Pith. sign in

REVIEW 3 major objections 4 minor 1 cited by

Inferring Structure via Duality for Photonic Inverse Design

T0 review · 3 major / 4 minor · reviewed 2026-08-16 · deepseek-v4-flash

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

Pith's one-line read Close the duality gap, and a photonic bound becomes a design seed.

desk verdict Genuinely new Sion-based lemmas for scattering QCQPs and an honest heuristic, but Lemma 0 has a fixable proof gap and scrape's convergence is asserted, not proven. read the letter →

arxiv 2504.14083 v1 pith:4NPTM4WP submitted 2025-04-18 math.OC physics.optics

Sean Molesky , Pengning Chao , Alessio Amaolo , Alejandro W. Rodriguez This is my paper · ORCID
classification math.OCphysics.optics MSC 90C2690C4690C2278A45
keywords inversedesignquadraticallyconstrainedquadraticprogramsSion'sminimaxtheoremLagrangedualityconvexrelaxationtopologyoptimizationphotonicdevicescatteringconstraints
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

The paper sets out to turn convex relaxation bounds on photonic inverse-design problems into actual device geometries, not just performance limits. Its central claim is that a scattering QCQP, a quadratic optimization problem generated by Maxwell's equations and a variable material profile, can be deformed through three steps, contract, expand, and scrape, until strong duality holds. The deformation is justified by a minimax result: as the relaxed solution approaches the boundary of a compact power-balance constraint, it increasingly satisfies every other physical constraint, so the duality gap closes. Once primal and dual agree, a linear algebraic formula converts the dual field into a dielectric structure that seeds a secondary local optimization. In a companion study, this initialization reportedly improves dipole power-extraction designs by roughly an order of magnitude over random adjoint starts for device areas above $10\,\lambda^2$.

What carries the argument

The central object is the Sion set $S_P = \{x \in F_\kappa \mid S(x) \neq -\infty\}$, the region of fields for which the Lagrangian's infimum over multipliers is finite, together with the convex hull $C_P$ of truly feasible fields. The load-bearing identity is Lemma 4: approaching the boundary $\partial F_\kappa$ of a compact constraint from inside $S_P$ forces violation of all other scattering constraints to shrink, so $S_P$ and $C_P$ agree on that boundary and strong duality holds. Carrying the argument is the inference relation $X_{x,lk} = x_k / ((G_\circ x)_k + (e_i)_k) \delta_{lk}$, which converts any dual field into a scattering potential that would make the field physical. The contract, expand, and scrape protocols are the mechanism that drives $x^\circledast$ toward $\partial F_\kappa$.

What would settle it

Run the contract, expand, and scrape loop on a small SCQP whose optimal duality gap is known, for example the subset-sum encoding given in the paper, and measure the distance of $x^\circledast$ to $\partial F_\kappa$ at every step; if the iteration converges to a point strictly inside $F_\kappa$ with a positive duality gap for a family of objectives, the no-trap assumption is false.

Watch

Extended reading notes

Core claim

On its own terms, the paper argues that strong duality for photonic design QCQPs is within reach: Lemma 4 states that a maximizer of the Sion program lying on the boundary of a positive-definite compact constraint automatically satisfies every constraint, so the Sion set and the convex hull of feasible points coincide there and primal and dual values are equal. The dual solution of the modified program is then not merely a bound; through eq. (3) it defines a scattering potential that realizes that field as a polarization, giving a globally informed initial design. The three protocols, contract, expand, and scrape, are built to move the relaxed solution to that boundary: contraction inflates the distinction between Sion set and convex hull, expansion returns the constraints toward the physical device, and scraping rotates the objective so the maximizer is pushed outward. If the protocol reaches the boundary rather than stalling at a high-curvature interior point of the Sion set, the resulting initial geometry is claimed to be near-optimal in the original QCQP.

Load-bearing premise

The scheme assumes that repeatedly shifting the objective toward the current relaxed solution pushes that solution to the boundary of a bounded constraint, and that it does not get trapped at a high-curvature interior point of the relaxed feasible region; the paper states this as an unproved condition.

Editorial extensions

If this is right

  • If Lemma 4 holds and the protocols place the relaxed maximizer on the boundary of a compact constraint, the dual value equals the primal value, so the computed bound is certified tight for the modified program.
  • The inferred structure from eq. (3) supplies a feasible starting point for adjoint topology optimization; in the companion implementation this yields roughly an order-of-magnitude improvement in extracted power for design areas above $10\,\lambda^2$.
  • Because the dual and Sion solutions can differ only within the kernel of the Hessian, the dual field's spatial structure carries meaningful information about the primal optimizer rather than being an artifact of the relaxation.
  • The bound in Lemma 4 gives a quantitative way to estimate how much adding a previously absent constraint will tighten a performance limit, using the eigenvalues of the constraint and the distance of the relaxed field from the compact boundary.

