * `is_cvg_pseries_diffs_equiv` + * `pseries_snd_diffs` + * `expRE` + * `dvg_riemannR` + +- in `numfun.v`: + + canonicals `fimfun_mul`, `fimfun_ring`, `fimfun_ringType`, definition `fimfun_ringMixin` + -> instances using `GRing.isMulClosed.Build` and `[SubZmodule_isSubRing of ... by <:]` + + definition `fimfun_comRingMixin`, canonical `fimfun_comRingType` -> + instance using `[SubRing_isSubComRing of ... by <:]` + +- in `measure.v` + + canonicals `salgebraType_eqType`, `salgebraType_choiceType`, `salgebraType_ptType` + -> instance using `Pointed.on` + + filter now explicit in: + * definitions `sigma_additive`, `semi_sigma_additive` + * lemmas `nondecreasing_cvg_mu`, `nonincreasing_cvg_mu` + + canonicals `ring_eqType`, `ring_choiceType`, `ring_ptType` -> instance using `Pointed.on` + +- in `lebesgue_measure.v`: + + filter now explicit in lemmas `emeasurable_fun_cvg`, + `ae_pointwise_almost_uniform` + +- in `lebesgue_integral.v`: + + canonical `mfun_subType` -> instance using `isSub.Build` + + definitions `mfuneqMixin`, `mfunchoiceMixin`, canonicals `mfuneqType`, + `mfunchoiceType` -> instance using `[Choice of ... by <:]` + + canonicals `mfun_add`, `mfun_zmod`, `mfun_mul`, `mfun_subring`, + `mfun_zmodType`, `mfun_ringType`, `mfun_comRingType`, + definitions `mfun_zmodMixin`, `mfun_ringMixin`, `mfun_comRingMixin`, + -> instances using `GRing.isSubringClosed.Build` and `[SubChoice_isSubComRing of ... <:]` + + canonical `sfun_subType` -> instance using `isSub.Build` + + definitions `sfuneqMixin`, `sfunchoiceMixin`, + canonicals `sfuneqType`, `sfunchoiceType` -> + instance using `[Choice of .. by <:]` + + canonicals `sfun_add`, `sfun_zmod`, `sfun_mul`, `sfun_subring`, `sfun_zmodType`, + `sfun_ringType`, `sfun_comRingType`, + definitions `sfun_zmodMixin`, `sfun_ringMixin`, `sfun_comRingMixin` + -> instances using `GRing.isSubringClosed.Build` and `[SubChoice_isSubComRing of ... by <:]` + + now use `cvgn`/`limn` instead of `cvg`/`lim`: + * lemmas `is_cvg_sintegral`, `nd_sintegral_lim_lemma`, `nd_sintegral_lim`, `nd_ge0_integral_lim`, + `dvg_approx`, `ecvg_approx` + + filter now explicit in: + * lemmas `approximation`, `approximation_sfun`, `cvg_monotone_convergence` + +- in `kernel.v`: + + notation `X --> Y` changed to `X @ F --< Y` + * `measurable_fun_xsection_integral` + + definition `prob_pointed` and canonical `probability_ptType` -> + instance using `isPointed.Build` + + canonicals `probability_eqType`, `probability_choiceType` -> + instance using `gen_eqMixin` and `gen_choiceMixin` + +- in `summability.v`: + + `totally` now uses `set_system` + +- in `altreals/discrete.v`: + + canonical `pred_sub_subTypeP` -> instance using `[isSub for ...]` + + definition `pred_sub_eqMixin` and canonical `pred_sub_eqType` -> + instance using `[Equality of ... by <:]` + + definition `pred_sub_choiceMixin` and canonical `pred_sub_choiceType` -> + instance using `[Choice of ... <:]` + + definition `pred_sub_countMixin` and `pred_sub_countType` -> + instance using `[Countable of ... by <:]` + + definitions `countable_countMixin` and `countable_countType` -> `countable_countMixin` + + definitions `countable_choiceMixin` and `countable_choiceType` -> `countable_choiceMixin` + +- in `altreals/xfinmap.v`: + + in lemmas `enum_fset0` and `enum_fset1`: notation `[fintype of ...]` -> type constraint `... : finType` + +- in `misc/uniform_bigO.v`: + + in definition `OuO`: `[filter of ...]` -> `nbhs ...` + +### Generalized + +- in `cantor.v`: + + in definition `cantor_space`: `product_uniformType` -> `prod_topology` + * instances using `Pointed.on`, `Nbhs.on`, `Topological.on` + +- in `topology.v`: + + now use `nbhsType` instead of `topologicalType` + * lemma `near_fun` + * definition `discrete_space` + * definition `discrete_uniform_mixin` + * definition `discrete_ball`, lemma `discrete_ball_center`, + definition `discrete_pseudometric_mixin` + +### Removed + +- in `mathcomp_extra.v`: + + coercion `choice.Choice.mixin` + + lemmas `bigminr_maxr`, definitions `AC_subdef`, `oAC`, `opACE`, canonicals `opAC_law`, `opAC_com_law` + + lemmas `some_big_AC`, `big_ACE`, `big_undup_AC`, `perm_big_AC`, `big_const_idem`, + `big_id_idem`, `big_mkcond_idem`, `big_split_idem`, `big_id_idem_AC`, `bigID_idem`, + `big_rem_AC`, `bigD1_AC`, `sub_big`, `sub_big_seq`, `sub_big_seq_cond`, `uniq_sub_big`, + `uniq_sub_big_cond`, `sub_big_idem`, `sub_big_idem_cond`, `sub_in_big`, `le_big_ord`, + `subset_big`, `subset_big_cond`, `le_big_nat`, `le_big_ord_cond` + + lemmas `bigmax_le`, `bigmax_lt`, `lt_bigmin`, `le_bigmin` + + lemmas `bigmax_mkcond`, `bigmax_split`, `bigmax_idl`, `bigmax_idr`, `bigmaxID` + + lemmas `sub_bigmax`, `sub_bigmax_seq`, `sub_bigmax_cond`, `sub_in_bigmax`, + `le_bigmax_nat`, `le_bigmax_nat_cond`, `le_bigmax_ord`, `le_bigmax_ord_cond`, + `subset_bigmax`, `subset_bigmax_cond` + + lemmas `bigmaxD1`, `le_bigmax_cond`, `le_bigmax`, `bigmax_sup`, `bigmax_leP`, + `bigmax_ltP`, `bigmax_eq_arg`, `eq_bigmax`, `le_bigmax2` + + lemmas `bigmin_mkcond`, `bigmin_split`, `bigmin_idl`, `bigmin_idr`, `bigminID` + + lemmas `sub_bigmin`, `sub_bigmin_cond`, `sub_bigmin_seq`, `sub_in_bigmin`, + `le_bigmin_nat`, `le_bigmin_nat_cond`, `le_bigmin_ord`, `le_bigmin_ord_cond`, + `subset_bigmin`, `subset_bigmin_cond` + + lemmas `bigminD1`, `bigmin_le_cond`, `bigmin_le`, `bigmin_inf`, `bigmin_geP`, + `bigmin_gtP`, `bigmin_eq_arg`, `eq_bigmin` + +- in `boolp.v`: + + definitions `dep_arrow_eqType`, `dep_arrow_choiceClass`, `dep_arrow_choiceType` + +- in `classical_sets.v`: + + notations `PointedType`, `[pointedType of ...]` + +- in `cardinality.v`: + + lemma `countable_setT_countMixin` + +- in `constructive_ereal.v`: + + canonicals `isLaw.Build`, `mine_comoid` + +- in `topology.v`: + + structure `source`, definition `source_filter` + + definition `filter_of`, notation `[filter of ...]