A function satisfying the parallelogram law is a quadratic form #
Let M and N be additive commutative groups, suppose doubling is injective on N,
and let f : M → N satisfy the parallelogram law
f (x + y) + f (x - y) = 2 • f x + 2 • f y.
Then f is a quadratic form: its polarisation QuadraticMap.polar f x y = f (x + y) - f x - f y
is biadditive, and f (n • x) = n ^ 2 • f x. This file proves that, and packages it as a
QuadraticMap ℤ M N. Both f 0 = 0 and evenness of f come free from the law rather than being
assumed.
Mathlib has the converse direction — QuadraticMap.polar_add_left and friends read biadditivity
off a QuadraticMap, and LinearMap.BilinMap.toQuadraticMap builds one from a bilinear map —
but nothing in the other direction from the parallelogram law alone. Its parallelogram_law and
parallelogram_law_with_norm are statements about inner product spaces, and the Jordan–von Neumann
construction in Analysis/InnerProductSpace/OfNorm.lean recovers an inner product from a norm on
a real or complex space, using continuity. Neither applies to a function on a bare abelian group.
The elementary helpers need less: evenness holds without any condition on doubling, and
zero and evenness need only a left-cancellative additive monoid as target. Natural quadratic
scaling needs only a right-cancellative additive monoid as target, while integer scaling needs
an additive group. Both scaling results allow arbitrary additive groups as sources and assume
f 0 = 0. Thus scaling also applies to targets with 2-torsion.
The quadratic-map construction requires absence of 2-torsion #
htwo : IsSMulRegular N 2 says that doubling is injective on N. It is sharp in both directions.
It is not IsAddTorsionFree N, which is strictly stronger and excludes codomains where the
conclusion holds: IsSMulRegular (ZMod 3) 2 is true even though ZMod 3 has 3-torsion.
Nor can it be weakened away. With M = N = ZMod 2 every function satisfies the parallelogram
law, because x - y = x + y and 2 • z = 0 there; the constant function 1 is then one that
satisfies it while failing even f 0 = 0, which every quadratic form obeys — and correspondingly
¬ IsSMulRegular (ZMod 2) 2. Even assuming f 0 = 0 is insufficient for biadditivity:
on (ZMod 2)³, the function f(x) = x₁x₂x₃ satisfies the law and preserves zero, but violates
the three-variable identity at the three standard basis vectors.
For a torsion-free codomain it is one term: smul_right_injective N two_ne_zero supplies it for
N = ℝ, the canonical height's target, and for N = ℤ, the degree form's.
Main results #
TauCeti.QuadraticMap.map_zero_of_parallelogram:f 0 = 0when doubling is injective.TauCeti.QuadraticMap.map_neg_of_parallelogram:f (-x) = f x.TauCeti.QuadraticMap.map_add_add_add_map_of_parallelogram: the three-variable identityf (x + y + z) + (f x + f y + f z) = f (x + y) + f (y + z) + f (z + x), stated exactly asQuadraticMap.map_add_add_add_map.TauCeti.QuadraticMap.polar_add_left_of_parallelogramandpolar_zsmul_left_of_parallelogram: the polarisation is additive andℤ-linear on the left.TauCeti.QuadraticMap.map_nsmul_of_parallelogram:f (n • x) = n ^ 2 • f xforn : ℕ, assumingf 0 = 0, for an additive groupMand right-cancellative additive monoidN.TauCeti.QuadraticMap.map_zsmul_of_parallelogram:f (n • x) = n ^ 2 • f xforn : ℤ, assumingf 0 = 0, without commutativity of either group or injectivity of doubling onN.TauCeti.QuadraticMap.ofParallelogram:fas aQuadraticMap ℤ M N, with the polarisation as its companion bilinear map.
Where this is used #
Two constructions in arithmetic geometry arrive at a function known to satisfy the parallelogram law and want it as a quadratic form.
The canonical height of an elliptic curve is one. It satisfies the parallelogram law exactly, and
its polarisation is the Néron–Tate height pairing up to a factor of two: by
QuadraticMap.polar_self the polarisation here has polar f x x = 2 • f x, whereas the pairing
whose Gram determinant on a basis of the free part of the Mordell–Weil group is the regulator is
normalised so that ⟨P, P⟩ is the height itself. A consumer wanting the regulator convention
halves this one; the choice is not made here, since halving is not available in a general abelian
group.
The degree form on End E is the other: its polarisation is the trace form, and non-negativity of
the degree gives the Hasse bound by Cauchy–Schwarz (Silverman, The Arithmetic of Elliptic
Curves, V.1.2).
Both take values in a torsion-free group — ℝ and ℤ respectively — so both satisfy the
hypothesis below with room to spare. Stated for a general abelian group so that neither carries
its own copy.
A parallelogram-law function preserves zero.
A parallelogram-law function is even.
Quadraticity: a parallelogram-law function that preserves zero satisfies
f (n • x) = n ^ 2 • f x for a natural number n, with an additive group source and a
right-cancellative additive monoid target.
Quadraticity: a parallelogram-law function that preserves zero satisfies
f (n • x) = n ^ 2 • f x for an integer n, even for noncommutative additive groups.
The three-variable identity satisfied by every quadratic form, in subtraction-free form.
The polarisation is additive in its left argument.
The polarisation is ℤ-linear in its left argument.
A function satisfying the parallelogram law is a quadratic form. Its companion bilinear
map is QuadraticMap.polarBilin of it, which is the polarisation.
Equations
- TauCeti.QuadraticMap.ofParallelogram htwo hf = QuadraticMap.ofPolar f ⋯ ⋯ ⋯
Instances For
ofParallelogram coerces back to the function it was built from.
ofParallelogram evaluated at a point is the original function there.