Pith. sign in

REVIEW 3 major objections 6 minor 71 references

A single open-vocabulary flow model designs 2–100-layer optical coatings from any material curves and wavelength grid, without retraining.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

A joint discrete–continuous flow-matching model designs open-vocabulary multilayer optical coatings from query-time n/k curves and spectra, validated on 224 tasks and four fabricated coolers.

T0 review reviewed 2026-07-10 challenge →

load-bearing objection Solid methods paper: open-vocabulary n/k tokens + joint discrete–continuous flow matching for coatings, with real fabrication; best-of-N TMM ranking is the main caveat, not a collapse of the claim. the 3 major comments →

arxiv 2607.08392 v1 pith:334GDQQG submitted 2026-07-09 physics.optics cs.LG

Joint Discrete-Continuous Flow Matching for Open-Vocabulary Inverse Design of Multilayer Optical Coatings

Zhiyi Li , Yuheng Jin , Yidan Huang , Nan Chen , Hongyan Fu , Yikun Bu This is my paper
classification physics.optics cs.LG
keywords query-based inverse designopen-vocabulary modelsflow matchingjoint discrete–continuous flow matchingmultilayer optical coatingsphotonicstransfer-matrix method
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

Amortized neural inverse design usually freezes the material menu, the wavelength grid, and a thickness quantization into the model, so a new film or a new band forces retraining. This paper argues that multilayer optical coatings need not pay that closed-world price. IrisFlow is a single 136M-parameter model that accepts, at query time, a target reflectance/transmittance spectrum, a wavelength grid, a local bank of material n(λ),k(λ) curves, and a requested layer count. Materials enter as wavelength-aware optical tokens rather than learned IDs; material sequences are generated by discrete flow matching over the query bank and thicknesses by continuous flow matching, jointly under one shared denoising backbone. The same checkpoint reconstructs in-distribution targets, keeps same-order accuracy on a held-out 15-material bank, handles bands beyond the training envelope, and designs four color-displaying radiative coolers that were fabricated and measured. If the interface works, a coating facility can drop in chamber-calibrated optical constants and get usable stacks without rebuilding the model.

Core claim

IrisFlow shows that open-vocabulary, query-conditioned joint discrete–continuous flow matching can amortize inverse design of multilayer optical coatings: one 136M-parameter checkpoint designs 2–100-layer stacks from user-supplied spectra, wavelength grids, and candidate n(λ),k(λ) curves, reconstructs in-distribution targets at median combined R,T RMSE about 4.6×10⁻², retains same-order accuracy on a held-out material bank without retraining, and produces process-calibrated cooler designs that were fabricated.

What carries the argument

Joint discrete–continuous flow matching under a shared denoising backbone: materials are scored by discrete flow matching (uniform CTMC) against a query-local bank of wavelength-aware n,k tokens; thicknesses are integrated by continuous flow matching without discretization; both heads condition on the full joint noisy state and the query.

Load-bearing premise

That ranking many random draws by exact transfer-matrix re-simulation, plus human or cooling-aware selection from that pool, is a fair stand-in for what the generative model itself delivers when a new material curve is simply plugged in.

What would settle it

Supply a genuinely novel chamber-measured n,k curve never near the training neighborhood, freeze the same checkpoint and best-of-N budget, and check whether the best TMM-verified design still lands within roughly 2–3× the matched in-distribution RMSE and, after deposition, within the claimed color and solar-NIR tolerances without swapping stacks or relaxing the selection rule.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

3 major / 6 minor

Summary. IrisFlow is a single 136M-parameter query-conditioned generative model for multilayer optical coatings that treats materials as wavelength-aware n(λ),k(λ) tokens (not fixed IDs), samples material sequences by discrete flow matching over a query-local candidate bank, and samples thicknesses by continuous flow matching without discretization. A staged curriculum trains one checkpoint for 2–100 layers. On a 224-task suite the model reconstructs in-distribution targets (median combined R,T RMSE ~4.6×10⁻²), retains same-order accuracy on a 15-material held-out bank without retraining (median matched-cell OOD/in-distribution ratio ~1.67×), extrapolates wavelength support beyond the 380–1400 nm training envelope, designs against analytic application targets, and outperforms OptoGPT on an 11-case shared subset using OptoGPT’s library as OOD curves. With process-corrected optical constants, four color-displaying coolers were designed and fabricated; the three chromatic devices reach CIEDE2000 3.1–5.2 with 93–95% solar NIR reflectance.