` (now replaced by `nbhs`), lemma `filter_of_filterE` + + definition `open_of_nbhs` + + definition `open_from`, lemma `open_fromT` + + canonical `eventually_filter_source` + + canonical `discrete_topological_mixin` + + canonical `set_filter_source` + + definitions `filtered_of_normedZmod`, `pseudoMetric_of_normedDomain` + + definitions `fct_UniformFamily` (use `uniform_fun_family` instead), canonicals `fct_UniformFamilyFilteredType`, + `fct_UniformFamilyTopologicalType`, `fct_UniformFamilyUniformType` + +- in `cantor.v`: + + definition `pointed_discrete` + +- in `normedtype.v`: + + `filtered_of_normedZmod` + + section `pseudoMetric_of_normedDomain` + * lemmas `ball_norm_center`, `ball_norm_symmetric`, `ball_norm_triangle`, `nbhs_ball_normE` + * definition `pseudoMetric_of_normedDomain` + + lemma `normrZ` + + canonical `matrix_normedZmodType` + + lemmas `eq_cvg`, `eq_is_cvg` + +- in `convex.v`: + + field `convexspacechoiceclass`, canonicals `conv_eqType`, `conv_choiceType`, `conv_choiceType` + +- in `measure.v`: + + field `ptclass` in mixin `isSemiRingOfSets` + + canonicals `ringOfSets_eqType`, `ringOfSets_choiceType`, + `ringOfSets_ptType`, `algebraOfSets_eqType`, `algebraOfSets_choiceType`, + `algebraOfSets_ptType`, `measurable_eqType`, `measurable_choiceType`, + `measurable_ptType` + + field `ptclass` in factory `isAlgebraOfSets` + + field `ptclass` in factory `isMeasurable` + +- in `lebesgue_measure.v`: + + no more "pointed class" argument in definition `ereal_isMeasurable` + +- in `lebesgue_stieltjes_measure.v` + + lemma `sigmaT_finite_lebesgue_stieltjes_measure` turned into a `Let` + +- in `altreals/discrete.v`: + + notation `[countable of ...]` ## [0.7.0] - 2024-01-19 diff --git a/CHANGELOG_UNRELEASED.md b/CHANGELOG_UNRELEASED.md index 5a37c0fc1..67bb43c3b 100644 --- a/CHANGELOG_UNRELEASED.md +++ b/CHANGELOG_UNRELEASED.md @@ -4,653 +4,16 @@ ### Added -- in `cantor.v`: - + definition `pointed_principal_filter`, instances using `Pointed.on` and `hasNbhs.Build` - + definition `pointed_discrete_topology` - + lemma `discrete_pointed` - + lemma `discrete_bool_compact` - -- in `charge.v`: - + `cscale` instances using `SigmaFinite_isFinite.Build` and `isAdditiveCharge.Build` - -- in `convex.v`: - + definition `convex_lmodType` with instances using `Choice.on` and `isConvexSpace.Build` - + definition `convex_realDomainType` with instance using `isConvexSpace.Build` - -- in `constructive_ereal.v`: - + definition `dEFin` - + notations `%:dE`, `%:E` (`ereal_dual_scope`) - + notation `\bar^d ...` (`type_scope`) for dual extended real numbers - + instance using `isNmodule.Build` for `\bar` - + instances using `Choice.on` and `isNmodule.Build` for `\bar^d` - + lemma `EFin_semi_additive` - + lemmas `dEFinE`, `dEFin_semi_additive` - + instance using `isSemiAdditive.Build` for `\bar^d` - + canonical `dEFin_snum` - -- in `derive.v`: - + lemma `cvg_at_rightE`, `cvg_at_leftE` - -- in `lebesgue_integral.v`: - + mixin `isNonNegFun`, notations `{nnfun _ >-> _}`, `[nnfun of _]` - + section `ring` - * lemmas `fimfun_mulr_closed`, instances using `GRing.isMulClosed.Build`, - `[SubZmodule_isSubRing of ... by <:]` - * lemmas `fimfunM`, `fimfun1`, `fimfun_prod`, `fimfunX`, - * lemma `indic_fimfun_subproof`, instance using `indic_fimfun_subproof` - * definition `indic_fimfun` - * instance using `FImFun.copy`, definition `scale_fimfun` - + section `comring` - * instance using `[SubRing_isSubComRing of ... by <:]` - * instance using `FImFun.copy` - + lemmas `fimfunE`, `fimfunEord`, `trivIset_preimage1`, `trivIset_preimage1_in` - + section `fimfun_bin` - * lemma `max_fimfun_subproof`, instance using `max_fimfun_subproof` - + factory `FiniteDecomp` - -- in `lebesgue_stieltjes_measure.v`: - + instance on `ocitv_type` using `Pointed.on` - -- in `normedtype.v`: - + definition `urysohnType` with instances using `Pointed.on` and `isUniform.Build` - -- in `reals.v`: - + definition `Rint_pred` - -- in `topology.v` - + definition `set_system`, identity coercion `set_system_to_set` - with instances using `Equality.on`, `Choice.on`, `Pointed.on`, - `isFiltered.Build` - + mixin `selfFiltered`, factory `hasNbhs`, structure `Nbhs`, - type `nbhsType` - + instance for matrices using `selfFiltered.Build` - + lemmas `cvg_in_ex`, `cvg_inP`, `cvg_in_toP`, `dvg_inP`, `cvg_inNpoint`, - `eq_is_cvg_in` - + notations `E @[ x \oo ]`, `limn`, `cvgn` - + definition `continuous_at` - + definitions `weak_topology`, `sup_topology`, `prod_topology` - + definition `prod_topo_apply` - + definition `discrete_topology` - + instead of `zmodType` using `isPointed.Build` - + definition `pointwise_cvgE`, - instance using `Uniform.copy` for `{ptws _ -> _}` - + definition `compact_open_of_nbhs`, lemmas `compact_openK_nbhsE_subproof`, - `compact_openK_openE_subproof` - ### Changed -- in `boolp.v`: - + in lemma `gen_choiceMixin`: `Choice.mixin_of` -> `hasChoice` - + in definition `gen_eqMixin`: `EqMixin` -> `hasDecEq.Build` - + canonical `dep_arrow_eqType` -> instance using `gen_eqMixin` - + canonical `dep_arrow_choiceType` -> instance using `gen_choiceMixin` - + canonical `Prop_eqType` -> instance using `gen_eqMixin` - + canonical `Prop_choiceType` -> instance using `gen_choiceMixin` - + canonical `classicType_eqType` -> instance using `gen_eqMixin` - + canonical `classicType_choiceType` -> instance using `gen_choiceMixin` - + canonical `eclassicType_eqType` -> instance using `Equality.copy` - + canonical `eclassicType_choiceType` -> instance using `gen_choiceMixin` - + definition `porderMixin` and canonical `porderType` -> instance using `isPOrder.Build` - + definition `latticeMixin` and canonical `latticeType` -> instance using `POrder_isLattice.Build` - -- in `cardinality.v`: - + canonical `rat_pointedType` -> instance using `isPointed.Build` - + canonical `fimfun_subType` -> instance using `isSub.Build` - + definition `fimfuneqMixin` and canonical `fimfuneqType` -> instance using `[Equality of ... by <:]` - + definition `fimfunchoiceMixin` and canonical `fimfunchoiceType` -> instance using `[Choice of ... by <:]` - + canonicals `fimfun_add`, `fimfun_zmod`, `fimfun_zmodType`, and definition `fimfun_zmodMixin` -> - instances using `isZmodClosed.Build` and `[SubChoice_isSubZmodule of ... <:]` - -- in `classical_sets.v`: - + canonicals - `setU_monoid`, `setU_comoid`, `setU_mul_monoid`, `setI_monoid`, - `setI_comoid`, `setI_mul_monoid`, `setU_add_monoid`, `setI_add_monoid` - -> instances using - `isComLaw.Build`, `isMulLaw.Build`, `isComLaw.Build`, `isMulLaw.Build`, `isAddLaw.Build`, `isAddLaw.Build` - + module `Pointed` (packed class) -> mixin `isPointed`, structure `Pointed` - + canonical `arrow_pointedType` and definition `dep_arrow_pointedType` -> instance using `isPointed.Build` - + canonicals - `unit_pointedType`, `bool_pointedType`, `Prop_pointedType`, `nat_pointedType`, - `prod_pointedType`, `matrix_pointedType`, `option_pointedType`, `pointed_fset` - -> instances using - `isPointed.Build` - + module `Empty` (packed class) -> mixin `isEmpty`, structure `Empty`, factories `Choice_isEmpty`, `Type_isEmpty` - + definition `False_emptyMixin` and canonicals `False_eqType`, - `False_choiceType`, `False_countType`, `False_finType`, `False_emptyType` -> - instance using `Type_isEmpty.Build` - + definition `void_emptyMixin` and canonical `void_emptyType` -> instance using `isEmpty.Build` - + definition `orderMixin` and canonicals `porderType`, `latticeType`, `distrLatticeType` -> - instances using `Choice.copy` and `isMeetJoinDistrLattice.Build` - + canonicals `bLatticeType`, `tbLatticeType`, `bDistrLatticeType`, `tbDistrLatticeType` -> - instances using `hasBottom.Build` and `hasTop.Build` - + canonical `cbDistrLatticeType` -> instance using `hasRelativeComplement.Build` - + canonical `ctbDistrLatticeType` -> instance using `hasComplement.Build` - -- in `functions.v`: - + notation `split` - + notation `\_` moved from `fun_scope` to `function_scope` - + notations `pinv`, `pPbij`, `pPinj`, `injpPfun`, `funpPinj` - + in definition `fct_zmodMixin`: `ZmodMixin` -> `isZmodule.Build` - + canonical `fct_zmodType` -> instance using `fct_zmodMixin` - + in definition `fct_ringMixin`: `RingMixin` -> `Zmodule_isRing.Build` - + canonical `fct_ringType` -> instance using `fct_ringMixin` - + canonical `fct_comRingType` -> - definition and instance using `Ring_hasCommutativeMul.Build` and `fct_comRingType` - + definition `fct_lmodMixin` and canonical `fct_lmodType` -> - definition `fct_lmodMixin` and instance using `fct_lmodMixin` - -- in `Rstruct.v`: - + canonicals `R_eqMixin`, `R_eqType` -> instance using `hasDecEq.Build` - + definition `R_choiceMixin` and canonical `R_choiceType` -> instance using `hasChoice.Build` - + definition `R_zmodMixin` and canonical `R_zmodType` -> instance using `isZmodule.Build` - + definition `R_ringMixin` and canonicals `R_ringType`, `R_comRingType` -> - instances using `Zmodule_isRing.Build`, `Ring_hasCommutativeMul.Build` - + canonicals `Radd_monoid`, `Radd_comoid` -> instance using `isComLaw.Build` - + canonicals `Rmul_monoid`, `Rmul_comoid` -> instance using `isComLaw.Build` - + canonical `Rmul_mul_law` -> instance using `isMulLaw.Build` - + canonical `Radd_add_law` -> instance using `isAddLaw.Build` - + definition `R_unitRingMixin` and canonical `R_unitRing` -> instance using `Ring_hasMulInverse.Build` - + canonicals `R_comUnitRingType` and `R_idomainType` -> - instance using `ComUnitRing_isIntegral.Build` - + in lemma `R_fieldMixin`: `GRing.Field.mixin_of` -> `GRing.field_axiom` - + definition `Definition` and canonical `R_fieldType` -> instance using `UnitRing_isField.Build` - + definition `R_numMixin`, canonicals `R_porderType`, `R_numDomainType`, `R_normedZmodType`, `R_numFieldType` - -> instance using `IntegralDomain_isNumRing.Build` - + in lemma `R_total`: `totalPOrderMixin` -> `total` - + canonicals `R_latticeType`, `R_distrLatticeType`, `R_orderType`, `R_realDomainType`, `R_realFieldType` - -> instance using `POrder_isTotal.Build` - + in lemmas `Rarchimedean_axiom`, `Rreal_closed_axiom`: `R_numDomainType` -> `[the numDomainType of R : Type]` - + canonical `R_realArchiFieldType` -> instance using `RealField_isArchimedean.Build` - + canonical `R_rcfType` -> instance using `RealField_isClosed.Build` - + definition `real_realMixin` and canonical `real_realType` -> instance using `ArchimedeanField_isReal.Build` - -- in `altreals/discrete.v`: - + canonical `pred_sub_subTypeP` -> instance using `[isSub for ...]` - + definition `pred_sub_eqMixin` and canonical `pred_sub_eqType` -> - instance using `[Equality of ... by <:]` - + definition `pred_sub_choiceMixin` and canonical `pred_sub_choiceType` -> - instance using `[Choice of ... <:]` - + definition `pred_sub_countMixin` and `pred_sub_countType` -> - instance using `[Countable of ... by <:]` - + definitions `countable_countMixin` and `countable_countType` -> `countable_countMixin` - + definitions `countable_choiceMixin` and `countable_choiceType` -> `countable_choiceMixin` - -- in `altreals/xfinmap.v`: - + in lemmas `enum_fset0` and `enum_fset1`: notation `[fintype of ...]` -> type constraint `... : finType` - -- in `cantor.v`: - + in definition `tree_of` and lemma `cantor_like_finite_prod`: - `pointed_discrete` -> `pointed_discrete_topology` - -- in `constructive_ereal.v`: - + definition `ereal_eqMixin` and canonical `ereal_eqType` -> instance using `hasDecEq.Build` - + definition `ereal_choiceMixin` and canonical `ereal_choiceType` -> instance using `Choice.copy` - + definition `ereal_countMixin` and `ereal_countType` -> instance using `PCanIsCountable` - + definition `ereal_porderMixin` and canonical `Choice.copy` -> instance using `isPOrder.Build` - + in lemma `le_total_ereal` : `le_total_ereal` -> `total` - + canonicals `ereal_latticeType`, `ereal_distrLatticeType`, `ereal_orderType`, `ereal_blatticeType`, - `ereal_tblatticeType`, lemmas `ereal_blatticeMixin`, `ereal_blatticeMixin` -> - instances using `POrder_isTotal.Build`, `hasBottom.Build`, `hasTop.Build` - + canonicals `adde_monoid`, `adde_comoid`, `mule_mulmonoid` -> instance using `isMulLaw.Build` - + notations `maxe`, `mine`: `fun_scope` -> `function_scope` - + canonicals `mule_monoid`, `mule_comoid` -> instance using `isComLaw.Build` - + canonicals `maxe_monoid`, `maxe_comoid` -> instance using `isLaw.Build` - -- in `derive.v` - + in notation `'d`, `differentiable`, `is_diff`: `[filter of ...]