Inverse Microparticle Design for Enhanced Optical Trapping and Detection Efficiency in All Six Degrees of Freedom

Benjamin A. Stickler; Moosung Lee; Sungkun Hong; Thomas Pertsch

REVIEW 3 major objections 4 minor 1 cited by

Inverse Microparticle Design for Enhanced Optical Trapping and Detection Efficiency in All Six Degrees of Freedom

T0 review · 3 major / 4 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read Inverse design with a modified Born series forward solver and the adjoint method yields printable microparticles with simultaneously enhanced optical trap stiffness and detection efficiency across all six motional degrees of freedom.

desk verdict A useful inverse-design framework for six-DOF optical trapping; the silica numbers are credible, the silicon numbers need an independent solver check before they carry the weight the paper puts on them. read the letter →

arxiv 2506.01837 v3 pith:2HBF74IA submitted 2025-06-02 physics.optics quant-ph

Moosung Lee , Benjamin A. Stickler , Thomas Pertsch , Sungkun Hong This is my paper

classification physics.opticsquant-ph

keywords inversedesignlevitatedoptomechanicsopticaltrappingadjointmethodmodifiedBornseriesmicroparticlesdetectionefficiencysixdegreesoffreedom

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 claims that inverse design, running a fast electromagnetic scattering solver inside a gradient-based optimization loop, can find microparticle shapes that are simultaneously stiffer and easier to read out than equal-volume spheres in a standing-wave optical trap. In simulations, optimized silica microstructures hold all six motional modes at trap frequencies between 90 and 280 kHz and return scattered-light detection efficiencies above 46 percent per mode. Optimized extruded silicon shapes reach librational frequencies above 1 MHz, and both materials remain stably trapped in size ranges where equivalent spheres are unstable. If this holds, levitated optomechanics could extend quantum motional control from sub-micrometer nanospheres to larger, printable particles.

What carries the argument

The machinery is a topology-optimization loop. The forward step solves the three-dimensional inhomogeneous wave equation for the total electric field with a modified Born series, a convergent iterative scattering solver; the inverse step uses the adjoint method to compute the gradient of an objective functional without differentiating every voxel. The objective sums squared angular trap frequencies for all three translational and three librational modes, adds detection efficiencies with a weighting factor, and includes a mass-regularization term centered on a target mass. Shape constraints, specifically Gaussian smoothing, adaptive binarization, mirror symmetry across three planes, and for extruded designs a fixed thickness, steer the solution to a connected binary monolithic particle that can be fabricated by lithography. Trap and detection quantities are computed from derivatives of the total field with respect to infinitesimal translations and rotations, with detection efficiency defined as the fraction of Fisher-information-carrying far-field radiation collected by the two objectives.

What would settle it

Re-run the optimized silicon geometries with an independent Maxwell solver, such as finite-difference time-domain or a full multipole method, at susceptibility 11.1 and compare the predicted trap frequencies and detection efficiencies; alternatively, fabricate the extruded silicon particle and measure its librational frequencies in the same standing-wave trap to check whether the predicted values above 1 MHz appear.

Watch

Extended reading notes

Core claim

The central discovery is that the inverse-design loop generates monolithic, lithography-compatible particle geometries whose trap stiffness and motional detection efficiency are simultaneously high in all six degrees of freedom, unlike microspheres, which have no optical torque and can exhibit weak or negative axial stiffness. For silica, optimized three-dimensional and extruded shapes achieve translational and librational trap frequencies of 90 to 280 kHz and detection efficiencies above 46 percent for every mode, and they remain trapped for volumes where the same-volume sphere is unstable in the standing-wave trap. For silicon, whose susceptibility is roughly ten times larger, the optimized extruded particles act as levitated optical resonators with pronounced resonance effects, yielding librational frequencies above 1 MHz and stable trapping across a wide range of sizes. The message is that geometry, not just material or trap power, can be engineered to satisfy the simultaneous stiffness and readout requirements of ground-state cooling protocols.

Load-bearing premise

The whole design loop presumes the modified Born series solver stays accurate for high-contrast silicon where resonances dominate, but the paper benchmarks the solver against Mie theory only for low-index silica spheres.

