Pith. sign in

REVIEW 3 major objections 5 minor 7 references

Impact of microlens shape on the performance of Laser Guide Star wavefront sensors for ELT-class telescopes

T0 review · 3 major / 5 minor · reviewed 2026-08-01 · deepseek-v4-flash

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

Pith's one-line read Real microlenses degrade LGS centroid accuracy by 1.8–2.4×, equivalent to a ~2× photon penalty.

desk verdict The single-subaperture result is solid and useful; the system-level 1.25–1.4× mitigation factor is an extrapolation that needs a direct end-to-end check. read the letter →

arxiv 2607.26543 v1 pith:KF4AUUIH submitted 2026-07-29 astro-ph.IM

Paul Rouquette , Benoit Neichel , Cédric Taisir Heritierc , Pierre Jouve , Anne Costille , Kjetil Dohlen , Thierry Fusco This is my paper
classification astro-ph.IM
keywords Shack-HartmannwavefrontsensormicrolensarraylaserguidestaradaptiveopticscentroidaccuracycenterofgravitytomographicAOStrehlratio
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 establishes that manufacturing errors in the microlens arrays of laser-guide-star wavefront sensors materially degrade centroiding: compared with an ideal lens, a real lens increases center-of-gravity variance by a factor of 1.8–2.4, equivalent to needing about twice as many photons for the same accuracy. The penalty shrinks as the laser spot becomes more elongated and as microlens sag increases, because the real lens's phase error mainly redistributes light from the central peak to the wings. The paper then translates this per-subaperture penalty into an effective flux loss and propagates it into six-laser-guide-star tomographic adaptive-optics simulations, finding that measurement redundancy and MMSE reconstructor weighting reduce the system-level K-band Strehl flux loss to only 1.25–1.4×. An optical-bench measurement on three prototype arrays gives excess-variance ratios in the same range (medians 1.8–2.35), and also demonstrates a fast full-array qualification method. If the analysis is right, single-subaperture photon budgets for ELT-class systems must be roughly doubled, while the total laser power does not need to double.

What carries the argument

The carrying mechanism is the 'per-subaperture effective flux-loss model': the tomographic simulation replaces each real microlens with an ideal lens whose photon count is divided by the high-flux asymptote of the single-lens CoG-variance ratio. The reduction is justified because CoG variance scales as 1/N in the photon-noise regime, so a phase-degraded spot behaves like a fainter ideal spot. The single-lens ratios are computed from an analytic weighted center-of-gravity variance expression (extended to elongated, asymmetric spots) and verified with Monte Carlo spot simulations using measured phase maps.

What would settle it

Directly simulate a tomographic AO loop inserting the measured phase maps (or a representative subset of lenslets) into the subapertures instead of using the scalar flux-loss model, and compare the K-band Strehl at N = 1000 photons/subaperture/frame with the paper's 1.25–1.4× loss prediction. Alternatively, on the bench, compare prototypes #2 and #3 at identical elongation and flux: if the lower-sag lens systematically beats the higher-sag lens, the sag trend used for the manufacturing specification is not robust.

Watch

Extended reading notes

Core claim

The central discovery is a quantitative bridge from microlens manufacturing error to adaptive-optics performance. Interferometric surface maps of prototype microlenses are converted to phase errors; the resulting spot PSFs are fed into weighted center-of-gravity variance calculations under photon and read-out noise. The real lens makes CoG variance 1.8–2.4 times larger than an ideal lens over the operating flux range, and this ratio is the same as requiring ~1.9× more photons. The ratio is not constant: it drops toward 1.3–1.5 for spot elongations above 8 arcsec and improves as lens sag grows. The authors then model each subaperture as an ideal lens with reduced throughput and show that a si

Load-bearing premise

The load-bearing assumption is that a real microlens's phase error behaves purely as a per-subaperture throughput reduction, so that system-level performance can be predicted by dividing each subaperture's photon count by a scalar factor read from the single-lens variance ratio — a step that is not validated by direct end-to-end simulation with the measured phase maps, and the manufacturing recommendation additionally rests on only three prototypes whose bench ranking does no

