Datasets:
Tasks:
Text Generation
Modalities:
Text
Sub-tasks:
language-modeling
Languages:
English
Size:
100K - 1M
License:
File size: 11,057 Bytes
4365a98 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 |
/-
Copyright (c) 2021 David Wärn. All rights reserved.
Released under Apache 2.0 license as described in the file LICENSE.
Authors: David Wärn
-/
import topology.stone_cech
import topology.algebra.semigroup
import data.stream.init
/-!
# Hindman's theorem on finite sums
We prove Hindman's theorem on finite sums, using idempotent ultrafilters.
Given an infinite sequence `a₀, a₁, a₂, …` of positive integers, the set `FS(a₀, …)` is the set
of positive integers that can be expressed as a finite sum of `aᵢ`'s, without repetition. Hindman's
theorem asserts that whenever the positive integers are finitely colored, there exists a sequence
`a₀, a₁, a₂, …` such that `FS(a₀, …)` is monochromatic. There is also a stronger version, saying
that whenever a set of the form `FS(a₀, …)` is finitely colored, there exists a sequence
`b₀, b₁, b₂, …` such that `FS(b₀, …)` is monochromatic and contained in `FS(a₀, …)`. We prove both
these versions for a general semigroup `M` instead of `ℕ+` since it is no harder, although this
special case implies the general case.
The idea of the proof is to extend the addition `(+) : M → M → M` to addition `(+) : βM → βM → βM`
on the space `βM` of ultrafilters on `M`. One can prove that if `U` is an _idempotent_ ultrafilter,
i.e. `U + U = U`, then any `U`-large subset of `M` contains some set `FS(a₀, …)` (see
`exists_FS_of_large`). And with the help of a general topological argument one can show that any set
of the form `FS(a₀, …)` is `U`-large according to some idempotent ultrafilter `U` (see
`exists_idempotent_ultrafilter_le_FS`). This is enough to prove the theorem since in any finite
partition of a `U`-large set, one of the parts is `U`-large.
## Main results
- `FS_partition_regular`: the strong form of Hindman's theorem
- `exists_FS_of_finite_cover`: the weak form of Hindman's theorem
## Tags
Ramsey theory, ultrafilter
-/
open filter
/-- Multiplication of ultrafilters given by `∀ᶠ m in U*V, p m ↔ ∀ᶠ m in U, ∀ᶠ m' in V, p (m*m')`. -/
@[to_additive "Addition of ultrafilters given by
`∀ᶠ m in U+V, p m ↔ ∀ᶠ m in U, ∀ᶠ m' in V, p (m+m')`." ]
def ultrafilter.has_mul {M} [has_mul M] : has_mul (ultrafilter M) :=
{ mul := λ U V, (*) <$> U <*> V }
local attribute [instance] ultrafilter.has_mul ultrafilter.has_add
/- We could have taken this as the definition of `U * V`, but then we would have to prove that it
defines an ultrafilter. -/
@[to_additive]
lemma ultrafilter.eventually_mul {M} [has_mul M] (U V : ultrafilter M) (p : M → Prop) :
(∀ᶠ m in ↑(U * V), p m) ↔ ∀ᶠ m in U, ∀ᶠ m' in V, p (m * m') := iff.rfl
/-- Semigroup structure on `ultrafilter M` induced by a semigroup structure on `M`. -/
@[to_additive "Additive semigroup structure on `ultrafilter M` induced by an additive semigroup
structure on `M`."]
