Lemma 7.3.2 — Hamiltonians in the commutant generate a Lie algebra acting only on the degeneracies (corrected: , not a direct sum)
ProvedBaconThesis.Commutant.lemma_7_3_2Let be an OSR (or SME) algebra in decomposed form, with unitary and , let with irrep projector , and let be a set of Hamiltonians (Hermitian operators) each of which commutes with every element of . Then:
- every acts on information encoded into the degeneracy of the -th irrep without taking it out of that irrep: , there is a Hermitian matrix with , and for every real ;
- the real Lie algebra generated by (Lie brackets and real linear combinations) has the reducible structure
that is, , where , the restriction of to the degeneracy of the -th irrep, is a (perhaps further reducible) Lie algebra of matrices, and is the identity on the -dimensional irrep factor.
The lemma is the Lie-algebraic form of Lemma 7.3.1: whatever a set of commutant Hamiltonians generates under the bracket still acts only on the degeneracies, so encoded information never leaves its irrep.
Formalization Note "Generate" is LieSubalgebra.lieSpan ℝ of with the commutator bracket (§7.3.3 contrasts it with the associative algebra). The commutant is Set.centralizer.
Differences from the source:
- correction: (7.9) L ≅ ⊕J L{n_J} ⊗ I_{d_J}, a direct sum of Lie algebras → every element of L has the form W†(⊕J Y_J ⊗ I{d_J})W, i.e. L ⊆ ⊕J L{n_J} ⊗ I_{d_J} with L_{n_J} the restriction of L to block J; reason: two irreps with n_J = d_J = 1 (A the diagonal 2 × 2 matrices) and S = {diag(1, 2)} give L = ℝ·i·diag(1, 2), which is not the direct sum ℝi ⊕ ℝi of its block restrictions (i·diag(1, 0) is in the sum but not in L)
- reading: "will act on this encoded information without taking this encoded information out of the Kth irrep" → for every H ∈ S: H P_K = P_K H, H P_K = W†(0 ⊕ (Y ⊗ I_{d_K}) ⊕ 0)W with Y Hermitian, and e^{−iHt} P_K = W†(0 ⊕ (e^{−iYt} ⊗ I_{d_K}) ⊕ 0)W for all real t; reason: the block form (7.2) that Lemma 7.3.1's proof invokes, for the operator and for its evolution
- encoding: "S will generate a Lie algebra L" → the real Lie span of {iH : H ∈ S} with the matrix commutator; reason: Hamiltonians H generate unitaries e^{−iHt}; §7.3.3 specifies Lie bracket and linear combination
- encoding: A ≅ ⊕J I{n_J} ⊗ M_{d_J} → A = {W† (⊕J I{n_J} ⊗ X_J) W} for an explicit unitary W; reason: the basis change of Theorem 5.4.1 made explicit; same instances
- encoding: irreps J ∈ 𝒥 of Theorem 5.4.1 (each irrep present, with a degeneracy n_J and a dimension d_J) → hypotheses ∀ J, 0 < n J and ∀ J, 0 < d J; reason: the book's irreps and degeneracies are nonzero; an empty block would contribute nothing to any operator, so no instance changes
import Mathlib import Definitions.Def_BaconThesis_Commutant_blockC import Definitions.Def_BaconThesis_Commutant_algebraA import Definitions.Def_BaconThesis_Commutant_irrepProj import Definitions.Def_BaconThesis_Commutant_lieGen
attribute [local instance 100] LieRing.ofAssociativeRing
namespace BaconThesis.Commutant
/-- **Lemma 7.3.2 (Hamiltonians in the commutant generate a Lie algebra acting only on the
degeneracies; corrected (7.9)).**
Bacon, *Decoherence, Control, and Symmetry in Quantum Computers*, arXiv:quant-ph/0305025v1,
§7.3.3, Lemma 7.3.2, p.103 (PDF p.120): "Suppose one is given an OSR or SME algebra
`A ≅ ⊕_{J∈𝒥} I_{n_J} ⊗ M_{d_J}` and information has been encoded into the degeneracy of the `K`th
irrep. A set of Hamiltonians `S` each of which commutes with all of the elements of `A` will act on
this encoded information without taking this encoded information out of the `K`th irrep.
