Adams operators and the first Eulerian idempotent on the planar algebra #
In the convolution ring WithConv (H k →ₗ[k] H k) the identity is a unit with inverse the
antipode; its convolution powers are the Adams operators Ψₙ = id^{⋆n}. The augmentation
J = id − u∘ε is locally nilpotent, so over a ℚ-algebra the convolution logarithm
e⁽¹⁾ = log_⋆ id is a finite sum on each forest.
Main results #
ConnesKreimer.adams,ConnesKreimer.toConv_adams_add,ConnesKreimer.adamsUnit,ConnesKreimer.adamsUnit_inv_val: Adams operators,Ψₘ ⋆ Ψₙ = Ψₘ₊ₙ, andΨ₋₁ = S.ConnesKreimer.Jc_pow_ofConv_sf_eq_zero:J^{⋆n}vanishes on forests with fewer thannnodes.ConnesKreimer.adams_primitive:Ψₙ x = n • xfor a primitivex.ConnesKreimer.eulerian1,ConnesKreimer.eulerian1_eq_logTrunc: the first Eulerian idempotent and its truncation formula.ConnesKreimer.eulerian1_primitive:e⁽¹⁾fixes every primitive element.ConnesKreimer.eulerian1_idem_dot:e⁽¹⁾ (e⁽¹⁾ •) = e⁽¹⁾ •on the one-node tree.
References #
- M. Aguiar, A. Lauve, The characteristic polynomial of the Adams operators on graded connected Hopf algebras, Algebra Number Theory 9 (2015) 547–583.
- J.-L. Loday, Opérations sur l'homologie cyclique des algèbres commutatives, Invent. Math. 96 (1989) 205–230.
- C. Marín, How a Tree Forgets Its Order: the Eulerian idempotent and the Adams spectrum on the
commutative Connes–Kreimer–Butcher Hopf algebra of rooted trees, machine-checked in Lean 4
(2026), doi:10.5281/zenodo.20774821. Ported from
ConnesKreimer.leanof karlesmarin/connes-kreimer-lean (Lean v4.30.0-rc2).
Adams operators Ψₙ = id^{⋆n} and the antipode as Ψ₋₁. #
The convolution ring `WithConv (H k →ₗ[k] H k)` (Mathlib `LinearMap.convRing`) turns the
identity into a ring element whose convolution powers are the Adams (Hopf-power) operators.
`Ψ₀ = u∘ε` (the convolution unit), `Ψ₁ = id`, the semigroup law `Ψₘ ⋆ Ψₙ = Ψₘ₊ₙ`, and the
identity is a convolution unit whose inverse is the antipode — i.e. `Ψ₋₁ = S`
(Aguiar–Lauve theory; the Eulerian idempotent `e⁽¹⁾ = log_⋆ id` is the next step, needing
connected-graded nilpotence).
The n-th Adams operator Ψₙ = id^{⋆n}: the n-th convolution power of the identity in the
convolution ring WithConv (H k →ₗ[k] H k).
Equations
Instances For
Ψ₀ = u∘ε is the convolution unit.
Ψ₁ = id.
The convolution-power semigroup law Ψₘ ⋆ Ψₙ = Ψₘ₊ₙ.
S ⋆ id = u∘ε: the antipode is a left convolution inverse of Ψ₁ = id.
id ⋆ S = u∘ε: the antipode is also a right convolution inverse of Ψ₁ = id.
Ψ₁ = id is a convolution UNIT, with the antipode as its (two-sided) inverse. This is the
precise sense in which Ψ₋₁ = S.
Equations
- ConnesKreimer.adamsUnit k = { val := WithConv.toConv LinearMap.id, inv := WithConv.toConv (ConnesKreimer.antipode k), val_inv := ⋯, inv_val := ⋯ }
Instances For
Ψ₋₁ = S: the inverse of the Adams unit Ψ₁ is the antipode.
Local nilpotency of J = id − u∘ε (the connected-graded engine for log_⋆). #
`J` kills the empty forest and is the identity on nonempty ones. Its convolution powers
`J^{⋆k}` vanish on every basis forest of fewer than `k` nodes (`Jc_pow_ofConv_sf_eq_zero`):
each of the `k` convolution factors must absorb at least one node, so `k` nodes are needed.
Hence the series `log_⋆(id) = Σ_{k≥1} ((-1)^{k+1}/k)·J^{⋆k}` is finite in each degree — the
connected-graded local nilpotency on which the Aguiar–Lauve Eulerian spectrum rests.
comul on a basis forest equals the combinatorial Δ₀.
Convolution-on-basis engine. (F ⋆ G) on a basis forest is the sum over cuts:
(F ⋆ G)(f) = Σ_{(p,r) ∈ coprodForest f} F(p)·G(r).
The augmentation projector J = id − u∘ε, as an element of the convolution ring.
Equations
Instances For
The convolution unit 1 = u∘ε on a basis forest: ε(f)·1.
Local nilpotency. J^{⋆n} = (id − u∘ε)^{⋆n} vanishes on every basis forest with fewer
than n nodes. Proof: induction on n via the convolution engine and degree conservation;
in a nonzero cut the left leg carries ≥1 node, so the right leg has < n−1 nodes and the
induction hypothesis kills it, while an empty left leg is killed by J directly.
