/- Copyright (c) 2020 Yury Kudryashov. All rights reserved. Released under Apache 2.0 license as described in the file LICENSE. Authors: Alexander Bentkamp, Yury Kudryashov -/ import analysis.convex.jensen import analysis.normed.group.pointwise import analysis.normed_space.finite_dimension import analysis.normed_space.ray import topology.path_connected import topology.algebra.affine /-! # Topological and metric properties of convex sets We prove the following facts: * `convex.interior` : interior of a convex set is convex; * `convex.closure` : closure of a convex set is convex; * `set.finite.compact_convex_hull` : convex hull of a finite set is compact; * `set.finite.is_closed_convex_hull` : convex hull of a finite set is closed; * `convex_on_norm`, `convex_on_dist` : norm and distance to a fixed point is convex on any convex set; * `convex_on_univ_norm`, `convex_on_univ_dist` : norm and distance to a fixed point is convex on the whole space; * `convex_hull_ediam`, `convex_hull_diam` : convex hull of a set has the same (e)metric diameter as the original set; * `bounded_convex_hull` : convex hull of a set is bounded if and only if the original set is bounded. * `bounded_std_simplex`, `is_closed_std_simplex`, `compact_std_simplex`: topological properties of the standard simplex; -/ variables {ι : Type*} {E : Type*} open metric set open_locale pointwise convex lemma real.convex_iff_is_preconnected {s : set ℝ} : convex ℝ s ↔ is_preconnected s := convex_iff_ord_connected.trans is_preconnected_iff_ord_connected.symm alias real.convex_iff_is_preconnected ↔ _ is_preconnected.convex /-! ### Standard simplex -/ section std_simplex variables [fintype ι] /-- Every vector in `std_simplex 𝕜 ι` has `max`-norm at most `1`. -/ lemma std_simplex_subset_closed_ball : std_simplex ℝ ι ⊆ metric.closed_ball 0 1 := begin assume f hf, rw [metric.mem_closed_ball, dist_zero_right], refine (nnreal.coe_one ▸ nnreal.coe_le_coe.2 $ finset.sup_le $ λ x hx, _), change |f x| ≤ 1, rw [abs_of_nonneg $ hf.1 x], exact (mem_Icc_of_mem_std_simplex hf x).2 end variable (ι) /-- `std_simplex ℝ ι` is bounded. -/ lemma bounded_std_simplex : metric.bounded (std_simplex ℝ ι) := (metric.bounded_iff_subset_ball 0).2 ⟨1, std_simplex_subset_closed_ball⟩ /-- `std_simplex ℝ ι` is closed. -/ lemma is_closed_std_simplex : is_closed (std_simplex ℝ ι) := (std_simplex_eq_inter ℝ ι).symm ▸ is_closed.inter (is_closed_Inter $ λ i, is_closed_le continuous_const (continuous_apply i)) (is_closed_eq (continuous_finset_sum _ $ λ x _, continuous_apply x) continuous_const) /-- `std_simplex ℝ ι` is compact. -/ lemma compact_std_simplex : is_compact (std_simplex ℝ ι) := metric.compact_iff_closed_bounded.2 ⟨is_closed_std_simplex ι, bounded_std_simplex ι⟩ end std_simplex /-! ### Topological vector space -/ section has_continuous_const_smul variables {𝕜 : Type*} [linear_ordered_field 𝕜] [add_comm_group E] [module 𝕜 E] [topological_space E] [topological_add_group E] [has_continuous_const_smul 𝕜 E] /-- If `s` is a convex set, then `a • interior s + b • closure