Furthermore this set of Hamiltonians `S` will generate a Lie algebra `L` which has a reducible
structure of the form `L ≅ ⊕_{J∈𝒥} L_{n_J} ⊗ I_{d_J}`, (7.9) where `L_{n_J}` is a (perhaps further
reducible) `n_J` dimensional Lie algebra and `I_{d_J}` represents the identity action on a `d_J`
dimensional space."
Let `A = {W† (⊕_J I_{n_J} ⊗ X_J) W}` as in `algebraA` (`W` unitary, `n_J, d_J ≥ 1`), let `K` be an
irrep label with projector `P_K`, and let `S` be a set of Hermitian operators each commuting with
every element of `A`. Then
1. every `H ∈ S` satisfies `H P_K = P_K H`, acts on the `K`-th irrep as `Y_H ⊗ I_{d_K}` for a
Hermitian `Y_H` on the degeneracy, and its evolution satisfies
`e^{−iHt} P_K = W† (0 ⊕ ⋯ ⊕ (e^{−iY_H t} ⊗ I_{d_K}) ⊕ ⋯ ⊕ 0) W` for every real `t`;
2. every element `X` of the real Lie algebra `L` generated by `{iH : H ∈ S}` (`lieGen S`) has the
form `X = W† (⊕_J Y_J ⊗ I_{d_J}) W` for some `n_J × n_J` matrices `Y_J`; that is,
`L ⊆ ⊕_J L_{n_J} ⊗ I_{d_J}`, where `L_{n_J} = {Y_J : X ∈ L}` (the restriction of `L` to the
degeneracy of the `J`-th irrep) is a real Lie algebra of `n_J × n_J` matrices.
**Formalization Note.** Correction of (7.9): read as a direct sum of Lie algebras, (7.9) is false
(two irreps with `n_J = d_J = 1`, `S = {diag(1, 2)}`: `L = ℝ · i diag(1, 2)` is not
`ℝi ⊕ ℝi`); clause 2 is the reducible block structure the lemma asserts. "Generate" is the real
Lie span with the matrix commutator (§7.3.3), not the associative algebra. -/
theorem lemma_7_3_2 {𝒥 ι : Type} [Fintype 𝒥] [DecidableEq 𝒥] [Fintype ι] [DecidableEq ι]
(n d : 𝒥 → ℕ) (hn : ∀ J, 0 < n J) (hd : ∀ J, 0 < d J)
(W : Matrix (Σ J : 𝒥, Fin (n J) × Fin (d J)) ι ℂ)
(hW₁ : W.conjTranspose * W = 1) (hW₂ : W * W.conjTranspose = 1)
(K : 𝒥) (S : Set (Matrix ι ι ℂ))
(hS : ∀ H ∈ S, H.IsHermitian ∧ H ∈ Set.centralizer (algebraA n d W)) :
(∀ H ∈ S, H * irrepProj n d W K = irrepProj n d W K * H ∧
∃ Y : Matrix (Fin (n K)) (Fin (n K)) ℂ, Y.IsHermitian ∧
H * irrepProj n d W K =
W.conjTranspose *
blockC n d (Pi.single (M := fun J => Matrix (Fin (n J)) (Fin (n J)) ℂ) K Y) * W ∧
∀ t : ℝ, NormedSpace.exp ((-(t : ℂ) * Complex.I) • H) * irrepProj n d W K =
W.conjTranspose *
blockC n d (Pi.single (M := fun J => Matrix (Fin (n J)) (Fin (n J)) ℂ) K
(NormedSpace.exp ((-(t : ℂ) * Complex.I) • Y))) * W) ∧
∀ X ∈ lieGen S, ∃ Y : (J : 𝒥) → Matrix (Fin (n J)) (Fin (n J)) ℂ,
X = W.conjTranspose * blockC n d Y * W := by sorry
end BaconThesis.Commutant
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.