Pith. sign in

REVIEW 3 major objections 6 minor 12 references

Scalable freeform optimization of wide-aperture 3D metalenses by zoned discrete axisymmetry

T0 review · 3 major / 6 minor · reviewed 2026-08-10 · deepseek-v4-flash

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

Pith's one-line read Zoned discrete axisymmetry makes full-wave inverse design of centimeter-scale metalenses scale nearly linearly with diameter, without locally periodic approximations.

desk verdict Zoned discrete axisymmetry is a genuine linear-scaling full-wave design idea, but the headline efficiencies rest on an unvalidated zone-independence assumption and need independent confirmation before the state-of-the-art claim is taken at face value. read the letter →

arxiv 2501.07979 v6 pith:BLXACM4M submitted 2025-01-14 physics.optics physics.comp-ph

Mengdi Sun , Ata Shakeri , Arvin Keshvari , Dimitrios Giannakopoulos , Qing Wang , Wei Ting Chen , Steven G. Johnson , Zin Lin This is my paper · ORCID
classification physics.opticsphysics.comp-ph
keywords zoneddiscreteaxisymmetrymetalensinversedesigntopologyoptimizationfull-waveFDTDfreeformmeta-opticsachromaticGPU-acceleratedsimulationcomputationalimaging
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

This paper introduces "zoned discrete axisymmetry" (ZDA), a way to split a wide-aperture metalens into a few concentric radial zones and give each zone its own degree of rotational symmetry, increasing outward. The central claim is that this makes full-wave, freeform inverse design scale nearly linearly with lens diameter, instead of with area, while keeping the full 3D design freedom that continuous axisymmetry sacrifices. On that basis the authors report topology-optimized designs of a 0.8-cm long-wave-infrared lens with 58.4% absolute focusing efficiency, a 1.05-mm RGB-achromatic lens with 33.1% average efficiency, and a 1.96-mm six-wavelength lens with 12% average efficiency, all exceeding their chosen state-of-the-art baselines. A sympathetic reader would care because full-wave Maxwell optimization at millimeter and centimeter scales was previously thought impractical without approximations such as the locally periodic approximation.

What carries the argument

The central mechanism is "zoned discrete axisymmetry" (ZDA): the lens is divided into concentric radial zones of width much larger than the wavelength, and the ℓ-th zone is made periodic in azimuth with period $2\pi/n_{\ell}$, where $n_{\ell}$ grows roughly linearly with radius. Each zone's simulation is a cylindrical-coordinate wedge terminated by absorbing layers, with Bloch mode expansions in the azimuthal coordinate; a differentiable FDTD solver with adjoint-based gradients then performs topology optimization inside each zone independently. This object carries the argument because it reduces the total computational domain from area-scaling to diameter-scaling while retaining genuinely three-dimensional freeform patterns.

What would settle it

Take one of the published designs (for example the 0.8-cm LWIR lens), run a full-lens simulation or experimental measurement with all zones present, and compare the focal-spot efficiency and point-spread function with the zone-decomposed prediction; a discrepancy beyond a few percent in focusing efficiency, or a visible change in focal position, would show the independent-zone assumption fails at the relevant numerical aperture.

Watch

Extended reading notes

Core claim

Using a GPU-accelerated finite-difference time-domain solver in cylindrical coordinates, the paper states that each supra-wavelength radial zone can be simulated independently with absorbing boundaries and azimuthal Bloch periodicity, so the simulated volume (and hence computational cost) grows roughly linearly with radius. The n-fold symmetry order n is chosen to grow approximately linearly with radius, keeping the azimuthal period subwavelength and suppressing spurious diffraction orders while adding about ten times more degrees of freedom than continuous axisymmetry. The paper's demonstration designs—a 0.8-cm LWIR lens at NA 0.3, a 1.05-mm RGB lens at NA 0.8, and a 1.96-mm six-wavelength lens at NA 0.3—are claimed to outperform prior large-area metalenses, and the LWIR design beats its continuously axisymmetric counterpart (58.4% vs 25.1% focusing efficiency).