Significance. If the results hold, this is a substantial interface contribution for amortized inverse design in photonics: open-vocabulary material entry via optical curves, query-conditioned wavelength grids, continuous thicknesses jointly generated with discrete material choices, and a single model spanning 2–100 layers. The evaluation is unusually thorough for the area (matched-cell OOD bank, nearest-training-sample audit against ~114M spectra, wavelength extrapolation, classical-optimizer budget comparison, partial-structure clamping, angle/polarization via bank tilt, and a fabrication loop). Appendix A’s consistency account of the joint discrete–continuous flow, the diversity analysis, and the process-corrected n/k fabrication path are genuine strengths. The work is relevant beyond coatings to other coupled discrete–continuous design problems with facility-local component banks.

major comments (3)
  1. Methods §4.10 and Appendix §G.2: essentially every headline fidelity number (Tiers 1–4, OptoGPT head-to-head, cooler pools) is best-of-N under oracle TMM re-simulation (N=100 or 500). Appendix §D.2 shows a single deterministic draw is far worse (mean RMSE ~0.59 vs ~0.11 at N=100). This is defensible for a one-to-many inverse map and is partially framed in Discussion as generator + cheap TMM filter, but the Abstract and Results still read as if the generative model alone delivers the reported RMSE. Please state the evaluation mode explicitly in the Abstract/Results, and report single-draw (or small-N) medians alongside best-of-N for at least Tier 1 and Tier 2 so readers can separate generative quality from selection quality.
  2. Results §2.8 and Appendix §N: the hardware claim is that open-vocabulary design is “carried through to fabricated coatings,” yet (i) designs are selected by a cooling-aware / practicality rule outside the model, (ii) selected designs sit at the ∆E00≤3 boundary with no deposition headroom, (iii) measured magenta is ∆E00=5.2 (above the stated tolerance), and (iv) yellow was swapped for a simpler in-pool stack. Process-corrected n/k is a real open-vocabulary strength; please qualify the fabrication claim to match the measured outcomes (e.g., three chromatic coolers with stated color/cooling metrics, one limiting near-black case) and separate model generation from post-hoc selection more clearly in the main text.
  3. Appendix §K / Table 2: the OptoGPT comparison is valuable as an OOD-material transfer test, but total draw budgets differ (500 for OptoGPT vs 19×500 for IrisFlow’s layer sweep) and models are not parameter-matched. The paper argues per-configuration N=500 matching and reports sizes; still, the 10–1 “wins” framing in Results §2.5 overstates a protocol that favors IrisFlow’s controllable depth axis. Soften the win language and lead with median/mean RMSE and the OOD-bank aspect rather than pairwise case counts.
minor comments (6)
  1. Figure 2 heatmaps: cell text is RMSE×10²; state this explicitly in the caption (it is easy to misread as raw RMSE).
  2. Table 1 vs Appendix Table I1: AR-5 handling (median of two variants) should be footnoted in the main-text table for consistency with §G.2.
  3. Methods §4.6: λ_st=0.4 and λ_soft=0.1 are fixed from a pilot; a one-sentence sensitivity note (or pointer to any ablation) would help reproducibility.
  4. Appendix §P: the p-polarization admittance proxy residual is quantified in SI but only briefly mentioned in Discussion; a short main-text sentence on residual magnitude would help practitioners.
  5. Code/data availability is “upon reasonable request.” For a methods paper of this type, releasing at least the benchmark case registry, effective n/k libraries, and evaluation scripts would strengthen the contribution.
  6. Minor notation: “h4 (LaTiO3)” appears without first expansion in some early figures; define once in main text.