Editorial extensions

If this is right

  • LGS wavefront sensor designs should budget roughly twice the photons per subaperture if microlens phase errors remain at prototype levels.
  • Microlens manufacturers get a concrete target: higher sag (closer to the ideal sphere) reduces the centroiding penalty, and the penalty is worst for the least-elongated spots near the launcher.
  • System-level AO performance is far less sensitive: with six LGS and MMSE reconstruction, the effective K-band Strehl flux loss is 1.25–1.4×, so the laser power budget does not need to double.
  • The full-array bench method (per-subaperture excess-variance ratio) can qualify production microlens arrays in a single measurement, replacing slow individual lenslet profiling.
  • The same flux-loss approach should apply to any low-flux Shack-Hartmann sensor where the spot is photon-noise limited.

Reading between the lines

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

  • Inference: The 1/N-based flux-loss mapping is only proven for the photon-noise regime; extending it to read-out-noise-dominated or undersampled spots would require re-deriving the loss factors, since the RON term couples the two axes.
  • Inference: The bench data hint that inter-lenslet scatter matters as much as mean sag — prototype #2 (sag 6.53 μm) shows a lower median excess ratio (1.81) than prototype #3 (sag 7.02 μm, median 1.90) — so a qualification metric based on the distribution, not just the mean, may be a better manufacturing spec.
  • Inference: Because the penalty decreases with elongation, a system designer could deliberately stretch spots (e.g., via a larger asterism or beam modulation) to mask microlens errors, trading some intrinsic elongation noise for reduced sensitivity to lens shape — a trade the paper does not explore.
  • Inference: A direct end-to-end check that inserts measured phase maps into a handful of representative subapertures of a full tomographic simulation would test whether the scalar throughput approximation misses correlated errors; such a test is now practical with the bench data in hand.
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 / 5 minor

Summary. The paper quantifies the impact of microlens surface-shape imperfections on the centroiding accuracy of laser-guide-star Shack–Hartmann wavefront sensors for ELT-class telescopes. Using interferometric surface maps of three prototype microlens arrays, the authors compute phase errors, simulate spot images, and evaluate weighted center-of-gravity (WCoG) variance as a function of flux and LGS elongation. They report a real-to-ideal CoG variance ratio of 1.8–2.4×, interpret this as an effective flux loss of ~2×, and propagate this loss into a six-LGS tomographic AO simulation by modeling each subaperture's throughput reduction. The system-level K-band Strehl-ratio penalty is claimed to be only 1.25–1.4×, thanks to redundancy and MMSE reconstructor weighting. Experimental validation on an optical bench at LAM measures per-subaperture excess variance ratios of 1.81–2.35 for the three prototypes. The paper concludes that higher sag improves performance and that the methodology is broadly applicable to low-flux Shack–Hartmann systems.

Significance. The single-subaperture characterization is a valuable and credible contribution. The combination of analytical WCoG variance expressions, Monte Carlo simulations, interferometric phase maps, and bench measurements gives a coherent account of the per-lenslet penalty, and the full-array bench characterization method is a practical advance for qualifying microlens arrays. If the system-level propagation were independently validated, the conclusion that the tomographic AO system sees only a 1.25–1.4× effective flux loss (rather than the 2× suggested by single-subaperture results) would be important for LGS AO design margins. However, the system-level result rests on an asserted equivalence that is not directly tested, and the experimental prototype ranking does not support the claimed monotonic sag trend. These issues are load-bearing for the paper's central claims.