s ⊆ interior s` for all `0 < a`, `0 ≤ b`, `a + b = 1`. See also `convex.combo_interior_self_subset_interior` for a weaker version. -/ lemma convex.combo_interior_closure_subset_interior {s : set E} (hs : convex 𝕜 s) {a b : 𝕜} (ha : 0 < a) (hb : 0 ≤ b) (hab : a + b = 1) : a • interior s + b • closure s ⊆ interior s := interior_smul₀ ha.ne' s ▸ calc interior (a • s) + b • closure s ⊆ interior (a • s) + closure (b • s) : add_subset_add subset.rfl (smul_closure_subset b s) ... = interior (a • s) + b • s : by rw is_open_interior.add_closure (b • s) ... ⊆ interior (a • s + b • s) : subset_interior_add_left ... ⊆ interior s : interior_mono $ hs.set_combo_subset ha.le hb hab /-- If `s` is a convex set, then `a • interior s + b • s ⊆ interior s` for all `0 < a`, `0 ≤ b`, `a + b = 1`. See also `convex.combo_interior_closure_subset_interior` for a stronger version. -/ lemma convex.combo_interior_self_subset_interior {s : set E} (hs : convex 𝕜 s) {a b : 𝕜} (ha : 0 < a) (hb : 0 ≤ b) (hab : a + b = 1) : a • interior s + b • s ⊆ interior s := calc a • interior s + b • s ⊆ a • interior s + b • closure s : add_subset_add subset.rfl $ image_subset _ subset_closure ... ⊆ interior s : hs.combo_interior_closure_subset_interior ha hb hab /-- If `s` is a convex set, then `a • closure s + b • interior s ⊆ interior s` for all `0 ≤ a`, `0 < b`, `a + b = 1`. See also `convex.combo_self_interior_subset_interior` for a weaker version. -/ lemma convex.combo_closure_interior_subset_interior {s : set E} (hs : convex 𝕜 s) {a b : 𝕜} (ha : 0 ≤ a) (hb : 0 < b) (hab : a + b = 1) : a • closure s + b • interior s ⊆ interior s := by { rw add_comm, exact hs.combo_interior_closure_subset_interior hb ha (add_comm a b ▸ hab) } /-- If `s` is a convex set, then `a • s + b • interior s ⊆ interior s` for all `0 ≤ a`, `0 < b`, `a + b = 1`. See also `convex.combo_closure_interior_subset_interior` for a stronger version. -/ lemma convex.combo_self_interior_subset_interior {s : set E} (hs : convex 𝕜 s) {a b : 𝕜} (ha : 0 ≤ a) (hb : 0 < b) (hab : a + b = 1) : a • s + b • interior s ⊆ interior s := by { rw add_comm, exact hs.combo_interior_self_subset_interior hb ha (add_comm a b ▸ hab) } lemma convex.combo_interior_closure_mem_interior {s : set E} (hs : convex 𝕜 s) {x y : E} (hx : x ∈ interior s) (hy : y ∈ closure s) {a b : 𝕜} (ha : 0 < a) (hb : 0 ≤ b) (hab : a + b = 1) : a • x + b • y ∈ interior s := hs.combo_interior_closure_subset_interior ha hb hab $ add_mem_add (smul_mem_smul_set hx) (smul_mem_smul_set hy) lemma convex.combo_interior_self_mem_interior {s : set E} (hs : convex 𝕜 s) {x y : E} (hx : x ∈ interior s) (hy : y ∈ s) {a b : 𝕜} (ha : 0 < a) (hb : 0 ≤ b) (hab : a + b = 1) : a • x + b • y ∈ interior s := hs.combo_interior_closure_mem_interior hx (subset_closure hy) ha hb hab lemma convex.combo_closure_interior_mem_interior {s : set E} (hs : convex 𝕜 s) {x y : E} (hx : x ∈ closure s) (hy : y ∈ interior s) {a b : 𝕜} (ha : 0 ≤ a) (hb : 0 < b) (hab : a + b = 1) : a • x + b • y ∈ interior s := hs.combo_closure_interior_subset_interior