# Torus Continuum Limit: OS Axioms Plan ## Goal Prove OS axioms OS0–OS3 for **both** the Gaussian and interacting P(φ)₂ continuum limit measures on the torus T²_L. The torus approach isolates the UV limit (a = L/N → 0) from IR issues by fixing the physical volume L. **Current state (2026-03-03):** | | Gaussian | Interacting | |--|----------|-------------| | Limit exists | PROVED (full sequence) | PROVED (subsequential) | | OS0 (Analyticity) | sorry | not started | | OS1 (Regularity) | sorry | not started | | OS2 (Translation+D4) | PROVED | not started | | OS3 (Reflection Positivity) | PROVED | not started | | `SatisfiesTorusOS` | assembled (2 sorries) | not started | 7 files in `TorusContinuumLimit/`, 15 torus axioms, 2 sorries. ## Architecture The OS axiom definitions (`SatisfiesTorusOS`) are measure-agnostic — they apply to any probability measure on `Configuration(TorusTestFunction L)`. The Gaussian and interacting cases share the same axiom bundle but have different proof paths. ``` SatisfiesTorusOS L μ (measure-agnostic bundle) ╱ ╲ Gaussian proof path Interacting proof path ─────────────────── ────────────────────── OS0: exp(-½Q) entire OS0: exponential moments OS1: ‖Z[J]‖ ≤ exp(c·q(J)) OS1: ‖Z[J]‖ ≤ exp(c·q(J)) OS2: G_L invariance OS2: V_a invariance OS3: rp_closed_under_weak_limit OS3: rp_closed_under_weak_limit + torusLattice_rp + interacting lattice RP + full sequence convergence + subsequential convergence ``` ## Phase 1: Complete Gaussian OS (fill 2 sorries) ### 1a. `torusGaussianLimit_os1` — OS1 (needs characteristic functional axiom) The complex generating functional `Z[f_re, f_im] = ∫ exp(iω(f_re) - ω(f_im)) dμ` satisfies `‖Z[f_re, f_im]‖ ≤ exp(c · (q(f_re) + q(f_im)))` for a continuous seminorm `q` on the torus test function space. For Gaussian μ with covariance G_L, the bound holds with `q(f) = G_L(f,f)` and `c = ½`, since `|exp(-½ G_L(f_re+if_im, f_re+if_im))| ≤ exp(½ G_L(f_im,f_im))`. ### 1b. `torusGaussianLimit_os0` — OS0 (straightforward, uses char. func. axiom) By `torusGaussianLimit_characteristic_functional`, `Z[Σ zᵢJᵢ]` = `exp(-½ Σᵢⱼ zᵢzⱼ G(Jᵢ,Jⱼ))`. Re of this is a real-analytic function of z ∈ ℝⁿ (composition of polynomial with exp). ## Phase 2: Interacting OS3 (Reflection Positivity) OS3 is the easiest interacting axiom because `rp_closed_under_weak_limit` is already proved and applies to any weak limit of RP measures. ### 2a. Interacting lattice RP Need: `IsRP (torusInteractingMeasure L N P mass hmass) (torusConfigReflection L) (torusRP_admissible L)` The interacting measure `dμ_P = (1/Z) e^{-V} dμ_{GFF}` is RP because: - The GFF is RP (transfer matrix `H ≥ 0`) - The Wick-ordered interaction `V = Σ_x :P(φ(x)):` decomposes as `V = V_+ + V_-` where `V_±` depend on the positive/negative time sites respectively - The Boltzmann weight `e^{-V} = e^{-V_+} · e^{-V_-}` factors, preserving the RP structure This is a new axiom (or provable from `torusLattice_rp` + interaction structure). **Reference:** Glimm-Jaffe §6.3, Simon Ch. II. ### 2b. Thread subsequence through RP closure `torusInteractingLimit_exists` gives `(φ, μ, StrictMono φ, IsProbabilityMeasure μ, weak convergence along φ)`. Need to apply `rp_closed_under_weak_limit` with the subsequential measures `fun n => torusInteractingMeasure L (φ n + 1) P mass hmass`. This is a straightforward adaptation of `torusGaussianLimit_os3` — the only difference is using subsequential convergence instead of full sequence convergence. ## Phase 3: Interacting OS1 (Regularity) `‖Z[f_re, f_im]‖ ≤ exp(c · (q(f_re) + q(f_im)))` for a continuous seminorm q. For the interacting case, the bound follows from Nelson's hypercontractive estimate, which controls exponential moments of the field. Needs an axiom for interacting exponential moments (similar to `continuum_exponential_moments`). ## Phase 4: Interacting OS2 (Symmetry) ### 4a. Translation invariance The lattice interaction is translation-invariant (`latticeAction_translation_invariant` is proved). The lattice GFF is translation-invariant. Therefore the interacting lattice measure is translation-invariant, and this transfers to the limit. Two approaches: 1. **Direct:** Prove the interacting lattice generating functional is translation-invariant, then transfer by weak convergence. 