major comments (3)
  1. [§3, Fig. 11] The central system-level claim depends on the assertion that a real microlens can be modeled as a per-subaperture throughput reduction, with flux-loss factors read from the high-flux asymptote of the CoG variance ratio. This equivalence is asserted but not derived or validated. Section 2.3.2 (Fig. 6b) shows that the CoG variance ratio varies with flux (1.8–2.4 across the flux range), and Eq. (6) contains both 1/N and 1/N^2 terms, so a single effective flux-loss factor may not capture the low-flux regime relevant to the tomographic simulation. The bench validation (Section 4) confirms only the per-subaperture variance penalty; it does not test the propagation through the MMSE reconstructor. Please add a direct end-to-end test using measured phase maps for at least a subset of subapertures, or a clear demonstration that the reduction to throughput is exact in the regime used. Without this,
  2. [§4.2, Fig. 16b] The paper claims a monotonic trend that higher sag improves performance, but the bench data contradict this ordering: prototype #2 (sag 6.53 μm) has a median excess-variance ratio of 1.81, while prototype #3 (sag 7.02 μm) has a median of 1.90 (Fig. 16b). The text states these results are consistent with the sag-dependent trend, but the ordering is reversed. No error bars or statistical significance are provided for the median ratios, so it is unclear whether the difference is meaningful. This inconsistency undermines the manufacturing specification proposed in §2.3.3 and §5. Please reassess whether sag is the controlling parameter or whether other surface-error metrics (e.g., mid-spatial-frequency ripple, residual asphericity) better explain the measured ranking.
  3. [§3, Fig. 11] The system-level Strehl-ratio curves in Fig. 11 are presented without error bars, sensitivity analysis, or comparison against a direct phase-map-inclusive simulation. The mitigation factors (1.25× and 1.4×) are quoted as precise values, but no uncertainty is propagated from the single-subaperture ratio measurements or from the model reduction. The claim that MMSE reconstructor weighting and multi-LGS redundancy explain the mitigation would be strengthened by a controlled run that isolates these effects (e.g., using a pure throughput reduction versus a phase-error map with the same variance). Please provide error bars or at least a sensitivity range.
minor comments (5)
  1. [§2.3.2] There are two subsections numbered 2.3.2: 'Effect of flux' and 'Effect of elongation'. Please renumber.
  2. [Abstract] Typo: 'Str ehl' should be 'Strehl'. Also, the abstract claims results are 'directly applicable to any ELT-class or future large-telescope LGS instrument', but the simulations are based on the MORFEO configuration; please temper the generality claim or provide supporting argument.
  3. [§2.3.1, Eq. (6)] The analytical variance expressions are derived for Gaussian spots, but the simulated sodium-layer profiles (Fig. 3) are not Gaussian. The paper states results are 'quite similar' but does not quantify the difference. A brief quantitative comparison would help.
  4. [§4.2, Fig. 16] The caption for Fig. 16 gives median ratios for the three arrays, but the text in §4.2 says prototype #2 'yields the best performance overall'. This is correct numerically, but it conflicts with the earlier statement in §2.3.3 that prototype #3 (highest sag) has the smallest penalty. Please reconcile these statements explicitly.
  5. [References] Some references lack page numbers or are incomplete (e.g., Ref. [2] and Ref. [7]). Please check journal formatting requirements.

Circularity Check

0 steps flagged · score 1.0 of 10

No significant circularity: the system-level flux-loss result is a forward-model extrapolation, not a reduction of the output to the input; self-citations are not load-bearing.

full rationale

The paper's derivation chain is largely self-contained. The single-lens CoG variance penalty (1.8–2.4×) is computed from interferometric surface measurements via a phase-to-PSF-to-centroid model, cross-checked against an analytic WCoG formula (Eq. 6) and experimentally validated on the LAM bench (Section 4). The system-level propagation in Section 3 is an explicit modeling approximation: the per-subaperture flux-loss factors are read from the Section 2 variance-ratio curves and injected as throughput reductions into a tomographic AO simulation. The resulting K-band Strehl flux losses (1.25–1.4×) are not defined to equal the injected single-lens factors (1.5–2×); they are a distinct output produced by the MMSE reconstructor's redundancy, so this is a forward-model extrapolation rather than a circular identity. The absence of a direct end-to-end simulation with real phase maps is a validation gap and a correctness risk, not a circularity. The paper's self-citations (e.g., Refs. [5], [7]) support specific bench or reconstructor properties but do not carry the central derivation; the WCoG variance formula is an established external result (Thomas et al. 2006). One internal inconsistency exists — bench data in Fig. 16b show prototype #2 (median 1.81) outperforming prototype #3 (median 1.90), contradicting the claimed 'higher sag is better' trend — but this is an empirical consistency issue, not a circular step. Overall, no load-bearing circularity was found; score 1 reflects only minor self-citation presence.

