Positive-definite functions on an involutive additive monoid #
A complex-valued function F on an additive monoid M equipped with an involution star
(an AddMonoid with a StarAddMonoid structure) is positive definite when, for every
finite family (cᵢ, aᵢ) of scalars cᵢ : ℂ and points aᵢ : M, the Hermitian form
∑_{i,j} cᵢ · conj(cⱼ) · F(aᵢ + aⱼ⋆) is a nonnegative real number. The involution aⱼ⋆ inside
the argument is what makes this the right notion on an involutive semigroup (Berg–Christensen–
Ressel): on a finite-dimensional real inner-product space with a⋆ = -a it specialises to the
classical translation-invariant positive-definiteness ∑ cᵢ conj(cⱼ) F(aᵢ - aⱼ) ≥ 0, and on the
product monoid ℝ≥0 × V it produces the BCR involution (t, a)⋆ = (t, -a).
This file introduces the predicate TauCeti.IsPositiveDefinite at this general level and develops
its basic algebraic API: the value at 0 is real and nonnegative, the function is conjugate
symmetric in the involution, it satisfies the Cauchy–Schwarz inequality coming from the 2 × 2
sub-form, and the class is closed under sums and nonnegative complex scalar multiples, with the
Schur pointwise product closure and nonnegative constants as examples.
This is the Objects and first API to develop slice of Part C of the OneParameterSemigroups
roadmap in TauCetiRoadmap: "positive-definite functions and Bochner's theorem". Mathlib has
related APIs for positive-semidefinite matrices, bilinear and linear maps, and RKHS kernels, but
not for positive-definite functions on an involutive monoid, so the predicate and its API are built
here. The continuity theory and Bochner's representation theorem are later milestones.
Main declarations #
TauCeti.IsPositiveDefinite: the positive-definiteness predicate forF : M → ℂ.TauCeti.IsPositiveDefinite.sum_nonneg: nonnegativity for arbitrary finite families.TauCeti.IsPositiveDefinite.quadForm_two_nonneg: nonnegativity of the2 × 2sub-form.TauCeti.IsPositiveDefinite.map_zero_nonneg:0 ≤ F 0.TauCeti.IsPositiveDefinite.map_zero_im:(F 0).im = 0.TauCeti.IsPositiveDefinite.map_zero_re_nonneg:0 ≤ (F 0).re.TauCeti.IsPositiveDefinite.map_zero_eq_ofReal_re:F 0 = ((F 0).re : ℂ).TauCeti.IsPositiveDefinite.map_zero_re_pos_of_ne_zero: ifF 0 ≠ 0, then0 < (F 0).re.TauCeti.IsPositiveDefinite.conj_symm:conj (F (b + a⋆)) = F (a + b⋆).TauCeti.IsPositiveDefinite.normSq_le: the Cauchy–Schwarz inequality‖F (a + b⋆)‖² ≤ (F (a + a⋆)).re * (F (b + b⋆)).re.TauCeti.IsPositiveDefinite.norm_apply_le_map_zero_re_of_add_star_eq_zero:‖F a‖ ≤ (F 0).rewhena + star a = 0, with the additive-group corollaryTauCeti.IsPositiveDefinite.norm_apply_le_map_zero_re_of_star_eq_negforstar a = -a.TauCeti.IsPositiveDefinite.apply_eq_zero_of_map_zero_re_eq_zero:(F 0).re = 0forcesFto vanish identically.TauCeti.IsPositiveDefinite.posSemidef: a positive-definite function gives the positive-definite kernelfun a b => F (a + star b).TauCeti.IsPositiveDefinite.of_posSemidef: conversely, if the kernelfun a b => F (a + star b)is positive definite thenFis positive definite.TauCeti.IsPositiveDefinite.add,TauCeti.IsPositiveDefinite.sum,TauCeti.IsPositiveDefinite.const_mul,TauCeti.IsPositiveDefinite.mul,TauCeti.IsPositiveDefinite.prod,TauCeti.isPositiveDefinite_const: closure properties and examples.
References #
- C. Berg, J. P. R. Christensen, P. Ressel, Harmonic Analysis on Semigroups (GTM 100, 1984), Chapter 3.
A function F : M → ℂ on an involutive additive monoid is positive definite when, for
every finite family of scalars c : Fin n → ℂ and points v : Fin n → M, the Hermitian form
∑_{i,j} c i · conj (c j) · F (v i + star (v j)) is a nonnegative real number (using the order on
ℂ for which 0 ≤ z means z is real and nonnegative).
Equations
Instances For
Positive-definiteness holds for an arbitrary finite index type, not just Fin n: for every
finite family of scalars c : ι → ℂ and points v : ι → M, the Hermitian form
∑_{i,j} c i · conj (c j) · F (v i + star (v j)) is a nonnegative real number.
The 2 × 2 Hermitian sub-form of a positive-definite function at the points a, b with
coefficients c₀, c₁ is nonnegative.
A positive-definite function takes a real, nonnegative value at every "norm point"
a + star a.
The value of a positive-definite function at a "norm point" a + star a has zero imaginary
part.
The value of a positive-definite function at a "norm point" a + star a has nonnegative real
part.
The value of a positive-definite function at 0 is real and nonnegative.
The value of a positive-definite function at 0 has zero imaginary part.
The real part of the value of a positive-definite function at 0 is nonnegative.
The value at the origin of a positive-definite function is equal to the real number
(F 0).re, viewed as a complex number.
If a positive-definite function is nonzero at the origin, then the real part of that value is strictly positive.
A positive-definite function is conjugate symmetric in the involution:
conj (F (b + star a)) = F (a + star b).
A positive-definite function F induces the positive-definite kernel K(a, b) = F(a + b⋆).
This is the forward half of the function ↔ kernel correspondence.
If the kernel K(a, b) = F(a + b⋆) is positive definite, then so is the function F. This is
the reverse half of the function ↔ kernel correspondence.
The Cauchy–Schwarz inequality for a positive-definite function: the squared norm of an off-diagonal value is bounded by the product of the two diagonal values.
If a + star a = 0, then a positive-definite function is bounded at a by its value at
zero.
A positive-definite function with (F 0).re = 0 vanishes identically.
No hypothesis on the point is required, and none on the ambient structure beyond an
involutive additive monoid (AddMonoid and StarAddMonoid).
If the involution negates the point a, then a positive-definite function is bounded at a
by its value at zero. In particular this applies on an additive group whose involution is
negation.
Positive-definite functions are closed under addition.
Positive-definite functions are closed under multiplication by a nonnegative complex scalar.
Positive-definite functions are closed under pointwise multiplication (Schur product).
A nonnegative real constant is a positive-definite function.
The zero function is positive definite.
Positive-definite functions are closed under finite sums.
Positive-definite functions are closed under finite products (Schur product).