` -> `nbhs F` - + canonical `mulr_linear` -> instance using `isLinear.Build` - + canonical `mulr_rev_linear` -> instance using `isLinear.Build` - + canonical `mulr_bilinear` -> lemma `mulr_is_bilinear` and instance using `bilinear_isBilinear.Build` - + `set (set ...)` -> `set_system ...` - -- in `ereal.v`: - + canonical `ereal_pointed` -> instance using `isPointed.Build` - + definitions `ereal_dnbhs`, `ereal_nbhs` -> now use `set_system` - + canonical `ereal_ereal_filter` -> instance using `hasNbhs.Build` - + definition `ereal_topologicalMixin`, canonical `ereal_topologicalType`, - definitions `ereal_pseudoMetricType_mixin`, `ereal_uniformType_mixin`, - canonicals `ereal_uniformType`, `ereal_pseudoMetricType` -> - instance using `Nbhs_isPseudoMetric.Build` - -- in `esum.v`: - + several occurrences of `cvg`/`lim` changed to `cvgn`/`limn` and - usages of the notation `X --> Y` changed to `X @ F --> Y` (with an explicit filter) - * `is_cvg_pseries_inside_norm` - * `is_cvg_pseries_inside` - * `pseries_diffs_equiv` - * `is_cvg_pseries_diffs_equiv` - * `pseries_snd_diffs` - * `expRE` - * `dvg_riemannR` - -- in `forms.v`: - + module `Bilinear` (packed class) -> mixin `isBilinear`, structure `Bilinear`, - definition `bilinear_for`, factory `bilinear_isBilinear`, new module `Bilinear` - containing the definition `map` - + canonical `mulmx_bilinear` -> lemma `mulmx_is_bilinear` and instance using - `bilinear_isBilinear.Build` - -- in `itv.v`: - + canonical `itv_subType` -> instance using `[isSub for ... ]` - + definitions `itv_eqMixin`, `itv_choiceMixin` and canonicals `itv_eqType`, `itv_choiceType` - -> instance using `[Choice of ... by <:]` - + definition `itv_porderMixin` and canonical `itv_porderType` -> instance using - `[SubChoice_isSubPOrder of ... by <: with ...]` - -- in `kernel.v`: - + notation `X --> Y` changed to `X @ F --< Y` - * `measurable_fun_xsection_integral` - + definition `prob_pointed` and canonical `probability_ptType` -> - instance using `isPointed.Build` - + canonicals `probability_eqType`, `probability_choiceType` -> - instance using `gen_eqMixin` and `gen_choiceMixin` - -- in `landau.v`: - + now use `set_system` - * structures `littleo_type`, `bigO_type`, `bigOmega_type`, `bigTheta_type` - * lemmas `littleo_class`, `littleoE`, `littleo`, `bigO_exP`, `bigO_class`, - `bigO_clone`, `bigOP`, `bigOE`, `bigOmegaP`, `bigThetaP` - * definitions `littleo_clone`, `the_littleo`, `littleoP`, `the_bigO`, - `bigOmega_clone`, `the_bigOmega`, `is_bigOmega`, `bigTheta_clone`, - `is_bigTheta` - * variants `littleo_spec`, `bigOmega_spec`, `bigTheta_spec` - * notation `PhantomF` - * facts `is_bigOmega_key`, `is_bigTheta_key` - * canonicals `the_littleo_littleo`, `the_bigO_bigO`, `the_littleo_bigO`, - `is_bigOmega_keyed`, `the_bigOmega_bigOmega`, `is_bigTheta_keyed`, - `the_bigTheta_bigTheta` - + canonical `littleo_subtype` -> instance using `[isSub for ...]` - + canonical `bigO_subtype` -> instance using `[isSub for ...]` - + in `linear_for_continuous`: - * `GRing.Scale.op s_law` -> `GRing.Scale.Law.sort` - * argument `s_law` removed - + canonical `bigOmega_subtype` -> instance using `[isSub for ...]` - + canonical `bigTheta_subtype` -> instance using `[isSub for ...]` - -- in `lebesgue_integral.v`: - + canonical `mfun_subType` -> instance using `isSub.Build` - + definitions `mfuneqMixin`, `mfunchoiceMixin`, canonicals `mfuneqType`, - `mfunchoiceType` -> instance using `[Choice of ... by <:]` - + canonicals `mfun_add`, `mfun_zmod`, `mfun_mul`, `mfun_subring`, - `mfun_zmodType`, `mfun_ringType`, `mfun_comRingType`, - definitions `mfun_zmodMixin`, `mfun_ringMixin`, `mfun_comRingMixin`, - -> instances using `GRing.isSubringClosed.Build` and `[SubChoice_isSubComRing of ... <:]` - + canonical `sfun_subType` -> instance using `isSub.Build` - + definitions `sfuneqMixin`, `sfunchoiceMixin`, - canonicals `sfuneqType`, `sfunchoiceType` -> - instance using `[Choice of .. by <:]` - + canonicals `sfun_add`, `sfun_zmod`, `sfun_mul`, `sfun_subring`, `sfun_zmodType`, - `sfun_ringType`, `sfun_comRingType`, - definitions `sfun_zmodMixin`, `sfun_ringMixin`, `sfun_comRingMixin` - -> instances using `GRing.isSubringClosed.Build` and `[SubChoice_isSubComRing of ... by <:]` - + now use `cvgn`/`limn` instead of `cvg`/`lim`: - * lemmas `is_cvg_sintegral`, `nd_sintegral_lim_lemma`, `nd_sintegral_lim`, `nd_ge0_integral_lim`, - `dvg_approx`, `ecvg_approx` - + filter now explicit in: - * lemmas `approximation`, `approximation_sfun`, `cvg_monotone_convergence` - -- in `lebesgue_measure.v`: - + filter now explicit in lemmas `emeasurable_fun_cvg`, - `ae_pointwise_almost_uniform` - -- in `measure.v` - + canonicals `salgebraType_eqType`, `salgebraType_choiceType`, `salgebraType_ptType` - -> instance using `Pointed.on` - + filter now explicit in: - * definitions `sigma_additive`, `semi_sigma_additive` - * lemmas `nondecreasing_cvg_mu`, `nonincreasing_cvg_mu` - + canonicals `ring_eqType`, `ring_choiceType`, `ring_ptType` -> instance using `Pointed.on` - -- in `misc/uniform_bigO.v`: - + in definition `OuO`: `[filter of ...]` -> `nbhs ...