Circularity Check

0 steps flagged

No significant circularity: TMM is the shared physics simulator for data and scoring, not a self-definitional reduction of the inverse map.

full rationale

IrisFlow is an amortized generative model trained on TMM-simulated stacks and evaluated by re-simulating its outputs with the same TMM solver. That is standard for inverse design and is not circular: the model never receives the answer stack as input; it receives only the query (target spectrum, wavelength grid, candidate n/k bank, layer count) and samples material sequences plus continuous thicknesses. Tier 3 analytic targets are not TMM-realized by construction, the nearest-training-sample audit shows the model improves on corpus retrieval rather than memorizing, and the fabrication loop uses process-remeasured n/k plus measured CIEDE2000 / solar-NIR reflectance as external checks. Best-of-N oracle ranking by TMM RMSE and post-hoc cooling/practicality selection are evaluation and deployment choices that inflate reported fidelity relative to a single draw; they do not make the generative claim equivalent to its inputs by definition. No self-definitional identity, fitted-parameter-as-prediction, load-bearing self-citation uniqueness theorem, or renamed known result was found in the derivation chain. Residual mild concern is only that headline numbers are selection-quality as much as generative quality, which is an evaluation caveat rather than circularity.

Axiom & Free-Parameter Ledger

8 free parameters · 6 axioms · 2 invented entities

The central claims rest on standard TMM optics and published flow-matching/DFM machinery, plus engineering choices (loss weights, curriculum, sampling budgets, fabrication selection rules) that are free parameters of the method rather than derived constants. No new physical particles or forces are postulated; the main invented objects are architectural (optical-curve tokens, joint denoiser interface).

free parameters (8)
  • stack loss weight λ_st = 0.4
    Fixed at 0.4 from a task-uncertainty-weighting pilot; balances continuous thickness FM loss against discrete material DFM loss and directly shapes joint training.
  • CTMC-soft reweight λ_soft = 0.1
    Set to 0.1 in the reported model; reweights nearly-clean discrete draws in the material objective.
  • discrete noise floor α_end = 1e-4
    Log-linear CTMC schedule anchors t=1 within 1e-4 of uniform over the local bank; a schedule hyperparameter of the discrete process.
  • thickness fabrication window [d_min, d_max] = 5–300 nm
    Affine map of physical thicknesses to model scale; hard constraint on generated designs (5–300 nm).
  • best-of-N evaluation budgets = 100 / 500
    N=100 (Tiers 1–2) and N=500 (Tiers 3–4 / OptoGPT); oracle selection strongly affects reported fidelity and is a free evaluation choice.
  • sampler reverse steps and adaptive-power p = 15 steps, p=2
    15-step adaptive-power schedule (p=2) chosen from ablations; changes candidate quality vs compute tradeoff.
  • cooler color tolerance ∆E00 and selection rule = ∆E00≤3.0; cooling-aware
    Cooling-aware selection (max solar NIR reflectance subject to ∆E00≤3.0) and practicality swap for yellow determine which designs are fabricated and thus the hardware claim.
  • process-corrected GSST-a visible n,k segment = chamber-remeasured visible segment
    Witness-film fit replaces handbook GSST-a over the visible; designs and measured transfer depend on this calibrated curve.
axioms (6)
  • domain assumption Transfer-matrix method (TMM) exactly computes R(λ),T(λ) for stratified stacks under the stated incidence/substrate assumptions.
    All training labels, evaluation scores, and design verification rest on TMM (Introduction; Methods §4.1, §4.10).
  • standard math Continuous flow matching with linear path and velocity target recovers the conditional mean velocity integrated by the reverse ODE at population optimum.
    Thickness branch cites Lipman et al. / Albergo et al.; Appendix A states consistency with the sampler.
  • standard math Uniform-kernel CTMC discrete flow matching with denoiser parameterization yields valid reverse categorical updates from clean-material posteriors.
    Material branch follows Campbell/Gat-style DFM; product-form reverse kernel is an explicit mean-field approximation (Appendix A.5).
  • domain assumption A material’s design-relevant identity is carried by its n(λ),k(λ) on the query grid sufficiently for open-vocabulary transfer without class embeddings.
    Core open-vocabulary claim; tested via Tier 2 held-out bank and process-corrected fabrication, not proved in general.
  • ad hoc to paper Staged layer-count curriculum with replay preserves short-stack competence while learning deep stacks.
    Training procedure choice (Methods §4.8; Appendix C); supported by checkpoint curves but not a universal necessity.
  • domain assumption Oblique/polarized design can be steered by analytic n/k bank tilt with exact angled-TMM ranking, without retraining.
    Appendix P; exact for s-pol, approximate admittance proxy for p-pol.
invented entities (2)
  • Wavelength-aware optical-curve tokens (spectrum and candidate n/k memory) independent evidence
    purpose: Replace fixed material-ID embeddings so candidates enter as functions on the user’s wavelength grid and are scored locally.
    Architectural construct central to the open-vocabulary interface; independent evidence is empirical (family clustering in Appendix Q, Tier 2 transfer), not a new physical object.
  • IrisFlow joint discrete–continuous denoiser interface independent evidence
    purpose: Couple material CTMC state and continuous thickness state under shared flow time and one Transformer backbone for mixed-integer coating design.
    Method-level invention; validated by benchmarks and fabrication rather than external prior measurement of the same system.