def ultrafilter.semigroup {M} [semigroup M] : semigroup (ultrafilter M) :=
{ mul_assoc := λ U V W, ultrafilter.coe_inj.mp $ filter.ext' $ λ p,
by simp only [ultrafilter.eventually_mul, mul_assoc]
..ultrafilter.has_mul }
local attribute [instance] ultrafilter.semigroup ultrafilter.add_semigroup
/- We don't prove `continuous_mul_right`, because in general it is false! -/
@[to_additive]
lemma ultrafilter.continuous_mul_left {M} [semigroup M] (V : ultrafilter M) : continuous (* V) :=
topological_space.is_topological_basis.continuous ultrafilter_basis_is_basis _ $
set.forall_range_iff.mpr $ λ s, ultrafilter_is_open_basic { m : M | ∀ᶠ m' in V, m * m' ∈ s }
namespace hindman
/-- `FS a` is the set of finite sums in `a`, i.e. `m ∈ FS a` if `m` is the sum of a nonempty
subsequence of `a`. We give a direct inductive definition instead of talking about subsequences. -/
inductive FS {M} [add_semigroup M] : stream M → set M
| head (a : stream M) : FS a a.head
| tail (a : stream M) (m : M) (h : FS a.tail m) : FS a m
| cons (a : stream M) (m : M) (h : FS a.tail m) : FS a (a.head + m)
/-- `FP a` is the set of finite products in `a`, i.e. `m ∈ FP a` if `m` is the product of a nonempty
subsequence of `a`. We give a direct inductive definition instead of talking about subsequences. -/
@[to_additive FS]
inductive FP {M} [semigroup M] : stream M → set M
| head (a : stream M) : FP a a.head
| tail (a : stream M) (m : M) (h : FP a.tail m) : FP a m
| cons (a : stream M) (m : M) (h : FP a.tail m) : FP a (a.head * m)
/-- If `m` and `m'` are finite products in `M`, then so is `m * m'`, provided that `m'` is obtained
from a subsequence of `M` starting sufficiently late. -/
@[to_additive]
lemma FP.mul {M} [semigroup M] {a : stream M} {m : M} (hm : m ∈ FP a) :
∃ n, ∀ m' ∈ FP (a.drop n), m * m' ∈ FP a :=
begin
induction hm with a a m hm ih a m hm ih,
{ exact ⟨1, λ m hm, FP.cons a m hm⟩, },
{ cases ih with n hn, use n+1, intros m' hm', exact FP.tail _ _ (hn _ hm'), },
{ cases ih with n hn, use n+1, intros m' hm', rw mul_assoc, exact FP.cons _ _ (hn _ hm'), },
end
@[to_additive exists_idempotent_ultrafilter_le_FS]
lemma exists_idempotent_ultrafilter_le_FP {M} [semigroup M] (a : stream M) :
∃ U : ultrafilter M, U * U = U ∧ ∀ᶠ m in U, m ∈ FP a :=
begin
let S : set (ultrafilter M) := ⋂ n, { U | ∀ᶠ m in U, m ∈ FP (a.drop n) },
obtain ⟨U, hU, U_idem⟩ := exists_idempotent_in_compact_subsemigroup _ S _ _ _,
{ refine ⟨U, U_idem, _⟩, convert set.mem_Inter.mp hU 0, },
{ exact ultrafilter.continuous_mul_left },
{ apply is_compact.nonempty_Inter_of_sequence_nonempty_compact_closed,
{ intros n U hU,
apply eventually.mono hU,
rw [add_comm, ←stream.drop_drop, ←stream.tail_eq_drop],
exact FP.tail _ },
{ intro n, exact ⟨pure _, mem_pure.mpr $ FP.head _⟩, },
{ exact (ultrafilter_is_closed_basic _).is_compact, },
{ intro n, apply ultrafilter_is_closed_basic, }, },
{ exact is_closed.is_compact (is_closed_Inter $ λ i, ultrafilter_is_closed_basic _) },
{ intros U hU V hV,
rw set.mem_Inter at *,
intro n,
rw [set.mem_set_of_eq, ultrafilter.eventually_mul],
apply eventually.mono (hU n),
intros m hm,
obtain ⟨n', hn⟩ := FP.mul hm,
apply eventually.mono (hV (n' + n)),
intros m' hm',
apply hn,
simpa only [stream.drop_drop] using hm', }
end
@[to_additive exists_FS_of_large]
lemma exists_FP_of_large {M} [semigroup M] (U : ultrafilter M) (U_idem : U * U = U)
(s₀ : set M) (sU : s₀ ∈ U) : ∃ a, FP a ⊆ s₀ :=
begin
/- Informally: given a `U`-large set `s₀`, the set `s₀ ∩ { m | ∀ᶠ m' in U, m * m' ∈ s₀ }` is also
`U`-large (since `U` is idempotent). Thus in particular there is an `a₀` in this intersection. Now
let `s₁` be the intersection `s₀ ∩ { m | a₀ * m ∈ s₀ }`. By choice of `a₀`, this is again `U`-large,
so we can repeat the argument starting from `s₁`, obtaining `a₁`, `s₂`, etc. This gives the desired
infinite sequence. -/
