Differences of bounded positive-definite functions #
On an involutive commutative additive monoid M, a bounded positive-definite function F
dominates each of its translates by a self-adjoint element: if star s = s, then
x ↦ F x - F (x + s)
is again positive definite. This is the function-level form of
TauCeti.posSemidef_sub_comp_shift; self-adjointness of s is exactly what makes translation by
s a symmetric shift of the kernel (a, b) ↦ F (a + star b).
That difference is the negative of Mathlib's forward difference Δ_[s] F, so iterating gives that
the alternating iterated differences (-1) ^ n Δ_[s]^[n] F are positive definite too, in operator
form and in the explicit binomial form
x ↦ ∑ k ≤ n, (-1) ^ k (n choose k) F (x + k • s).
Positive definiteness is a statement about quadratic forms, not a pointwise sign; the values
themselves are nonnegative at the "norm points" a + star a.
Boundedness is only ever assumed at the "norm points" a + star a, the diagonal of the kernel:
Cauchy--Schwarz propagates a bound there to every point of the form p + star q. It is essential
and is not a technical artefact: t ↦ exp t is positive definite on the involutive monoid
(ℝ≥0, +) with the trivial involution, and it increases. The statement is
proved by a purely numerical moment-problem estimate on the finite quadratic forms; no topology,
measurability, or representing measure is assumed or produced here.
This advances TauCetiRoadmap/OneParameterSemigroups/README.md, Part C: it is the closure property
of the generic IsPositiveDefinite predicate that the Berg--Christensen--Ressel representation
(Milestone 2) runs on. The zero-spatial time axis t ↦ F (t, 0) of a bounded positive-definite
function on ℝ≥0 × V is such a function, for the trivial involution on ℝ≥0, and
TauCeti/Analysis/PositiveDefinite/SemigroupGroup/Time/Difference.lean reads its complete
monotonicity off the results below. The two-variable statements on ℝ≥0 × V itself are not
instances of them — the involutive-monoid wrapper carrying the Berg--Christensen--Ressel involution
is private to TauCeti/Analysis/PositiveDefinite/SemigroupGroup/Basic.lean — and are proved there
from the Kolmogorov decomposition of the Berg--Christensen--Ressel kernel; only the
forward-difference bookkeeping below is shared.
Main declarations #
TauCeti.neg_one_pow_mul_fwdDiff_iter_succ: one differencing step advances the alternating iterated forward difference by one.TauCeti.neg_one_pow_mul_fwdDiff_iter_eq_alternating_sum: the binomial expansion of the alternating iterated forward difference.TauCeti.IsPositiveDefinite.norm_apply_add_star_le: a bound at the "norm points"a + star abounds a positive-definite function at every pointp + star q.TauCeti.IsPositiveDefinite.sub_shift: the difference of a bounded positive-definite function and its translate by a self-adjoint element is positive definite.TauCeti.IsPositiveDefinite.neg_one_pow_mul_fwdDiff_iterandTauCeti.IsPositiveDefinite.alternating_sum: the alternating iterated differences are positive definite, in forward-difference and in binomial form.TauCeti.IsPositiveDefinite.sub_shift_add_star_self_nonneg,TauCeti.IsPositiveDefinite.neg_one_pow_mul_fwdDiff_iter_add_star_self_nonnegandTauCeti.IsPositiveDefinite.alternating_sum_add_star_self_nonneg: the resulting nonnegativity at the "norm points"a + star a.
References #
- C. Berg, J. P. R. Christensen, P. Ressel, Harmonic Analysis on Semigroups (GTM 100, 1984), Chapter 4.
Forward-difference bookkeeping #
Pure identities about Mathlib's forward difference operator; no positive definiteness is involved.
Differencing once advances the alternating iterated forward difference by one step. This is the bookkeeping behind every induction on the number of differencing steps.
The alternating iterated forward difference as an explicit alternating binomial sum.
Differences of bounded positive-definite functions #
A positive-definite function bounded at the "norm points" a + star a is bounded at every
point of the form p + star q, by Cauchy--Schwarz for its kernel.
A bounded positive-definite function dominates its translate by a self-adjoint element.
Translation by s with star s = s is a symmetric shift of the kernel (a, b) ↦ F (a + star b),
so the bounded-kernel estimate TauCeti.posSemidef_sub_comp_shift applies; only the diagonal of
that kernel, the values at the "norm points" a + star a, has to be bounded. Boundedness cannot be
dropped: t ↦ exp t is positive definite on ℝ≥0 with the trivial involution and increases.
The difference of a bounded positive-definite function and its translate by a self-adjoint
element is nonnegative at every "norm point" a + star a.
The alternating iterated differences of a bounded positive-definite function are positive
definite. This is the iterate of IsPositiveDefinite.sub_shift: each differencing step doubles
the admissible bound, which is harmless because only the existence of some bound is used.
The alternating iterated difference of a bounded positive-definite function is nonnegative at
every "norm point" a + star a.
The alternating iterated differences, expanded as binomial sums, are positive definite.
This is IsPositiveDefinite.neg_one_pow_mul_fwdDiff_iter with the forward-difference operator
resolved into the explicit alternating sum over an arithmetic progression of translates.
The alternating binomial sums of a bounded positive-definite function are nonnegative at every
"norm point" a + star a: complete monotonicity in the finite-difference sense, along the
arithmetic progression starting there.