reviewed 2026-07-10 · how reviews work

0 comments
Cite this review

Pith. "Pith review of Joint Discrete-Continuous Flow Matching for Open-Vocabulary Inverse Design of Multilayer Optical Coatings." pith.science (2026). https://pith.science/paper/334GDQQG

@misc{pith2026260708392,
  author       = {Pith},
  title        = {Pith review of: Joint Discrete-Continuous Flow Matching for Open-Vocabulary Inverse Design of Multilayer Optical Coatings},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/334GDQQG}},
  note         = {Machine review of arXiv:2607.08392}
}
Share X Bluesky LinkedIn Reddit HN
read the original abstract

Amortized neural inverse design typically remains closed-world: component choices are fixed vocabulary tokens, coordinate grids are frozen at training time, and continuous variables are discretized into sequence tokens. Multilayer optical coatings are an industrially important instance, coupling material sequence, layer thickness and wavelength-dependent response. We present IrisFlow, a query-based, open-vocabulary flow-matching framework instantiated in coatings: the target reflectance/transmittance spectrum, wavelength grid, candidate-material optical constants and layer count are supplied at query time. Candidate materials enter as wavelength-aware optical tokens rather than learned identities; material sequences are sampled by discrete flow matching over the query's candidate bank, thicknesses by continuous flow matching without discretization. A single 136M-parameter model designs 2-100-layer stacks. Across a 224-task benchmark it reconstructs in-distribution targets faithfully and retains same-order accuracy on a 15-material held-out bank without retraining; it reconstructs bands up to 1100 nm beyond its training envelope, designs against analytic application specifications and outperforms an autoregressive baseline on that baseline's material library. With optical constants calibrated to our deposition process, IrisFlow designs four color-displaying coolers, fabricated by ion-assisted evaporation: the three chromatic devices reach a CIEDE2000 color error of 3.1-5.2 while retaining 93-95% solar near-infrared reflectance, demonstrating open-vocabulary design carried through to fabricated coatings.

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

71 extracted references · 71 canonical work pages · 3 internal anchors

  1. [1]

    A.Thin-Film Optical Filters5th edn (CRC Press, Boca Raton, 2017)

    Macleod, H. A.Thin-Film Optical Filters5th edn (CRC Press, Boca Raton, 2017)

    work page 2017

  2. [2]

    V., Trubetskov, M

    Tikhonravov, A. V., Trubetskov, M. K. & DeBell, G. W. Application of the needle optimization technique to the design of optical coatings.Applied Optics 35, 5493–5508 (1996)

    work page 1996

  3. [3]

    Dobrowolski, J. A. & Kemp, R. A. Refinement of optical multilayer systems with different optimization procedures.Applied Optics29, 2876–2893 (1990). 24

    work page 1990

  4. [4]

    & Schoenauer, M

    Martin, S., Rivory, J. & Schoenauer, M. Synthesis of optical multilayer systems using genetic algorithms.Applied Optics34, 2247–2254 (1995)

    work page 1995

  5. [5]

    Rabady, R. I. & Ababneh, A. Global optimal design of optical multilayer thin-film filters using particle swarm optimization.Optik125, 548–553 (2014)

    work page 2014

  6. [6]

    & Price, K

    Storn, R. & Price, K. Differential evolution – a simple and efficient heuristic for global optimization over continuous spaces.Journal of Global Optimization11, 341–359 (1997)

    work page 1997

  7. [7]

