The Weyl operators of Schwartz kernels form a non-unital algebra #
Category: 1-Mathlib (CSD-free; staged as a Mathlib-upstream candidate). BACKLOG #127, out of #126.
#126 proved Op(K₁) ∘ Op(K₂) = Op(K₁ ⋆ K₂) and claimed nothing about the structure. This is the
structure, and the keystone is a theorem the row did not have:
- ★★★
weylOpK_injective— the kernel is determined by the operator.Opis faithful on the Schwartz kernel class, so a statement about operators can be pulled back to a statement about kernels; - ★★★
weylCompKernel_assoc—⋆is associative, and ★★weylCompKernel_add_left/_right,_smul_left/_right— it is bilinear. All four are corollaries of faithfulness and the corresponding fact about operators, which is where the work went; - ★★★
not_weylOpK_eq_idand ★★★weylCLM_ne_one— there is no unit. No Schwartz kernel represents the identity operator, so this is a genuinely non-unital algebra.
The correction this row records #
#127's own entry said associativity would follow from "associativity of operator composition and
the injectivity of weylKernelCLM". That route does not close. weylKernelCLM is the change of
coordinates (x, y) ↦ ((x+y)/2, x − y), and its injectivity says nothing about whether two
different kernels can give the same operator — which is exactly what is needed to turn an identity
between operators into an identity between kernels. The ingredient is faithfulness of Op, and that
is a theorem rather than bookkeeping: ★★★ weylOpK_injective is proved by feeding the operator the
conjugate of its own kernel slice (SchwartzMap.conjCLM of #121(i)'s slice), which turns the
pairing into ∫ ‖k(x, y)‖² dy; that vanishes only if the slice vanishes almost everywhere, and a
continuous function vanishing almost everywhere for Lebesgue measure vanishes.
Honest scope #
⚠️ No Mul instance, deliberately. Declaring ⋆ as the Mul of 𝓢(ℝ × ℝ, ℂ) would commit the
type globally to the Weyl product, when the pointwise product is at least as natural a choice and a
NonUnitalAlgebra instance would then fix which one every downstream file means. The facts are
stated as theorems; a bundled NonUnitalAlgHom is one letI away for a consumer who wants it, and
nothing in the corpus does.
⚠️ Nothing about the symbol product. Faithfulness here is of K ↦ Op(K) on kernels; that the
symbol composes by the Moyal star product still needs #121(ii), and the ℏ² expansion is #64.
⚠️ The non-unitality is stated for this class only. That no Schwartz kernel gives the identity is proved; a wider class (distributional kernels, where the identity's kernel is a delta) is a different setting and is not formalised.
References: WeylComposition.lean (#126, weylCompKernel, weylCLM_comp),
SchwartzSlice.lean (#121(i), slice),
WeylSmooth.lean (#122, weylCLM); specs/BACKLOG.md #127, #126, #122, #64.
Conjugation on Schwartz space #
Needed to feed the operator the conjugate of its own kernel slice. Conjugation is ℝ-linear and not
ℂ-linear, which is why this is a map over ℝ; only its values matter below.
Complex conjugation on Schwartz space, as a continuous ℝ-linear map.
Equations
Instances For
The Weyl operator is linear in its kernel #
Faithfulness: the kernel is determined by the operator #
The change of coordinates, in the form the injectivity argument uses.
The change of coordinates is injective, because it is composition with a surjection.
The squared modulus of a Schwartz function is integrable, which is what the pairing below produces.
★★★ The kernel is determined by the operator. A kernel whose operator annihilates every
state is zero — fed the conjugate of its own slice, the operator returns ∫ ‖k(x, y)‖² dy, which
vanishes only if the slice does. This is what #127's entry was missing, and it is a theorem rather
than bookkeeping.
★★★ Op is faithful on the Schwartz kernel class.
★★ The same, for the packaged operators of #122.
The algebra laws #
★★★ ⋆ is associative. Operator composition is associative and Op is faithful, so the
identity descends to the kernels.
★★ ⋆ is additive in its left argument.
★★ ⋆ is additive in its right argument. The right slot is the inner operator, so this one
goes through the operators' action rather than through linearity in the kernel alone.
★★ ⋆ is homogeneous in its left argument.
★★ ⋆ is homogeneous in its right argument.
There is no unit #
The identity operator is not the Weyl operator of any Schwartz kernel. A unit would have to reproduce
every state's value at every point from an integral against a bounded kernel, and a shrinking bump
has value 1 at the centre while its integral against anything integrable tends to 0.
A bump at 0 of outer radius 2/(n+1), as a complex Schwartz function: value 1 at 0 for
every n, supported in a ball that shrinks to nothing.
Equations
Instances For
★★★ No Schwartz kernel represents the identity, so the algebra of #126 is genuinely
non-unital. The witness is the shrinking bump: it keeps the value 1 at the origin while its
pairing with the kernel's slice tends to 0.
★★★ The same, for the packaged operator: no K has weylCLM K = 1.