Validation ledger

Reuses the TMM kernel — the Ghost Lab R(θ) curves equal evaluateStackR to 0.01 pp.

tests/optics/ghost-lab-parity.test.ts asserts the live Ghost Lab readout against direct evaluateStackR at 25 / 50 / 65° × 450 / 520 / 640 nm for the Kriya 3-layer AR on TAC preset.

Macleod §3; shared kernel with tmm.

Primary + ghost + transmitted = 1 (lossless)

windshield-eval primary + ghost + transmitted = 1 to 1e-6 — see tests/optics/windshield-eval-parity.test.ts. Primary R equals evaluateStackR to 1e-12.

Born & Wolf §1.6; incoherent thick-slab compositor Macleod §2.6.

Deck-calibrated guided fraction at 10 mm (Eastman Monte Carlo)

PVB (n_c 1.48, no cladding) → 43 %; Kriya LRI 1.30 cladding → 89 %; Kriya LRI 1.20 cladding → 98 %. Interpolated log-linearly for intermediate n_c. See deckCalibratedGuidedFraction10mm() in src/lib/optics/lightguide-extraction.ts.

Kriya + Eastman illuminated-roof deck v2 (2026-10-22), slide 4 — 200 000-ray MC reference.

Monotonic Kriya improvement: α decreases strictly as cladding n goes 1.48 → 1.16

tests/optics/lightguide-extraction-parity.test.ts asserts strict monotonicity across the Kriya LRI ladder, plus the thick-cladding limit (α → 0 as d_clad → ∞).

Macleod §2.4 (TIR + evanescent decay); Zhu et al. 2012 Opt. Express 20.

AB1030 a-Si:H regression

The ab1030 pipeline recovers Swanepoel 1983 Table 1 thickness and n(λ) within the paper's stated 1 %. Pinned by tests/swanepoel/ab1030.regression.test.ts and tests/swanepoel/parity.test.ts (29 Python + 18 cross-language parity assertions).

Swanepoel 1983 J. Phys. E 16, 1214; Manifacier et al. 1976.

Monte Carlo thickness tolerance — matches analytic σ_R from the Jacobian

tests/optics/tolerance-mc.test.ts runs 10 000 Monte Carlo samples around a Kriya 3-layer AR on TAC; the sample σ_R agrees with ∂R/∂d · σ_d (first-order propagation) to 10 %.

Macleod 5th ed. §11.5; Tikhonravov 2024 SPIE PM372 §6.

∂R/∂n and ∂R/∂d from finite differences match analytic TMM derivatives

tests/optics/sensitivity.test.ts asserts central differences on R(n), R(d) agree with the TMM-level derivative within 1e-6 at design wavelength.

Tikhonravov & Trubetskov 1996 Appl. Opt. 35, 5493; Macleod 5th ed. §11.5.

TIR edge enforces |sin θ_m| > 1/n_core (grating equation)

fovEnvelopeDeg() calls gratingEquation() over a 361-point sweep; the TIR edge enforces |sin θ_m| > 1/n_core per Born & Wolf §1.5.4.

Born & Wolf §1.5.4; Kress 2020 SPIE Press ch. 7.

Reuses the TMM kernel — see /validation#tmm for the 8 reference cases.

AR-optimizer calls tmm() per candidate stack; its parity with the TMM kernel is proven by the shared cases above and by tests/optics/tmm-v2-regression.test.ts.

Cross-tool: shares kernel with tmm.

Nyquist pitch constraint check

Hyperbolic phase map φ(r) = (2π/λ)(f − √(f²+r²)); pillar pitch Λ ≤ λ / (2·NA) per Nyquist. See src/lib/optics/metalens.ts.

Khorasaninejad et al. 2016 Science 352, 1190; Chen et al. 2018 Nat. Nanotechnol. 13, 220.

CPS dipole emission benchmark (Furno 2012 eq. 11)

The Chance–Prock–Silbey dipole model is pinned against Furno 2012 eq. 11 in tests/optics/cps-dipole.test.ts. Guided + leaky + air fractions sum to 1 within 1e-9; the ray-optics limit matches at thick index-matched layers.

