refactor(algebra/matrix): rename theorems, split proof of transposition theorem

This commit is contained in:
Rob Lewis 2016-03-24 15:45:33 -04:00 committed by Leonardo de Moura
parent 85f8f7df57
commit e721cf9c79

View file

@ -3,7 +3,7 @@ Copyright (c) 2016 Robert Y. Lewis. All rights reserved.
Released under Apache 2.0 license as described in the file LICENSE.
Author: Robert Y. Lewis
Basic properties of matrices and vectors. Proof of Farkas' lemma.
Basic properties of matrices and vectors. Proof of Motzkin's transposition theorem.
-/
import data.real data.fin algebra.module
@ -13,8 +13,8 @@ namespace matrix
definition matrix [reducible] (A : Type) (m n : ) := fin m → fin n → A
notation `M[`A`]_(`m`, `n`)` := matrix A m n
notation `M_(`m`, `n`)` := matrix _ m n
notation `M[`A`]_(`m`, `n`)` := matrix A m n
-- should rvector and cvector just be notation?
@ -22,8 +22,8 @@ definition rvector [reducible] (A : Type) (n : ) := M[A]_(1, n)
definition cvector [reducible] (A : Type) (m : ) := M[A]_(m, 1)
/-definition val_of_singleton_matrix [coercion] {A : Type} (M : matrix A 1 1) : A :=
M !fin.zero !fin.zero-/ -- this coercion fails
definition mval {A : Type} (M : matrix A 1 1) : A :=
M !fin.zero !fin.zero
-- useful for testing
definition cvector_of_list {A : Type} (l : list A) : cvector A (list.length l) :=
@ -45,7 +45,11 @@ definition r_ith (v : rvector A n) (i : fin n) := v !fin.zero i
definition c_ith (v : cvector A m) (i : fin m) := v i !fin.zero
definition nonneg [weak_order A] [has_zero A] (M : matrix A m n) := ∀ i j, M i j ≥ 0
definition nonneg [reducible] [weak_order A] [has_zero A] (M : matrix A m n) := ∀ i j, M i j ≥ 0
definition le [reducible] [has_le A] (M N : matrix A m n) := ∀ i j, M i j ≤ N i j
definition lt [reducible] [has_lt A] (M N : matrix A m n) := ∀ i j, M i j < N i j
definition r_dot [semiring A] (u v : rvector A n) := Suml (fin.upto n) (λ i, r_ith u i * r_ith v i)
@ -54,7 +58,6 @@ definition c_dot [semiring A] (u v : cvector A m) := Suml (fin.upto m) (λ i, c_
definition transpose (M : matrix A m n) : matrix A n m := λ a b, M b a
notation M`^Tr` := transpose M
definition dot [semiring A] (u : rvector A n) (v : cvector A n) := c_dot (u^Tr) v
theorem transpose_transpose (M : matrix A m n) : (M^Tr)^Tr = M := rfl
@ -161,7 +164,7 @@ theorem Suml_assoc [semiring A] (la : list B) (lb : list C) (f : B → C → A)
rewrite Suml_cons}}
end
theorem big_sum_dist_left [semiring A] (a : A) (f : T → A) :
theorem mul_Suml [semiring A] (a : A) (f : T → A) :
a * Suml l f = Suml l (λ k, a * f k) :=
begin
induction l with h tt ih,
@ -169,8 +172,8 @@ theorem big_sum_dist_left [semiring A] (a : A) (f : T → A) :
rewrite [*Suml_cons, left_distrib, ih]
end
theorem big_sum_dist_right [semiring A] (a : A) (f : T → A) :
Suml l f * a= Suml l (λ k, f k * a) :=
theorem Suml_mul [semiring A] (a : A) (f : T → A) :
(Suml l f) * a = Suml l (λ k, f k * a) :=
begin
induction l with h tt ih,
rewrite [Suml_nil, zero_mul],
@ -262,21 +265,13 @@ theorem Suml_le_of_le_strong [ordered_semiring A] (f g : T → A)
assumption}
end
theorem inner_sum_assoc [semiring A] (l1 : list B) (l2 : list C) (f : B → A) (g : B → C → A) :
theorem Suml_mul_Suml_eq_Suml_Suml_mul [semiring A] (l1 : list B) (l2 : list C) (f : B → A) (g : B → C → A) :