Editorial extensions

If this is right

Printable silica microparticles can be stably trapped and read out on all three translational and all three rotational axes in a standard standing-wave trap, without needing multi-particle assemblies.
Measurement-based ground-state cooling becomes plausible for these larger particles, since the optimized detection efficiencies exceed the roughly 20 percent threshold identified in the feedback-cooling literature while stiffness remains high.
Silicon particles, which are often unstable to trap as spheres in a high-NA focus, can be made stably trappable by shape optimization, with librational frequencies above 1 MHz that favor sideband-resolved cooling.
The design constraints cost little performance: extruded quasi-2D shapes perform nearly as well as fully 3D optimized shapes, so the gains are compatible with standard lithography.
The same optimization opens size regimes where spherical particles cannot be trapped, enlarging the mass range available for macroscopic quantum experiments.

Reading between the lines

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

Because the paper's detection-efficiency metric counts only information radiation collected by the lenses, real experiments that include mode matching, homodyne detection, and stray light may see lower effective efficiencies than the reported values; applying the same optimization with a more complete measurement model is an untested extension.
The same adjoint-plus-Born-series loop could be retargeted at trapping at intensity minima rather than at the focus, or at designing the trap field itself through wavefront shaping, both directions the paper names as future work.
If the solver's resonant-regime predictions survive independent cross-checking, the silicon results suggest that internal resonances of optimized particles can be used deliberately as a design resource rather than being avoided as an instability risk.

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.

Desk Editor's Note

read the letter

This paper deserves a serious read. The genuinely new piece is the combination of a modified Born series forward solver with adjoint topology optimization to target all six motional degrees of freedom at once, including librational stiffness and detection efficiency, while enforcing lithography-friendly extrude-and-thickness constraints. Earlier inverse design in this space focused on translational trap stiffness, and forward detection analysis was mostly limited to spheres and simple shapes. Here the optimization genuinely outputs particle geometries that improve on same-volume spheres and remain stably trapped in size regimes where spheres are unstable. The silica results are the strongest part: the derivations in the SI are self-contained, the solver is benchmarked against Mie theory for silica spheres, and the reported 90–280 kHz trap frequencies and >46% detection efficiencies across all six modes are plausible enough to take seriously. The code is on GitHub, which helps reproducibility.

The soft spot is the silicon section. The high-index (chi_e = 11.1) results—librational frequencies over 1 MHz and stable trapping where spheres are unstable—come from the same Born-series solver that was only validated on silica (chi_e = 1.07). The paper itself flags 'pronounced resonance effects' in this regime, which is exactly where a truncated Born series and 50 nm voxel grid can be most delicate. Without a benchmark for a silicon sphere against Mie theory, or a cross-check with an independent method on an optimized silicon geometry, those headline numbers are not established. A second, also real but more minor, issue is that the quantitative claims come from single optimization runs with no multi-seed variation or error bars. That is common in inverse-design work, but it matters more here because the paper quotes sharp specific frequencies. On detection efficiency, the authors define it as the collected information radiation fraction and say they are neglecting mode-matching and specific measurement schemes, so that limitation is acknowledged rather than hidden.

These are addressable weaknesses, not fatal ones. The method is coherent, the silica results are credible, and the silicon claims are a request for extra evidence, not a reason to dismiss the paper. I would take it for review with two demands: an independent solver validation in the high-index regime, and a robustness check of the optimization (different seeds or at least a statement about variance). The paper is aimed at levitated optomechanics researchers, especially those pushing toward quantum control of larger, non-spherical particles. It does not deserve a desk reject; it deserves a careful referee with the right experimental and computational background.

Referee Report

3 major / 4 minor

Summary. The paper presents an inverse-design framework for levitated optomechanics that combines a modified Born series forward solver with the adjoint method to optimize the 3D shape of a microparticle in a standing-wave optical trap. The optimization targets simultaneously enhanced trap stiffness and motional detection efficiency across all six motional degrees of freedom (three translations and three librations), subject to shape constraints that enforce fabricability (smoothing, binarization, mirror symmetry, and extrusion). The authors benchmark the forward solver against Mie theory for SiO2 spheres and then apply the method to SiO2 and Si particles. They report that optimized SiO2 microstructures achieve translational and librational trap frequencies of 90-280 kHz and detection efficiencies above 46% across all degrees of freedom, while optimized extruded Si particles reach librational frequencies exceeding 1 MHz and remain stably trapped in size regimes where equivalent spheres are unstable. The paper includes a detailed supplement with derivations of the objective functional, gradient via the adjoint method, shape constraints, and pseudocode, and it makes the optimization code available on GitHub.