Reading between the lines

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

  • One testable extension is to use the distance of $x^\circledast$ from $\partial F_\kappa$ as a stopping criterion and confidence measure in numerical solvers: small distance certifies near-strong duality, while a stalled positive distance flags the interior high-curvature trap the paper concedes.
  • The scrape update is a form of homotopy on the objective, so one could connect it to gradient flows on the Sion function and investigate convergence rates on small random SCQPs, including the subset-sum encoding, where the duality gap is exactly measurable.
  • Because Lemma 4 needs only one positive-definite constraint, a practical route to stronger bounds is to add a single artificial compact constraint engineered to touch the Sion set, rather than increasing the number of physical witnesses.
  • The same contract, expand, and scrape logic may transfer to any nonconvex QCQP with a compact constraint, but the transfer is only as sound as the assumption that the Sion set has no trapping interior high-curvature point; characterizing that set is the key open problem.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 4 minor

Summary. The manuscript studies Lagrange duality for quadratically constrained quadratic programs with the 'scattering' structure that arises in photonic inverse design. It defines a class of scattering-constrained quadratic programs (SCQPs), introduces the Sion program via Sion's minimax theorem, and proves four lemmas, the most important of which is Lemma 4: a maximizer of the Sion function that lies on the boundary of a positive-definite compact constraint must satisfy every constraint, so that strong duality holds. It then proposes a 'verlan' heuristic consisting of contract, expand, and scrape steps, aiming to transform a given SCQP into a strongly dual program and to use the dual solution together with the inference relation Eq. (3) to construct a material geometry that seeds a secondary local optimization. Numerical evidence is deferred to a companion paper (ref. [54]).

Significance. If the lemmas are correct, Lemma 4 provides a clean structural criterion for strong duality in this class of nonconvex QCQPs, and the Sion-set perspective offers a plausible explanation for why duality gaps shrink as device size grows. The paper is honest about the heuristic character of the verlan scheme and explicitly disclaims rigorous optimality; the companion numerical results are not needed to judge the theoretical core. The inference relation Eq. (3) is an explicit, construction-based map rather than a fitted quantity, which is a strength. However, the formal section contains proof gaps, most notably in Lemma 0, and the convergence of the scrape/contract loop is asserted rather than proved; the manuscript therefore needs revision before the theoretical claims can be regarded as established.

major comments (3)
  1. [I.A, Lemma 0 proof] The proof contains a non-sequitur: from the displayed concavity inequality L_phi[tx+(1-t)y] >= tL_phi(x)+(1-t)L_phi(y) the text concludes 'tx+(1-t)y in F_kappa'. This does not follow; membership of the convex combination in F_kappa is exactly the convexity of F_kappa, which is not implied by compactness and is not derived from the definition F_kappa = {x | f_kappa(x) >= 0}. Since Sion's theorem requires a compact convex F_kappa, the manuscript must either state convexity as an explicit standing assumption of the SCQP definition or prove it for the scattering-constraint superlevel sets. The rest of Lemma 0 is sound once this is supplied.
  2. [I.A, Lemma 2 proof] The proof asserts that x_subset = A_psi^+ s_psi is 'then a solution of max_{x in F_kappa} L(phi_subset, x)' even when A_psi_subset is singular. This is false in general: for a concave quadratic with A singular and s not in the range of A, the unconstrained stationary point does not exist and the maximizer over the compact convex set F_kappa lies on its boundary. A correct argument can likely be obtained by decomposing any maximizer as x = x0 + u + v with u in range(A), v in ker(A), showing the quadratic part forces u=0 and the remaining linear term acts only on ker(A); this would still yield the stated kernel conclusion, but the proof as written is invalid and should be rewritten.
  3. [II.A, Verlan Scheme] The termination of the scrape protocol rests on the unproved assertion that 'So long as x_subset is not trapped at a high curvature point of the Sion set in the interior of F_kappa, it will eventually push x_subset to partial F_kappa', together with the claim that under contraction such a trap is 'effectively ruled out'. The contract step (Pa) is defined procedurally, and no argument shows that a finite decrease of V contracts the Sion set toward CP in a way that eliminates the interior trap. Because Lemma 4's payoff requires exact boundary contact, this is a load-bearing gap for the scheme as stated. The paper explicitly labels the method a heuristic and disclaims rigorous optimality; the gap should be formalized as an open condition and, ideally, accompanied by either a proof of a sufficient condition or a numerical study of when the trap occurs.
minor comments (4)
  1. [I.B, Eq. (3)] The denominator (G0 x)_k + (e_i)_k can vanish for arbitrary fields x; the inference map should state a nondegeneracy assumption or a regularization for those components.
  2. [I.A, Lemma 1 corollary proof] In the corollary proof, the expression 'f_o(x_k)' appears to contain a typo: the index should be j, and the sentence 'holds generally' should spell out that S(x_j)=f_o(x_j) for feasible x_j even when f_o is nonlinear.
  3. [I.A, Definitions] The symbol F_kappa is used both for the feasible set of the composite constraint f_kappa and for the domain of the Sion function; the notation should be adjusted or explicitly identified to avoid ambiguity.
  4. [III.C, Semi-Definite Relaxation] The homogeneous reformulation divides by x_tilde_{n+1} to recover x; the text should note that the case x_tilde_{n+1}=0 is excluded or handled separately.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the lemmas follow from Sion's theorem, and eq. (3) is an explicit construction, not a fitted prediction.