` - -- in `normedtype.v`: - + module `PseudoMetricNormedZmodule` (packed class) -> - mixin `NormedZmod_PseudoMetric_eq` (with field `pseudo_metric_ball_norm`), - structure `PseudoMetricNormedZmod` - + now use `set_system`: - * definitions `pinfty_nbhs`, `ninfty_nbhs`, `dominated_by`, `strictly_dominated_by`, - `bounded_near`, `sub_klipschitz`, `lipschitz_on`, `sub_lipschitz` - * lemmas `cvgrnyP`, `cvgenyP`, `fcvgrPdist_lt`, `cvgrPdist_lt`, `cvgrPdistC_lt`, - `cvgr_dist_lt`, `cvgr_distC_lt`, `cvgr_dist_le`, `cvgr_distC_le`, `cvgr0Pnorm_lt`, - `cvgr0_norm_lt`, `cvgr0_norm_le`, `cvgrPdist_le`, `cvgrPdist_ltp`, `cvgrPdist_lep`, - `cvgrPdistC_le`, `cvgrPdistC_ltp`, `cvgrPdistC_lep`, `cvgr0Pnorm_le`, `cvgr0Pnorm_ltp`, - `cvgr0Pnorm_lep`, `cvgr_norm_lt`, `cvgr_norm_le`, `cvgr_norm_gt`, `cvgr_norm_ge`, - `cvgr_neq0`, `real_cvgr_lt`, `real_cvgr_le`, `real_cvgr_gt`, `real_cvgr_ge`, `cvgr_lt`, - `cvgr_le`, `cvgr_gt`, `cvgr_ge`, `sub_dominatedl`, `sub_dominatedr`, `ex_dom_bound`, - `ex_strict_dom_bound`, `sub_boundedr`, `sub_boundedl`, `ex_bound`, `ex_strict_bound`, - `ex_strict_bound_gt0`, `klipschitzW`, `cvg_bounded`, `fcvgr2dist_ltP`, `cvgr2dist_ltP`, - `cvgr2dist_lt` - + module `NormedModule` (packed class) -> - mixin `PseudoMetricNormedZmod_Lmodule_isNormedModule`, - structure `NormedModule` - + module and section `regular_topology` -> section `regular_topology` - with instances using `Num.NormedZmodule.on`, `NormedZmod_PseudoMetric_eq.Build`, - `seudoMetricNormedZmod_Lmodule_isNormedModule.Build` - + in module `numFieldNormedType` - * `realType` instances using `GRing.ComAlgebra.copy`, `Vector.copy`, `NormedModule.copy` - * `rcfType` instances using `GRing.ComAlgebra.copy`, `Vector.copy`, `NormedModule.copy` - * `archiFieldType` instances using `GRing.ComAlgebra.copy`, `Vector.copy`, `NormedModule.copy` - * `realFieldType` instances using `GRing.ComAlgebra.copy`, `Vector.copy`, `NormedModule.copy`, `Num.RealField.on` - * `numClosedFieldType` instances using `GRing.ComAlgebra.copy`, `Vector.copy`, `NormedModule.copy`, `Num.ClosedField.on` - * `numFieldType` instances using `GRing.ComAlgebra.copy`, `Vector.copy`, `NormedModule.copy`, `Num.NumField.on` - + in lemma `norm_lim_id`: now explicit use of `nbhs` - + definition `matrix_PseudoMetricNormedZmodMixin` and canonical `matrix_normedModType` - -> instance using `PseudoMetricNormedZmod_Lmodule_isNormedModule.Build` - + definition `prod_pseudoMetricNormedZmodMixin` and canonical `prod_normedModType` - -> instance using `PseudoMetricNormedZmod_Lmodule_isNormedModule.Build` - + module `CompleteNormedModule` (packed class) -> structure `CompleteNormedModule` - + canonicals `R_regular_completeType`, `R_regular_CompleteNormedModule` -> - instance using `Uniform_isComplete.Build` - + canonicals `R_completeType` and `R_CompleteNormedModule` -> instance using `Complete.on` - + now use `cvgn` instead of `cvg`: - * lemma `cvg_seq_bounded` - -- "moved" from `normedtype.v` to `Rstruct.v`: - + canonicals `R_pointedType`, `R_filteredType`, `R_topologicalType`, `R_uniformType`, - `R_pseudoMetricType` -> instance using `PseudoMetric.copy` - -- in `numfun.v`: - + canonicals `fimfun_mul`, `fimfun_ring`, `fimfun_ringType`, definition `fimfun_ringMixin` - -> instances using `GRing.isMulClosed.Build` and `[SubZmodule_isSubRing of ... by <:]` - + definition `fimfun_comRingMixin`, canonical `fimfun_comRingType` -> - instance using `[SubRing_isSubComRing of ... by <:]` - -- in `prodnormedzmodule.v`: - + definition `normedZmodMixin` and canonical `normedZmodType` -> - instance using `Num.Zmodule_isNormed.Build` - -- in `realfun.v`: - + now explicitly display the filter in the notation `X --> Y`: - * lemma s`cvg_at_rightP`, `cvg_at_leftP`, `cvge_at_rightP`, `cvge_at_leftP` - -- in `reals.v`: - + module `Real` (packed class) -> mixin `ArchimedeanField_isReal` - with fields `sup_upper_bound_subdef`, `sup_adherent_subdef`, - structure `Real` - + canonicals `Rint_keyed`, `Rint_opprPred`, `Rint_addrPred`, `Rint_mulrPred`, `Rint_zmodPred`, - `Rint_semiringPred`, `Rint_smulrPred`, `Rint_subringPred` - -> instance using `GRing.isSubringClosed.Build` - -- in `sequences.v`: - + the lemmas and the notations (in particular, bigop notations) that were using - `cvg` or `cvg (... @ \oo)`/`lim` are now using `cvgn`/`limn` - and now explicitly mention the filter in the notation `X --> Y` - -- in `signed.v`: - + definitions `signed_subType`, `signed_choiceMixin`, `signed_porderMixin`, - canonicals `signed_eqMixin`, `signed_eqType`, `signed_choiceType`, `signed_porderType` - -> instances using `[isSub for ...]` and `[POrder of ... by <:]` - + in lemma `signed_le_total`: `totalPOrderMixin` -> `total` - + canonicals `signed_latticeType`, `signed_distrLatticeType`, `signed_orderType` - -> instance using `Order.POrder_isTotal.Build` - -- in `summability.v`: - + `totally` now uses `set_system` - -- in `trigo.v`: - + now make explicit mention of the filter: - * definitions `sin`, `cos` - * lemmas `cvg_series_cvg_series_group`, `lt_sum_lim_series`, `is_cvg_series_sin_coeff`, - `sinE`, `cvg_sin_coeff'`, `is_cvg_series_cos_coeff`, `cosE`, `cvg_cos_coeff'` - -- in `topology.v`: - + canonicals `linear_eqType`, `linear_choiceType` -> - instances using `gen_eqMixin`, `gen_choiceMixin` - + canonical `gen_choiceMixin` -> instance using `isPointed.Build` - + module `Filtered` (packed class) -> mixin `isFiltered` with field `nbhs`, - structure `Filtered` - + now use `set_system`: - * definitions `filter_from`, `filter_prod`, `cvg_to`, `type_of_filter`, `lim_in`, - `Build_ProperFilter`, `filter_ex`, `fmap`, `fmapi`, `globally`, `in_filter_prod`, - `within`, `subset_filter`, `powerset_filter_from`, `principal_filter`, `locally_of`, - `sup_subbase`, `cluster`, `compact`, `near_covering`, `near_covering_within`, - `compact_near`, `nbhs_`, `weak_ent`, `sup_ent`, `cauchy`, `cvg_cauchy`, `cauchy_ex.Build`, - `cauchy_ball` - * classes `Filter`, `ProperFilter'`, `UltraFilter` - * instances `fmap_proper_filter`, `fmapi_filter`, `fmapi_proper_filter`, - `filter_prod_filter`, `filter_prod1`, `filter_prod2` - * record `in_filter` - * structure `filter_on` - * variant `nbhs_subspace_spec` - * lemmas `nearE`, `eq_near`, `nbhs_filterE`, `cvg_refl`, `cvg_trans`, `near2_curry`, - `near_swap`, `filterP_strong`, `filter_nbhsT`, `nearT`, `filter_not_empty_ex`, - `filter_ex_subproof`, `filter_getP`, `near`, `nearW`, `filterE`, `filter_app`, - `filter_app2`, `filter_app3`, `filterS2`, `filterS3`, `nearP_dep`, `filter2P`, - `filter_ex2`, `filter_fromP`, `filter_fromTP`, `filter_bigIP`, `filter_forall`, - `filter_imply`, `fmapEP`, `fmapiE`, `cvg_id`, `appfilterP`, `cvg_app`, `cvgi_app`, - `cvg_comp`, `cvgi_comp`, `near_eq_cvg`, `eq_cvg`, `neari_eq_loc`, `cvg_near_const`, - `near_pair`, `near_map`, `near_map2`, `near_mapi`, `filter_pair_set`, - `filter_pair_near_of`, `cvg_pair`, `cvg_comp2`, `near_powerset_map`, - `near_powerset_map_monoE`, `cvg_fmap`, `continuous_cvg`, `continuous_is_cvg`, - `continuous2_cvg`, `cvg_near_cst`, `is_cvg_near_cst`, `cvg_cst`, `is_cvg_cst`, - `fmap_within_eq`, `cvg_image`, `cvg_fmap2`, `cvg_within_filter`, `cvg_app_within`, - `meets_openr`, `meets_openl`, `meetsxx`, `proper_meetsxx`, `ultra_cvg_clusterE`, - `ultraFilterLemma`, `compact_ultra`, `proper_image`, `in_ultra_setVsetC`, - `ultra_image`, `filter_finI`, `close_cvg`, `discrete_cvg`, `nbhs_E`, `cvg_closeP`, - `cvg_mx_entourageP`, `cvg_fct_entourageP`, `fcvg_ball2P`, `cvg_ball2P`, `cauchy_cvgP`, - `mx_complete`, `Uniform_isComplete.Build`, `cauchy_ballP`, `cauchy_exP`, `cauchyP`, - `compact_cauchy_cvg`, `pointwise_cvgE`, `pointwise_uniform_cvg`, `cvg_sigL`, `uniform_restrict_cvg`, - `cvg_uniformU`, `cvg_uniform_set0`, `fam_cvgP`, `family_cvg_subset`, - `family_cvg_finite_covers`, `fam_cvgE`, `Nbhs_isTopological`, `compact_open_fam_compactP`, - `compact_cvg_within_compact`, `nbhs_subspace`, `subspace_cvgP`, `uniform_limit_continuous`, - `uniform_limit_continuous_subspace`, `pointwise_compact_cvg` - * `t` in module type `PropInFilterSig` - + canonical `matrix_filtered` -> instance using `isFiltered.Build` - + now use `nbhs` instead of `[filter of ...]` - * notations `-->`, `E @[ x --> F ]`, `f @ F`, ```E `@[ x --> F ]```, ```f `@ G```, - `{ptws, F --> f }` - + notation `lim` is now a definition - + canonical `filtered_prod` -> instances using `isFiltered.Build`, `selfFiltered.Build` - + now use `set_system` and also `nbhsType` instead of `filteredType ...` - * lemmas `cvg_ex`, `cvgP`, `cvg_toP`, `dvgP`, `cvgNpoint`, `eq_is_cvg` - + canonicals `filter_on_eqType`, `filter_on_choiceType`, `filter_on_PointedType`, - `filter_on_FilteredType` -> instances using `gen_eqMixin`, `gen_choiceMixin`, - `isPointed.Build`, `isFiltered.Build` - + canonical `bool_discrete_filter` -> instance using `hasNbhs.Build` - + module `Topological` (packed class) -> mixin `Nbhs_isTopological`, - structure `Topological`, type `topologicalType` - * definition `open` now a field of the mixin - + notation `continuous` now uses definition `continuous_at` - + section `TopologyOfFilter` -> factory `Nbhs_isNbhsTopological` - + section `TopologyOfOpen` -> factory `Pointed_isOpenTopological` - + section `TopologyOfBase` -> factory `Pointed_isBaseTopological` - + section `TopologyOfSubbase` -> factory `Pointed_isSubBaseTopological` - + definition `Pointed_isSubBaseTopological`, - canonicals `nat_filteredType`, `nat_topologicalType` -> - instance using `Pointed_isBaseTopological.Build` - + filter now explicit in the notation `X --> Y` - * lemmas `cvg_addnr`, `cvg_subnr`, `cvg_mulnl`, `cvg_mulnr`, `cvg_divnr` - + definition `prod_topologicalTypeMixin`, canonical `prod_topologicalType` -> - instances using `hasNbhs.Build`, `Nbhs_isNbhsTopological.Build` - + definition `matrix_topologicalTypeMixin`, canonical `matrix_topologicalType` -> - instance using `Nbhs_isNbhsTopological.Build` - + definitions `weak_topologicalTypeMixin`, `weak_topologicalType` -> - instances using `Pointed.on`, `Pointed_isOpenTopological.Build` - + definitions `sup_topologicalTypeMixin`, `sup_topologicalType` -> - instances using `Pointed.on` and `Pointed_isSubBaseTopological.Build` - + definition `product_topologicalType` -> definition `product_topology_def` - and instance using `Topological.copy` - + in `lim_id`: `nbhs` now explicit - + canonical `bool_discrete_topology` -> instance using `bool_discrete_topology` - + module `Uniform` (packed class) -> mixin `Nbhs_isUniform_mixin`, - structure `Uniform`, type `uniformType`, - factories `Nbhs_isUniform`, `isUniform` - + definition `prod_uniformType_mixin`, canonical `prod_uniformType` -> - instance using `Nbhs_isUniform.Build` - + definition `matrix_uniformType_mixin`, canonical `matrix_uniformType` -> - instance using `Nbhs_isUniform.Build` - + definitions `weak_uniform_mixin`, `weak_uniformType` -> instance - using `Nbhs_isUniform.Build` - + definitions `fct_uniformType_mixin`, `fct_topologicalTypeMixin`, - `generic_source_filter`, `fct_topologicalType`, `fct_uniformType` -> - definition `arrow_uniform` and instance using `arrow_uniform` - + definitions `sup_uniform_mixin`, `sup_uniformType` -> instance using - `Nbhs_isUniform.Build` - + definition `product_uniformType` -> instance using `Uniform.copy` - + definition `discrete_uniformType` -> instance using `Choice.on`, - `Choice.on`, `discrete_uniform_mixin` - + module `PseudoMetric` (packed class) -> factory `Nbhs_isPseudoMetric` - + definition `ball` now a field of factory `Nbhs_isPseudoMetric` - + definition `matrix_pseudoMetricType_mixin`, canonical `matrix_pseudoMetricType` -> - instance using `Uniform_isPseudoMetric.Build` - + definition `prod_pseudoMetricType_mixin`, canonical `prod_pseudoMetricType` -> - instance using `Uniform_isPseudoMetric.Build` - + definition `fct_pseudoMetricType_mixin`, canonical `fct_pseudoMetricType` -> - instance using `Uniform_isPseudoMetric.Build` - + canonical `quotient_subtype` -> instance using `Quotient.copy` - + canonical `quotient_eq` -> instance using `[Sub ... of ... by %/]` - + canonical `quotient_choice` -> instance using `[Choice of ... by <:]` - + canonical `quotient_pointed` -> instance using `isPointed.Build` - + in definition `quotient_topologicalType_mixin`: - `topologyOfOpenMixin` -> `Pointed_isOpenTopological.Build` - + canonical `quotient_topologicalType` -> instance using `quotient_topologicalType_mixin` - + lemma `repr_comp_continuous` uses the notation `\pi_` instead of `... == ... %[mod ...]