    Peurifoy, J.et al.Nanophotonic particle simulation and inverse design using artificial neural networks.Science Advances4, eaar4206 (2018)

    work page 2018

  8. [8]

    Molesky, S.et al.Inverse design in nanophotonics.Nature Photonics12, 659–670 (2018)

    work page 2018

  9. [9]

    & Rho, J

    So, S., Badloe, T., Noh, J., Bravo-Abad, J. & Rho, J. Deep learning enabled inverse design in nanophotonics.Nanophotonics9, 1041–1057 (2020)

    work page 2020

  10. [10]

    & Guo, L

    Ma, T., Ma, M. & Guo, L. J. Optical multilayer thin film structure inverse design: from optimization to deep learning.iScience28, 112222 (2025)

    work page 2025

  11. [11]

    Liu, D., Tan, Y., Khoram, E. & Yu, Z. Training deep neural networks for the inverse design of nanophotonic structures.ACS Photonics5, 1365–1369 (2018)

    work page 2018

  12. [12]

    & Strangi, G

    Lininger, A., Hinczewski, M. & Strangi, G. General inverse design of layered thin- film materials with convolutional neural networks.ACS Photonics8, 3641–3650 (2021)

    work page 2021

  13. [13]

    & Zheng, Y

    Unni, R., Yao, K. & Zheng, Y. Deep convolutional mixture density network for inverse design of layered photonic structures.ACS Photonics7, 2703–2712 (2020)

    work page 2020

  14. [14]

    & Liu, Y

    Ma, W., Cheng, F., Xu, Y., Wen, Q. & Liu, Y. Probabilistic representation and inverse design of metamaterials based on a deep generative model with semi- supervised learning strategy.Advanced Materials31, 1901111 (2019)

    work page 2019

  15. [15]

    & Marquardt, F

    Luce, A., Mahdavi, A., Wankerl, H. & Marquardt, F. Investigation of inverse design of multilayer thin-films with conditional invertible neural networks. Machine Learning: Science and Technology4, 015014 (2023)

    work page 2023

  16. [16]

    & Guo, L

    Ma, T., Wang, H. & Guo, L. J. OptoGPT: a foundation model for inverse design in optical multilayer thin film structures.Opto-Electronic Advances7, 240062 (2024)

    work page 2024

  17. [17]

    Sahoo, S. S.et al. Simple and effective masked diffusion language models. In Advances in Neural Information Processing Systems, Vol. 37, 130136–130184 (2024). 25

    work page 2024

  18. [18]

    Preprint at arXiv:2604.01106 (2026)

    Schaible, J.et al.Inverse design of optical multilayer thin films using robust masked diffusion models. Preprint at arXiv:2604.01106 (2026)

    work page arXiv 2026

  19. [19]

    & Guo, L

    Wang, H., Zheng, Z., Ji, C. & Guo, L. J. Automated multi-layer optical design via deep reinforcement learning.Machine Learning: Science and Technology2, 025013 (2021)

    work page 2021

  20. [20]

    L., Mahdavi, A., Eichler, C

    Wankerl, H., Stern, M. L., Mahdavi, A., Eichler, C. & Lang, E. W. Parameterized reinforcement learning for optical system optimization.Journal of Physics D: Applied Physics54, 305104 (2021)

    work page 2021

  21. [21]

    Wu, H.et al.Dual-decoder transformer for inverse design of optical multilayer films.Intelligent Opto-Electronics2, 250018 (2026)

    work page 2026

  22. [22]

    Wang, R., Xue, R., Chen, B. & Wu, H. PRISM: position-encoded regres- sive inverse spectral model for multilayer thin-film design. Preprint at arXiv:2605.26502 (2026)

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  23. [23]

    Lipman, Y., Chen, R. T. Q., Ben-Hamu, H., Nickel, M. & Le, M.Flow matching for generative modeling. InInternational Conference on Learning Representations (2023)

    work page 2023

  24. [24]

    Stochastic Interpolants: A Unifying Framework for Flows and Diffusions

    Albergo, M. S., Boffi, N. M. & Vanden-Eijnden, E. Stochastic interpolants: a unifying framework for flows and diffusions. Preprint at arXiv:2303.08797 (2023)