Load-bearing premise

The method assumes that adjacent radial zones interact only weakly through the lens, so each zone can be optimized on its own with artificial absorbing edges; if near-field coupling across zone boundaries is actually significant, the assembled lens will not perform as the zone-by-zone simulations predict.

Editorial extensions

If this is right

  • Full-wave inverse design becomes practical for metalenses thousands of wavelengths across, without invoking the locally periodic approximation.
  • Adding azimuthal degrees of freedom raises achievable focusing efficiency: in the paper's comparison, the discrete-axisymmetry LWIR lens reaches 58.4% versus 25.1% for the continuous-axisymmetry lens under the same parameters.
  • Achromatic and poly-achromatic visible metalenses with diameters around 1600–3000 wavelengths can be designed at average focusing efficiencies of 33% and 12%, outperforming LPA-based designs at similar diameters.
  • All target wavelengths are extracted from one forward and one adjoint time-domain simulation, so adding colors costs little extra compute.
  • The optimized lenses can produce super-oscillatory focal spots (FWHM 1.34λ versus 1.6λ for the ideal Airy disk), which could aid computational super-resolution imaging.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • An implication the authors leave implicit: if inter-zone independence is as clean as claimed, the method could be parallelized across zones almost without limit, so wall-clock time should drop further with more GPUs.
  • A testable extension would apply ZDA to an end-to-end imaging objective rather than focal-spot intensity; the freeform, non-Fresnel patterns typical of computational imaging are precisely where LPA is least reliable, so ZDA's advantage should be largest there.
  • The same zone-with-growing-n construction could be adapted to discrete rotational symmetries for polarization or orbital-angular-momentum sensing, though the paper does not demonstrate those devices.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 6 minor

Summary. The manuscript introduces "zoned discrete axisymmetry" (ZDA) as a symmetry-reduction and domain-decomposition strategy for full-wave inverse design of large-area metalenses. The lens is divided into concentric supra-wavelength annular zones, and the azimuthal symmetry order n is increased roughly linearly with radius so that the simulated wedge area per zone is nearly independent of radius and the total computational cost scales almost linearly with lens diameter. Each zone is simulated independently in a GPU-accelerated cylindrical-coordinate FDTD solver with Bloch boundary conditions and PML terminations, and a differentiable adjoint/automatic-differentiation framework is used for freeform topology optimization. The authors report three designs: a 0.8-cm-diameter LWIR metalens (NA=0.3, 58.4% absolute focusing efficiency), a 1.05-mm RGB-achromatic metalens (NA=0.8, 33.1% average efficiency), and a 1.96-mm six-wavelength metalens (NA=0.3, 12% average efficiency), and they argue that these results outperform state-of-the-art metalenses. Permittivity profiles for all designs are provided as supplementary files.

Significance. If the ZDA approximation is valid, this is a substantial methodological contribution: it offers a route to full-wave, non-LPA topology optimization of millimeter- and centimeter-scale metalenses with near-linear cost scaling, and it provides concrete, reproducible designs with stated efficiencies. The GPU cylindrical-FDTD implementation and the hybrid time/frequency-domain adjoint-AD integration are valuable building blocks for the community. The main caveat is that the reported performance numbers are produced by the same zone-decomposed PML-terminated solver used for optimization, and the accuracy of this decomposition is not independently verified for the specific high-NA, wide-aperture parameter ranges. The significance of the efficiency claims therefore depends on closing that validation gap.