full rationale

I find no circular step. The four lemmas are proved from Sion's minimax theorem (eq. 9) together with the SCQP definitions; for example, Lemma 4 constructs ζ = φ_j + δγ ∈ Ψ_P and shows that L(nζ/k, x⊛) < 0 for large n, contradicting the positivity of S(x⊛) that follows from feasibility, without assuming the conclusion. The structure-inference map eq. (3) defines X_x so that (X_x^{-1} − G°)x = e_i; this is an explicit construction, and the paper itself calls it 'by no means unique,' so it is not a fitted parameter disguised as a prediction. The verlan protocols are heuristic and the paper candidly states, 'We have not tested enough scenarios to provide useful rules of thumb... and are unaware of any rigorous optimality.' The load-bearing but unproved claim that scrape eventually reaches ∂F_κ ('So long as x⊛ is not trapped at a high curvature point of the Sion set in the interior of F_κ, it will eventually push x⊛ to ∂F_κ') is a correctness gap, not circularity, because it is not derived from the desired conclusion. Self-citations to the companion article [54] supply empirical motivation for the heuristic's practical value, but they are not used to prove the lemmas, and the numerical results are externally checkable. No 'prediction' reduces by construction to an input.

Assumptions & free parameters 3 free parameters · 5 assumptions · 1 invented entities

The mathematical derivation rests on Sion's theorem and on the existence of compact convex feasible sets; the proposed algorithm adds unproven convergence assumptions and hyperparameters, while the inference relation introduces a possibly unphysical operator X_x. No constants are fitted to data in this paper.

free parameters (3)
  • scrape parameters sigma and gamma = not specified
    Update rule r^(n)=r^(o)+sigma x^(o)_⊛ or r^(n)=r^(o)+gamma x^(o)_⊛; no values or tuning guidance are given, and the protocol's behavior depends on them.
  • expansion parameter epsilon = not specified
    Controls the rate at which V returns to X• during expand; no concrete rule for setting it is provided.
  • near-boundary tolerance = not specified
    Strong duality is checked by x⊛ being within some tolerance of ∂Fκ; the tolerance is never formalized.
assumptions (5)
  • standard math Sion's minimax theorem
    Applied to L(φ,x) on ΨP×Fκ; the theorem itself is a standard result [42,61].
  • domain assumption Existence of a compact constraint and a non-empty feasible set in every scattering QCQP
    Abstract and SCQP definition; physically motivated by resistive power balance in passive devices, but not proven for every photonic design problem.
  • ad hoc to paper Fκ is compact and convex
    Needed for Sion's theorem; stated in Definitions with ref. [59] but not derived, and the Lemma 0 proof's concavity argument does not actually establish convexity.
  • domain assumption Scattering constraints encoded by witnesses Qj commuting with IX
    Section 'QCQPs in Photonic Device Design' assumes local polarization response and Qj commuting with IX to obtain eq. (2); the nonlocal case is deferred to the appendix.
  • ad hoc to paper Scrape, contract, and expand preserve near-optimality
    The paper states there is no rigorous optimality and only conjectures that locally rational updates lead to good global initialization, making this the load-bearing heuristic assumption.
invented entities (1)
  • inferred scattering operator X_x
    purpose: Maps an arbitrary field x to a scattering potential via X_x^{lk} = x_k / ((G◦x)_k + (ei)_k) delta_lk, used to convert dual solutions into material geometries.
    The operator may be highly unphysical; the paper itself says the jump from implied potential to actual design can be very large and must be tamed.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Inferring Structure via Duality for Photonic Inverse Design." pith.science (2026). https://pith.science/paper/4NPTM4WP

@misc{pith2026250414083,
  author       = {Pith},
  title        = {Pith review of: Inferring Structure via Duality for Photonic Inverse Design},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/4NPTM4WP}},
  note         = {Machine review of arXiv:2504.14083}
}
abstract

Led by a result derived from Sion's minimax theorem concerning constraint violation in quadratically constrained quadratic programs (QCQPs) with at least one constraint bounding the possible solution magnitude, we propose a heuristic scheme for photonic inverse design unifying core ideas from adjoint optimization and convex relaxation bounds. Specifically, through a series of alterations to the underlying constraints and objective, the QCQP associated with a given design problem is gradually transformed so that it becomes strongly dual. Once equivalence between primal and dual programs is achieved, a material geometry is inferred from the solution of the modified QCQP. This inferred structure, due to the complementary relationship between the dual and primal programs, encodes overarching features of the optimization landscape that are otherwise difficult to synthesize, and provides a means of initializing secondary optimization methods informed by the global problem context. An exploratory implementation of the framework, presented in a partner manuscript, is found to achieve dramatic improvements for the exemplary photonic design task of enhancing the amount of power extracted from a dipole source near the boundary of a structured material region -- roughly an order of magnitude compared to randomly initialized adjoint-based topology optimization for areas surpassing $10~\lambda^{2}$.

Figures

Figures reproduced from arXiv: 2504.14083 by the authors.

Figure 1
Figure 1. FIG. 1 [PITH_FULL_IMAGE:figures/full_fig_p005_1.png] view at source ↗
Figure 2
Figure 2. FIG. 2 [PITH_FULL_IMAGE:figures/full_fig_p008_2.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Indexed singular value bounds on scattering operators: How many channels can a photonic device support?

    physics.optics 2025-10 conditional novelty 7.0 of 10

    A Courant-Fischer-Weyl min-max principle combined with convex relaxations yields computable upper bounds on each individual singular value of the electromagnetic Green operator for arbitrary linear scatterers.