` - + definition `discrete_pseudoMetricType` -> instead using `discrete_pseudometric_mixin` - + module `Complete` (packed class) -> mixin `Uniform_isComplete`, - structure `Complete`, type `completeType` - * lemma `cauchy_cvg` now a mixin field - + canonical `matrix_completeType` -> instance using `Uniform_isComplete.Build` - + canonical `fun_completeType` -> instance using `Uniform_isComplete.Build` - + module `CompletePseudoMetric` (packed class) -> structure `CompletePseudoMetric` - + matrix instance using `Uniform_isComplete.Build` - + function instance using `Uniform_isComplete.Build` - + module `regular_topology` -> instances using ` Pointed.on`, - `hasNbhs.Build`, `Nbhs_isPseudoMetric.Build` - + in module `numFieldTopology`: - * `realType`, `rcfType`, `archiFieldType`, `realFieldType`, `numClosedFieldType`, - `numFieldType` instances using `PseudoMetric.copy` - + definition `fct_RestrictedUniform`, `fct_RestrictedUniformTopology`, canonical - `fct_RestrictUniformFilteredType`, `fct_RestrictUniformTopologicalType`, `fct_restrictedUniformType` - -> - definition `uniform_fun`, instance using `Unifom.copy` for ```{uniform` _ -> _}``` - + definitions `fct_Pointwise`, `fct_PointwiseTopology`, canonicals `fct_PointwiseFilteredType`, - `fct_PointwiseTopologicalType` -> definition `pointwise_fun`, instance using `Topological.copy` - + definition `compact_openK_topological_mixin`, canonical `compact_openK_filter`, - `compact_openK_topological` -> instances using `Pointed.on`, - `hasNbhs.Build`, `compact_openK_openE_subproof` for `compact_openK` - + canonical `compact_open_pointedType`, definition `compact_open_topologicalType`, - canonicals `compact_open_filtered`, `compact_open_filtered` -> definition `compact_open_def`, - instances using `Pointed.on`, `Nbhs.copy`, `Pointed.on`, `Nbhs_isTopological` - + definitions `weak_pseudoMetricType_mixin`, `weak_pseudoMetricType` -> lemmas - `weak_pseudo_metric_ball_center`, `weak_pseudo_metric_entourageE`, instance using `niform_isPseudoMetric.Build` - + definition `countable_uniform_pseudoMetricType_mixin` -> module `countable_uniform` - with definition `type`, instances using `Uniform.on`, `Uniform_isPseudoMetric.Build`, - lemma `countable_uniform_bounded`, notation `countable_uniform` - + definitions `sup_pseudoMetric_mixin`, `sup_pseudoMetricType`, `product_pseudoMetricType` -> - instances using `PseudoMetric.on`, `PseudoMetric.copy` - + definitions `subspace_pointedType`, `subspace_topologicalMixin`, canonicals `subspace_filteredType`, - `subspace_topologicalType` -> instance using `Choice.copy`, `isPointed.Build`, `hasNbhs.Build`, - lemmas `nbhs_subspaceP_subproof`, `nbhs_subspace_singleton`, `nbhs_subspace_nbhs`, instance using - `Nbhs_isNbhsTopological.Build` - + definition `subspace_uniformMixin`, canonical `subspace_uniformType` -> instance using - `Nbhs_isUniform_mixin.Build` - + definition `subspace_pseudoMetricType_mixin`, canonical `subspace_pseudoMetricType` -> lemmas `subspace_pm_ball_center`, - `subspace_pm_ball_sym`, `subspace_pm_ball_triangle`, `subspace_pm_entourageE`, instance using - `Uniform_isPseudoMetric.Build` - + section `gauges` -> module `gauge` - * `gauge_pseudoMetricType` -> `gauge.type` (instances using `Uniform.on`, `PseudoMetric.on`) - * `gauge_uniformType` -> `gauge.type` - ### Renamed ### Generalized -- in `cantor.v`: - + in definition `cantor_space`: `product_uniformType` -> `prod_topology` - * instances using `Pointed.on`, `Nbhs.on`, `Topological.on` - -- in `topology.v`: - + now use `nbhsType` instead of `topologicalType` - * lemma `near_fun` - * definition `discrete_space` - * definition `discrete_uniform_mixin` - * definition `discrete_ball`, lemma `discrete_ball_center`, - definition `discrete_pseudometric_mixin` - ### Deprecated ### Removed -- in `mathcomp_extra.v`: - + coercion `choice.Choice.mixin` - + lemmas `bigminr_maxr`, definitions `AC_subdef`, `oAC`, `opACE`, canonicals `opAC_law`, `opAC_com_law` - + lemmas `some_big_AC`, `big_ACE`, `big_undup_AC`, `perm_big_AC`, `big_const_idem`, - `big_id_idem`, `big_mkcond_idem`, `big_split_idem`, `big_id_idem_AC`, `bigID_idem`, - `big_rem_AC`, `bigD1_AC`, `sub_big`, `sub_big_seq`, `sub_big_seq_cond`, `uniq_sub_big`, - `uniq_sub_big_cond`, `sub_big_idem`, `sub_big_idem_cond`, `sub_in_big`, `le_big_ord`, - `subset_big`, `subset_big_cond`, `le_big_nat`, `le_big_ord_cond` - + lemmas `bigmax_le`, `bigmax_lt`, `lt_bigmin`, `le_bigmin` - + lemmas `bigmax_mkcond`, `bigmax_split`, `bigmax_idl`, `bigmax_idr`, `bigmaxID` - + lemmas `sub_bigmax`, `sub_bigmax_seq`, `sub_bigmax_cond`, `sub_in_bigmax`, - `le_bigmax_nat`, `le_bigmax_nat_cond`, `le_bigmax_ord`, `le_bigmax_ord_cond`, - `subset_bigmax`, `subset_bigmax_cond` - + lemmas `bigmaxD1`, `le_bigmax_cond`, `le_bigmax`, `bigmax_sup`, `bigmax_leP`, - `bigmax_ltP`, `bigmax_eq_arg`, `eq_bigmax`, `le_bigmax2` - + lemmas `bigmin_mkcond`, `bigmin_split`, `bigmin_idl`, `bigmin_idr`, `bigminID` - + lemmas `sub_bigmin`, `sub_bigmin_cond`, `sub_bigmin_seq`, `sub_in_bigmin`, - `le_bigmin_nat`, `le_bigmin_nat_cond`, `le_bigmin_ord`, `le_bigmin_ord_cond`, - `subset_bigmin`, `subset_bigmin_cond` - + lemmas `bigminD1`, `bigmin_le_cond`, `bigmin_le`, `bigmin_inf`, `bigmin_geP`, - `bigmin_gtP`, `bigmin_eq_arg`, `eq_bigmin` - -- in `boolp.v`: - + definitions `dep_arrow_eqType`, `dep_arrow_choiceClass`, `dep_arrow_choiceType` - -- in `cardinality.v`: - + lemma `countable_setT_countMixin` - -- in `classical_sets.v`: - + notations `PointedType`, `[pointedType of ...]