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  25. [25]

    A continuous time framework for discrete denoising models

    Campbell, A.et al. A continuous time framework for discrete denoising models. InAdvances in Neural Information Processing Systems, Vol. 35, 28266–28279 (2022)

    work page 2022

  26. [26]

    Discrete flow matching

    Gat, I.et al. Discrete flow matching. InAdvances in Neural Information Processing Systems, Vol. 37, 133345–133385 (2024)

    work page 2024

  27. [27]

    & Jaakkola, T.Generative flows on discrete state-spaces: enabling multimodal flows with applications to protein co- design

    Campbell, A., Yim, J., Barzilay, R., Rainforth, T. & Jaakkola, T.Generative flows on discrete state-spaces: enabling multimodal flows with applications to protein co- design. InProceedings of the 41st International Conference on Machine Learning, Vol. 235 ofProceedings of Machine Learning Research, 5453–5512 (2024)

    work page 2024

  28. [28]

    P., Anoma, M

    Raman, A. P., Anoma, M. A., Zhu, L., Rephaeli, E. & Fan, S. Passive radiative cooling below ambient air temperature under direct sunlight.Nature515, 540– 544 (2014)

    work page 2014

  29. [29]

    & Fan, S

    Yin, X., Yang, R., Tan, G. & Fan, S. Terrestrial radiative cooling: using the cold universe as a renewable and sustainable energy source.Science370, 786–791 (2020)

    work page 2020

  30. [30]

    J., Kim, Y

    Lee, G. J., Kim, Y. J., Kim, H. M., Yoo, Y. J. & Song, Y. M. Colored, day- time radiative coolers with thin-film resonators for aesthetic purposes.Advanced 26 Optical Materials6, 1800707 (2018)

    work page 2018

  31. [31]

    Chen, Y.et al.Colored and paintable bilayer coatings with high solar-infrared reflectance for efficient cooling.Science Advances6, eaaz5413 (2020)

    work page 2020

  32. [32]

    A., Blandre, E., Joulain, K

    Yal¸ cın, R. A., Blandre, E., Joulain, K. & Dr´ evillon, J. Colored radiative cooling coatings with nanoparticles.ACS Photonics7, 1312–1322 (2020)

    work page 2020

  33. [33]

    L.et al.De novo design of protein structure and function with RFdiffusion.Nature620, 1089–1100 (2023)

    Watson, J. L.et al.De novo design of protein structure and function with RFdiffusion.Nature620, 1089–1100 (2023)

    work page 2023

  34. [34]

    Zeni, C.et al.A generative model for inorganic materials design.Nature639, 624–632 (2025)

    work page 2025

  35. [35]

    Attention is all you need

    Vaswani, A.et al. Attention is all you need. InAdvances in Neural Information Processing Systems, Vol. 30 (2017)

    work page 2017

  36. [36]

    R., Pawar, S

    Henry, A., Dachapally, P. R., Pawar, S. S. & Chen, Y.Query-key normalization for transformers. InFindings of the Association for Computational Linguistics: EMNLP 2020, 4246–4253 (2020)

    work page 2020

  37. [37]

    & Courville, A.FiLM: visual rea- soning with a general conditioning layer

    Perez, E., Strub, F., de Vries, H., Dumoulin, V. & Courville, A.FiLM: visual rea- soning with a general conditioning layer. InProceedings of the AAAI Conference on Artificial Intelligence, Vol. 32 (2018)

    work page 2018

  38. [38]

    Neurocomputing568, 127063 (2024)

    Su, J.et al.RoFormer: enhanced transformer with rotary position embedding. Neurocomputing568, 127063 (2024)

    work page 2024

  39. [39]

    D., Ho, J., Tarlow, D

    Austin, J., Johnson, D. D., Ho, J., Tarlow, D. & van den Berg, R.Struc- tured denoising diffusion models in discrete state-spaces. InAdvances in Neural Information Processing Systems, Vol. 34, 17981–17993 (2021)

    work page 2021

  40. [40]

    & Cipolla, R.Multi-task learning using uncertainty to weigh losses for scene geometry and semantics