Reference graph

Works this paper leans on

85 extracted references · 68 canonical work pages · cited by 1 Pith paper

  1. [40]

    On performance bounds for topology optimization

    Anna Dalklint, Rasmus E Christiansen, and Ole Sig- mund. On performance bounds for topology optimiza- tion. arXiv preprint arXiv:2410.20375 , 2024

    work page Pith review arXiv 2024

  2. [54]

    Bounds as blueprints: To- wards optimal and accelerated photonics inverse design

    Pengning Chao, Alessio Amaolo, Sean Molesky, and Alejandro W Rodriguez. Bounds as blueprints: To- wards optimal and accelerated photonics inverse design. arXiv:2504.10469, 2025

    arXiv 2025

  3. [1]

    Computing power network: The architecture of convergence of com- puting and networking towards 6g requirement

    Xiongyan Tang, Chang Cao, Youxiang Wang, Shuai Zhang, Ying Liu, Mingxuan Li, and Tao He. Computing power network: The architecture of convergence of com- puting and networking towards 6g requirement. China communications, 18(2):175–185, 2021

    2021

  4. [2]

    Com- pute trends across three eras of machine learning

    Jaime Sevilla, Lennart Heim, Anson Ho, Tamay Be- siroglu, Marius Hobbhahn, and Pablo Villalobos. Com- pute trends across three eras of machine learning. In 2022 International Joint Conference on Neural Networks (IJCNN), pages 1–8. IEEE, 2022

    2022

  5. [3]

    Ai and ml accelerator survey and trends

    Albert Reuther, Peter Michaleas, Michael Jones, Vijay Gadepally, Siddharth Samsi, and Jeremy Kepner. Ai and ml accelerator survey and trends. In 2022 IEEE High Performance Extreme Computing Conference (HPEC) , pages 1–10. IEEE, 2022

    2022

  6. [4]

    Future of hpc: Diversifying hetero- geneity

    Dejan Milojicic, Paolo Faraboschi, Nicolas Dube, and Duncan Roweth. Future of hpc: Diversifying hetero- geneity. In 2021 Design, Automation & Test in Europe Conference & Exhibition (DATE), pages 276–281. IEEE, 2021

    2021

  7. [5]

    Ultrafast 3d nanofabrication via digital holography

    Wenqi Ouyang, Xiayi Xu, Wanping Lu, Ni Zhao, Fei Han, and Shih-Chi Chen. Ultrafast 3d nanofabrication via digital holography. Nature Communications, 14(1): 1716, 2023

    2023

  8. [6]

    Two-photon poly- merization lithography for imaging optics

    Hao Wang, Cheng-Feng Pan, Chi Li, Kishan S Menghra- jani, Markus A Schmidt, Aoling Li, Fu Fan, Yu Zhou, Wang Zhang, Hongtao Wang, et al. Two-photon poly- merization lithography for imaging optics. International Journal of Extreme Manufacturing , 6(4):042002, 2024

    work page 2024

Show all 85 references
  1. [7]

    Nanofabrication for nanophotonics

    Younghwan Yang, Youngsun Jeon, Zhaogang Dong, Joel KW Yang, Mahsa Haddadi Moghaddam, Dai-Sik Kim, Dong Kyo Oh, Jihae Lee, Mario Hentschel, Harald Giessen, et al. Nanofabrication for nanophotonics. ACS nano, 2025

    2025

  2. [8]

    High speed simulation and freeform optimization of nanophotonic devices with physics-augmented deep learning

    Mingkun Chen, Robert Lupoiu, Chenkai Mao, Der-Han Huang, Jiaqi Jiang, Philippe Lalanne, and Jonathan A Fan. High speed simulation and freeform optimization of nanophotonic devices with physics-augmented deep learning. ACS Photonics, 9(9):3110–3123, 2022

    2022

  3. [9]

    Fullwave design of cm-scale cylindrical metasurfaces via fast direct solvers

    Wenjin Xue, Hanwen Zhang, Abinand Gopal, Vladimir Rokhlin, and Owen D Miller. Fullwave design of cm-scale cylindrical metasurfaces via fast direct solvers. arXiv preprint arXiv:2308.08569, 2023

    2023 arXiv

  4. [10]

    Physics-enhanced deep surrogates for partial differential equations

    Rapha¨ el Pestourie, Youssef Mroueh, Chris Rackauckas, Payel Das, and Steven G Johnson. Physics-enhanced deep surrogates for partial differential equations. Nature Machine Intelligence, 5(12):1458–1465, 2023

    2023

  5. [11]

    In- verse design in nanophotonics

    Sean Molesky, Zin Lin, Alexander Y Piggott, Weiliang Jin, Jelena Vuˇ ckovi´ c, and Alejandro W Rodriguez. In- verse design in nanophotonics. Nature Photonics, 12(11): 659–670, 2018. 12

    2018

  6. [12]

    Multi-task topol- ogy optimization of photonic devices in low-dimensional fourier domain via deep learning

    Simei Mao, Lirong Cheng, Houyu Chen, Xuanyi Liu, Zi- han Geng, Qian Li, and Hongyan Fu. Multi-task topol- ogy optimization of photonic devices in low-dimensional fourier domain via deep learning. Nanophotonics, 12(5): 1007–1018, 2023

    2023

  7. [13]

