The IsPoly
predicate #
WittVector.IsPoly
is a (type-valued) predicate on functions f : Π R, 𝕎 R → 𝕎 R
.
It asserts that there is a family of polynomials φ : ℕ → MvPolynomial ℕ ℤ
,
such that the n
th coefficient of f x
is equal to φ n
evaluated on the coefficients of x
.
Many operations on Witt vectors satisfy this predicate (or an analogue for higher arity functions).
We say that such a function f
is a polynomial function.
The power of satisfying this predicate comes from WittVector.IsPoly.ext
.
It shows that if φ
and ψ
witness that f
and g
are polynomial functions,
then f = g
not merely when φ = ψ
, but in fact it suffices to prove
∀ n, bind₁ φ (wittPolynomial p _ n) = bind₁ ψ (wittPolynomial p _ n)
(in other words, when evaluating the Witt polynomials on φ
and ψ
, we get the same values)
which will then imply φ = ψ
and hence f = g
.
Even though this sufficient condition looks somewhat intimidating,
it is rather pleasant to check in practice;
more so than direct checking of φ = ψ
.
In practice, we apply this technique to show that the composition of WittVector.frobenius
and WittVector.verschiebung
is equal to multiplication by p
.
Main declarations #
WittVector.IsPoly
,WittVector.IsPoly₂
: two predicates that assert that a unary/binary function on Witt vectors is polynomial in the coefficients of the input values.WittVector.IsPoly.ext
,WittVector.IsPoly₂.ext
: two polynomial functions are equal if their families of polynomials are equal after evaluating the Witt polynomials on them.WittVector.IsPoly.comp
(+ many variants) show that unary/binary compositions of polynomial functions are polynomial.WittVector.idIsPoly
,WittVector.negIsPoly
,WittVector.addIsPoly₂
,WittVector.mulIsPoly₂
: several well-known operations are polynomial functions (for Verschiebung, Frobenius, and multiplication byp
, see their respective files).
On higher arity analogues #
Ideally, there should be a predicate IsPolyₙ
for functions of higher arity,
together with IsPolyₙ.comp
that shows how such functions compose.
Since mathlib does not have a library on composition of higher arity functions,
we have only implemented the unary and binary variants so far.
Nullary functions (a.k.a. constants) are treated
as constant functions and fall under the unary case.
Tactics #
There are important metaprograms defined in this file:
the tactics ghost_simp
and ghost_calc
and the attribute @[ghost_simps]
.
These are used in combination to discharge proofs of identities between polynomial functions.
The ghost_calc
tactic makes use of the IsPoly
and IsPoly₂
typeclass and its instances.
(In Lean 3, there was an @[is_poly]
attribute to manage these instances,
because typeclass resolution did not play well with function composition.
This no longer seems to be an issue, so that such instances can be defined directly.)
Any lemma doing "ring equation rewriting" with polynomial functions should be tagged
@[ghost_simps]
, e.g.
@[ghost_simps]
lemma bind₁_frobenius_poly_wittPolynomial (n : ℕ) :
bind₁ (frobenius_poly p) (wittPolynomial p ℤ n) = (wittPolynomial p ℤ (n+1))
Proofs of identities between polynomial functions will often follow the pattern
ghost_calc _
<minor preprocessing>
ghost_simp
References #
A function f : Π R, 𝕎 R → 𝕎 R
that maps Witt vectors to Witt vectors over arbitrary base rings
is said to be polynomial if there is a family of polynomials φₙ
over ℤ
such that the n
th
coefficient of f x
is given by evaluating φₙ
at the coefficients of x
.
See also WittVector.IsPoly₂
for the binary variant.
The ghost_calc
tactic makes use of the IsPoly
and IsPoly₂
typeclass and its instances.
(In Lean 3, there was an @[is_poly]
attribute to manage these instances,
because typeclass resolution did not play well with function composition.
This no longer seems to be an issue, so that such instances can be defined directly.)
- mk' :: (
- poly : ∃ (φ : ℕ → MvPolynomial ℕ ℤ), ∀ ⦃R : Type u_2⦄ [inst : CommRing R] (x : WittVector p R), (f x).coeff = fun (n : ℕ) => (MvPolynomial.aeval x.coeff) (φ n)
- )
Instances
The identity function on Witt vectors is a polynomial function.
Equations
- WittVector.IsPoly.instInhabitedId p = { default := ⋯ }
The composition of polynomial functions is polynomial.
A binary function f : Π R, 𝕎 R → 𝕎 R → 𝕎 R
on Witt vectors
is said to be polynomial if there is a family of polynomials φₙ
over ℤ
such that the n
th
coefficient of f x y
is given by evaluating φₙ
at the coefficients of x
and y
.
See also WittVector.IsPoly
for the unary variant.
The ghost_calc
tactic makes use of the IsPoly
and IsPoly₂
typeclass and its instances.
(In Lean 3, there was an @[is_poly]
attribute to manage these instances,
because typeclass resolution did not play well with function composition.
This no longer seems to be an issue, so that such instances can be defined directly.)
- mk' :: (
- poly : ∃ (φ : ℕ → MvPolynomial (Fin 2 × ℕ) ℤ), ∀ ⦃R : Type u_2⦄ [inst : CommRing R] (x y : WittVector p R), (f x y).coeff = fun (n : ℕ) => WittVector.peval (φ n) ![x.coeff, y.coeff]
- )
Instances
The composition of polynomial functions is polynomial.
The composition of a polynomial function with a binary polynomial function is polynomial.
The diagonal fun x ↦ f x x
of a polynomial function f
is polynomial.
The additive negation is a polynomial function on Witt vectors.
The function that is constantly zero on Witt vectors is a polynomial function.
The coefficients of 1 : 𝕎 R
as polynomials.
Equations
- WittVector.onePoly n = if n = 0 then 1 else 0
Instances For
The function that is constantly one on Witt vectors is a polynomial function.
Addition of Witt vectors is a polynomial function.
Multiplication of Witt vectors is a polynomial function.
Equations
- WittVector.IsPoly₂.instInhabitedHAdd = { default := ⋯ }
A macro for a common simplification when rewriting with ghost component equations.
Equations
- One or more equations did not get rendered due to their size.
Instances For
ghost_calc
is a tactic for proving identities between polynomial functions.
Typically, when faced with a goal like
∀ (x y : 𝕎 R), verschiebung (x * frobenius y) = verschiebung x * y
you can
- call
ghost_calc
- do a small amount of manual work -- maybe nothing, maybe
rintro
, etc - call
ghost_simp
and this will close the goal.
ghost_calc
cannot detect whether you are dealing with unary or binary polynomial functions.
You must give it arguments to determine this.
If you are proving a universally quantified goal like the above,
call ghost_calc _ _
.
If the variables are introduced already, call ghost_calc x y
.
In the unary case, use ghost_calc _
or ghost_calc x
.
ghost_calc
is a light wrapper around type class inference.
All it does is apply the appropriate extensionality lemma and try to infer the resulting goals.
This is subtle and Lean's elaborator doesn't like it because of the HO unification involved,
so it is easier (and prettier) to put it in a tactic script.
Equations
- One or more equations did not get rendered due to their size.