Suml l1 (λ s, f s * Suml l2 (λ t, g s t)) = Suml l1 (λ s, Suml l2 (λ t, f s * g s t)) :=
begin
congruence,
apply funext,
intro s,
rewrite big_sum_dist_left
end
theorem Suml_distrib [semiring A] (f : T → A) (a : A) :
Suml l (λ x, f x * a) = (Suml l f) * a :=
begin
induction l,
{rewrite [*Suml_nil, zero_mul]},
{rewrite [*Suml_cons, right_distrib], congruence, assumption}
rewrite mul_Suml
end
/- ------------------------- -/
@ -286,9 +281,9 @@ theorem m_mul_assoc [semiring A] {o p : } (M : matrix A m n) (N : matrix A n
begin
rewrite ↑mul,
repeat (apply funext; intro),
rewrite [↑c_dot, ↑c_ith, ↑row_of, ↑transpose, ↑col_of, inner_sum_assoc, Suml_assoc],
rewrite [↑c_dot, ↑c_ith, ↑row_of, ↑transpose, ↑col_of, Suml_mul_Suml_eq_Suml_Suml_mul, Suml_assoc],
congruence, apply funext, intro,
rewrite -Suml_distrib,
rewrite Suml_mul,
congruence, apply funext, intro,
rewrite mul.assoc
end
@ -372,11 +367,15 @@ theorem dot_lt_dot_of_nonneg_of_nonzero_of_lt [linear_ordered_ring A]
exact !not_lt_self Hilt
end
/- instances -/
end matrix_defs
section inst_dec
theorem matrix_inhabited [instance] (A : Type) [HA : inhabited A] (m n : ) :
/- instances -/
variables (A : Type) (m n : )
theorem matrix_inhabited [instance] [HA : inhabited A] :
inhabited (M[A]_(succ m, succ n)) :=
inhabited.rec_on HA (λ a, inhabited.mk (λ s t, a))
@ -393,9 +392,23 @@ definition matrix_decidable_eq [instance] [decidable_eq A] : decidable_eq (matri
decidable.inl (funext (λ i, funext (λ j, H i j)))
else
decidable.inr (begin intro Heq, apply H, intros, congruence, exact Heq end)
end inst_dec
end matrix_defs
/- matrix A m n is not a strict order if either m or n is 0. If it would be useful,
we could prove
order_pair (matrix A (succ m) n)
and
order_pair (matrix A m (succ n)).
-/
definition is_weak_order [instance] [order_pair A] : weak_order (matrix A m n) :=
begin
fapply weak_order.mk,
{exact le},
{unfold le, intros, apply le.refl},
{unfold le, intros, apply le.trans !a_1 !a_2},
{unfold le, intros, repeat (apply funext; intro), apply eq_of_le_of_ge !a_1 !a_2}
end
end inst_dec
section matrix_arith_m_n
@ -463,7 +476,7 @@ theorem neg_matrix [ring A] (m n : ) (M : matrix A m n) : -M = (λ a b, -M a
theorem zero_matrix [ring A] (m n : ) : (0 : matrix A m n) = (λ a b, 0) := rfl
theorem m_add_app [ring A] (m n : ) (M N : matrix A m n) (a : fin m) (b : fin n) :
theorem matrix_add_app [ring A] (m n : ) (M N : matrix A m n) (a : fin m) (b : fin n) :
(M + N) a b = M a b + N a b := rfl
theorem neg_smul [ring A] (m n o : ) (M : matrix A m n) (N : matrix A n o) : (-M) ⬝ N = - (M ⬝ N) :=
@ -497,9 +510,44 @@ theorem smul_zero [ring A] (l m n : ) (M : matrix A l m) : M ⬝ (0 : matrix
apply Suml_zero
end
theorem le_iff_forall_row [weak_order A] {n : } (u v : rvector A n) :
u ≤ v ↔ ∀ i, r_ith u i ≤ r_ith v i :=
begin
replace u ≤ v with le u v,
unfold le,
apply iff.intro,
{intros H i, apply H},
{intros H i j, rewrite fin.fin_one_eq_zero, apply H}
end
theorem le_iff_forall_col [weak_order A] {n : } (u v : cvector A n) :
u ≤ v ↔ ∀ i, c_ith u i ≤ c_ith v i :=
begin
replace u ≤ v with le u v,
unfold le,
apply iff.intro,
{intros H i, apply H},
{intros H i j, rewrite fin.fin_one_eq_zero, apply H}
end
-- why doesn't rfl work to prove this?