    Roadmapping the next generation of silicon photonics

    Sudip Shekhar, Wim Bogaerts, Lukas Chrostowski, John E Bowers, Michael Hochberg, Richard Soref, and Bhavin J Shastri. Roadmapping the next generation of silicon photonics. Nature Communications , 15(1):751, 2024

    2024

  8. [14]

    Roadmap for optical metasurfaces

    Arseniy I Kuznetsov, Mark L Brongersma, Jin Yao, Mu Ku Chen, Uriel Levy, Din Ping Tsai, Nikolay I Zhe- ludev, Andrei Faraon, Amir Arbabi, Nanfang Yu, et al. Roadmap for optical metasurfaces. ACS photonics , 11 (3):816–865, 2024

    2024

  9. [15]

    Convex restrictions in physical design

    Guillermo Angeris, Jelena Vuˇ ckovi´ c, and Stephen Boyd. Convex restrictions in physical design. Scientific Reports, 11(1):1–10, 2021. doi:10.1038/s41598-021-92451-1

    2021 doi

  10. [16]

    Physical lim- its in electromagnetism

    Pengning Chao, Benjamin Strekha, Rodrick Kuate Defo, Sean Molesky, and Alejandro W Rodriguez. Physical lim- its in electromagnetism. Nature Reviews Physics , 4(8): 543–559, 2022

    2022

  11. [17]

    Upper bounds on absorption and scat- tering

    Mats Gustafsson, Kurt Schab, Lukas Jelinek, and Miloslav Capek. Upper bounds on absorption and scat- tering. New Journal of Physics, 22(7):073013, September

  12. [18]

    Heuristic methods and performance bounds for photonic design

    Guillermo Angeris, Jelena Vuˇ ckovi´ c, and Stephen Boyd. Heuristic methods and performance bounds for photonic design. Optics Express, 29(2):2827–2854, 2021

    2021

  13. [19]

    Fundamental limits to ra- diative heat transfer: theory

    Sean Molesky, Prashanth S Venkataram, Weiliang Jin, and Alejandro W Rodriguez. Fundamental limits to ra- diative heat transfer: theory. Physical Review B, 101(3): 035408, 2020

    2020

  14. [20]

    Hanwen Zhang, Zeyu Kuang, Shruti Puri, and Owen D. Miller. Conservation law based global bounds to quan- tum optimal control. Physical Review Letters , 127: 110506, Sep 2021. doi:10.1103/PhysRevLett.127.110506

    2021 doi

  15. [21]

    Fun- damental bounds for multi-port antennas

    Miloslav Capek, Lukas Jelinek, and Michal Masek. Fun- damental bounds for multi-port antennas. In 2021 15th European Conference on Antennas and Propagation (Eu- CAP), pages 1–4. IEEE, 2021

    2021

  16. [22]

    Rodriguez

    Sean Molesky, Pengnging Chao, Jewel Mohajan, Wes- ley Reinhart, Heng Chi, and Alejandro W. Rodriguez. T-operator limits on optical communication: Metaop- tics, computation, and input-output transformations. Physical Review Research , 4:013020, Jan 2022. doi: 10.1103/PhysRevRes...

    2022 doi

  17. [23]

    Fundamental limits to multi-functional and tun- able nanophotonic response

    Hyungki Shim, Zeyu Kuang, Zin Lin, and Owen D Miller. Fundamental limits to multi-functional and tun- able nanophotonic response. Nanophotonics, 13(12): 2107–2116, 2024

    2024

  18. [24]

    Maximum gain, effective area, and directivity

    Mats Gustafsson and Miloslav Capek. Maximum gain, effective area, and directivity. IEEE Transactions on Antennas and Propagation , 67(8):5282, 2019. doi: 10.1109/TAP.2019.2916760

    2019

  19. [25]

    Upper bounds on focusing efficiency

    Kurt Schab, Lukas Jelinek, Miloslav Capek, and Mats Gustafsson. Upper bounds on focusing efficiency. Optics Express, 30(25):45705–45723, 2022

    2022

  20. [26]

    Perfect absorption at the ultimate thickness limit in planar films

    Zarko Sakotic, Alexander Ware, Michelle Povinelli, and Daniel Wasserman. Perfect absorption at the ultimate thickness limit in planar films. ACS Photonics , 10(12): 4244–4251, 2023

    2023

  21. [27]

    How thin and efficient can a metasurface reflector be? universal bounds on reflection for any direction and polarization

    Mohamed Ismail Abdelrahman and Francesco Monti- cone. How thin and efficient can a metasurface reflector be? universal bounds on reflection for any direction and polarization. Advanced Optical Materials, 11(4):2201782, 2023

    2023

  22. [28]

    Transmission efficiency limit for nonlocal metalenses

    Shiyu Li and Chia Wei Hsu. Transmission efficiency limit for nonlocal metalenses. Laser & Photonics Reviews , 17 (9):2300201, 2023

    2023

  23. [29]

    Exploring the fundamental limits of integrated beam splitters with arbitrary phase via topology optimization

    Abhishek Nanda, Michael Kues, and Antonio Cal` a Lesina. Exploring the fundamental limits of integrated beam splitters with arbitrary phase via topology optimization. Optics Letters, 49(5):1125–1128, 2024

    2024

  24. [30]

    Fundamental limits to near-field opti- cal response over any bandwidth

    Hyungki Shim, Lingling Fan, Steven G Johnson, and Owen D Miller. Fundamental limits to near-field opti- cal response over any bandwidth. Physical Review X , 9 (1):011043, 2019. doi:10.1103/PhysRevX.9.011043