` - -- in `altreals/discrete.v`: - + notation `[countable of ...]` - -- in `cantor.v`: - + definition `pointed_discrete` - -- in `convex.v`: - + field `convexspacechoiceclass`, canonicals `conv_eqType`, `conv_choiceType`, `conv_choiceType` - -- in `constructive_ereal.v`: - + canonicals `isLaw.Build`, `mine_comoid` - -- in `lebesgue_measure.v`: - + no more "pointed class" argument in definition `ereal_isMeasurable` - -- in `measure.v`: - + field `ptclass` in mixin `isSemiRingOfSets` - + canonicals `ringOfSets_eqType`, `ringOfSets_choiceType`, - `ringOfSets_ptType`, `algebraOfSets_eqType`, `algebraOfSets_choiceType`, - `algebraOfSets_ptType`, `measurable_eqType`, `measurable_choiceType`, - `measurable_ptType` - + field `ptclass` in factory `isAlgebraOfSets` - + field `ptclass` in factory `isMeasurable` - -- in `normedtype.v`: - + `filtered_of_normedZmod` - + section `pseudoMetric_of_normedDomain` - * lemmas `ball_norm_center`, `ball_norm_symmetric`, `ball_norm_triangle`, `nbhs_ball_normE` - * definition `pseudoMetric_of_normedDomain` - + lemma `normrZ` - + canonical `matrix_normedZmodType` - + lemmas `eq_cvg`, `eq_is_cvg` - -- in `topology.v`: - + structure `source`, definition `source_filter` - + definition `filter_of`, notation `[filter of ...]` (now replaced by `nbhs`), lemma `filter_of_filterE` - + definition `open_of_nbhs` - + definition `open_from`, lemma `open_fromT` - + canonical `eventually_filter_source` - + canonical `discrete_topological_mixin` - + canonical `set_filter_source` - + definitions `filtered_of_normedZmod`, `pseudoMetric_of_normedDomain` - + definitions `fct_UniformFamily` (use `uniform_fun_family` instead), canonicals `fct_UniformFamilyFilteredType`, - `fct_UniformFamilyTopologicalType`, `fct_UniformFamilyUniformType` - -- in `lebesgue_stieltjes_measure.v` - + lemma `sigmaT_finite_lebesgue_stieltjes_measure` turned into a `Let` - ### Infrastructure ### Misc diff --git a/INSTALL.md b/INSTALL.md index 088565513..d3b05ba62 100644 --- a/INSTALL.md +++ b/INSTALL.md @@ -2,10 +2,10 @@ ## Requirements -- [The Coq Proof Assistant version ≥ 8.15](https://coq.inria.fr) -- [Mathematical Components version ≥ 1.17.0](https://github.com/math-comp/math-comp) -- [Finmap library version ≥ 1.5.1](https://github.com/math-comp/finmap) -- [Hierarchy builder version >= 1.2.0](https://github.com/math-comp/hierarchy-builder) +- [The Coq Proof Assistant version ≥ 8.16](https://coq.inria.fr) +- [Mathematical Components version ≥ 2.0.0](https://github.com/math-comp/math-comp) +- [Finmap library version ≥ 2.0.0](https://github.com/math-comp/finmap) +- [Hierarchy builder version >= 1.4.0](https://github.com/math-comp/hierarchy-builder) - [bigenough >= 1.0.0](https://github.com/math-comp/bigenough) These requirements can be installed in a custom way, or through @@ -48,7 +48,7 @@ $ opam install coq-mathcomp-analysis ``` To install a precise version, type, say ``` -$ opam install coq-mathcomp-analysis.0.7.0 +$ opam install coq-mathcomp-analysis.1.0.0 ``` 4. Everytime you want to work in this same context, you need to type ``` @@ -71,28 +71,28 @@ using [proof general for emacs](https://github.com/ProofGeneral/PG) ## Break-down of phase 3 of the installation procedure step by step -With the example of Coq 8.15.0 and MathComp 1.17.0. For other versions, update the +With the example of Coq 8.16.0 and MathComp 2.0.0. For other versions, update the version numbers accordingly. -1. Install Coq 8.15.0 +1. Install Coq 8.16.0 ``` -$ opam install coq.8.15.0 +$ opam install coq.8.16.0 ``` 2. Install the Mathematical Components ``` -$ opam install coq-mathcomp-ssreflect.1.17.0 -$ opam install coq-mathcomp-fingroup.1.17.0 -$ opam install coq-mathcomp-algebra.1.17.0 -$ opam install coq-mathcomp-solvable.1.17.0 -$ opam install coq-mathcomp-field.1.17.0 +$ opam install coq-mathcomp-ssreflect.2.0.0 +$ opam install coq-mathcomp-fingroup.2.0.0 +$ opam install coq-mathcomp-algebra.2.0.0 +$ opam install coq-mathcomp-solvable.2.0.0 +$ opam install coq-mathcomp-field.2.0.0 ``` 3. Install the Finite maps library ``` -$ opam install coq-mathcomp-finmap.1.5.1 +$ opam install coq-mathcomp-finmap.2.0.0 ``` 4. Install the Hierarchy Builder ``` -$ opam install coq-hierarchy-builder.1.2.0 +$ opam install coq-hierarchy-builder.1.6.0 ``` 5. Download and compile `coq-mathcomp-analysis` without installing ``` diff --git a/README.md b/README.md index 1ead77e71..60e861c4c 100644 --- a/README.md +++ b/README.md @@ -22,6 +22,7 @@ the Coq proof-assistant and using the Mathematical Components library. - Author(s): - Reynald Affeldt (initial) + - Alessandro Bruni - Yves Bertot - Cyril Cohen (initial) - Marie Kerjean @@ -69,7 +70,6 @@ make # or make -j make install ``` - ## Disclaimer This library is still at an experimental stage. Contents may @@ -80,7 +80,7 @@ own risk. ## Documentation Each file is documented in its header -([coqdoc presentation for the last version](https://math-comp.github.io/analysis/htmldoc_0_7_0/index.html)). +([coqdoc presentation for the last version](https://math-comp.github.io/analysis/htmldoc_1_0_0/index.html)). Changes are documented in [CHANGELOG.md](CHANGELOG.md) and [CHANGELOG_UNRELEASED.md](CHANGELOG_UNRELEASED.md). diff --git a/coq-mathcomp-analysis.opam b/coq-mathcomp-analysis.opam index 12286a782..99659934e 100644 --- a/coq-mathcomp-analysis.opam +++ b/coq-mathcomp-analysis.opam @@ -51,6 +51,7 @@ tags: [ ] authors: [ "Reynald Affeldt" + "Alessandro Bruni" "Yves Bertot" "Cyril Cohen" "Marie Kerjean" diff --git a/coq-mathcomp-classical.opam b/coq-mathcomp-classical.opam index 455c334d8..dc3e54ccf 100644 --- a/coq-mathcomp-classical.opam +++ b/coq-mathcomp-classical.opam @@ -38,6 +38,7 @@ tags: [ ] authors: [ "Reynald Affeldt" + "Alessandro Bruni" "Yves Bertot" "Cyril Cohen" "Marie Kerjean"