================================================================================ CONTRARIAN AUDIT REPORT v2 — SECOND PASS ON EQUATIONS_SPEC_v2.txt ================================================================================ Date: 2026-03-20 Auditor: Claude (contrarian second-pass) Input: EQUATIONS_SPEC_v2.txt (revised spec) Reference: AUDIT_REPORT_v1.txt (20 original findings + gotchas) Sources: theory.txt lines 140-210, mathematical_framework.txt lines 248-320, math_framework_approach_B_results.txt, 24cell_research.txt Result: 17 RESOLVED, 2 PARTIALLY RESOLVED, 1 NOT RESOLVED, 2 NEW ISSUES ================================================================================ FINDING-BY-FINDING VERIFICATION ================================================================================ ---------------------------------------------------------------------- #1 [CRITICAL] Missing potential term V(psi,A) ---------------------------------------------------------------------- VERDICT: RESOLVED The v2 spec now includes the full potential term: d²ψ/dt² = c²∇²ψ - ω₀²ψ(1 + A(x)/A_ref) + J(x,t) This matches the Euler-Lagrange equation at math_framework.txt line 293: d²ψ/dt² - c²∇²ψ + ω₀²ψ(1+A/A_ref) = J(x,t) Sign convention is consistent (v2 moves the potential to the RHS and flips sign, which is correct). The Lagrangian (1b) is provided and matches approach_B_results.txt line 80 exactly. Fix is correct. ---------------------------------------------------------------------- #2 [HIGH] Amplitude coupling half-implemented ---------------------------------------------------------------------- VERDICT: RESOLVED v2 Section 5 now specifies that A(x) appears in TWO places: the potential term and the source amplitude. The phase separation (Phase 1: uniform A; Phase 2: CN-coupled A) is clearly documented and physically reasonable. ---------------------------------------------------------------------- #3 [MEDIUM] nabla^4 notation means biharmonic ---------------------------------------------------------------------- VERDICT: RESOLVED v2 uses ∇² throughout and explicitly states "(NOT ∇⁴ which is biharmonic)." The 4D discrete Laplacian is written out as eq. (2). Correct. ---------------------------------------------------------------------- #4 (not in original numbering — skipped) ---------------------------------------------------------------------- ---------------------------------------------------------------------- #5 [HIGH] Form A/B normalization mismatch ---------------------------------------------------------------------- VERDICT: PARTIALLY RESOLVED v2 correctly identifies Form A at radius 1.0 and Form B at radius √2, and states Form B must be divided by √2 to normalize to unit sphere. This is correct. HOWEVER: v2 labels Form A as "cell-center form" and Form B as "edge-midpoint form." The 24cell_research.txt does not use these labels. The standard labels in the literature are: - Form A = Hurwitz quaternion coordinates (unit quaternion form) - Form B = D4 root system coordinates The labels "cell-center" and "edge-midpoint" are not standard and could cause confusion if someone tries to verify against a reference. This is cosmetic, not mathematical — the coordinates themselves are correct. STATUS: The normalization fix is correct. The naming is non-standard but harmless. RESOLVED on the substance. ---------------------------------------------------------------------- #6 [LOW] Self-dual rotation involves scaling ---------------------------------------------------------------------- VERDICT: RESOLVED v2 now states: "This is a rotation-plus-scaling, not a pure rotation. The 45° angle is exact; the scaling is √2." This matches 24cell_research.txt line 716-718 which shows the dual maps Form B to a scaled Form A. Correct. ---------------------------------------------------------------------- #7 [HIGH] C_potential uses energy density, theory says curvature ---------------------------------------------------------------------- VERDICT: PARTIALLY RESOLVED — see detailed analysis below in Additional Checks. v2 now uses κ(x) = ∇²⟨|ψ(x)|²⟩ / max(|∇²⟨|ψ|²⟩|), which is the Laplacian of the energy density, normalized. This is a defensible choice for "curvature" but has a subtlety — see NEW ISSUE A below. ---------------------------------------------------------------------- #8 [MEDIUM] CFL lacks safety factor and grid assumptions ---------------------------------------------------------------------- VERDICT: RESOLVED v2 provides: - Explicit formula: dt = 0.9 × h / (c_max × 2) - Safety factor (Courant number = 0.9) - Grid spacing specified (h = 6.0/N for each phase) - Modified CFL with potential term computed numerically All correct. The √4 = 2 