    2019 doi

  25. [31]

    Maximum electromagnetic local density of states via material structuring.Nanophotonics, 12(3):549–557, 2023

    Pengning Chao, Rodrick Kuate Defo, Sean Molesky, and Alejandro Rodriguez. Maximum electromagnetic local density of states via material structuring.Nanophotonics, 12(3):549–557, 2023

    2023

  26. [32]

    Suppressing electromagnetic local density of states via slow light in lossy quasi-one-dimensional gratings

    Benjamin Strekha, Pengning Chao, Rodrick Kuate Defo, Sean Molesky, and Alejandro W Rodriguez. Suppressing electromagnetic local density of states via slow light in lossy quasi-one-dimensional gratings. Physical Review A, 109(4):L041501, 2024

    2024

  27. [33]

    Physical limits on raman scattering: The critical role of pump and signal co-design

    Alessio Amaolo, Pengning Chao, Thomas J Maldonado, Sean Molesky, and Alejandro W Rodriguez. Physical limits on raman scattering: The critical role of pump and signal co-design. Physical Review A, 110(6):L061501, 2024

    2024

  28. [34]

    Limitations on bandwidth-integrated passive cloaking

    Benjamin Strekha, Alessio Amaolo, Jewel Mohajan, Pengning Chao, Sean Molesky, and Alejandro W Ro- driguez. Limitations on bandwidth-integrated passive cloaking. Physical Review A, 110(6):063513, 2024

    2024

  29. [35]

    The power of convex relaxation: Near-optimal matrix completion

    Emmanuel J Cand` es and Terence Tao. The power of convex relaxation: Near-optimal matrix completion. IEEE transactions on information theory , 56(5):2053– 2080, 2010

    2010

  30. [36]

    Topological nature of edge states for one-dimensional systems without symmetry protection

    Janet Zhong, Heming Wang, Alexander N Poddubny, and Shanhui Fan. Topological nature of edge states for one-dimensional systems without symmetry protection. arXiv preprint arXiv:2412.10526 , 2024

    2024 arXiv

  31. [37]

    Passivity constraints on the relations between transmission, reflection, and ab- sorption eigenvalues

    Cheng Guo and Shanhui Fan. Passivity constraints on the relations between transmission, reflection, and ab- sorption eigenvalues. Physical Review B, 110(20):205431, 2024

    2024

  32. [38]

    Time-reversal-symmetry bounds on electromagnetic fields

    Wenchao Ma, Rapha¨ el Pestourie, and Steven G John- son. Time-reversal-symmetry bounds on electromagnetic fields. arXiv preprint arXiv:2503.03342 , 2025

    2025 arXiv

  33. [39]

    Many photonic design problems are sparse qcqps

    Shai Gertler, Zeyu Kuang, Colin Christie, Hao Li, and Owen D Miller. Many photonic design problems are sparse qcqps. Science Advances, 11(1):eadl3237, 2025

    2025

  34. [41]

    Please contact us if you are aware of relevant references that we have overlooked concerning this point

    Note1. Please contact us if you are aware of relevant references that we have overlooked concerning this point

  35. [42]

    On general minimax theorems

    Maurice Sion. On general minimax theorems. Pacific Journal of Mathematics , 8(1):171–176, 1958

    1958

  36. [43]

    Computational bounds for photonic de- sign

    Guillermo Angeris, Jelena Vuˇ ckovi´ c, and Stephen P Boyd. Computational bounds for photonic de- sign. ACS Photonics , 6(5):1232, 2019. doi: 10.1021/acsphotonics.9b00154

    2019 doi

  37. [44]

    Ro- driguez

    Sean Molesky, Pengning Chao, and Alejandro W. Ro- driguez. Hierarchical mean-field T operator bounds 13 on electromagnetic scattering: Upper bounds on near-field radiative purcell enhancement. Physi- cal Review Research , 2:043398, Dec 2020. doi: 10.1103/PhysRevResearch.2.043398

    2020 doi

  38. [45]

    Zeyu Kuang and Owen D. Miller. Computational bounds to light–matter interactions via local conservation laws. Physical Review Letters , 125:263607, Dec 2020. doi: 10.1103/PhysRevLett.125.263607

    2020 doi

  39. [46]

    Sub- structure limits to optical phenomena

    L Jelinek, M Gustafsson, M Capek, and K Schab. Sub- structure limits to optical phenomena. In 2020 Four- teenth International Congress on Artificial Materials for Novel Wave Phenomena (Metamaterials) , pages 400–

    2020

  40. [47]

    The text does not describe the most general means of forming a QCQP from eq

    Note2. The text does not describe the most general means of forming a QCQP from eq. (1) via scattering theory. For example, any fixed design can be taken as a background, vacuum is only the simplest choice

  41. [48]

    Following a procedure similar to ref

    Note3. Following a procedure similar to ref. [84], these equations can also be extended to treat multi-material design scenarios. A similar result, in relation to treating continuous media, is also effectively achieved by limiting imposed Qj operators to volumes that are subst...