    Kendall, A., Gal, Y. & Cipolla, R.Multi-task learning using uncertainty to weigh losses for scene geometry and semantics. InProceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 7482–7491 (2018)

    work page 2018

  41. [41]

    & Hutter, F.Decoupled weight decay regularization

    Loshchilov, I. & Hutter, F.Decoupled weight decay regularization. InInternational Conference on Learning Representations(2019)

    work page 2019

  42. [42]

    & Dalal, E

    Sharma, G., Wu, W. & Dalal, E. N. The CIEDE2000 color-difference formula: implementation notes, supplementary test data, and mathematical observations. Color Research & Application30, 21–30 (2005)

    work page 2005

  43. [43]

    ASTM G173-03(2012): standard tables for reference solar spectral irradiances: direct normal and hemispherical on 37°tilted surface

    ASTM International. ASTM G173-03(2012): standard tables for reference solar spectral irradiances: direct normal and hemispherical on 37°tilted surface. ASTM International, West Conshohocken, PA (2012). 27

    work page 2012

  44. [44]

    Powell, M. J. D. An efficient method for finding the minimum of a function of several variables without calculating derivatives.The Computer Journal7, 155–162 (1964)

    work page 1964

  45. [45]

    GLU Variants Improve Transformer

    Shazeer, N. GLU variants improve transformer. Preprint at arXiv:2002.05202 (2020)

    work page internal anchor Pith review Pith/arXiv arXiv 2002

  46. [46]

    Polyanskiy, M. N. Refractiveindex.info database of optical constants.Scientific Data11, 94 (2024)

    work page 2024

  47. [47]

    Aguilar, O., de Castro, S., Godoy, M. P. F. & Rebello Sousa Dias, M. Optoelec- tronic characterization of Zn 1−xCdxO thin films as an alternative to photonic crystals in organic solar cells.Optical Materials Express9, 3638–3648 (2019)

    work page 2019

  48. [48]

    Vangelista, S.et al.Structural, chemical and optical properties of cerium dioxide film prepared by atomic layer deposition on TiN and Si substrates.Thin Solid Films636, 78–84 (2017)

    work page 2017

  49. [49]

    Measurement of the optical constants of yttrium oxide.Japanese Journal of Applied Physics7, 404–408 (1968)

    Nigara, Y. Measurement of the optical constants of yttrium oxide.Japanese Journal of Applied Physics7, 404–408 (1968)

    work page 1968

  50. [50]

    & Liu, L

    Zhang, X., Qiu, J., Li, X., Zhao, J. & Liu, L. Complex refractive indices mea- surements of polymers in visible and near-infrared bands.Applied Optics59, 2337–2344 (2020)

    work page 2020

  51. [51]

    Werner, W. S. M., Glantschnig, K. & Ambrosch-Draxl, C. Optical constants and inelastic electron-scattering data for 17 elemental metals.Journal of Physical and Chemical Reference Data38, 1013–1092 (2009)

    work page 2009

  52. [52]

    & Rubin, M

    von Rottkay, K. & Rubin, M. Optical indices of pyrolytic tin-oxide glass.MRS Proceedings426, 449 (1996)

    work page 1996

  53. [53]

    Minenkov, A.et al.Monitoring the electrochemical failure of indium tin oxide electrodes via operando ellipsometry complemented by electron microscopy and spectroscopy.ACS Applied Materials & Interfaces16, 9517–9531 (2024)

    work page 2024

  54. [54]

    E.et al.Optical design and fabrication of fully sputtered CdTe/CdS solar cells.Journal of Physics: Conference Series286, 012038 (2011)

    Treharne, R. E.et al.Optical design and fabrication of fully sputtered CdTe/CdS solar cells.Journal of Physics: Conference Series286, 012038 (2011)

    work page 2011

  55. [55]

    Synowicki, R. A. & Tiwald, T. E. Optical properties of bulk c-ZrO 2, c-MgO and a-As2S3 determined by variable angle spectroscopic ellipsometry.Thin Solid Films455–456, 248–255 (2004)

    work page 2004

  56. [56]

    & Eberhart, R.Particle swarm optimization

    Kennedy, J. & Eberhart, R.Particle swarm optimization. InProceedings of ICNN’95 — International Conference on Neural Networks, Vol. 4, 1942–1948 (IEEE, 1995). 28

    work page 1942

  57. [57]