for 4D is standard. ---------------------------------------------------------------------- #9 [HIGH] "No additional terms" checklist masks MISSING terms ---------------------------------------------------------------------- VERDICT: RESOLVED The "What the engine does NOT include" section (§11) now explicitly lists what the engine DOES include, separating "not imposed" from "actively present." The audit checklist in §13 covers all active terms. Good fix. ---------------------------------------------------------------------- #10 [MEDIUM] Energy conservation impossible with active sources ---------------------------------------------------------------------- VERDICT: RESOLVED v2 never claims energy conservation. Instead it specifies energy BALANCE tracking: E_injected, E_field, E_overflow, E_boundary (PML absorbed). The measurement section asks for energy time series. This is the correct approach for a driven system. ---------------------------------------------------------------------- #11 [CRITICAL] Boundary conditions never specified ---------------------------------------------------------------------- VERDICT: RESOLVED v2 Section 7 specifies PML with: - N_pml = 10 grid points - Quadratic conductivity profile σ(d) = σ_max × (d/N_pml)² - σ_max = 3c / (2h × N_pml) (standard Berenger calibration) - Secondary test with periodic BCs for comparison This is a standard PML specification. See Additional Checks for 4D validity. ---------------------------------------------------------------------- #12 [HIGH] Grid resolution, domain size unspecified ---------------------------------------------------------------------- VERDICT: RESOLVED v2 Section 9 provides three resolution phases (N=32, 48, 64), domain [-3,3]^4, memory estimates, and runtime estimates. Source-to-grid interpolation method is specified (4D trilinear, 16-point stencil). ---------------------------------------------------------------------- #13 [HIGH] CN(x) undefined for continuum field ---------------------------------------------------------------------- VERDICT: RESOLVED v2 Section 5 defers CN coupling to Phase 2 and provides a concrete algorithm: identify local maxima of time-averaged intensity, count neighbors within cutoff. CN_ref = 8 (BCC coordination). For Phase 1, A(x) = A_base (uniform). This is physically reasonable and implementable. ---------------------------------------------------------------------- #14 [MEDIUM] r=0.5 ceiling citation wrong ---------------------------------------------------------------------- VERDICT: RESOLVED v2 now cites "math_framework B.1 line 296: Collapse at r=0.5: active/rest phase symmetry (exact)." This matches mathematical_framework.txt line 296 verbatim. Correct citation. ---------------------------------------------------------------------- #15 [MEDIUM] Rectangular window creates Gibbs artifacts ---------------------------------------------------------------------- VERDICT: RESOLVED v2 Section 3 adds a raised cosine ramp with N_ramp = max(4, 0.05 × T/dt). The spec acknowledges the rectangular window is the theory's specification and the ramp is a numerical necessity, with sensitivity testing recommended. Good balance of fidelity and numerical pragmatism. ---------------------------------------------------------------------- #16 [LOW] c_4D = 1+sin(45°) has no derivation ---------------------------------------------------------------------- VERDICT: RESOLVED v2 Section 10 now labels 1.707 as "HEURISTIC candidate, not derived from theory" and explicitly says "connection is ad-hoc; included in sweep for completeness, NOT as a prediction." The 1.625 = 13/8 Fibonacci prediction is properly labeled as derived. Correct. ---------------------------------------------------------------------- #17 [HIGH] "Does not include" omits spillover and misframes phi ---------------------------------------------------------------------- VERDICT: RESOLVED v2 splits this into three sub-items: #17a: 5D overflow is now implemented (Section 8) #17b: phi is explicitly NOT imposed — it's a prediction to test Section 11 now has separate "does not" and "does" lists. ---------------------------------------------------------------------- #18 [MEDIUM] Audit checklist missing 9 items ---------------------------------------------------------------------- VERDICT: RESOLVED v2 Section 13 has a comprehensive 24-item checklist covering equations, geometry, numerics, physics, outputs, and units. All major items covered. ---------------------------------------------------------------------- #19 [MEDIUM] Three conflicting source placement options ---------------------------------------------------------------------- VERDICT: RESOLVED v2 Section 4 designates