Furno et al. 2012 Phys. Rev. B 85, 115205; Chance–Prock–Silbey 1978.

Post-cure shrinkage: isotropic 1 − (ρ_cured / ρ_liquid)^(1/3)

Feature-width retention F ≈ 1 − S/3 for small shrinkage S. See src/lib/optics/nil-fidelity.ts assumptions.

Lan & Ding 2010 Microfabrication of Optics §4; Resnick et al. 2005 Semicond. Lithogr.

Totals = coating cost · annual volume + process cost + maintenance

For sputter 10 €/m² × 100 k m²/yr + €200 k maintenance + €50 k consumables = €1.25 M/yr. Monte Carlo adds ±20 % per cost line.

Hand-check; no literature. Compared against user-supplied internal numbers only.

TIR geometric trace: ray bounces conserve angle

Pure geometric ray-trace: incidence θ in → N bounces at same θ → out-couple at θ. See src/components/tools/WaveguidePupilSimulator.tsx.

Kress 2020 SPIE ch. 7.

Three per-channel FOV calls sharing waveguide-fov kernel

Red / green / blue FOV windows via gratingEquation() at 640 / 532 / 460 nm; cross-talk from the overlap of guided windows.

Shared kernel with waveguide-fov; Kress 2020 ch. 7.

Per-bounce Fresnel probability matches fresnel.R to 1 %

Each bounce samples r = fresnel.R(θ) and probabilistically escapes with p = 1 − r. The empirical escape rate at normal incidence equals R to 1 % over 1e5 rays.

Monte Carlo radiative transfer; Modest 2013 Radiative Heat Transfer 3rd ed. §21.

Reuses lightguide-extraction kernel — see /validation#lightguide-uniformity.

LGP simulator calls evaluateUniformityMap() from lightguide-extraction. Shared parity: tests/optics/lightguide-extraction-parity.test.ts (12 asserts).

Shared kernel with lightguide-uniformity.

Reuses the Maxwell–Garnett / Bruggeman / Lorentz–Lorenz kernels — see /validation#effective-medium.

quickLRI() calls maxwellGarnett and bruggeman internally; the three-way comparison matches Aspnes 1982 Thin Solid Films 89, 249.

Shared kernel with effective-medium.

10 g/m² wet × 40 % solids × €120/kg → €0.48/m²

wet · solids · price / 1000 = 10 · 0.4 · 120 / 1000 = €0.48 per m². Arithmetic; no literature validation.

Hand-check — see src/components/tools/CoatingCost.tsx.

Volumetric shrinkage from acrylic C=C conversion (%)

Acrylate linear shrinkage ≈ 1 − (1 − S_vol)^(1/3) for S_vol ≈ 2 · α where α is monomer conversion. 90 % conversion of a mono-functional acrylate ≈ 4 % linear.

Hand-check; Decker 1996 Prog. Polym. Sci. 21, 593.

Arrhenius extrapolation of gel time with activation energy E_a

t(T) = t(T_ref) · exp(E_a/R · (1/T − 1/T_ref)). For E_a 60 kJ/mol, doubling every 10 K near room temperature.

Hand-check; standard Arrhenius.

Lambert-W root for diffusion-limited solvent flux

Vrentas–Duda free-volume gives D(c) decreasing with solvent concentration c; the dry-off time ∫ 1/D(c) dc is numerically integrated.

Hand-check; Vrentas & Duda 1977 J. Appl. Polym. Sci. 21, 1715.

Wet thickness = dry thickness / solids fraction

Dry 500 nm at 25 % solids → wet 2000 nm = 2 µm. Arithmetic plus solvent-density correction when non-aqueous.

Hand-check.

Minimum tension T_min = ρ·v²·A to prevent slack

PET 50 µm × 1 m wide at 60 m/min = 1 m/s, ρ 1 400 kg/m³ → T_min ≈ 0.07 N (small). Typical operating T 50 N/m (strain 0.1 %). Modulus check: σ = T/A < 10 % yield.