major comments (3)
  1. [Zoned discrete axisymmetry (ZDA); Results and Discussion] The headline efficiencies (58.4% LWIR, 33.1% RGB, 12% six-wavelength) are all computed from the same zone-decomposed, PML-terminated solver that generated the designs. The accuracy of the decomposition is imported from Ref. 13, whose <1% error bound is cited for >10λ-wide zones, but the conditions of that test are not reproduced here. The present designs use zone widths of 25λ and 47λ, thicknesses of roughly 0.6λ–1λ, and NAs up to 0.8, and the assembled lenses contain abrupt changes in the symmetry order n at zone boundaries that are not represented when each zone is independently terminated by PMLs. No full-lens simulation or experimental measurement is reported for any assembled design. Because the "outperform the state of the art" claim rests on these efficiency values, an independent validation—for example, a full-wave simulation of a smaller lens with the same NA and zone widths, or a fabricated and measured sub-aperture—is needed before the central claim is established.
  2. [Table 1; Millimeter-scale poly-achromatic metalenses in the visible] The state-of-the-art comparison in Table 1 is not yet on a common footing. The absolute focusing efficiency is defined in the LWIR section as the fraction of total incident power within 3 FWHMs, but the RGB and six-wavelength efficiencies are reported without restating this definition, and the table does not define the column header R or the normalization (incident vs. transmitted power). Literature entries in Table 1 use a variety of efficiency definitions, so the assertion that the present designs outperform the state of the art needs a side-by-side comparison with the same integration aperture and normalization for each cited work. At minimum, please specify the integration radius in wavelengths for every design and state whether each value is absolute or relative efficiency.
  3. [Centimeter-scale metalens at long-wave infrared] The claim that discrete axisymmetry is "critical" and that the performance gain over continuous axisymmetry is due to the enlarged design space is supported by a single comparison (58.4% vs. 25.1%). The text states that the two designs share diameter, NA, and materials, but it does not report the optimization schedules (number of iterations, filter/binarization ramps, starting points) for the two runs. Without evidence that both designs were optimized to comparable convergence, the attribution of the 2.3x efficiency gap specifically to the additional azimuthal degrees of freedom is not fully established.
minor comments (6)
  1. [Table 1] The column header R is not defined; from the text it appears to denote lens radius rather than diameter, and the caption should state this explicitly.
  2. [Differentiable FDTD in cylindrical coordinates] The implementation of the continuously axisymmetric first zone (n approaching infinity) is not described; please clarify how this limit is represented in the Bloch-mode FDTD solver and how the m=±1 excitation is handled near r=0.
  3. [Introduction and Abstract] The term "full-wave" is used for a method that still relies on the zone-independence approximation; please qualify it (e.g., "full-wave within each zone") so that readers do not infer that the assembled lens is simulated without any domain-decomposition approximation.
  4. [Millimeter-scale poly-achromatic metalenses in the visible] There are two occurrences of "for for" in this subsection; please correct the typos.
  5. [Centimeter-scale metalens at long-wave infrared] The paper does not state how the FDTD throughput of one billion voxels per second was measured; a sentence describing the benchmark procedure would improve reproducibility.
  6. [Results and Discussion] The text refers to the optimized variables as "pixels" and to the designs as "3D freeform" and "volumetric"; please clarify explicitly whether the permittivity varies along z within the metalens layer or whether the design is a 2D patterning in (r, phi) with fixed thickness.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the ZDA scaling is a consequence of the construction, the reported efficiencies are simulation outputs rather than fitted inputs, and the one self-cited decomposition benchmark is an externally falsifiable published numerical test.

full rationale

The derivation chain is not circular. The ZDA linear-scaling claim follows directly from the construction itself: with n increasing roughly proportionally to radius, the simulated wedge area in each zone stays approximately constant, so the total simulated area grows linearly with diameter; the paper also reports concrete timings consistent with that scaling, making this an analytic property plus a benchmark, not a circular prediction. The reported focusing efficiencies (58.4%, 33.1%, 12%) are full-wave FDTD outputs of the topology-optimized permittivity profiles, not fitted parameters: the optimization objective is the focal intensity, while the reported metric is encircled power within 3 FWHMs, so the two are not identical by construction, and the design files are released for independent simulation. The one load-bearing self-citation is the quantitative error claim for zone-PML decomposition, imported from Ref. 13 (Lin & Johnson). That cited result is a published numerical comparison against full simulation, i.e., a parameter-free, externally falsifiable test, and the present zone widths (25–47λ) satisfy its stated >10λ condition, so under the review rules it counts as real evidence rather than circularity. The absence of a monolithic full-lens simulation or experiment at NA=0.8 in the visible is a legitimate validation/completeness caveat, but it does not reduce any equation to its own input or rename a fitted parameter as a prediction.