  42. [49]

    In practice, the QCQP used to determined bounds is often furthered relaxed by only considering some spe- cial physically motivated subset of choices of Q, e.g

    Note4. In practice, the QCQP used to determined bounds is often furthered relaxed by only considering some spe- cial physically motivated subset of choices of Q, e.g. Q corresponding to the conservation of power

  43. [50]

    Fundamental lim- its on radiativeχ (2) second harmonic generation

    Jewel Mohajan, Pengning Chao, Weiliang Jin, Sean Molesky, and Alejandro W Rodriguez. Fundamental lim- its on radiativeχ (2) second harmonic generation. Optics Express, 31(26):44212–44223, 2023

    2023

  44. [51]

    Sum-of-squares bounds on surface-enhanced raman scattering

    Pengning Chao, Ian M Hammond, and Steven G John- son. Sum-of-squares bounds on surface-enhanced raman scattering. arXiv preprint arXiv:2502.09821 , 2025

    2025 arXiv

  45. [52]

    Maximum shannon capacity of pho- tonic structures

    Alessio Amaolo, Pengning Chao, Benjamin Strekha, Ste- fan Clarke, Jewel Mohajan, Sean Molesky, and Alejan- dro W Rodriguez. Maximum shannon capacity of pho- tonic structures. arXiv preprint arXiv:2409.02089, 2024

    2024 arXiv

  46. [53]

    Modes, bounds, and synthesis of opti- mal electromagnetic scatterers

    Mats Gustafsson. Modes, bounds, and synthesis of opti- mal electromagnetic scatterers. New Journal of Physics , 26(10):103039, 2024

    2024

  47. [55]

    witnessing

    Note5. Connecting with §Appendix: Non-Local Scatter- ing Constraints, the witness terminology is based on the notion of “witnessing” a certain design parameter, in the scattering form, of the overarching device optimization problem

  48. [56]

    It is plausible that the majority of the ideas we have investigated here could be extended to a wider class of objective functions

    Note6. It is plausible that the majority of the ideas we have investigated here could be extended to a wider class of objective functions. Namely, we suspect that analogous behaviour holds whenever the dual transformation is well defined

  49. [57]

    Reducibility among combinatorial problems

    Richard M Karp. Reducibility among combinatorial problems. In 50 Years of Integer Programming 1958- 2008: from the Early Years to the State-of-the-Art , pages 219–241. Springer, 2009

    1958

  50. [58]

    Np-complete problems and physical re- ality

    Scott Aaronson. Np-complete problems and physical re- ality. ACM SIGACT News , 36(1):30–52, 2005

    2005

  51. [59]

    Topology and geometry , volume 139

    Glen E Bredon. Topology and geometry , volume 139. Springer Science & Business Media, 2013

    2013

  52. [60]

    A similar argument shows that kerAϕ⊛̸= 0, then s◦̸∈ kerAϕ⊛ unless inf ϕ∈ΨPD (ϕ) =D (0) [44]

    Note7. A similar argument shows that kerAϕ⊛̸= 0, then s◦̸∈ kerAϕ⊛ unless inf ϕ∈ΨPD (ϕ) =D (0) [44]

  53. [61]

    Elementary proof for sion’s minimax theorem

    Hidetoshi Komiya. Elementary proof for sion’s minimax theorem. Kodai Mathematical Journal, 11(1):5–7, 1988

    1988

  54. [62]

    Real analysis: modern techniques and their applications, volume 40

    Gerald B Folland. Real analysis: modern techniques and their applications, volume 40. John Wiley & Sons, 1999

    1999

  55. [63]

    It may also be sensible to normalize r(n) or in- clude some removal of r(o) in the update, i.e

    Note8. It may also be sensible to normalize r(n) or in- clude some removal of r(o) in the update, i.e. r(n) = (1−γ)r(o) +γx(o) ⊛

  56. [64]

    If it is not clear that this assumption holds, an additional normalization step can be added to scrape

    Note9. If it is not clear that this assumption holds, an additional normalization step can be added to scrape

  57. [65]

    Global T operator bounds on elec- tromagnetic scattering: upper bounds on far-field cross sections

    Sean Molesky, Pengning Chao, Weiliang Jin, and Alejan- dro W Rodriguez. Global T operator bounds on elec- tromagnetic scattering: upper bounds on far-field cross sections. Physical Review Research, 2(3):033172, 2020

    2020

  58. [66]

    Algorithm 887: Cholmod, supernodal sparse cholesky factorization and update/downdate

    Yanqing Chen, Timothy A Davis, William W Hager, and Sivasankaran Rajamanickam. Algorithm 887: Cholmod, supernodal sparse cholesky factorization and update/downdate. ACM Transactions on Mathematical Software (TOMS), 35(3):1–14, 2008

    2008

  59. [67]

    Accelerating sparse cholesky factorization on gpus

    Steven C Rennich, Darko Stosic, and Timothy A Davis. Accelerating sparse cholesky factorization on gpus. In 2014 4th Workshop on Irregular Applications: Architec- tures and Algorithms (IAˆ 3) , pages 9–16. IEEE, 2014

    2014

  60. [68]

    A frame- work to exploit data sparsity in tile low-rank cholesky factorization

    Qinglei Cao, Rabab Alomairy, Yu Pei, George Bosilca, Hatem Ltaief, David Keyes, and Jack Dongarra. A frame- work to exploit data sparsity in tile low-rank cholesky factorization. In 2022 IEEE International Parallel and Distributed Processing Symposium (IPDPS) , pages 414–