a PRIMARY configuration (Form A only, 24 sources) and SECONDARY configurations (Form B only; all 48). The primary test is the scientifically meaningful one (observe whether dual structure emerges). ---------------------------------------------------------------------- #20 [MEDIUM] No units system defined ---------------------------------------------------------------------- VERDICT: RESOLVED v2 has a dedicated Units System section specifying dimensionless natural units with c_3D = 1, λ₀ = 1, f₀ = 1. Sweep values are dimensionless ratios. Clear and sufficient. ---------------------------------------------------------------------- ADDITIONAL GOTCHAS FROM v1 AUDIT ---------------------------------------------------------------------- GOTCHA: 5D overflow dimension needed when r→0.5 VERDICT: RESOLVED — Section 8 implements this fully. GOTCHA: c_eff must be uniform across grid VERDICT: RESOLVED — The spec uses a single c per sweep run (not position- dependent). The potential term modifies effective dynamics but c itself is uniform. GOTCHA: CFL in 4D with c=1.8 will produce tiny dt — runtime concern VERDICT: RESOLVED — Runtime estimates are provided. dt ≈ 0.03 is not problematically small. Phase 1 runs in ~3 seconds. See Additional Checks for full runtime analysis. ================================================================================ ADDITIONAL CHECKS — DEEP TECHNICAL REVIEW ================================================================================ ---------------------------------------------------------------------- CHECK A: 5D overflow — is the 1D wave equation physically sound? ---------------------------------------------------------------------- The spec proposes a 1D wave equation along w at each 4D position: d²ψ₅/dt² = c₅² ∂²ψ₅/∂w² This is computationally clever but physically questionable. CONCERN: The 5D layer has NO lateral coupling between adjacent 4D positions. Two neighboring points in 4D that both overflow will inject energy into independent 1D channels with no communication between them. Real overflow would spread in all 5 dimensions, not just the 5th. MITIGATING FACTOR: This is explicitly a thin capture layer (N_5 = 8), not a full 5D simulation. Its purpose is to TRACK overflow energy, not to simulate 5D physics faithfully. The spec reports 5D energy separately. VERDICT: Acceptable for Phase 1 as an energy accounting mechanism. If the 5D layer shows significant energy accumulation, a coupled 5D simulation would be needed. The spec should document this limitation. SEVERITY: LOW — design choice, not error. The c_5D = 21/8 = 2.625 value is correctly derived from the Fibonacci pair {8, 13} per mathematical_framework.txt line 1907. This is consistent with the theory's dimensional framerate scaling. ---------------------------------------------------------------------- CHECK B: Curvature κ(x) = ∇²⟨|ψ|²⟩ — correct second derivative? ---------------------------------------------------------------------- NEW ISSUE A: AMBIGUITY IN CURVATURE DEFINITION The theory (theory.txt lines 157-165) says: "the curve is Lagrangian" and "time slows as the bandwidth curve steepens." The v2 spec interprets "steepness = curvature = second derivative" and uses the Laplacian: κ(x) = ∇²⟨|ψ(x)|²⟩ Three possible interpretations of "curvature" exist: (a) ∇²⟨|ψ|²⟩ — Laplacian (what v2 uses). This measures concavity of the energy density profile. Positive at valleys, negative at peaks. (b) |∇⟨|ψ|²⟩| — gradient magnitude. This measures steepness directly. The theory says "steepens" which is arguably gradient, not curvature. (c) The full curvature tensor κ_ij = ∂²⟨|ψ|²⟩/∂x_i∂x_j, or its trace (which is the Laplacian, i.e., option a). The theory says "the space of decoherence INCREASES at the peak of the curve." With the Laplacian: at a peak of ⟨|ψ|²⟩, ∇² < 0 (concave down). The normalized κ would be negative at peaks. Then r(x) = r₀ + α×κ(x) would DECREASE r at peaks (if α > 0), meaning LESS decoherence at peaks. But the theory says decoherence INCREASES at the peak. This means either: - α must be NEGATIVE (so negative κ → increased r), OR - The "peak of the curve" refers to the curvature peak, not the energy peak (as v2 line 193 claims), OR - The gradient magnitude (option b) should be used instead RECOMMENDATION: The spec should explicitly state the sign convention for α and verify that r(x) increases at locations where the theory predicts more decoherence. As written, the sign may be inverted relative to the theory's intent. SEVERITY: MEDIUM — the physics could be inverted. ---------------------------------------------------------------------- CHECK C: Runtime estimates — realistic for VPS? ---------------------------------------------------------------------- Phase 1 (N=32): 1M points, ~3,300 steps, ~3 seconds per run - This seems reasonable. 