Assumptions & free parameters 4 free parameters · 4 assumptions · 0 invented entities

No new physical entities are proposed. The framework introduces a computational symmetry constraint (ZDA) rather than a new force, particle, or material. The load-bearing content is the locality assumption for supra-wavelength zones, imported from the authors' earlier work, and the hand-selected symmetry schedule, both listed above.

free parameters (4)
  • Symmetry schedule n(ℓ) = C·ℓ for zone ℓ = C=720 (LWIR), C=360 (RGB), C=1440 (six-wavelength)
    Chosen by hand so the azimuthal period r·τ_φ stays below λ, suppressing propagating higher diffraction orders while keeping per-zone simulated wedge area roughly constant.
  • Zone radial width = 25λ (LWIR and RGB), 47λ (six-wavelength)
    Selected as supra-wavelength widths; the locality justification from ref 13 requires >10λ, and the chosen values affect the number of zones and cost.
  • Conic filter radius = 0.08λ (LWIR), 0.056λ (visible designs)
    Hand-chosen length scale for binarization and manufacturability; influences achievable efficiency and feature size.
  • Spatial discretization = δr=δz=λ/50 and rδϕ<0.021λ (LWIR); δ=λ/36 (visible)
    Convergence settings chosen per standard FDTD practice; not fitted to performance but determines the effective design DOF count.
assumptions (4)
  • standard math Bloch theorem applies to n-fold azimuthally periodic structures, permitting the field expansion E = Σ_m u_m(r,ϕ,z) e^{imϕ}.
    Invoked in Eqs. (1)-(5); standard Fourier-Bloch decomposition in cylindrical coordinates.
  • domain assumption Near-field inter-zone coupling is negligible when transverse zone size is much larger than lens thickness; PML-terminated zones reproduce full-lens far field to <1% error.
    Stated in the ZDA section and justified by the authors' prior ref 13; underpins all domain decomposition and final assembled-performance estimates.
  • domain assumption Choosing r·τ_ϕ < λ prevents higher-order angular-momentum diffraction orders (m' = m + k n) from propagating, so stray light is suppressed.
    Invoked in the LWIR design section ('we choose a sufficiently high n... such that rτϕ < λ'); assumes a local grating-like diffraction condition in cylindrical coordinates.
  • standard math A normally incident plane wave can be represented as m = ±1 azimuthal components in cylindrical coordinates.
    Mentioned in Results and Discussion with refs 14,17; standard cylindrical wave decomposition.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Scalable freeform optimization of wide-aperture 3D metalenses by zoned discrete axisymmetry." pith.science (2026). https://pith.science/paper/BLXACM4M

@misc{pith2026250107979,
  author       = {Pith},
  title        = {Pith review of: Scalable freeform optimization of wide-aperture 3D metalenses by zoned discrete axisymmetry},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/BLXACM4M}},
  note         = {Machine review of arXiv:2501.07979}
}
abstract

We introduce a novel framework for design and optimization of 3D freeform metalenses that attains nearly linear scaling of computational cost with diameter, by breaking the lens into a sequence of radial "zones" with $n$-fold discrete axisymmetry, where $n$ increases with radius. This allows vastly more design freedom than imposing continuous axisymmetry, while avoiding the compromises of the locally periodic approximation (LPA) or scalar diffraction theory. Using a GPU-accelerated finite-difference time-domain (FDTD) solver in cylindrical coordinates, we perform full-wave simulation and topology optimization within each supra-wavelength zone. We validate our approach by designing millimeter and centimeter-scale, poly-achromatic, 3D freeform metalenses which outperform the state of the art. By demonstrating the scalability and resulting optical performance enabled by our "zoned discrete axisymmetry" (ZDA) and supra-wavelength domain decomposition, we highlight the potential of our framework to advance large-scale meta-optics and next-generation photonic technologies.