    2022

  61. [69]

    Towards general neural sur- rogate solvers with specialized neural accelerators

    Chenkai Mao, Robert Lupoiu, Tianxiang Dai, Mingkun Chen, and Jonathan A Fan. Towards general neural sur- rogate solvers with specialized neural accelerators. arXiv preprint arXiv:2405.02351, 2024

    2024 arXiv

  62. [70]

    Fast approximate solvers for meta- materials design in electromagnetism

    Raphael Pestourie. Fast approximate solvers for meta- materials design in electromagnetism. arXiv preprint arXiv:2503.20065, 2025

    2025 arXiv

  63. [71]

    Optimal input excitations for suppressing nonlinear instabilities in mul- timode fibers

    Kabish Wisal, Chun-Wei Chen, Zeyu Kuang, Owen D Miller, Hui Cao, and A Douglas Stone. Optimal input excitations for suppressing nonlinear instabilities in mul- timode fibers. Optica, 11(12):1663–1672, 2024

    2024

  64. [72]

    Classically computing performance bounds on depolarized quantum circuits

    Sattwik Deb Mishra, Miguel Fr´ ıas-P´ erez, and Rahul Trivedi. Classically computing performance bounds on depolarized quantum circuits. PRX Quantum , 5(2): 020317, 2024

    2024

  65. [73]

    Calculation of electromagnetic sensitiv- ities in the time domain

    Andreas H¨ ordt. Calculation of electromagnetic sensitiv- ities in the time domain. Geophysical Journal Interna- tional, 133(3):713–720, 1998

    1998

  66. [74]

    Exci- tation waveforms for maximum instantaneous power de- livery

    Jakub Liska, Lukas Jelinek, and Miloslav Capek. Exci- tation waveforms for maximum instantaneous power de- livery. arXiv preprint arXiv:2409.16293 , 2024

    2024 arXiv

  67. [75]

    High- performance hybrid time/frequency-domain topology op- timization for large-scale photonics inverse design.Optics Express, 30(3):4467–4491, 2022

    Alec M Hammond, Ardavan Oskooi, Mo Chen, Zin Lin, Steven G Johnson, and Stephen E Ralph. High- performance hybrid time/frequency-domain topology op- timization for large-scale photonics inverse design.Optics Express, 30(3):4467–4491, 2022

    2022

  68. [76]

    Time-domain topology optimization of ar- bitrary dispersive materials for broadband 3d nanopho- tonics inverse design

    Johannes Gedeon, Emadeldeen Hassan, and Antonio Cal` a Lesina. Time-domain topology optimization of ar- bitrary dispersive materials for broadband 3d nanopho- tonics inverse design. ACS Photonics, 10(11):3875–3887, 2023. 14

    2023

  69. [77]

    Inverse-designed dispersive time-modulated nanostruc- tures

    Puneet Garg, Jan David Fischbach, Aristeidis G Lampri- anidis, Xuchen Wang, Mohammad S Mirmoosa, Viktar S Asadchy, Carsten Rockstuhl, and Thomas J Sturges. Inverse-designed dispersive time-modulated nanostruc- tures. arXiv preprint arXiv:2409.04551 , 2024

    2024 arXiv

  70. [78]

    Matrix rank minimization with applica- tions

    Maryam Fazel. Matrix rank minimization with applica- tions. PhD thesis, PhD thesis, Stanford University, 2002

    2002

  71. [79]

    Semidefinite relaxation of quadratic optimization problems

    Zhi-Quan Luo, Wing-Kin Ma, Anthony Man-Cho So, Yinyu Ye, and Shuzhong Zhang. Semidefinite relaxation of quadratic optimization problems. IEEE Signal Pro- cessing Magazine, 27(3):20–34, 2010

    2010

  72. [80]

    Convex relaxation for optimal power flow problem: Mesh networks

    Ramtin Madani, Somayeh Sojoudi, and Javad Lavaei. Convex relaxation for optimal power flow problem: Mesh networks. IEEE Transactions on Power Systems , 30(1): 199–211, 2014

    2014

  73. [81]

    Chordal graphs and semidefinite optimization

    Lieven Vandenberghe and Martin S Andersen. Chordal graphs and semidefinite optimization. Foundations and Trends® in Optimization, 1(4):241–433, 2015

    2015

  74. [82]

    Mathematical methods of classical mechanics, volume 60

    Vladimir Igorevich Arnol’d. Mathematical methods of classical mechanics, volume 60. Springer Science & Busi- ness Media, 2013

    2013

  75. [83]

    As in the main text, the quadratic equations re- sulting from eq

    Note10. As in the main text, the quadratic equations re- sulting from eq. (16) can be adapted to continuous design parameters by replacing equalities with inequalities

  76. [84]

    Can photonic heterostructures provably outperform single- material geometries? Nanophotonics, 13(3):283–288, 2024

    Alessio Amaolo, Pengning Chao, Thomas J Maldon- ado, Sean Molesky, and Alejandro W Rodriguez. Can photonic heterostructures provably outperform single- material geometries? Nanophotonics, 13(3):283–288, 2024

    2024

  77. [2020]

    doi:10.1088/1367-2630/ab83d3

    ISSN 1367-2630. doi:10.1088/1367-2630/ab83d3

    doi

Pith tools

Reviewed August 16, 2026 · model on record in the stance chip above.