Assumptions & free parameters 2 free parameters · 6 assumptions · 0 invented entities

The per-subaperture CoG penalty is grounded in measured surface maps and Monte Carlo simulation, so the ledger is light. The main added assumption is the effective flux-loss equivalence used to scale from one lenslet to the full tomographic system; the rest are standard AO/optics modeling choices (Gaussian noise, uniform sodium layer, MMSE reconstructor). No new physical entities are introduced.

free parameters (2)
  • WCoG weighting-to-spot FWHM ratio β = 2
    Chosen as a typical operating value (§2.3.1); affects the analytical variance normalization and all ratio predictions.
  • Effective subaperture flux-loss factor per prototype/elongation = ≈1.5× (#3), ≈2× (#1) single-subaperture; system-level 1.25×/1.4×
    Read from the high-flux asymptote of the same CoG-variance-ratio model (§3) and applied as throughput in the E2E simulation; it is a derived model parameter, not independently measured at system level.
assumptions (6)
  • domain assumption WCoG variance can be computed analytically assuming Gaussian spot and Gaussian weighting, with Poisson photon noise and Gaussian RON.
    Equations (3)–(6) in §2.3.1; used as the baseline for all CoG variance predictions and for the experimental excess-ratio R.
  • domain assumption Real-lens phase error is obtained by Δ=(n−1)(z_real−z_ideal) and propagated by Fourier optics to the PSF.
    Section 2.1; assumes measured surface height fully describes the optical error and that scalar diffraction applies.
  • domain assumption LGS spot can be modeled by a uniform sodium-layer profile, with 14×14 pixels of ~1.2″ per subaperture.
    Section 2.2 and Table 1; the elongation profile drives the relative insensitivity of elongated spots to lens errors.
  • ad hoc to paper In the photon-noise regime CoG variance scales as 1/N, so a real microlens is equivalent to a throughput reduction.
    Section 3, first paragraph; this is the load-bearing reduction that turns phase maps into flux-loss factors and is not independently validated against direct phase-map E2E.
  • domain assumption MMSE tomographic reconstructor downweights noisy subapertures via its noise covariance matrix; six-LGS redundancy mitigates the per-subaperture penalty.
    Section 3 and Ref. [7]; standard AO reconstructor behavior, but the mitigation factor depends on this assumption.
  • domain assumption Atmosphere represented by a 9-layer Cerro Armazones profile with r0=13.85 cm, L0=25 m at zenith 30°.
    Table 1; standard site modeling, but the system-level Strehl numbers depend on it.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Impact of microlens shape on the performance of Laser Guide Star wavefront sensors for ELT-class telescopes." pith.science (2026). https://pith.science/paper/KF4AUUIH

@misc{pith2026260726543,
  author       = {Pith},
  title        = {Pith review of: Impact of microlens shape on the performance of Laser Guide Star wavefront sensors for ELT-class telescopes},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/KF4AUUIH}},
  note         = {Machine review of arXiv:2607.26543}
}
read the original abstract