ha hb hab $ add_mem_add (smul_mem_smul_set hx) (smul_mem_smul_set hy) lemma convex.combo_self_interior_mem_interior {s : set E} (hs : convex 𝕜 s) {x y : E} (hx : x ∈ s) (hy : y ∈ interior s) {a b : 𝕜} (ha : 0 ≤ a) (hb : 0 < b) (hab : a + b = 1) : a • x + b • y ∈ interior s := hs.combo_closure_interior_mem_interior (subset_closure hx) hy ha hb hab lemma convex.open_segment_interior_closure_subset_interior {s : set E} (hs : convex 𝕜 s) {x y : E} (hx : x ∈ interior s) (hy : y ∈ closure s) : open_segment 𝕜 x y ⊆ interior s := begin rintro _ ⟨a, b, ha, hb, hab, rfl⟩, exact hs.combo_interior_closure_mem_interior hx hy ha hb.le hab end lemma convex.open_segment_interior_self_subset_interior {s : set E} (hs : convex 𝕜 s) {x y : E} (hx : x ∈ interior s) (hy : y ∈ s) : open_segment 𝕜 x y ⊆ interior s := hs.open_segment_interior_closure_subset_interior hx (subset_closure hy) lemma convex.open_segment_closure_interior_subset_interior {s : set E} (hs : convex 𝕜 s) {x y : E} (hx : x ∈ closure s) (hy : y ∈ interior s) : open_segment 𝕜 x y ⊆ interior s := begin rintro _ ⟨a, b, ha, hb, hab, rfl⟩, exact hs.combo_closure_interior_mem_interior hx hy ha.le hb hab end lemma convex.open_segment_self_interior_subset_interior {s : set E} (hs : convex 𝕜 s) {x y : E} (hx : x ∈ s) (hy : y ∈ interior s) : open_segment 𝕜 x y ⊆ interior s := hs.open_segment_closure_interior_subset_interior (subset_closure hx) hy /-- If `x ∈ closure s` and `y ∈ interior s`, then the segment `(x, y]` is included in `interior s`. -/ lemma convex.add_smul_sub_mem_interior' {s : set E} (hs : convex 𝕜 s) {x y : E} (hx : x ∈ closure s) (hy : y ∈ interior s) {t : 𝕜} (ht : t ∈ Ioc (0 : 𝕜) 1) : x + t • (y - x) ∈ interior s := by simpa only [sub_smul, smul_sub, one_smul, add_sub, add_comm] using hs.combo_interior_closure_mem_interior hy hx ht.1 (sub_nonneg.mpr ht.2) (add_sub_cancel'_right _ _) /-- If `x ∈ s` and `y ∈ interior s`, then the segment `(x, y]` is included in `interior s`. -/ lemma convex.add_smul_sub_mem_interior {s : set E} (hs : convex 𝕜 s) {x y : E} (hx : x ∈ s) (hy : y ∈ interior s) {t : 𝕜} (ht : t ∈ Ioc (0 : 𝕜) 1) : x + t • (y - x) ∈ interior s := hs.add_smul_sub_mem_interior' (subset_closure hx) hy ht /-- If `x ∈ closure s` and `x + y ∈ interior s`, then `x + t y ∈ interior s` for `t ∈ (0, 1]`. -/ lemma convex.add_smul_mem_interior' {s : set E} (hs : convex 𝕜 s) {x y : E} (hx : x ∈ closure s) (hy : x + y ∈ interior s) {t : 𝕜} (ht : t ∈ Ioc (0 : 𝕜) 1) : x + t • y ∈ interior s := by simpa only [add_sub_cancel'] using hs.add_smul_sub_mem_interior' hx hy ht /-- If `x ∈ s` and `x + y ∈ interior s`, then `x + t y ∈ interior s` for `t ∈ (0, 1]`. -/ lemma convex.add_smul_mem_interior {s : set E} (hs : convex 𝕜 s) {x y : E} (hx : x ∈ s) (hy : x + y ∈ interior s) {t : 𝕜} (ht : t ∈ Ioc (0 : 𝕜) 1) : x + t • y ∈ interior s := hs.add_smul_mem_interior' (subset_closure hx) hy ht /-- In a topological vector space, the interior of a convex set is convex. -/ protected lemma convex.interior {s : set E} (hs : convex 𝕜 s) : convex 𝕜 (interior