Significance. If the reported results hold, this work would be a valuable step toward quantum control of levitated particles beyond the Rayleigh regime. It is, to my knowledge, the first application of topology-optimization-style inverse design to the full six-degree-of-freedom optomechanical problem, and it explicitly addresses the practical need for lithographically printable geometries via extrusion constraints. The use of the modified Born series with the adjoint method is a sensible and potentially efficient combination, and the authors provide machine-readable code and a self-contained supplement. The validation against Mie theory for silica is a genuine strength. However, the quantitative claims for silicon—the most impressive part of the paper—rest on a solver that is not validated in the high-index, resonant regime, and the optimization results are drawn from single runs with no demonstration of reproducibility. These gaps must be addressed before the central claims can be considered established.

major comments (3)

[Shape optimization of printable silicon particles (Fig. 5)] The silicon performance claims (librational frequencies exceeding 1 MHz and stable trapping over a wide size range) are computed with the modified Born series forward solver, but the solver is validated only for silica with susceptibility χ_e = 1.07. No benchmark is shown for Si with χ_e = 11.1, even though these particles have size parameters kR ≈ 1.5–2.4 and in-material wavelengths near 440 nm, placing them in a resonance-rich regime where the paper itself notes 'pronounced resonance effects.' Because the optimization gradients are obtained from the same solver, any systematic error in the high-index resonant regime can be amplified, potentially yielding geometries that exploit numerical artifacts. Please add a validation benchmark for high-index particles—for example, comparing the modified Born series against Mie theory for Si spheres of the same volumes—and report the convergence of the Born series (number of iterations and residual) and a voxel-size convergence study for the optimized geometries.
[Benchmarking performance with 3D silica particle design (Figs. 2, 3, 5)] All quantitative results are based on a single random initialization and one optimization trajectory per condition. The initial condition is 'a small ellipsoid at the center with surrounding regions populated by random Gaussian noise,' and the non-convex objective combined with the binarization schedule can plausibly lead to different local optima for different seeds. The reported headline numbers (90–280 kHz, >46% detection efficiency, >1 MHz Si libration) are therefore not established as representative. Please run the optimization for at least several independent random seeds and report the spread in the resulting trap frequencies and detection efficiencies, or otherwise demonstrate that the presented geometries are not outliers.
[Eq. (2) and SI Table S2 (pseudo-code)] The objective functional contains several regularization and shape-constraint hyperparameters whose values and sensitivity are not reported: τ_η, τ_m, τ_R, τ_m,2, s_th, the smoothing radius σ_s, the binarization increment γ(Δp), and the background noise σ0. The paper's decision to optimize only Ω and to accept η above a threshold (20%) is a practical choice that could depend strongly on these weights. Please provide the actual hyperparameter values used for each reported optimization and include a brief sensitivity study (e.g., how the optimized structure and its performance change when the most important weights are varied by a factor of two).

minor comments (4)

[Throughout] There are several typographical issues: 'regulation term' should be 'regularization term' (near Eq. 2); in the Discussion, 'tap stability' should be 'trap stability'; and in the SI, Eq. (S7) appears to mistakenly repeat the definition of J_j (torque derivative) instead of defining the moment of inertia I_j, which was previously denoted with the same symbol J_j elsewhere.
[SI Supplementary Note 1] The finite-difference step sizes δx and δθ used to compute the trap stiffness and the adjoint fields are not specified. Please state the chosen values and justify their accuracy, since too-large steps cause truncation error and too-small steps cause cancellation error.
[Abstract and Fig. 2] The abstract claims 'detection efficiencies exceeding 46% across all motional degrees of freedom,' but the main text later states that for size-dependent optimizations only a 20% threshold is enforced. Clarify that the 46% figure refers specifically to the representative optimized structure in Fig. 2, not to all optimized structures.
[Fig. 5] The silicon structures are described as exhibiting 'pronounced resonance effects,' but no spectral information or quality factors are shown. A plot of the scattering cross-section or a resonance spectrum for the optimized Si particles would help substantiate this interpretation.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: optimized shapes are genuine outputs of a first-principles scattering objective, benchmarked externally against Mie theory; the author self-citation is non-load-bearing.

full rationale

The paper's central claims are not circular. The optimized geometries are produced by an iterative adjoint-method optimization of an objective functional (Eq. 2) built from trap frequencies and detection efficiencies that are themselves computed from the Maxwell scattering problem (Eqs. S1-S16). No target trap frequency, detection efficiency, or optimized geometry is supplied as a fitting input; the only regularization target is the mass m0, which is a constraint rather than a prediction. The optimized structures are then evaluated against an external benchmark, the equivalent-volume sphere, with the paper explicitly reporting consistency of its forward simulations with Mie theory for spheres (translational trap frequencies and motional detections). The one direct author self-citation, the modified Born series solver from Lee et al. (ref. 28), is not load-bearing in a circular way: it is a separately published numerical method, and in this paper the solver is checked against Mie-theory results for spherical particles. Any concern about the solver's accuracy for high-index silicon (chi_e = 11.1) is a correctness/validation risk, not a circularity, because the silicon performance numbers are genuine outputs of the stated scattering computation rather than consequences of the inputs. Accordingly, no step in the derivation chain reduces by definition or by self-citation to its own inputs.

