refactor(library/algebra/ordered_ring): add workarounds to improve performance

This commit is contained in:
Leonardo de Moura 2014-12-13 13:12:25 -08:00
parent 6f775be1b6
commit 628faa10eb

View file

@ -34,6 +34,14 @@ section
variables (a b c d e : A)
include s
-- TODO: remove after we short-circuit class-graph
definition ordered_semiring.to_mul [instance] [priority 100000] : has_mul A :=
has_mul.mk (@ordered_semiring.mul A s)
definition ordered_semiring.to_lt [instance] [priority 100000] : has_lt A :=
has_lt.mk (@ordered_semiring.lt A s)
definition ordered_semiring.to_zero [instance] [priority 100000] : has_zero A :=
has_zero.mk (@ordered_semiring.zero A s)
theorem mul_le_mul_of_nonneg_left {a b c : A} (Hab : a ≤ b) (Hc : 0 ≤ c) :
c * a ≤ c * b := !ordered_semiring.mul_le_mul_left Hab Hc
@ -95,6 +103,14 @@ section
variables (a b c : A)
include s
-- TODO: remove after we short-circuit class-graph
definition linear_ordered_semiring.to_mul [instance] [priority 100000] : has_mul A :=
has_mul.mk (@linear_ordered_semiring.mul A s)
definition linear_ordered_semiring.to_lt [instance] [priority 100000] : has_lt A :=
has_lt.mk (@linear_ordered_semiring.lt A s)
definition linear_ordered_semiring.to_zero [instance] [priority 100000] : has_zero A :=
has_zero.mk (@linear_ordered_semiring.zero A s)
theorem lt_of_mul_lt_mul_left {a b c : A} (H : c * a < c * b) (Hc : c ≥ 0) : a < b :=
lt_of_not_le
(assume H1 : b ≤ a,
@ -190,6 +206,14 @@ section
variables (a b c : A)
include s
-- TODO: remove after we short-circuit class-graph
definition ordered_ring.to_mul [instance] [priority 100000] : has_mul A :=
has_mul.mk (@ordered_ring.mul A s)
definition ordered_ring.to_lt [instance] [priority 100000] : has_lt A :=
has_lt.mk (@ordered_ring.lt A s)
definition ordered_ring.to_zero [instance] [priority 100000] : has_zero A :=
has_zero.mk (@ordered_ring.zero A s)
theorem mul_le_mul_of_nonpos_left {a b c : A} (H : b ≤ a) (Hc : c ≤ 0) : c * a ≤ c * b :=
have Hc' : -c ≥ 0, from iff.mp' !neg_nonneg_iff_nonpos Hc,
have H1 : -c * b ≤ -c * a, from mul_le_mul_of_nonneg_left H Hc',
@ -249,6 +273,14 @@ section
theorem le_add_of_nonneg_right {a b : A} (H : b ≥ 0) : a ≤ a + b := !add.right_id ▸ add_le_add_left H a
theorem le_add_of_nonneg_left {a b : A} (H : b ≥ 0) : a ≤ b + a := !add.left_id ▸ add_le_add_right H a
-- TODO: remove after we short-circuit class-graph
definition linear_ordered_ring.to_mul [instance] [priority 100000] : has_mul A :=
has_mul.mk (@linear_ordered_ring.mul A s)
definition linear_ordered_ring.to_lt [instance] [priority 100000] : has_lt A :=
has_lt.mk (@linear_ordered_ring.lt A s)
definition linear_ordered_ring.to_zero [instance] [priority 100000] : has_zero A :=
has_zero.mk (@linear_ordered_ring.zero A s)
/- TODO: a good example of a performance bottleneck.