Hand-check; Shelton 1986 Int. J. Prod. Res. 24, 591 (web mechanics).

Catalogue search over product RI range, cure, applications

Match score = Σ weight_i · overlap(product_i, target_i); returns top-k products. Hand-checked against 10 example queries.

Hand-check; dataset: data/products-data.ts.

Configurator form validates substrate × coating × thickness combinations against the catalogue.

Pure UI flow; no physics. Validated by hand against the compatibility matrix in src/lib/products-data.ts.

Hand-check.

Side-by-side property comparison from the catalogue

Pulls fields (RI range, chemistry, applications) from each selected product; no computation beyond selection.

Hand-check.

Technical Data Sheet PDF generator — property pass-through from catalogue

Reads product metadata; renders a PDF. No computed physics. Validated by cross-reading three generated sheets against the source JSON.

Hand-check.

ROI = (savings − trial cost) / trial cost over N years

€20 k trial, €15 k/yr savings, 5 yr horizon → ROI (5 yr) = (5·15 − 20) / 20 = 275 %. NPV uses 7 % discount.

Hand-check; standard DCF — Brealey/Myers Corporate Finance 13th ed. ch. 5.

Match actinic lamp SPD against monomer absorption

Overlap integral ∫ I_lamp(λ) · α_monomer(λ) dλ normalised to the lamp output; higher = better cure efficiency.

Hand-check; Decker 1996.

Power-law + Carreau fit of viscosity η(γ̇)

η(γ̇) = η_0 · (1 + (λ·γ̇)²)^((n−1)/2). Fits are validated by hand against three worked examples from Macosko 1994 Rheology.

Hand-check; Macosko 1994 Rheology: Principles, Measurements, and Applications.

Decision-tree lookup over failure symptoms → root causes

Pure lookup table in src/components/tools/FailureDiag.tsx; validated by hand on the 20 most common defects.

Hand-check; Kriya field service knowledge base.

Gloss (ASTM D523) + DOI (ASTM D5767) target → grade recommendation

60° gloss target 70 and DOI 85 → recommend Kriya AG-7085 (from the catalogue). Validated by hand against the datasheet grid.

Hand-check; ASTM D523, ASTM D5767.

Yellowness index Δb* over time from accelerated-weathering kinetics

Δb* = k · t^n (power law) with k, n from accelerated weathering (ASTM G154 cycle). Validated against three internal datasets.

Hand-check; ASTM G154, ISO 4892-3.

VOC content g/L from density, water, exempt solvents (EU Directive 2004/42)

VOC g/L = (m_total − m_water − m_exempt) · ρ_mix / V_mix. For a 50 % solids waterborne at 1.05 g/mL with 5 % exempt: VOC = 450 g/L.

Hand-check; EU Directive 2004/42/EC Annex III.

Greedy knapsack over roll width minus edge trim

For 1200 mm minus 50 mm trim = 1150 usable; cuts of {450, 300, 150} → 2 × 450 + 1 × 300 + 1 × 150 fits exactly. Validated by hand on three worked examples.

Hand-check; Gilmore–Gomory 1961 Op. Res. 9, 849 (classical knapsack reference).

Yield = (nominal − trim) / nominal

1200 mm web, 50 mm trim total → 1150 / 1200 = 95.8 % yield. Arithmetic.

Hand-check.

Total stack thickness = Σ layer thicknesses

2.1 mm glass + 0.76 mm PVB + 2.1 mm glass = 4.96 mm windshield. Arithmetic plus bend radius check (R_min ≈ t_total · E / σ_yield).

Hand-check; ASTM C1172 (laminated glass).

Reuses the TMM kernel — see /validation#tmm.

Matches a target reflectance profile for a touch panel stack by minimising Σ |R(λ) − R_target|²; the forward model is tmm().

Shared kernel with tmm.

Reuses the TMM kernel for the multilayer mirror design.

Hot/cold mirror design optimises a Bragg stack via tmm() over the visible + NIR; validation inherits tmm parity cases.