Assumptions & free parameters 3 free parameters · 5 assumptions · 0 invented entities

The central results depend on several hand-chosen optimization hyperparameters whose values are not reported, plus a single random initialization per design. These do not constitute curve-fitting to data, but they mean the reported geometries are one realization of a stochastic optimization, not a unique or robust optimum. The paper introduces no new physical entities; it relies on standard electromagnetic theory and a numerical solver from prior work.

free parameters (3)

objective regularization weights (τ_η, τ_m, τ_R, τ_m,2, s_th) = not disclosed
Control the balance among Ω_j^2, η_j, mass deviation, and rotational weight in Eq. (2) and Table S2; chosen ad hoc, values absent from the text, and they determine which local optimum the optimizer converges to.
shape constraint parameters (smoothing radius σ_s, binarization schedule β increment γ(Δp), background noise σ0) = not disclosed
Set the minimum feature size and the convergence path in Supplementary Note 4 and Table S2; no values or sensitivity study provided, so the printed designs are tied to unstated choices.
random initialization seed = not disclosed
Every design starts from a small ellipsoid plus random Gaussian noise; a single seed per design is used, so reported performance may not be representative of the method's typical output.

assumptions (5)

domain assumption The modified Born series renders converged, accurate solutions of the integral equation (S1) for all particle shapes and susceptibilities used, including Si with χ_e = 11.1.
Forward solver basis for every reported quantity; validation in the paper is only shown against Mie theory for low-index SiO2 spheres.
standard math The total-field expressions for force and torque (Eqs. S2-S3) are exact for lossless nonmagnetic dielectrics and are valid at these particle sizes and field intensities.
Standard electrodynamics result used to derive trap stiffness and gradients throughout the SI.
domain assumption The incident standing-wave field from the vectorized angular spectrum method (NA=0.8, λ=1550 nm) is a faithful model of the physical trap.
The incident field sets the entire optimization landscape; the method is cited from the authors' prior work and not cross-checked in this paper.
domain assumption Detection efficiency defined as the collected fraction of information radiation (Eq. S10) is a sufficient proxy for measurement-based cooling performance.
The authors explicitly neglect mode-matching and homodyne detection details, which could change the practical efficiency.
ad hoc to paper A single random initialization and one optimization trajectory reaches a representative optimum.
Random Gaussian noise initialization is used; no multiple seeds or restarts are analyzed, so the reported designs may be seed-dependent.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Inverse Microparticle Design for Enhanced Optical Trapping and Detection Efficiency in All Six Degrees of Freedom." pith.science (2026). https://pith.science/paper/2HBF74IA

@misc{pith2026250601837,
  author       = {Pith},
  title        = {Pith review of: Inverse Microparticle Design for Enhanced Optical Trapping and Detection Efficiency in All Six Degrees of Freedom},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/2HBF74IA}},
  note         = {Machine review of arXiv:2506.01837}
}

read the original abstract

Achieving quantum-limited motional control of optically trapped particles beyond the sub-micrometer scale is an outstanding problem in levitated optomechanics. A key obstacle is solving the light scattering problem and identifying particle geometries that allow stable trapping and efficient motional detection of their center of mass and rotational motion in three dimensions. Here, we present a computational framework that combines an efficient electromagnetic scattering solver with the adjoint method to inversely design printable microparticles tailored for levitated optomechanics. Our method allows identifying optimized geometries, characterized by enhanced optical trapping and detection efficiencies compared to conventional microspheres. This improves the feasibility of quantum-limited motional control of all translational and rotational degrees of freedom in a standard standing-wave optical trap.

Discussion (0). Sign in 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

Cooling of an optically levitated nanoparticle via measurement-free coherent feedback
quant-ph 2025-06 conditional novelty 6.0 of 10

Coherent, measurement-free optical feedback cools a levitated nanoparticle to about 344 phonons, with phase noise identified as the main barrier to ground-state cooling.