Figures

Figures reproduced from arXiv: 2501.07979 by the authors.

Figure 1
Figure 1. An artistic rendition (not drawn to scale) of a wide-aperture (diameter > 3000 [PITH_FULL_IMAGE:figures/full_fig_p006_1.png] view at source ↗
Figure 2
Figure 2. Our zoned discrete axisymmetry (ZDA) scheme. Only the red-shaded domains are [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗
Figure 3
Figure 3. The schematic of the structure of the inverse-designed 8-mm diameter monochro [PITH_FULL_IMAGE:figures/full_fig_p011_3.png] view at source ↗
Figures from the paper (5 more)
Figure 4
Figure 4. Figure 4: Focal spot properties of the 8-mm-diameter and NA = 0.3 metalens at a wave [PITH_FULL_IMAGE:figures/full_fig_p014_4.png]
Figure 5
Figure 5. Figure 5: The schematic of the structure of the inverse-designed 1-mm diameter RGB [PITH_FULL_IMAGE:figures/full_fig_p016_5.png]
Figure 6
Figure 6. Figure 6: Simulated focal intensity distribution of the 1-mm-diameter and NA=0.8 RGB [PITH_FULL_IMAGE:figures/full_fig_p017_6.png]
Figure 7
Figure 7. Figure 7: The schematic of the structure of the inverse-designed 2-mm diameter poly [PITH_FULL_IMAGE:figures/full_fig_p018_7.png]
Figure 8
Figure 8. Figure 8: Simulated focal intensity distribution of the 2-mm-diameter and NA=0.3 poly [PITH_FULL_IMAGE:figures/full_fig_p019_8.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

12 extracted references · 10 canonical work pages

  1. [1]

    T.; Devlin, R

    (1) Khorasaninejad, M.; Chen, W. T.; Devlin, R. C.; Oh, J.; Zhu, A. Y.; Capasso, F. Metal- enses at visible wavelengths: Diffraction-limited focusing and subwavelength resolution imaging. Science 2016, 352, 1190–1194. (2) Chen, W. T.; Zhu, A. Y.; Sanjeev, V.; Khorasaninejad, M.; Shi, Z.; Lee, E.; Capasso, F. A broadband achromatic metalens for focusing an...

    work page 2016

  2. [7]

    S.; Sigmund, O

    (33) Jensen, J. S.; Sigmund, O. Topology optimization for nano-photonics.Laser & Photon- ics Reviews 2011, 5, 308–321. (34) Shin, W.; Fan, S. Choice of the perfectly matched layer boundary condition for frequency-domain Maxwell’s equations solvers. Journal of Computational Physics 2012, 231, 3406–3431. (35) Kunz, K. S.; Luebbers, R. J.The finite differenc...

    work page 2011

  3. [11]

    (45) Li, L.; Li, F.; Cui, T. J. Computational superoscillation imaging beyond the Rayleigh limit from far-field measurements.Optics Express 2014, 22, 5431–5441. (46) Gbur, G. Using superoscillations for superresolved imaging and subwavelength focusing. Nanophotonics 2019, 8, 205–225. (47) Hou, M.; Chen, Y.; Li, J.; Yi, F. Single 5-centimeter-aperture meta...

    work page 2014

  4. [15]

    On causality and dynamic stability of perfectly matched layers for FDTD simulations.IEEE Transactions on Microwave Theory and Techniques1999, 47, 775–785

    (27) Teixeira, F.; Chew, W. On causality and dynamic stability of perfectly matched layers for FDTD simulations.IEEE Transactions on Microwave Theory and Techniques1999, 47, 775–785. (28) Hammond, A. M.; Oskooi, A.; Chen, M.; Lin, Z.; Johnson, S. G.; Ralph, S. E. High- performance hybrid time/frequency-domain topology optimization for large-scale pho- ton...

    arXiv 2022

  5. [56]