The Adams eigenvalue on primitives: Ψₙ(x) = n·x for primitive x. #
The first piece of the Aguiar–Lauve spectrum, valid on ANY connected graded bialgebra (no
commutativity needed): the primitives are the eigenspace of every Adams operator `Ψₙ` with
eigenvalue `n`. Proof is at the element level (not the basis): on a primitive `x`, the recursion
`Ψₙ₊₁ = id ⋆ Ψₙ` and `Δx = x⊗1 + 1⊗x` give `Ψₙ₊₁(x) = x + Ψₙ(x)`, and `Ψₙ` fixes the unit.
A primitive element: Δx = x⊗1 + 1⊗x.
Equations
- ConnesKreimer.IsPrimitive k x = ((ConnesKreimer.Δ k) x = x ⊗ₜ[k] 1 + 1 ⊗ₜ[k] x)
Instances For
Convolution at the element level: (F ⋆ G)(x) = μ((F⊗G)(Δx)).
Ψₙ₊₁ = id ⋆ Ψₙ evaluated: Ψₙ₊₁(x) = μ((id⊗Ψₙ)(Δx)).
Adams eigenvalue on primitives: for primitive x, Ψₙ₊₁(x) = (n+1)·x. The
primitives are the eigenspace of Ψₙ with eigenvalue n, and S = Ψ₋₁ acts as −1,
recovering S(x) = −x.
The first Eulerian idempotent e⁽¹⁾ = log_⋆(id) as a linear map. #
Over a `ℚ`-algebra the series `log_⋆(id) = Σ_{k≥1} ((-1)^{k+1}/k)·J^{⋆k}` is, by the local
nilpotency above, finite on each basis forest (only `k ≤ |f|` contribute). We define `e⁽¹⁾`
by that finite per-degree sum, mirroring how `Δ` and the antipode are defined on the basis.
`e⁽¹⁾` kills the unit and fixes primitives (eigenvalue `1`) — the connected-graded projector
onto primitives that begins the Aguiar–Lauve spectral decomposition.
Coefficient (-1)^{j+1}/j of the convolution-log series, pushed into k. j = 0 ↦ 0.
Equations
- ConnesKreimer.eulerCoef k j = (algebraMap ℚ k) ((-1) ^ (j + 1) / ↑j)
Instances For
e⁽¹⁾ on a basis forest: the finite log series Σ_{j=0}^{|f|} ((-1)^{j+1}/j)·J^{⋆j}(f)
(the j=0 term is 0; terms with j > |f| are absent — and would vanish anyway).
Equations
- ConnesKreimer.eulerian1F k f = (List.map (fun (j : ℕ) => ConnesKreimer.eulerCoef k j • (ConnesKreimer.Jc k ^ j).ofConv (ConnesKreimer.sf k f)) (List.range (ConnesKreimer.sizeF f + 1))).sum
Instances For
Idempotency on the primitive •: e⁽¹⁾(e⁽¹⁾ •) = e⁽¹⁾ •. Immediate from eulerian1_dot
(since e⁽¹⁾ fixes •); likewise e⁽¹⁾ is idempotent on anything it fixes or kills.
NOTE (honest scope): the GENERAL projector identity e⁽¹⁾∘e⁽¹⁾ = e⁽¹⁾ is false on this
planar (Foissy) Hopf algebra — it is neither commutative nor cocommutative, and the Eulerian
idempotents are orthogonal idempotents only in the (co)commutative case (Patras, Loday).
On the cherry, e⁽¹⁾∘e⁽¹⁾ and e⁽¹⁾ differ in the •·ℓ₂ vs ℓ₂·• terms (computed by
evaluation, not proved here) — exactly the planar non-commutativity. Genuine idempotency (the
Hodge/Eulerian decomposition) lives on the commutative quotient
(Formalizations.RingTheory.HopfAlgebra.ConnesKreimer.Eulerian), not here.
e⁽¹⁾ fixes every primitive (element-level primitive facts). #
For primitive `x`: `ε(x)=0`, `J(x)=x`, and `J^{⋆k}(x)=0` for `k ≥ 2` (each higher factor must
absorb the unit, which `J` kills). These feed the operator identity below.
A primitive element has zero counit (forced by the counit axiom (ε⊗id)∘Δ = 1⊗·).
J fixes a primitive: J(x) = x.
e⁽¹⁾ fixes every primitive (the operator identity). #
`e⁽¹⁾` agrees on degree-`≤N` elements with the finite truncation `Σ_{j≤N} c_j J^{⋆j}` (extra
terms vanish by nilpotency); on a primitive only the `j=1` term survives, giving `e⁽¹⁾(x)=x`.
Bridge: a List.range map-sum equals the corresponding Finset.range sum.
The degree-N truncation of log_⋆(id), as a linear map.
Equations
- ConnesKreimer.logTrunc k N = ∑ j ∈ Finset.range (N + 1), ConnesKreimer.eulerCoef k j • (ConnesKreimer.Jc k ^ j).ofConv
Instances For
A linear endomorphism of H is the basis-sum of its values on the singletons (proved by
exact, sidestepping the MonoidAlgebra/Finsupp syntactic wall in rw).
e⁽¹⁾ fixes every primitive: e⁽¹⁾(x) = x for primitive x. Only the j=1 term of the
log series survives (J(x)=x; j=0 has coefficient 0; j≥2 vanish), so e⁽¹⁾(x)=x. The
primitives are exactly the eigenvalue-1 eigenspace of e⁽¹⁾.