1M multiply-adds per step at ~1ms is realistic for a single-threaded Python/NumPy implementation on a modern VPS. Phase 2 (N=48): 5.3M points, ~25 seconds per run - 5.3M points × ~3,300 steps = 17.5 billion operations. At ~1 GFlop/s (NumPy on single core), this is ~17 seconds. 25 seconds is a reasonable estimate accounting for memory overhead. - Memory: 127 MB for field arrays. With PML, CN arrays, 5D layer, and temporaries, realistically 300-500 MB. Feasible on a VPS with 4+ GB. Phase 3 (N=64): 16.8M points per step - Memory: 403 MB stated, but with all auxiliary arrays, more like 1-1.5 GB. - Runtime: ~80 seconds per run (13-run sweep = ~17 minutes). Feasible. The 5D overflow layer adds N_5 = 8 points per 4D position, but only at positions where r ≥ 0.5. In practice, few positions will overflow, so the overhead is negligible. VERDICT: Estimates are realistic. No concerns for a Hetzner VPS with 8+ GB RAM. Phase 1 is cheap enough for rapid iteration. ---------------------------------------------------------------------- CHECK D: PML validity in 4D ---------------------------------------------------------------------- PML (Perfectly Matched Layer) was originally derived for 2D and 3D electromagnetic waves (Berenger, 1994). The mathematical principle — a complex coordinate stretching — generalizes to any dimension and any second-order wave equation. The key requirement is: 1. The wave equation must be hyperbolic (yes — it's a wave equation) 2. The PML conductivity must be applied independently per dimension 3. The CFL condition must account for the PML modification For a scalar wave equation (not Maxwell's equations), the split-field formulation is not needed. A simple auxiliary differential equation (ADE) or unsplit PML works. The quadratic conductivity profile σ(d) = σ_max × (d/N_pml)² is standard and dimension-independent. The calibration σ_max = 3c/(2h×N_pml) is the standard formula from Berenger. It assumes a target reflection coefficient of ~10⁻³, which is appropriate. CONCERN: With the potential term ω₀²ψ(1+A/A_ref), the PML must damp BOTH the propagating component AND any quasi-static component from the mass-like term. If ω₀ is large relative to the wave's spatial frequency, the PML may not absorb evanescent modes efficiently. However, for ω₀ = 2π and the spatial scales in this problem, this is unlikely to be significant. VERDICT: PML in 4D is valid. The formulation is standard. No issues. ---------------------------------------------------------------------- CHECK E: Modified CFL with potential term ---------------------------------------------------------------------- The spec gives: dt < h / √(4c² + ω₀²h²) Derivation check: For the equation d²ψ/dt² = c²∇²ψ - ω₀²ψ, the dispersion relation with FDTD discretization is: [2sin(ωdt/2)/dt]² = Σᵢ c²[2sin(kᵢh/2)/h]² + ω₀² The worst case (maximum frequency) occurs when sin(kᵢh/2) = 1 for all 4 dimensions: [2/dt]² ≥ 4 × c²[2/h]² + ω₀² 4/dt² ≥ 16c²/h² + ω₀² dt² ≤ 4 / (16c²/h² + ω₀²) dt ≤ 2 / √(16c²/h² + ω₀²) dt ≤ 2h / √(16c² + ω₀²h²) The spec says dt < h / √(4c² + ω₀²h²). Let me check: h / √(4c² + ω₀²h²) vs 2h / √(16c² + ω₀²h²) Factor out from the correct expression: 2h / √(16c² + ω₀²h²) The spec's formula: h / √(4c² + ω₀²h²) = h / √(4c² + ω₀²h²) = 2h / √(16c² + 4ω₀²h²) These are NOT the same: 2h/√(16c² + ω₀²h²) vs 2h/√(16c² + 4ω₀²h²). The spec's formula has 4ω₀²h² where it should have ω₀²h². NEW ISSUE B: MODIFIED CFL FORMULA IS SLIGHTLY WRONG The spec's formula dt < h / √(4c² + ω₀²h²) is equivalent to dt < 2h / √(16c² + 4ω₀²h²), but the correct formula is dt < 2h / √(16c² + ω₀²h²). The spec's formula is MORE restrictive (smaller dt), so it is CONSERVATIVE — it will not cause instability. However, it over-constrains the timestep by a small amount. Numerically: with c=1.8, ω₀=2π, h=0.125: Spec: √(4×3.24 + 39.48×0.0156) = √(12.96 + 0.616) = √13.58 → dt < 0.034 Correct: √(16×3.24 + 39.48×0.0156)/2 = √(51.84+0.616)/2 = √52.46/2 → dt < 0.034 Wait — let me redo this more carefully: Correct: dt < 2h/√(16c² + ω₀²h²) = 2(0.125)/√(16(3.24) + 39.48(0.0156)) = 0.25/√(51.84 + 0.616) = 0.25/√52.46 = 0.25/7.24 = 0.0345 Spec: dt < h/√(4c² + ω₀²h²) = 0.125/√(12.96 + 0.616) = 0.125/√13.58 = 0.125/3.686 = 0.0339 The difference is 0.0345 vs 0.0339 — less than 2%. Since both are above the stated dt = 0.03125 (from the basic CFL with safety factor 0.9), and since the safety factor already provides margin, this error has NO practical impact. SEVERITY: LOW — formula has an extra factor of 4 on the ω₀²h² term but the error is conservative and negligible with