theorem nonneg_iff_le_zero_row [ordered_ring A] {n : } (u : rvector A n) :
nonneg u ↔ (0 : rvector A n) ≤ u :=
begin
change nonneg u ↔ le (λ a b, 0) u,
reflexivity
end
theorem nonneg_iff_le_zero_col [ordered_ring A] {n : } (u : cvector A n) :
nonneg u ↔ (0 : cvector A n) ≤ u :=
begin
change nonneg u ↔ le (0 : cvector A n) u,
reflexivity
end
end matrix_arith_m_n
section motzkin_inversion
section motzkin_transposition
variables {A : Type} [linear_ordered_ring A]
/-
@ -507,6 +555,9 @@ variables {A : Type} [linear_ordered_ring A]
a strict ineqs P : M[A]_(a, n), p : cvector a
b weak ineqs Q : M[A]_(b, n), q : cvector b
c eqs R : M[A]_(c, n), r : cvector c
There are various ways to express Hsum equivalently: eg,
(y ⬝ P) + (z ⬝ Q) + (t ⬝ R) = (0 : rvector A n).
-/
variables {a b c n : }
(P : M[A]_(a, n)) (Q : M[A]_(b, n)) (R : M[A]_(c, n))
@ -514,20 +565,14 @@ variables {a b c n : }
(y : rvector A a) (z : rvector A b) (t : rvector A c)
(Hnny : nonneg y) (Hnnz : nonneg z)
(Hsum : ∀ i : fin n, r_ith (y ⬝ P) i + r_ith (z ⬝ Q) i + r_ith (t ⬝ R) i = 0)
include Hnny Hnnz Hsum
theorem motzkin_inversion_with_equalities
(Hor : (c_dot (y^Tr) p + c_dot (z^Tr) q + c_dot (t^Tr) r < 0)
((∃ i, r_ith y i > 0) ∧ c_dot (y^Tr) p + c_dot (z^Tr) q + c_dot (t^Tr) r ≤ 0)) :
¬ ∃ x : cvector A n, (∀ i, c_ith (P ⬝ x) i < c_ith p i) ∧
(∀ i, c_ith (Q ⬝ x) i ≤ c_ith q i) ∧
(∀ i, c_ith (R ⬝ x) i = c_ith r i) :=
include Hsum
-- .3 seconds to elaborate
private theorem dot_eq_zero {x : cvector A n} (HsatR : ∀ i : fin c, c_ith (R⬝x) i = c_ith r i) :
dot y (P ⬝ x) + dot z (Q ⬝ x) + dot t r = 0 :=
begin
intro Hsat,
cases Hsat with x Hsat,
cases Hsat with HsatP Hsat,
cases Hsat with HsatQ HsatR,
have Hneg : (y ⬝ P) + (z ⬝ Q) + (t ⬝ R) = 0, begin
have Hneg : (y ⬝ P) + (z ⬝ Q) + (t ⬝ R) = (0 : rvector A n), begin
change (λ a b, (y⬝P + z⬝Q + t⬝R) a b) = (λ a b, 0),
repeat (apply funext; intro),
rewrite fin.fin_one_eq_zero,
@ -538,18 +583,33 @@ theorem motzkin_inversion_with_equalities
rewrite fin.fin_one_eq_zero,
apply HsatR
end,
have Hneg' : ((y ⬝ P) + (z ⬝ Q) + (t ⬝ R)) ⬝ x = 0 ⬝ x, by rewrite Hneg,
have Hneg' : ((y ⬝ P) + (z ⬝ Q) + (t ⬝ R)) ⬝ x = 0 ⬝ x, by rewrite Hneg, -- this line and the following take about .1 sec
rewrite [2 distrib_smul_right at Hneg', -3 m_mul_assoc at Hneg', Hr at Hneg',
zero_smul at Hneg', zero_matrix at Hneg'],
have Hneg'' : (λ a b : fin 1, (y⬝(P⬝x) + z⬝(Q⬝x) + t⬝r) a b) = (λ a b : fin 1, 0), from Hneg',