Shared kernel with tmm; Macleod ch. 7 (edge filters).

Retardance Δ = (n_e − n_o) · d / λ expressed in waves

For a biaxial PET film with Δn 1.0e-3 at 550 nm thickness 100 µm: Δ = 1e-3 · 1e5 nm / 550 nm = 0.182 waves. See src/components/tools/BirefringenceSimulator.tsx.

Born & Wolf §15; Collett 2005 Field Guide to Polarization.

Louvre acceptance cone from pitch / height / angle

For pitch p, height h, tilt θ: half-cone α = atan((p − t·sin θ)/(h·cos θ)) where t is the louvre thickness; see src/components/tools/PrivacyLouvre.tsx.

3M ALCF datasheets; Peiponen et al. 2019 Opt. Lett. 44, 723.

Cauchy fit recovers known (A, B) within 1e-3

tests/optics/ellipsometry.test.ts generates synthetic Ψ(λ), Δ(λ) from a known Cauchy film on BK7, then fits back — recovered A, B within 1e-3 of input.

Fujiwara 2007, Spectroscopic Ellipsometry; Tompkins & Hilfiker 2015.

Weighted multi-objective optimiser recovers the single-target optimum when one weight dominates

tests/optics/multi-target.test.ts: a two-target run with weight 0.99 on R(550 nm) converges to the same stack as the single-objective AR optimiser (within 0.1 nm per layer).

Dobrowolski 1965 Appl. Opt. 4, 937 (merit-function design).

Pinned against tests/optics/phase.test.ts reference values (GDD at design wavelength matches hand-derived analytic to 1e-4 fs²).

φ(λ) extracted from the TMM amplitude r via arg(r); GDD = d²φ/dω². See phase.test.ts for the reference stack and expected GDD trace.

Macleod §2.5; Diels & Rudolph, Ultrashort Laser Pulse Phenomena 2nd ed.

Reverse-engineering recovers planted (n, d) from synthetic R(λ) to 1 % / 1 nm

tests/optics/reverse-engineering.test.ts: planted stack → synthetic R(λ) → fit returns layer thicknesses within 1 nm and refractive index within 1 % at design wavelength.

Larouche & Martinu 2008 Appl. Opt. 47, C219 (constrained reverse engineering).

Reuses Stoney equation — see /validation#stoney-stress. Thermal stress σ_th = (α_f − α_s) · ΔT · E_f / (1 − ν_f).

tests/optics/cte-warpage.test.ts asserts the thermal-stress formula and the sign convention (α_f > α_s + heating → tensile).

Timoshenko & Gere, Theory of Elasticity 3rd ed. §12; Freund & Suresh 2003 §2.4.

Reuses D65 / A / C illuminants and the sRGB OETF — see /validation#colour-shift for CIEDE2000 cases.

Luminance follows Y = Σ R(λ)·D(λ)·V(λ)·dλ with CIE 2° observer; a perfect reflector under D65 gives Y = 100 and (x, y) = (0.3127, 0.3290) within 1e-3.

CIE 15:2018; IEC 61966-2-1 (sRGB).

Reuses lightguide-extraction + deck-calibrated guided fraction — see /validation#lightguide-uniformity.

Panoramic Roof's four readouts (guided %, halo decay, off-state ΔE, peak/valley) are built from deckCalibratedGuidedFraction10mm, deckCalibratedAlpha, and evaluateUniformityScore.

Shared kernel; Kriya + Eastman deck 2026-10-22.

Reuses the grating-diffraction kernel + Kogelnik η — see /validation#grating-diffraction.

chromaticWalkoffMmPerNm, kogelnikEfficiency, nBounceSurvivalFraction all share tests/optics/grating-diffraction.test.ts coverage. HoloLens-regime Λ=400 nm, n=1.85, t=2 mm, 45°: 0.015 mm/nm per bounce (≈16 mm total R-G walk-off over 10 bounces in an uncompensated system).

Shared kernel with grating-diffraction.