s) := convex_iff_open_segment_subset.mpr $ λ x y hx hy, hs.open_segment_closure_interior_subset_interior (interior_subset_closure hx) hy /-- In a topological vector space, the closure of a convex set is convex. -/ protected lemma convex.closure {s : set E} (hs : convex 𝕜 s) : convex 𝕜 (closure s) := λ x y hx hy a b ha hb hab, let f : E → E → E := λ x' y', a • x' + b • y' in have hf : continuous (λ p : E × E, f p.1 p.2), from (continuous_fst.const_smul _).add (continuous_snd.const_smul _), show f x y ∈ closure s, from mem_closure_of_continuous2 hf hx hy (λ x' hx' y' hy', subset_closure (hs hx' hy' ha hb hab)) end has_continuous_const_smul section has_continuous_smul variables [add_comm_group E] [module ℝ E] [topological_space E] [topological_add_group E] [has_continuous_smul ℝ E] /-- Convex hull of a finite set is compact. -/ lemma set.finite.compact_convex_hull {s : set E} (hs : s.finite) : is_compact (convex_hull ℝ s) := begin rw [hs.convex_hull_eq_image], apply (compact_std_simplex _).image, haveI := hs.fintype, apply linear_map.continuous_on_pi end /-- Convex hull of a finite set is closed. -/ lemma set.finite.is_closed_convex_hull [t2_space E] {s : set E} (hs : s.finite) : is_closed (convex_hull ℝ s) := hs.compact_convex_hull.is_closed open affine_map /-- If we dilate the interior of a convex set about a point in its interior by a scale `t > 1`, the result includes the closure of the original set. TODO Generalise this from convex sets to sets that are balanced / star-shaped about `x`. -/ lemma convex.closure_subset_image_homothety_interior_of_one_lt {s : set E} (hs : convex ℝ s) {x : E} (hx : x ∈ interior s) (t : ℝ) (ht : 1 < t) : closure s ⊆ homothety x t '' interior s := begin intros y hy, have hne : t ≠ 0, from (one_pos.trans ht).ne', refine ⟨homothety x t⁻¹ y, hs.open_segment_interior_closure_subset_interior hx hy _, (affine_equiv.homothety_units_mul_hom x (units.mk0 t hne)).apply_symm_apply y⟩, rw [open_segment_eq_image_line_map, ← inv_one, ← inv_Ioi (@one_pos ℝ _ _), ← image_inv, image_image, homothety_eq_line_map], exact mem_image_of_mem _ ht end /-- If we dilate a convex set about a point in its interior by a scale `t > 1`, the interior of the result includes the closure of the original set. TODO Generalise this from convex sets to sets that are balanced / star-shaped about `x`. -/ lemma convex.closure_subset_interior_image_homothety_of_one_lt {s : set E} (hs : convex ℝ s) {x : E} (hx : x ∈ interior s) (t : ℝ) (ht : 1 < t) : closure s ⊆ interior (homothety x t '' s) := (hs.closure_subset_image_homothety_interior_of_one_lt hx t ht).trans $ (homothety_is_open_map x t (one_pos.trans ht).ne').image_interior_subset _ /-- If we dilate a convex set about a point in its interior by a scale `t > 1`, the interior of the result includes the closure of the original set. TODO Generalise this from convex sets to sets that are balanced / star-shaped about `x`. -/ lemma convex.subset_interior_image_homothety_of_one_lt {s : set E} (hs : convex ℝ s) {x : E} (hx : x ∈ interior s) (t : ℝ) (ht : 