the safety factor applied. ---------------------------------------------------------------------- CHECK F: Hetzner VPS deployment concerns ---------------------------------------------------------------------- Memory: Phase 1: ~50 MB total (comfortable) Phase 2: ~300-500 MB total (comfortable on 4+ GB VPS) Phase 3: ~1-1.5 GB total (needs 4+ GB VPS) CPU: Single-threaded NumPy is sufficient for Phase 1-2. Phase 3 may benefit from NumPy with OpenBLAS/MKL for vectorized ops. No GPU needed. Disk: Raw output per sweep: 13 runs × ~50 MB (Phase 2) = ~650 MB Including numpy arrays and JSON: ~1 GB per sweep Manageable on any VPS. VERDICT: No concerns for Hetzner VPS with 4+ GB RAM. Phase 3 at N=64 is the tightest, but still feasible. ================================================================================ NEW ISSUES INTRODUCED BY REVISION ================================================================================ ---------------------------------------------------------------------- NEW ISSUE A [MEDIUM]: Sign ambiguity in curvature-decoherence coupling ---------------------------------------------------------------------- (Detailed above in Check B) The Laplacian ∇²⟨|ψ|²⟩ is NEGATIVE at energy peaks (concave down) and POSITIVE at energy valleys (concave up). The theory says decoherence INCREASES at "the peak of the curve." If "the curve" means the energy density, then r(x) should increase where ∇² < 0, requiring α < 0. If "the curve" means the curvature itself, the sign could go either way. The spec normalizes κ to [-1, 1] but does not specify the sign of α. Without this, the coupling direction is ambiguous and could invert the intended physics. RECOMMENDATION: Add a sentence specifying α > 0 or α < 0, with justification from the theory. Test both signs in the sweep if uncertain. ---------------------------------------------------------------------- NEW ISSUE B [LOW]: Modified CFL formula has minor algebraic error ---------------------------------------------------------------------- (Detailed above in Check E) The formula dt < h/√(4c² + ω₀²h²) should be dt < 2h/√(16c² + ω₀²h²), which simplifies differently. The error is conservative (produces a slightly smaller dt) and is absorbed by the safety factor. No practical impact. RECOMMENDATION: Either correct the formula or add a note that the stated formula is a conservative upper bound. ---------------------------------------------------------------------- NOTE (not a new issue): 4D trilinear interpolation terminology ---------------------------------------------------------------------- The spec calls the 4D source interpolation "trilinear" with a "16-point stencil." In 4D, the correct term is "quadrilinear interpolation" with a 2⁴ = 16-point stencil. "Trilinear" specifically means 3D (2³ = 8 points). This is a terminology issue only — the 16-point stencil count is correct. ================================================================================ SUMMARY SCORECARD ================================================================================ # Severity Status Notes --- ---------- -------------------- ------------------------------------ 1 CRITICAL RESOLVED Potential term now correct 2 HIGH RESOLVED Dual amplitude coupling documented 3 MEDIUM RESOLVED ∇² notation corrected 5 HIGH RESOLVED Normalization fixed, coords correct 6 LOW RESOLVED Scaling acknowledged 7 HIGH PARTIALLY RESOLVED Curvature used, but sign ambiguous 8 MEDIUM RESOLVED CFL fully specified 9 HIGH RESOLVED "Does/does not" lists separated 10 MEDIUM RESOLVED Energy balance, not conservation 11 CRITICAL RESOLVED PML boundaries specified 12 HIGH RESOLVED Three resolution phases defined 13 HIGH RESOLVED CN deferred; algorithm provided 14 MEDIUM RESOLVED Citation corrected 15 MEDIUM RESOLVED Smooth ramp added 16 LOW RESOLVED Heuristic label added 17 HIGH RESOLVED 5D overflow + phi framing fixed 18 MEDIUM RESOLVED Comprehensive checklist 19 MEDIUM RESOLVED Primary config designated 20 MEDIUM RESOLVED Units system defined NEW ISSUES: A MEDIUM Sign of α in κ-to-r coupling is ambiguous B LOW Modified CFL formula minor algebraic error (conservative) GOTCHAS: All 3 resolved. OVERALL VERDICT: The v2 spec is substantially improved. 17 of 20 original findings are fully resolved. The 2 partially resolved items and 2 new issues are all LOW-MEDIUM severity and addressable with minor clarifications. The spec is ready for implementation with the caveat that the sign of α (New Issue A) must be determined before running — it controls whether organized regions get MORE or LESS decoherence, which is the central physical mechanism. ================================================================================