    J.; Bagheri, M.; Faraon, A

    (38) Arbabi, A.; Horie, Y.; Ball, A. J.; Bagheri, M.; Faraon, A. Subwavelength-thick lenses with high numerical apertures and large efficiency based on high-contrast transmitar- rays. Nature Communications 2015, 6,

    work page 2015

  6. [60]

    (16) Chung, H.; Miller, O. D. High-NA achromatic metalenses by inverse design.Opt. Ex- press 2020, 28, 6945–6965. (17) Christiansen, R. E.; Lin, Z.; Roques-Carmes, C.; Salamin, Y.; Kooi, S. E.; Joannopou- los, J. D.; Soljačić, M.; Johnson, S. G. Fullwave Maxwell inverse design of axisymmetric, tunable, and multi-scale multi-wavelength metalenses.Opt. Expr...

    work page 2020

  7. [85]

    GPU Occupancy Prediction of Deep Learning Models Using Graph Neural Network

    (53) Mei, H.; Qu, H.; Sun, J.; Gao, Y.; Lin, H.; Sun, G. GPU Occupancy Prediction of Deep Learning Models Using Graph Neural Network. 2023 IEEE International Conference on Cluster Computing (CLUSTER). 2023; pp 318–329. (54) Presutti, F.; Monticone, F. Focusing on bandwidth: achromatic metalens limits.Optica 2020, 7, 624–631. (55) Arya, G.; Li, W. F.; Roqu...

    work page 2023

  8. [1662]

    Metaoptic Computa- tional Imaging.ACS Photonics 2025, 12, 1722–1733

    (43) Roques-Carmes, C.; Wang, K.; Yang, Y.; Majumdar, A.; Lin, Z. Metaoptic Computa- tional Imaging.ACS Photonics 2025, 12, 1722–1733. (44) Peng, Y.; Fu, Q.; Heide, F.; Heidrich, W. The diffractive achromat full spectrum com- putational imaging with diffractive optics.ACM Trans. Graph.2016, 35,

    work page 2025

Show all 12 references
  1. [1991]

    Advances in optical metalenses.Nature Photonics 2023, 17, 16–25

    24 (20) Arbabi, A.; Faraon, A. Advances in optical metalenses.Nature Photonics 2023, 17, 16–25. (21) Lalanne, P.; Chavel, P. Metalenses at visible wavelengths: past, present, perspectives. Laser & Photonics Reviews2017, 11, 1600295. (22) Xue, W.; Zhang, H.; Gopal, A.; Rokhlin,...

    2023 arXiv

  2. [2004]

    Wave Scattering from Rough Surfaces; Springer Berlin Heidelberg: Berlin, Heidelberg, 1999; pp 109–145

    (11) Voronovich, A. Wave Scattering from Rough Surfaces; Springer Berlin Heidelberg: Berlin, Heidelberg, 1999; pp 109–145. (12) Zhou, Y.; Mao, C.; Gershnabel, E.; Chen, M.; Fan, J. A. Large-Area, High-Numerical- Aperture, Freeform Metasurfaces.Laser & Photonics Reviews2024, 18...

    1999

  3. [2409]

    (6) Lin, Z.; Liu, V.; Pestourie, R.; Johnson, S. G. Topology optimization of freeform large- area metasurfaces.Opt. Express 2019, 27, 15765–15775. (7) Pestourie, R.; Pérez-Arancibia, C.; Lin, Z.; Shin, W.; Capasso, F.; Johnson, S. G. Inverse design of large-area metasurfaces.O...

    2019

  4. [7069]

    Nature Photonics2022, 16, 171–173

    (39) Engelberg,J.; Levy,U.Standardizingflatlenscharacterization. Nature Photonics2022, 16, 171–173. 26 (40) Shim, H.; Chung, H.; Miller, O. D. Maximal Free-Space Concentration of Electromag- netic Waves.Phys. Rev. Appl.2020, 14, 014007. (41) Menon, R.; Sensale-Rodriguez, B. In...

    2020

Pith tools

Reviewed August 10, 2026 · model on record in the stance chip above.