Source
Baez (2020), LIT-819, read in NOTE-630: Theorems 3, 4, 8 and 10 with their proofs, the remark on bilinearity after Theorem 3, and Theorem 12 (Alfsen and Shultz, quoted).
What was actually shown
Let a, b each generate a flow: the equation d/dt F_t^a(b) = {a, F_t^a(b)}, F_0^a(b) = b, has a unique solution. Then F_t^a(b) = b for all t iff {a, F_t^a(b)} = 0 for all t iff {a, b} = 0, the last step because the constant path solves the equation when {a, b} = 0 and solutions are unique. Antisymmetry turns {a, b} = 0 into {b, a} = 0, and the chain runs back with a and b exchanged. This is proved for Poisson algebras (Theorem 3), for self-adjoint elements of complex *-algebras with the bracket i[a, b] (Theorem 4), for a bare vector space with a bilinear bracket in which each element conserves itself (Theorem 8), and for Banach–Lie algebras, where flows always exist and are unique (Theorem 10). For a bilinear bracket, antisymmetry is equivalent to {a, a} = 0, the self-conservation principle.
Where observables and generators are different kinds of thing (a Jordan algebra O and a Lie algebra L), the equivalence needs a map ψ : O → L. In complex quantum mechanics ψ(a) = ia. For real and quaternionic matrices dim O ≠ dim L and no invariant map exists. For unital JB-algebras, Baez quotes Alfsen and Shultz (his Theorem 12): such a map with ψ_a(a) = 0 and a second condition exists exactly when O is the self-adjoint part of a C*-algebra. The record has not read that theorem, so it is reported as quoted, not checked (QUESTION-142).
What could have come out otherwise: Theorem 8 could have needed the Jacobi identity or an associative product; it does not. The proofs are short and were checked in the reading.
What this does not say
- It does not say symmetries give conservation laws in general. The equivalence needs reversible one-parameter groups and an antisymmetric bracket between the two quantities. For Markov semigroups, a diagonal observable and the generator have no such bracket, and a conserved mean does not give a symmetry (THEORY-158). Reading this THEORY as “Noether’s theorem holds for any dynamics with a generator” is the error it invites.
- It does not cover the Lagrangian form of Noether’s theorem or field theories; the link back is cited, not shown.
- It does not include Alfsen and Shultz’s theorem. The JB-algebra characterisation is quoted by Baez without proof, and the record has not read it. It is QUESTION-142.
- It does not settle why observables should be generators. Alfsen and Shultz’s theorem says what follows if they are; Baez’s reading of the second condition as “inverse temperature is imaginary time” is an interpretation he calls incomplete.
- Existence and uniqueness of flows is assumed, not proved, outside bounded settings (compact Poisson manifolds, C*- and Banach–Lie algebras).
Note of 2026-10-19: the bracket equivalence, re-derived
The promotion audit’s batch 008 re-derived the equivalence (ADR-047 v2). Assume that, for each a, the flow equation d/dt F_t(b) = {a, F_t(b)}, F_0(b) = b, has a unique solution.
- If F_t^a(b) = b for all t, differentiating at t = 0 gives {a, b} = 0.
- If {a, b} = 0, the constant path c(t) = b satisfies c′(t) = 0 = {a, b} = {a, c(t)} and c(0) = b. By uniqueness, F_t^a(b) = b for all t. So “a generates symmetries of b”, meaning F_t^a(b) = b for all t, holds iff {a, b} = 0.
- Antisymmetry gives {a, b} = 0 iff {b, a} = 0, so the relation is symmetric.
- For a bilinear bracket over a field of characteristic other than 2, antisymmetry is equivalent to {a, a} = 0 for all a. Antisymmetry gives {a, a} = −{a, a}. Conversely, expanding {a + b, a + b} = 0 gives {a, b} + {b, a} = 0.
The Jacobi identity and any associative product are used nowhere, as Theorem 8 has it.