have exists_elem : ∀ {s : set M} (hs : s ∈ U), (s ∩ { m | ∀ᶠ m' in U, m * m' ∈ s }).nonempty :=
λ s hs, ultrafilter.nonempty_of_mem (inter_mem hs $ by { rw ←U_idem at hs, exact hs }),
let elem : { s // s ∈ U } → M := λ p, (exists_elem p.property).some,
let succ : { s // s ∈ U } → { s // s ∈ U } := λ p, ⟨p.val ∩ { m | elem p * m ∈ p.val },
inter_mem p.2 $ show _, from set.inter_subset_right _ _ (exists_elem p.2).some_mem⟩,
use stream.corec elem succ (subtype.mk s₀ sU),
suffices : ∀ (a : stream M) (m ∈ FP a), ∀ p, a = stream.corec elem succ p → m ∈ p.val,
{ intros m hm, exact this _ m hm ⟨s₀, sU⟩ rfl, },
clear sU s₀,
intros a m h,
induction h with b b n h ih b n h ih,
{ rintros p rfl,
rw [stream.corec_eq, stream.head_cons],
exact set.inter_subset_left _ _ (set.nonempty.some_mem _), },
{ rintros p rfl,
refine set.inter_subset_left _ _ (ih (succ p) _),
rw [stream.corec_eq, stream.tail_cons], },
{ rintros p rfl,
have := set.inter_subset_right _ _ (ih (succ p) _),
{ simpa only using this },
rw [stream.corec_eq, stream.tail_cons], },
end
/-- The strong form of **Hindman's theorem**: in any finite cover of an FP-set, one the parts
contains an FP-set. -/
@[to_additive FS_partition_regular]
lemma FP_partition_regular {M} [semigroup M] (a : stream M) (s : set (set M)) (sfin : s.finite)
(scov : FP a ⊆ ⋃₀ s) : ∃ (c ∈ s) (b : stream M), FP b ⊆ c :=
let ⟨U, idem, aU⟩ := exists_idempotent_ultrafilter_le_FP a in
let ⟨c, cs, hc⟩ := (ultrafilter.finite_sUnion_mem_iff sfin).mp (mem_of_superset aU scov) in
⟨c, cs, exists_FP_of_large U idem c hc⟩
/-- The weak form of **Hindman's theorem**: in any finite cover of a nonempty semigroup, one of the
parts contains an FP-set. -/
@[to_additive exists_FS_of_finite_cover]
lemma exists_FP_of_finite_cover {M} [semigroup M] [nonempty M] (s : set (set M))
(sfin : s.finite) (scov : ⊤ ⊆ ⋃₀ s) : ∃ (c ∈ s) (a : stream M), FP a ⊆ c :=
let ⟨U, hU⟩ := exists_idempotent_of_compact_t2_of_continuous_mul_left
(@ultrafilter.continuous_mul_left M _) in
let ⟨c, c_s, hc⟩ := (ultrafilter.finite_sUnion_mem_iff sfin).mp (mem_of_superset univ_mem scov) in
⟨c, c_s, exists_FP_of_large U hU c hc⟩
@[to_additive FS_iter_tail_sub_FS]
lemma FP_drop_subset_FP {M} [semigroup M] (a : stream M) (n : ℕ) :
FP (a.drop n) ⊆ FP a :=
begin
induction n with n ih, { refl },
rw [nat.succ_eq_one_add, ←stream.drop_drop],
exact trans (FP.tail _) ih,
end
@[to_additive]
lemma FP.singleton {M} [semigroup M] (a : stream M) (i : ℕ) : a.nth i ∈ FP a :=
by { induction i with i ih generalizing a, { apply FP.head }, { apply FP.tail, apply ih } }
@[to_additive]
lemma FP.mul_two {M} [semigroup M] (a : stream M) (i j : ℕ) (ij : i < j) :
a.nth i * a.nth j ∈ FP a :=
begin
refine FP_drop_subset_FP _ i _,
rw ←stream.head_drop,
apply FP.cons,
rcases le_iff_exists_add.mp (nat.succ_le_of_lt ij) with ⟨d, hd⟩,
have := FP.singleton (a.drop i).tail d,
rw [stream.tail_eq_drop, stream.nth_drop, stream.nth_drop] at this,
convert this,
rw [hd, add_comm, nat.succ_add, nat.add_succ],
end
@[to_additive]
lemma FP.finset_prod {M} [comm_monoid M] (a : stream M) (s : finset ℕ) (hs : s.nonempty) :
s.prod (λ i, a.nth i) ∈ FP a :=
begin
refine FP_drop_subset_FP _ (s.min' hs) _,
induction s using finset.strong_induction with s ih,
rw [←finset.mul_prod_erase _ _ (s.min'_mem hs), ←stream.head_drop],
cases (s.erase (s.min' hs)).eq_empty_or_nonempty with h h,
{ rw [h, finset.prod_empty, mul_one], exact FP.head _ },
{ apply FP.cons, rw [stream.tail_eq_drop, stream.drop_drop, add_comm],
refine set.mem_of_subset_of_mem _ (ih _ (finset.erase_ssubset $ s.min'_mem hs) h),
have : s.min' hs + 1 ≤ (s.erase (s.min' hs)).min' h :=
nat.succ_le_of_lt (finset.min'_lt_of_mem_erase_min' _ _ $ finset.min'_mem _ _),
cases le_iff_exists_add.mp this with d hd,
rw [hd, add_comm, ←stream.drop_drop],
apply FP_drop_subset_FP }
end
end hindman
|