Operators of order two and three: Choie–Zagier criterion, Popa–Zagier's projection and exactness #
The group PSL(2, ℤ) is generated by S of order two and U of order three, and
T = U S. Popa and Zagier's proof of the Eichler–Selberg trace formula works in the rational
vector space ℛ = ℚ[ℳ] spanned by integral matrices of positive determinant modulo ±1, on which
PSL(2, ℤ) acts on both sides. The idempotents π_S = (1 + S) / 2 and π_U = (1 + U + U²) / 3
generate the right ideal
𝓘 = π_S ℛ + π_U ℛ
(their (5)). This file records the linear algebra behind their Lemmas 1 and 3 and the exactness
step in the proof of their §3 Theorem 2, for endomorphisms σ and υ of a module M with
σ ^ 2 = 1 and υ ^ 3 = 1, which are the left multiplications by S and U; the right
multiplications by S and U give a second such pair σ', υ', commuting with the first.
The ideal 𝓘 becomes the submodule range (1 + σ) ⊔ range (1 + υ + υ ^ 2): when 2
and 3 are invertible the unnormalised operators have the same ranges as the idempotents, and the
unnormalised form needs no invertibility where none is used.
Main results #
TauCeti.End.one_sub_apply_mem_range_one_sub_mul_iff: the Choie–Zagier criterion (Popa–Zagier, Lemma 1): if2and3are invertible andker (1 + σ)andker (1 + υ + υ ^ 2)are disjoint, thenx ∈ range (1 + σ) ⊔ range (1 + υ + υ ^ 2)exactly when(1 - σ) x ∈ range (1 - υ * σ), that is,ξ ∈ 𝓘 ↔ (1 - S) ξ ∈ (1 - T) ℛ. The forward direction,TauCeti.End.one_sub_apply_mem_range_one_sub_mul_of_mem_sup, needs neither hypothesis.TauCeti.End.exists_mem_sup_one_add_apply_sub_mem: the existence half of Popa–Zagier's Lemma 3 for the idempotents(1 + σ) / 2and(1 + υ + υ ^ 2) / 3, stated with the unnormalised operators: if(1 + σ) ξand(1 + υ + υ ^ 2) ξlie inP ⊔ Q, for submodulesPandQstable under1 + υ + υ ^ 2and1 + σ, thenξcan be corrected by an elementιofP ⊔ Qso that(1 + σ) (ξ - ι) ∈ Qand(1 + υ + υ ^ 2) (ξ - ι) ∈ P.TauCeti.End.mem_sup_range_one_add_mul_one_sub: exactness in Popa–Zagier's §3 Theorem 2: if2and3are invertible andker (1 + σ'),ker (1 + υ' + υ' ^ 2)are disjoint, then a vectorξ ∈ range (1 - σ') ⊔ range (1 - υ')with(1 + σ') ξ ∈ range (1 + υ + υ ^ 2)and(1 + υ' + υ' ^ 2) ξ ∈ range (1 + σ)lies inrange ((1 + σ) (1 - σ')) ⊔ range ((1 + υ + υ ^ 2) (1 - υ')), that is, an element of theirℬ(6) whose right coset sums vanish lies in𝒥 = π_S ℛ (1 - π_S) + π_U ℛ (1 - π_U)(7).TauCeti.one_sub_mul_one_add_add_sq: the ring identity(1 - σ) (1 + υ + υ ^ 2) = (1 - υ σ) (-σ (1 + υ + υ ^ 2))forσ ^ 2 = 1andυ ^ 3 = 1, and its mirror imageTauCeti.one_add_add_sq_mul_one_sub, for right multiplications.TauCeti.one_sub_mul_one_add_of_sq_eq_one,TauCeti.one_add_mul_one_sub_of_sq_eq_one,TauCeti.one_sub_mul_one_add_add_sq_of_pow_three_eq_one,TauCeti.one_add_add_sq_mul_one_sub_of_pow_three_eq_one:(1 ∓ σ) (1 ± σ) = 0forσ ^ 2 = 1, and(1 - υ) (1 + υ + υ ^ 2) = (1 + υ + υ ^ 2) (1 - υ) = 0forυ ^ 3 = 1.TauCeti.End.range_one_add_eq_ker_one_sub,TauCeti.End.range_one_add_add_sq_eq_ker_one_sub: the fixed vectors ofσand ofυare the ranges of1 + σand of1 + υ + υ ^ 2. The inclusionsTauCeti.End.range_one_add_le_ker_one_subandTauCeti.End.range_one_add_add_sq_le_ker_one_subneed no invertibility, and the reverse inclusions hold inside any submoduleYwithout the order hypotheses:TauCeti.End.inf_ker_one_sub_le_map_one_addandTauCeti.End.inf_ker_one_sub_le_map_one_add_add_sq(Y ⊓ ker (1 - σ) ≤ Y.map (1 + σ)when2is invertible, and likewise with3).TauCeti.End.range_one_sub_le_ker_one_add,TauCeti.End.range_one_sub_le_ker_one_add_add_sq: dually,1 + σkillsrange (1 - σ)and1 + υ + υ ^ 2killsrange (1 - υ).