1 < t) : s ⊆ interior (homothety x t '' s) := subset_closure.trans $ hs.closure_subset_interior_image_homothety_of_one_lt hx t ht /-- A nonempty convex set is path connected. -/ protected lemma convex.is_path_connected {s : set E} (hconv : convex ℝ s) (hne : s.nonempty) : is_path_connected s := begin refine is_path_connected_iff.mpr ⟨hne, _⟩, intros x x_in y y_in, have H := hconv.segment_subset x_in y_in, rw segment_eq_image_line_map at H, exact joined_in.of_line affine_map.line_map_continuous.continuous_on (line_map_apply_zero _ _) (line_map_apply_one _ _) H end /-- A nonempty convex set is connected. -/ protected lemma convex.is_connected {s : set E} (h : convex ℝ s) (hne : s.nonempty) : is_connected s := (h.is_path_connected hne).is_connected /-- A convex set is preconnected. -/ protected lemma convex.is_preconnected {s : set E} (h : convex ℝ s) : is_preconnected s := s.eq_empty_or_nonempty.elim (λ h, h.symm ▸ is_preconnected_empty) (λ hne, (h.is_connected hne).is_preconnected) /-- Every topological vector space over ℝ is path connected. Not an instance, because it creates enormous TC subproblems (turn on `pp.all`). -/ protected lemma topological_add_group.path_connected : path_connected_space E := path_connected_space_iff_univ.mpr $ convex_univ.is_path_connected ⟨(0 : E), trivial⟩ end has_continuous_smul /-! ### Normed vector space -/ section normed_space variables [seminormed_add_comm_group E] [normed_space ℝ E] {s t : set E} /-- The norm on a real normed space is convex on any convex set. See also `seminorm.convex_on` and `convex_on_univ_norm`. -/ lemma convex_on_norm (hs : convex ℝ s) : convex_on ℝ s norm := ⟨hs, λ x y hx hy a b ha hb hab, calc ∥a • x + b • y∥ ≤ ∥a • x∥ + ∥b • y∥ : norm_add_le _ _ ... = a * ∥x∥ + b * ∥y∥ : by rw [norm_smul, norm_smul, real.norm_of_nonneg ha, real.norm_of_nonneg hb]⟩ /-- The norm on a real normed space is convex on the whole space. See also `seminorm.convex_on` and `convex_on_norm`. -/ lemma convex_on_univ_norm : convex_on ℝ univ (norm : E → ℝ) := convex_on_norm convex_univ lemma convex_on_dist (z : E) (hs : convex ℝ s) : convex_on ℝ s (λ z', dist z' z) := by simpa [dist_eq_norm, preimage_preimage] using (convex_on_norm (hs.translate (-z))).comp_affine_map (affine_map.id ℝ E - affine_map.const ℝ E z) lemma convex_on_univ_dist (z : E) : convex_on ℝ univ (λz', dist z' z) := convex_on_dist z convex_univ lemma convex_ball (a : E) (r : ℝ) : convex ℝ (metric.ball a r) := by simpa only [metric.ball, sep_univ] using (convex_on_univ_dist a).convex_lt r lemma convex_closed_ball (a : E) (r : ℝ) : convex ℝ (metric.closed_ball a r) := by simpa only [metric.closed_ball, sep_univ] using (convex_on_univ_dist a).convex_le r lemma convex.thickening (hs : convex ℝ s) (δ : ℝ) : convex ℝ (thickening δ s) := by { rw ←add_ball_zero, exact hs.add (convex_ball 0 _) } lemma convex.cthickening (hs : convex ℝ s) (δ : ℝ) : convex ℝ (cthickening δ s) := begin obtain hδ | hδ := le_total 0 δ, { rw cthickening_eq_Inter_thickening hδ, exact convex_Inter₂ (λ _ _, hs.thickening _) }, { rw cthickening_of_nonpos hδ, exact hs.closure } end /-- If `s`, `t` are disjoint convex sets, `s` is compact and `t` is closed then we can find open disjoint convex sets containing them. -/ lemma disjoint.exists_open_convexes (disj : disjoint s t) (hs₁ : convex ℝ s) (hs₂ : is_compact s) (ht₁ : convex ℝ t) (ht₂ : is_closed t) : ∃ u v, is_open u ∧ is_open v ∧ convex ℝ u ∧ convex ℝ v ∧ s ⊆ u ∧ t ⊆ v ∧ disjoint u v := let ⟨δ, hδ, hst⟩ := disj.exists_thickenings hs₂ ht₂ in ⟨_, _, is_open_thickening, is_open_thickening, hs₁.thickening _, ht₁.thickening _, self_subset_thickening hδ _, self_subset_thickening hδ _, hst⟩ /-- Given a point `x` in the convex hull of `s` and a point `y`, there exists a point of `s` at distance at least `dist x y` from `y`. -/ lemma convex_hull_exists_dist_ge {s : set E} {x : E} (hx : x ∈ convex_hull ℝ s) (y : E) : ∃ x' ∈ s, dist x y ≤ dist x' y := (convex_on_dist y (convex_convex_hull ℝ _)).exists_ge_of_mem_convex_hull hx /-- Given a point `x` in the convex hull of `s` and a point `y` in the convex hull of `t`, there exist points `x' ∈ s` and `y' ∈ t` at distance at least `dist x y`. -/ lemma convex_hull_exists_dist_ge2 {s t : set E} {x y : E} (hx : x ∈ convex_hull ℝ s) (hy : y ∈ convex_hull ℝ t) : ∃ (x' ∈ s) (y' ∈ t), dist x y ≤ dist x' y' := begin rcases convex_hull_exists_dist_ge hx y with ⟨x', hx', Hx'⟩, rcases convex_hull_exists_dist_ge hy x' with ⟨y', hy', Hy'⟩, use [x', hx', y', hy'], exact le_trans Hx' (dist_comm y x' ▸ dist_comm y' x' ▸ Hy') end /-- Emetric diameter of the convex hull of a set `s` equals the emetric diameter of `s. -/ @[simp] lemma convex_hull_ediam (s : set E) : emetric.diam (convex_hull ℝ s) = emetric.diam s := begin refine (emetric.diam_le $ λ x hx y hy, _).antisymm (emetric.diam_mono $ subset_convex_hull ℝ s), rcases convex_hull_exists_dist_ge2 hx hy with ⟨x', hx', y', hy', H⟩, rw edist_dist, apply le_trans (ennreal.of_real_le_of_real H), rw ← edist_dist, exact emetric.edist_le_diam_of_mem hx' hy' end /-- Diameter of the convex hull of a set `s` equals the emetric diameter of `s. -/ @[simp] lemma convex_hull_diam (s : set E) : metric.diam (convex_hull ℝ s) = metric.diam s := by simp only [metric.diam, convex_hull_ediam] /-- Convex hull of `s` is bounded if and only if `s` is bounded. -/ @[simp] lemma bounded_convex_hull {s : set E} : metric.bounded (convex_hull ℝ s) ↔ metric.bounded s := by simp only [metric.bounded_iff_ediam_ne_top, convex_hull_ediam] @[priority 100] instance normed_space.path_connected : path_connected_space E := topological_add_group.path_connected @[priority 100] instance normed_space.loc_path_connected : loc_path_connected_space E := loc_path_connected_of_bases (λ x, metric.nhds_basis_ball) (λ x r r_pos, (convex_ball x r).is_path_connected $ by simp [r_pos]) lemma dist_add_dist_of_mem_segment {x y z : E} (h : y ∈ [x -[ℝ] z]) : dist x y + dist y z = dist x z := begin simp only [dist_eq_norm, mem_segment_iff_same_ray] at *, simpa only [sub_add_sub_cancel', norm_sub_rev] using h.norm_add.symm end end normed_space