have Heqz : (y⬝(P⬝x) + z⬝(Q⬝x) + t⬝r) !fin.zero !fin.zero = 0, by rewrite Hneg'',
have Heqz' : (y⬝(P⬝x)) !fin.zero !fin.zero + (z⬝(Q⬝x)) !fin.zero !fin.zero + (t⬝r) !fin.zero !fin.zero = 0,
from Heqz,
rewrite -*dot_eq_mul at Heqz',
rewrite *dot_eq_mul,
exact Heqz'
end
include Hnny Hnnz
theorem motzkin_transposition_with_equalities_lt
(Hlt : (c_dot (y^Tr) p + c_dot (z^Tr) q + c_dot (t^Tr) r < 0)) :
¬ ∃ x : cvector A n, (∀ i, c_ith (P ⬝ x) i < c_ith p i) ∧
(∀ i, c_ith (Q ⬝ x) i ≤ c_ith q i) ∧
(∀ i, c_ith (R ⬝ x) i = c_ith r i) :=
begin
intro Hsat,
cases Hsat with x Hsat,
cases Hsat with HsatP Hsat,
cases Hsat with HsatQ HsatR,
note Heqz' := dot_eq_zero P Q R r y z t Hsum HsatR,
apply not_lt_self (0 : A),
rewrite -Heqz' at {1},
cases Hor with Hor1 Hor2,
{have Hdlt : dot y (P⬝x) + dot z (Q⬝x) + dot t r ≤ dot y p + dot z q + dot t r, begin
have Hdlt : dot y (P⬝x) + dot z (Q⬝x) + dot t r ≤ dot y p + dot z q + dot t r, begin
apply add_le_add_right,
apply add_le_add,
apply dot_le_dot_of_nonneg_of_le,
@ -563,8 +623,23 @@ theorem motzkin_inversion_with_equalities
end,
apply lt_of_le_of_lt,
exact Hdlt,
exact Hor1},
{cases Hor2 with Hj Hor2,
exact Hlt
end
theorem motzkin_transposition_with_equalities_le
(Hyp : (∃ i, r_ith y i > 0) ∧ c_dot (y^Tr) p + c_dot (z^Tr) q + c_dot (t^Tr) r ≤ 0) :
¬ ∃ x : cvector A n, (∀ i, c_ith (P ⬝ x) i < c_ith p i) ∧
(∀ i, c_ith (Q ⬝ x) i ≤ c_ith q i) ∧
(∀ i, c_ith (R ⬝ x) i = c_ith r i) :=
begin
intro Hsat,
cases Hsat with x Hsat,
cases Hsat with HsatP Hsat,
cases Hsat with HsatQ HsatR,
note Heqz' := dot_eq_zero P Q R r y z t Hsum HsatR,
apply not_lt_self (0 : A),
rewrite -Heqz' at {1},
cases Hyp with Hj Hor2,
cases Hj with j Hj,
have Hdlt : dot y (P⬝x) + dot z (Q⬝x) + dot t r < dot y p + dot z q + dot t r, begin
apply add_lt_add_right,
@ -579,33 +654,24 @@ theorem motzkin_inversion_with_equalities
end,
apply lt_of_lt_of_le,
exact Hdlt,
exact Hor2}
exact Hor2
end
/- these are more useful statements of the above, without the disjunctive hypothesis. -/
theorem motzkin_inversion_with_equalities_lt
(Hlt : (c_dot (y^Tr) p + c_dot (z^Tr) q + c_dot (t^Tr) r < 0)) :
¬ ∃ x : cvector A n, (∀ i, c_ith (P ⬝ x) i < c_ith p i) ∧
(∀ i, c_ith (Q ⬝ x) i ≤ c_ith q i) ∧
(∀ i, c_ith (R ⬝ x) i = c_ith r i) :=
theorem motzkin_transposition_with_equalities
(Hor : (c_dot (y^Tr) p + c_dot (z^Tr) q + c_dot (t^Tr) r < 0)
((∃ i : fin a, r_ith y i > 0) ∧ c_dot (y^Tr) p + c_dot (z^Tr) q + c_dot (t^Tr) r ≤ 0)) :