Implementation notes #
The ring identities are proved in an arbitrary ring and then applied in Module.End; the
acyclicity hypothesis is stated as Disjoint (ker (1 + σ)) (ker (1 + υ + υ ^ 2)). The existence
half of Lemma 3 for arbitrary idempotents is
TauCeti.End.exists_mem_sup_apply_sub_mem_of_isIdempotentElem, and the exactness step for
arbitrary commuting endomorphisms is TauCeti.End.mem_inf_ker_sup_inf_ker_iff.
References #
- A. Popa and D. Zagier, An elementary proof of the Eichler–Selberg trace formula, J. Reine Angew. Math. 762 (2020), 105–122, arXiv:1711.00327, Section 3, Lemmas 1 and 3 and the proof of Theorem 2.
- Y. J. Choie and D. Zagier, Rational period functions for
PSL(2, ℤ), Contemp. Math. 143 (1993), 89–108.
(1 - σ) (1 + υ + υ²) lies in (1 - υ σ) A when σ ^ 2 = 1 and υ ^ 3 = 1.
This is Popa–Zagier's relation (1 - S) π_U = (1 - T⁻¹) π_U for T = U S, rewritten through
1 - T⁻¹ = (1 - T) (-T⁻¹), and it is the forward direction of the Choie–Zagier criterion.
(1 + υ + υ²) (1 - σ) = -((1 + υ + υ²) σ) (1 - σ υ) for σ ^ 2 = 1 and υ ^ 3 = 1: the
mirror image of TauCeti.one_sub_mul_one_add_add_sq, for right multiplications.
For σ ^ 2 = 1, the vectors (1 + σ) x are fixed by σ.
For σ ^ 2 = 1, the vectors (1 - σ) x are killed by 1 + σ.
When 2 is invertible, a vector y ∈ Y fixed by σ is (1 + σ) (y / 2).
The fixed vectors of an involution are the range of 1 + σ, when 2 is invertible.
When 3 is invertible, a vector y ∈ Y fixed by υ is (1 + υ + υ²) (y / 3).
The fixed vectors of an endomorphism υ with υ ^ 3 = 1 are the range of
1 + υ + υ², when 3 is invertible.
The Choie–Zagier criterion, forward direction (Popa–Zagier, Lemma 1, ⇒): for
σ ^ 2 = 1 and υ ^ 3 = 1, if x ∈ range (1 + σ) ⊔ range (1 + υ + υ²) then
(1 - σ) x ∈ range (1 - υ σ). No invertibility is needed.
The ranges of 1 - σ and 1 - υ are disjoint when σ ^ 2 = 1, υ ^ 3 = 1 and
ker (1 + σ), ker (1 + υ + υ²) are disjoint.
The Choie–Zagier criterion (Popa–Zagier, Lemma 1): for σ ^ 2 = 1 and υ ^ 3 = 1 with
2 and 3 invertible and ker (1 + σ), ker (1 + υ + υ²) disjoint, a vector x lies in
range (1 + σ) ⊔ range (1 + υ + υ²) exactly when (1 - σ) x ∈ range (1 - υ σ).
Exactness in Popa–Zagier's §3 Theorem 2: for σ ^ 2 = σ' ^ 2 = 1 and υ ^ 3 = υ' ^ 3 = 1,
with σ and υ commuting with σ' and υ', 2 and 3 invertible and ker (1 + σ'),
ker (1 + υ' + υ'²) disjoint: if ξ ∈ range (1 - σ') ⊔ range (1 - υ'),
(1 + σ') ξ ∈ range (1 + υ + υ²) and (1 + υ' + υ'²) ξ ∈ range (1 + σ), then
ξ ∈ range ((1 + σ) (1 - σ')) ⊔ range ((1 + υ + υ²) (1 - υ')).
For the left multiplications σ, υ and the right multiplications σ', υ' by S, U on ℛ, the
hypotheses on ξ say that ξ ∈ ℛ (1 - S) + ℛ (1 - U), the elements whose right coset sums vanish,
and that ξ lies in their ℬ = {ξ | ξ π_S ∈ π_U ℛ, ξ π_U ∈ π_S ℛ} (6); the conclusion is
ξ ∈ 𝒥 = π_S ℛ (1 - π_S) + π_U ℛ (1 - π_U) (7).
Popa–Zagier's projection for operators of order two and three (Lemma 3, existence half):
for σ ^ 2 = 1 and υ ^ 3 = 1 with 2 and 3 invertible, P stable under 1 + υ + υ² and Q
under 1 + σ, if (1 + σ) ξ and (1 + υ + υ²) ξ lie in P ⊔ Q, then there is ι ∈ P ⊔ Q with
(1 + σ) (ξ - ι) ∈ Q and (1 + υ + υ²) (ξ - ι) ∈ P. This is
TauCeti.End.exists_mem_sup_apply_sub_mem_of_isIdempotentElem for the idempotents (1 + σ) / 2 and
(1 + υ + υ²) / 3; the ring R need not be commutative, since 2 and 3 are central.