2. **Via covariance:** If the limit measure is determined by its Schwinger functions, and the Schwinger functions inherit lattice translation invariance, then the limit is translation-invariant. Approach 1 is cleaner. Need axiom or proof: - `torusInteractingMeasure_translation_invariant`: `Z_P[T_v f] = Z_P[f]` for all lattice translation vectors v. ### 4b. D4 invariance Similar to translation invariance: the lattice action is D4-invariant (nearest-neighbor interactions are symmetric under the square lattice point group). Need axiom or proof: - `torusInteractingMeasure_D4_invariant`: `Z_P[σf] = Z_P[f]` for σ ∈ D4. ### 4c. Transfer to continuum Weak convergence: if `Z_N[σf] = Z_N[f]` for all N, and `Z_N → Z` pointwise, then `Z[σf] = Z[f]`. This is immediate from the convergence. ## Phase 5: Interacting OS0 (Analyticity) This is the hardest axiom for the interacting case. OS0 requires `Re(Z[Σ zᵢJᵢ])` to be real-analytic on ℝⁿ. For the Gaussian this was trivial (exp of quadratic). For the interacting case, this requires showing the measure has all exponential moments: `∫ exp(c|ω(f)|) dμ < ∞` for all c > 0, f ∈ TorusTestFunction L. On the finite torus, this should follow from: 1. The interacting measure has density `(1/Z) e^{-V}` w.r.t. the GFF 2. V is bounded below (proved: `latticeInteraction_bounded_below`) 3. The GFF has all Gaussian moments 4. On the torus, the interaction is a polynomial in finitely many Gaussian variables, so all moments are finite The transfer to the continuum limit requires uniform moment bounds (analogue of `continuum_exponential_moments`). **Difficulty:** Medium-Hard. Needs Nelson-type estimates on the torus. ## Phase 6: Assembly ### 6a. `IsTorusInteractingContinuumLimit` predicate Define a predicate for the interacting continuum limit, analogous to `IsTorusGaussianContinuumLimit`. Should include: - `IsProbabilityMeasure μ` - Z₂ symmetry (inherited from the even polynomial P) - Weak limit of interacting lattice measures ### 6b. `torusInteractingLimit_satisfies_OS` Assemble OS0–OS3 for the interacting case. ### 6c. `torusInteractingOS_exists` Master existence theorem: ``` theorem torusInteractingOS_exists (P : InteractionPolynomial) (mass : ℝ) (hmass : 0 < mass) : ∃ (μ : Measure (Configuration (TorusTestFunction L))) (_ : IsProbabilityMeasure μ), @SatisfiesTorusOS L hL μ ‹_› ``` ## New Axioms Needed for Interacting OS | Axiom | Difficulty | Phase | |-------|-----------|-------| | Interacting lattice RP | Easy-Med | 2a | | Interacting exponential moments (torus) | Med | 3 | | Interacting translation invariance (lattice) | Easy | 4a | | Interacting D4 invariance (lattice) | Easy | 4b | | Interacting exponential moments (continuum transfer) | Med-Hard | 5 | Total: ~5 new axioms for the interacting case. ## Dependency Graph ``` Phase 1 (Gaussian sorries) independent, do first ├─ 1a: OS1 (straightforward, uses char. func. axiom) └─ 1b: OS0 (straightforward, uses char. func. axiom) Phase 2 (Interacting OS3) needs: interacting lattice RP axiom ├─ 2a: interacting lattice RP └─ 2b: thread subsequence Phase 3 (Interacting OS1) needs exponential moment axiom Phase 4 (Interacting OS2) needs: lattice symmetry axioms ├─ 4a: translation invariance ├─ 4b: D4 invariance └─ 4c: transfer to limit Phase 5 (Interacting OS0) hardest, needs moment bounds └─ exponential moments Phase 6 (Assembly) depends on all above ``` ## Implementation Order 1. Fill Gaussian OS1 sorry (Phase 1a) — straightforward 2. Fill Gaussian OS0 sorry (Phase 1b) — straightforward 3. Add interacting lattice RP axiom + prove OS3 (Phase 2) 4. Prove interacting OS1 (Phase 3) — needs exponential moment axiom 5. Add symmetry axioms + prove OS2 (Phase 4) 6. Add moment axiom + prove OS0 (Phase 5) 7. Assembly + master theorem (Phase 6) ## Success Criteria - `torusGaussianLimit_satisfies_OS` compiles with no sorry - `torusInteractingLimit_satisfies_OS` compiles (with axioms, no sorry) - `torusInteractingOS_exists` master theorem - `lake build` passes - Total new axiom count ≤ 4