Library Coq.Numbers.NatInt.NZSqrt
Square Root Function
Interface of a sqrt function, then its specification on naturals
Module Type Sqrt (Import A : Typ).
Parameter Inline sqrt : t -> t.
End Sqrt.
Module Type SqrtNotation (A : Typ)(Import B : Sqrt A).
Notation "√ x" := (sqrt x) (at level 6).
End SqrtNotation.
Module Type Sqrt' (A : Typ) := Sqrt A <+ SqrtNotation A.
Module Type NZSqrtSpec (Import A : NZOrdAxiomsSig')(Import B : Sqrt' A).
Axiom sqrt_spec : forall a, 0<=a -> √a * √a <= a < S (√a) * S (√a).
Axiom sqrt_neg : forall a, a<0 -> √a == 0.
End NZSqrtSpec.
Module Type NZSqrt (A : NZOrdAxiomsSig) := Sqrt A <+ NZSqrtSpec A.
Module Type NZSqrt' (A : NZOrdAxiomsSig) := Sqrt' A <+ NZSqrtSpec A.
Derived properties of power
Module Type NZSqrtProp
(Import A : NZOrdAxiomsSig')
(Import B : NZSqrt' A)
(Import C : NZMulOrderProp A).
First, sqrt is non-negative
The spec of sqrt indeed determines it
Hence sqrt is a morphism
An alternate specification
Lemma sqrt_spec_alt : forall a, 0<=a -> exists r,
a == (√a)² + r /\ 0 <= r <= 2*√a.
Lemma sqrt_unique' : forall a b c, 0<=c<=2*b ->
a == b² + c -> √a == b.
Sqrt is exact on squares
Sqrt and predecessors of squares
Sqrt is a monotone function (but not a strict one)
No reverse result for <=, consider for instance √2 <= √1
When left side is a square, we have an equivalence for <=
When right side is a square, we have an equivalence for <
Sqrt and basic constants
Lemma sqrt_0 : √0 == 0.
Lemma sqrt_1 : √1 == 1.
Lemma sqrt_2 : √2 == 1.
Lemma sqrt_pos : forall a, 0 < √a <-> 0 < a.
Lemma sqrt_lt_lin : forall a, 1<a -> √a<a.
Lemma sqrt_le_lin : forall a, 0<=a -> √a<=a.
Sqrt and multiplication.
Due to rounding error, we don't have the usual √(a*b) = √a*√b
but only lower and upper bounds.
Lemma sqrt_mul_below : forall a b, √a * √b <= √(a*b).
Lemma sqrt_mul_above : forall a b, 0<=a -> 0<=b -> √(a*b) < S (√a) * S (√b).
And we can't find better approximations in general.
Sqrt and successor :
- The lower bound is exact for squares
- Concerning the upper bound, for any c>0, take a=b=c²-1, then √(a*b) = c² -1 while S √a = S √b = c
- the sqrt function climbs by at most 1 at a time
- otherwise it stays at the same value
- the +1 steps occur for squares
Lemma sqrt_succ_le : forall a, 0<=a -> √(S a) <= S (√a).
Lemma sqrt_succ_or : forall a, 0<=a -> √(S a) == S (√a) \/ √(S a) == √a.
Lemma sqrt_eq_succ_iff_square : forall a, 0<=a ->
(√(S a) == S (√a) <-> exists b, 0<b /\ S a == b²).
Sqrt and addition
convexity inequality for sqrt: sqrt of middle is above middle
of square roots.
Lemma add_sqrt_le : forall a b, 0<=a -> 0<=b -> √a + √b <= √(2*(a+b)).
End NZSqrtProp.
Module Type NZSqrtUpProp
(Import A : NZDecOrdAxiomsSig')
(Import B : NZSqrt' A)
(Import C : NZMulOrderProp A)
(Import D : NZSqrtProp A B C).
For once, we define instead of axiomatizing, thanks to sqrt
Definition sqrt_up a :=
match compare 0 a with
| Lt => S √(P a)
| _ => 0
end.
Lemma sqrt_up_eqn0 : forall a, a<=0 -> √°a == 0.
Lemma sqrt_up_eqn : forall a, 0<a -> √°a == S √(P a).
Lemma sqrt_up_spec : forall a, 0<a -> (P √°a)² < a <= (√°a)².
First, sqrt_up is non-negative
sqrt_up is a morphism
The spec of sqrt_up indeed determines it
sqrt_up is exact on squares
sqrt_up and successors of squares
Basic constants
Links between sqrt and sqrt_up
Lemma le_sqrt_sqrt_up : forall a, √a <= √°a.
Lemma le_sqrt_up_succ_sqrt : forall a, √°a <= S (√a).
Lemma sqrt_sqrt_up_spec : forall a, 0<=a -> (√a)² <= a <= (√°a)².
Lemma sqrt_sqrt_up_exact :
forall a, 0<=a -> (√a == √°a <-> exists b, 0<=b /\ a == b²).
sqrt_up is a monotone function (but not a strict one)
No reverse result for <=, consider for instance √°3 <= √°2
When left side is a square, we have an equivalence for <
When right side is a square, we have an equivalence for <=
Lemma sqrt_up_le_square : forall a b, 0<=a -> 0<=b -> (a <= b² <-> √°a <= b).
Lemma sqrt_up_pos : forall a, 0 < √°a <-> 0 < a.
Lemma sqrt_up_lt_lin : forall a, 2<a -> √°a < a.
Lemma sqrt_up_le_lin : forall a, 0<=a -> √°a<=a.
sqrt_up and multiplication.
Due to rounding error, we don't have the usual √(a*b) = √a*√b
but only lower and upper bounds.
Lemma sqrt_up_mul_above : forall a b, 0<=a -> 0<=b -> √°(a*b) <= √°a * √°b.
Lemma sqrt_up_mul_below : forall a b, 0<a -> 0<b -> (P √°a)*(P √°b) < √°(a*b).
And we can't find better approximations in general.
sqrt_up and successor :
- The upper bound is exact for squares
- Concerning the lower bound, for any c>0, take a=b=c²+1, then √°(a*b) = c²+1 while P √°a = P √°b = c
- the sqrt_up function climbs by at most 1 at a time
- otherwise it stays at the same value
- the +1 steps occur after squares
Lemma sqrt_up_succ_le : forall a, 0<=a -> √°(S a) <= S (√°a).
Lemma sqrt_up_succ_or : forall a, 0<=a -> √°(S a) == S (√°a) \/ √°(S a) == √°a.
Lemma sqrt_up_eq_succ_iff_square : forall a, 0<=a ->
(√°(S a) == S (√°a) <-> exists b, 0<=b /\ a == b²).
sqrt_up and addition
Convexity-like inequality for sqrt_up: sqrt_up of middle is above middle
of square roots. We cannot say more, for instance take a=b=2, then
2+2 <= S 3