Reference graph

Works this paper leans on

11 extracted references · 11 canonical work pages · cited by 1 Pith paper

[1]

& Vellekoop, I

Osnabrugge, G., Leedumrongwatthanakun, S. & Vellekoop, I. M. A convergent Born series for solving the inhomogeneous Helmholtz equation in arbitrarily large media. Journal of Computational Physics 322, 113–124 (2016)

work page 2016
[2]

& Kienle, A

Krüger, B., Brenner, T. & Kienle, A. Solution of the inhomogeneous Maxwell’s equations using a Born series. Opt. Express, OE 25, 25165–25182 (2017)

work page 2017
[3]

Landau, L. D. et al. Electrodynamics of Continuous Media. vol. 8 (elsevier, 2013)

work page 2013
[4]

Hüpfl, J. et al. Optimal cooling of multiple levitated particles: Theory of far-field wavefront shaping. Phys. Rev. A 107, 023112 (2023). 31

work page 2023
[5]

& Romero -Isart, O

Maurer, P., Gonzalez -Ballestero, C. & Romero -Isart, O. Quantum theory of light interaction with a Lorenz -Mie particle: Optical detection and three-dimensional ground-state cooling. Phys. Rev. A 108, 033714 (2023)

work page 2023
[6]

Wang, L. et al. Optimal position detection of an optically levitated Mie particle. J. Opt. Soc. Am. B, JOSAB 42, 645–653 (2025)

work page 2025
[7]

& Stickler, B

Rudolph, H., Delić, U., Aspelmeyer, M., Hornberger, K. & Stickler, B. A. Force -Gradient Sensing and Entanglement via Feedback Cooling of Interacting Nanoparticles. Phys. Rev. Lett. 129, 193602 (2022)

work page 2022
[8]

& Park, Y

Lee, M., Hugonnet, H. & Park, Y. Inverse problem solver for multiple light scattering using modified Born series. Optica, OPTICA 9, 177–182 (2022)

work page 2022

Show all 11 references

[9]

Petersen, K. B. & Pedersen, M. S. The matrix cookbook. Technical University of Denmark 7, 510 (2008)

2008
[10]

W., Minkov, M., Williamson, I

Hughes, T. W., Minkov, M., Williamson, I. A. D. & Fan, S. Adjoint Method and Inverse Design for Nonlinear Nanophotonic Devices. ACS Photonics 5, 4781–4787 (2018)

2018
[11]

Christiansen, R. E. & Sigmund, O. Inverse design in photonics by topology optimization: tutorial. J. Opt. Soc. Am. B, JOSAB 38, 496– 509 (2021)

2021

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.

[1] [1]

& Vellekoop, I

Osnabrugge, G., Leedumrongwatthanakun, S. & Vellekoop, I. M. A convergent Born series for solving the inhomogeneous Helmholtz equation in arbitrarily large media. Journal of Computational Physics 322, 113–124 (2016)

work page 2016

[2] [2]

& Kienle, A

Krüger, B., Brenner, T. & Kienle, A. Solution of the inhomogeneous Maxwell’s equations using a Born series. Opt. Express, OE 25, 25165–25182 (2017)

work page 2017

[3] [3]

Landau, L. D. et al. Electrodynamics of Continuous Media. vol. 8 (elsevier, 2013)

work page 2013

[4] [4]

Hüpfl, J. et al. Optimal cooling of multiple levitated particles: Theory of far-field wavefront shaping. Phys. Rev. A 107, 023112 (2023). 31

work page 2023

[5] [5]

& Romero -Isart, O

Maurer, P., Gonzalez -Ballestero, C. & Romero -Isart, O. Quantum theory of light interaction with a Lorenz -Mie particle: Optical detection and three-dimensional ground-state cooling. Phys. Rev. A 108, 033714 (2023)

work page 2023

[6] [6]

Wang, L. et al. Optimal position detection of an optically levitated Mie particle. J. Opt. Soc. Am. B, JOSAB 42, 645–653 (2025)

work page 2025

[7] [7]

& Stickler, B

Rudolph, H., Delić, U., Aspelmeyer, M., Hornberger, K. & Stickler, B. A. Force -Gradient Sensing and Entanglement via Feedback Cooling of Interacting Nanoparticles. Phys. Rev. Lett. 129, 193602 (2022)

work page 2022

[8] [8]

& Park, Y

Lee, M., Hugonnet, H. & Park, Y. Inverse problem solver for multiple light scattering using modified Born series. Optica, OPTICA 9, 177–182 (2022)

work page 2022