    & Kennedy, J

    Clerc, M. & Kennedy, J. The particle swarm — explosion, stability, and conver- gence in a multidimensional complex space.IEEE Transactions on Evolutionary Computation6, 58–73 (2002)

    work page 2002

  58. [58]

    Kolobov, A. V. & Elliott, S. R. Photodoping of amorphous chalcogenides by metals.Advances in Physics40, 625–684 (1991)

    work page 1991

  59. [59]

    Zhang, Y.et al.Broadband transparent optical phase change materials for high- performance nonvolatile photonics.Nature Communications10, 4279 (2019)

    work page 2019

  60. [60]

    CIE 015:2018 Colorimetry, 4th Edition

    CIE TC 1-85. CIE 015:2018 Colorimetry, 4th Edition. International Commission on Illumination (CIE), Vienna (2018)

    work page 2018

  61. [61]

    IEC 61966-2-1:1999: multimedia sys- tems and equipment — colour measurement and management — part 2-1: colour management — default RGB colour space — sRGB

    International Electrotechnical Commission. IEC 61966-2-1:1999: multimedia sys- tems and equipment — colour measurement and management — part 2-1: colour management — default RGB colour space — sRGB. IEC, Geneva (1999). URL https://webstore.iec.ch/en/publication/6169. Appendix / SI The SI appendices are organized as follows.§A gives the formal account, def...

    work page 1999

  62. [62]

    Better vs 1000

    (×4.0), dropout 0.1 Normalization / position query–key normalization; RoPE over layer index (base 1000) Target-spectrum encoder curve-token encoder (flat mode): multilayer perceptron (MLP; hidden 1024, depth 2) over the flattened 128-point two-channel curve plus per-point wavelength features→one conditioning token; 2 channels (R, T) Candidaten, kencoder s...

    work page 1907

  63. [63]

    # targets

    local table 310.0– 2500.0 380–1400 1.2149– 2.1059 0.0095– 0.2092 HfO2 High-index oxides Measured in our laboratory This work local measured table 410.0– 2500.0 380–1400 1.9540– 2.0566 0.0060– 0.0060 PMMA Polymers Zhang-Mitsubishi, refractiveindex.info [50]organic/(C5H8O2)n - poly(methyl methacrylate)/nk/ Zhang-Mitsubishi.yml 400.0– 19900.0 380–1400 1.4770...

    work page 2092

  64. [64]

    (insertion plus local refinement), GA [4] (population 100×10,000 generations), PSO [56] (swarm 100×10,000 iterations, constriction form [57]) and uniform ran- dom search (10 6 trials). All methods share the identical targetR, Ton the same 400–1400 nm 128-point grid, the same five-material candidate pool (Ag, SiO 2, Ti, the process-corrected GSST-a curve o...

    work page

  65. [65]

    Open material vocabulary: a new candidate material enters as data (ann, kcurve or simulator table) at query/run time, with no retraining or vocabulary change

    work page

  66. [66]

    Wavelength-grid flexibility: targets may be specified on arbitrary user grids, including sub-bands and stitched non-contiguous bands, without retraining

    work page

  67. [67]

    Continuous thickness: layer thickness is a continuous output coordinate, not snapped to a discretization grid by the method itself

    work page

  68. [68]

    Variable layer count (one model): a single model or procedure serves variable stack depths natively

    work page

  69. [69]

    Constrained / partial queries: an arbitrary subset of layer materials and/or thick- nesses can be pinned exactly while the method completes the rest, with the free coordinates conditioned on the pinned ones

    work page

  70. [70]

    Solution diversity: one query yields multiple structurally distinct candidate designs

    work page

  71. [71]

    Direct-inverse nets, 71 diversity = 1

    Amortized inference: designs are produced by forward inference without a per- target optimization run. Table M1 Interface-capability rubric(0 = absent, 1 = partial, 2 = full). Axis IrisFlow OptoGPT Direct-inverse nets Classical optimizers Open material vocabulary 2 0 0 2 Wavelength-grid flexibility 2 0 0 2 Continuous thickness 2 0 2 2 Variable layer count...

    work page

This paper was first reviewed by grok-4.5 on July 10, 2026.