Without any of the "proof ... qed" pairs, it exceeds the unifier maximum number of steps.
@ -265,72 +297,27 @@ section
theorem pos_and_pos_or_neg_and_neg_of_mul_pos (Hab : a * b > 0) :
(a > 0 ∧ b > 0) (a < 0 ∧ b < 0) :=
lt_or_eq_or_lt_cases
-- @lt_or_eq_or_lt_cases A s 0 a ((a > 0 ∧ b > 0) (a < 0 ∧ b < 0))
(assume Ha : 0 < a,
proof
lt_or_eq_or_lt_cases
-- @lt_or_eq_or_lt_cases A s 0 b ((a > 0 ∧ b > 0) (a < 0 ∧ b < 0))
(assume Hb : 0 < b, or.inl (and.intro Ha Hb))
(assume Hb : 0 = b,
have H : 0 > 0, from !mul_zero_eq ▸ Hb⁻¹ ▸ Hab,
absurd H (lt.irrefl 0))
absurd (!mul_zero_eq ▸ Hb⁻¹ ▸ Hab) (lt.irrefl 0))
(assume Hb : b < 0,
have Hab' : a * b < 0, from mul_neg_of_pos_of_neg Ha Hb,
absurd Hab (not_lt_of_lt Hab'))
qed)
absurd Hab (not_lt_of_lt (mul_neg_of_pos_of_neg Ha Hb))))
(assume Ha : 0 = a,
proof
have H : 0 > 0, from !zero_mul_eq ▸ Ha⁻¹ ▸ Hab,
absurd H (lt.irrefl 0)
qed)
absurd (!zero_mul_eq ▸ Ha⁻¹ ▸ Hab) (lt.irrefl 0))
(assume Ha : a < 0,
proof
lt_or_eq_or_lt_cases
-- @lt_or_eq_or_lt_cases A s 0 b ((a > 0 ∧ b > 0) (a < 0 ∧ b < 0))
(assume Hb : 0 < b,
have Hab' : a * b < 0, from mul_neg_of_neg_of_pos Ha Hb,
absurd Hab (not_lt_of_lt Hab'))
absurd Hab (not_lt_of_lt (mul_neg_of_neg_of_pos Ha Hb)))
(assume Hb : 0 = b,
have H : 0 > 0, from !mul_zero_eq ▸ Hb⁻¹ ▸ Hab,
absurd H (lt.irrefl 0))
(assume Hb : b < 0, or.inr (and.intro Ha Hb))
qed)
absurd (!mul_zero_eq ▸ Hb⁻¹ ▸ Hab) (lt.irrefl 0))
(assume Hb : b < 0, or.inr (and.intro Ha Hb)))
/-
This version, with tactics, is also slow.
Also, I do not understand why I cannot succesfully return to "proof mode"
in the proof below. I think "s" is not visible there. Can I make it visible?
Here is the really strange thing: compiling the file with both proofs seems faster than
compiling the file with just one of them.
-/
theorem pos_and_pos_or_neg_and_neg_of_mul_pos' (Hab : a * b > 0) :
(a > 0 ∧ b > 0) (a < 0 ∧ b < 0) :=
begin
apply (@lt_or_eq_or_lt_cases _ _ 0 a),
intro Ha,
apply (@lt_or_eq_or_lt_cases _ _ 0 b),
intro Hb, apply (or.inl (and.intro Ha Hb)),
intro Hb,
/-
proof -- gives invalid local context
have H : 0 > 0, from !mul_zero_eq ▸ (eq.symm Hb) ▸ Hab,
absurd H (lt.irrefl 0)
qed
-/
apply (absurd (!mul_zero_eq ▸ (eq.symm Hb) ▸ Hab) (lt.irrefl 0)),
intro Hb,
apply (absurd Hab (not_lt_of_lt (mul_neg_of_pos_of_neg Ha Hb))),
intro Ha,
apply (absurd (!zero_mul_eq ▸ Ha⁻¹ ▸ Hab) (lt.irrefl 0)),
intro Ha,
apply (@lt_or_eq_or_lt_cases _ _ 0 b),
intro Hb,
apply (absurd Hab (not_lt_of_lt (mul_neg_of_neg_of_pos Ha Hb))),
intro Hb,
apply (absurd (!mul_zero_eq ▸ (eq.symm Hb) ▸ Hab) (lt.irrefl 0)),
intro Hb, apply (or.inr (and.intro Ha Hb))
end
set_option pp.coercions true
set_option pp.implicit true
set_option pp.notation false
-- print definition pos_and_pos_or_neg_and_neg_of_mul_pos
-- TODO: use previous and integral domain
theorem noneg_and_nonneg_or_nonpos_and_nonpos_of_mul_nonneg (Hab : a * b ≥ 0) :