¬ ∃ x : cvector A n, (∀ i : fin a, c_ith (P ⬝ x) i < c_ith p i) ∧
(∀ i : fin b, c_ith (Q ⬝ x) i ≤ c_ith q i) ∧
(∀ i : fin c, c_ith (R ⬝ x) i = c_ith r i) :=
begin
apply motzkin_inversion_with_equalities,
exact Hnny, exact Hnnz, exact Hsum,
left, exact Hlt
cases Hor with Hor1 Hor2,
{apply motzkin_transposition_with_equalities_lt,
exact Hnny, exact Hnnz, exact Hsum, exact Hor1},
{apply motzkin_transposition_with_equalities_le,
exact Hnny, exact Hnnz, exact Hsum, exact Hor2}
end
theorem motzkin_inversion_with_equalities_le
(Hyp : (∃ i, r_ith y i > 0) ∧ c_dot (y^Tr) p + c_dot (z^Tr) q + c_dot (t^Tr) r ≤ 0) :
¬ ∃ x : cvector A n, (∀ i, c_ith (P ⬝ x) i < c_ith p i) ∧
(∀ i, c_ith (Q ⬝ x) i ≤ c_ith q i) ∧
(∀ i, c_ith (R ⬝ x) i = c_ith r i) :=
begin
apply motzkin_inversion_with_equalities,
exact Hnny, exact Hnnz, exact Hsum,
right, exact Hyp
end
end motzkin_inversion
end motzkin_transposition
section matrix_arith_square
open fin
@ -621,7 +687,7 @@ theorem Suml_map {A B C : Type} [add_monoid C] (l : list A) (f : A → B) (g : B
end
-- this is annoying!
theorem b_vec_dot {A : Type} [semiring A] {m : } (k : fin m) (f : fin m → A) :
theorem dot_basis_vec {A : Type} [semiring A] {m : } (k : fin m) (f : fin m → A) :
Suml (upto m) (λ j, (f j) * if j = k then 1 else 0) = f k :=
begin
induction m,
@ -669,14 +735,14 @@ theorem b_vec_dot' {A : Type} [semiring A] {m : } (k : fin m) (f : fin m →
rewrite [if_pos Heq, if_pos (eq.symm Heq), mul_one, one_mul],
rewrite [if_neg Hneq, if_neg (ne.symm Hneq), mul_zero, zero_mul]
end,
rewrite [H, b_vec_dot]
rewrite [H, dot_basis_vec]
end
theorem sq_matrix_mul_one {A : Type} [semiring A] {m : } (M : matrix A m m) : mul M (sq_matrix_id A m) = M :=
begin
rewrite [↑sq_matrix_id, ↑mul],
repeat (apply funext; intro),
rewrite [↑row_of, ↑transpose, ↑col_of, ↑c_dot, ↑c_ith, b_vec_dot]
rewrite [↑row_of, ↑transpose, ↑col_of, ↑c_dot, ↑c_ith, dot_basis_vec]
end
theorem sq_matrix_one_mul {A : Type} [semiring A] {m : } (M : matrix A m m) : mul (sq_matrix_id A m) M = M :=
@ -722,7 +788,7 @@ definition m_square_ring [instance] [reducible] (A : Type) [ring A] (m : ) :
end matrix_arith_square
/-
section test
open list rat int -- why aren't rats pretty printing right?
definition c1 : cvector _ := cvector_of_list [3, 4, 2]
@ -783,6 +849,6 @@ definition map : matrix 0 3 := λ a b, 1
eval (mep ⬝ map) !fin.zero !fin.zero
end test
end test-/
end matrix