Laser Guide Star (LGS) adaptive optics systems on extremely large telescopes (ELTs) rely on Shack-Hartmann wavefront sensors (SHWFS) equipped with large-format microlens arrays. Manufacturing imperfections in the microlens surface profile degrade spot quality and reduce centroiding accuracy, yet this effect is rarely quantified in the context of full AO system performance. This paper presents a comprehensive characterization of the impact of microlens shape on LGS wavefront sensing, using the MORFEO-HARMONI LGS wavefront sensor design as the primary test case, results are directly applicable to any ELT-class or future large-telescope LGS instrument, including systems on the GMT and TMT. Starting from interferometric surface profile measurements of prototype microlenses, we derive the induced phase errors and compute the degradation of center-of-gravity (CoG) spot detection accuracy as a function of LGS elongation and flux. A real microlens degrades CoG variance by a factor of 1.8 to 2.4 compared to an ideal lens, equivalent to requiring approximately twice the photon flux to maintain the same measurement accuracy. This effect is shown to decrease with increasing LGS elongation, and to improve with higher microlens sag. The per-subaperture flux-loss model is then propagated into tomographic AO end-to-end simulations, where measurement redundancy across multiple LGS and MMSE reconstructor weighting partially mitigate the penalty, reducing the effective Strehl ratio flux loss to factors of 1.25-1.4. Experimental validation is provided with the LGS wavefront sensor optical bench prototype at LAM, which also enables full-array characterization in a single measurement. The methodology is broadly applicable to any Shack-Hartmann system operating in a low-flux regime.

Figures

Figures reproduced from arXiv: 2607.26543 by the authors.

Figure 1
Figure 1. (a) Altitude profile of a prototype microlens; lens center at (x,y) [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. (a) PSF of the ideal lens. (b) PSF of the real lens. (c) Comparison of PSF profiles along the x [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. LGS intensity profiles used in simulations: (a) non [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗
Figures from the paper (10 more)
Figure 4
Figure 4. Figure 4: Light distribution on the detector before pixel sampling, for ideal and real lenses, with non [PITH_FULL_IMAGE:figures/full_fig_p004_4.png]
Figure 5
Figure 5. Figure 5: Sampled light distribution on detector pixels (14×14 per subaperture) for ideal and real lenses, non [PITH_FULL_IMAGE:figures/full_fig_p004_5.png]
Figure 7
Figure 7. Figure 7: CoG variance as a function of x-axis elongation for a Gaussian spot, with ideal and real microlenses: (a) along the elongation axis; (b) along the perpendicular axis [PITH_FULL_IMAGE:figures/full_fig_p007_7.png]
Figure 8
Figure 8. Figure 8: a), b) as [PITH_FULL_IMAGE:figures/full_fig_p008_8.png]
Figure 10
Figure 10. Figure 10: LGS spot elongation pattern across the 68×68 SH WFS pupil. Elongation ranges from near [PITH_FULL_IMAGE:figures/full_fig_p010_10.png]
Figure 12
Figure 12. Figure 12: LGS wavefront sensor optical bench at LAM. Main subsystems: beam expander, spectral filter (589 [PITH_FULL_IMAGE:figures/full_fig_p011_12.png]
Figure 13
Figure 13. Figure 13: C-BLUE camera images: (a) full detector showing the complete pupil (68×68 subapertures); (b) zoom on a 5×5 sub￾array [PITH_FULL_IMAGE:figures/full_fig_p012_13.png]
Figure 14
Figure 14. Figure 14: (a) Mean image of one subaperture averaged over 1000 frames at high flux. (b) CoG variance vs. flux for that [PITH_FULL_IMAGE:figures/full_fig_p012_14.png]
Figure 15
Figure 15. Figure 15: Pupil maps (one dot per subaperture): (a) flux (photons/subaperture/frame); (b) FWHM along x [PITH_FULL_IMAGE:figures/full_fig_p013_15.png]
Figure 16
Figure 16. Figure 16: Excess CoG variance ratio (measured/theoretical) for three microlens arrays: (a) spatial map across the pupil; (b) histograms for arrays #1 (red, median = 2.35), #2 (purple, median = 1.81), and #3 (green, median = 1.9). 5. CONCLUSION We have presented a comprehensive …

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

7 extracted references

  1. [1]

    HARMONI at ELT: designing a laser guide star wavefront sensor for the ELT,

    A. Costille et al., "HARMONI at ELT: designing a laser guide star wavefront sensor for the ELT," Proc. SPIE 12185, Adaptive Optics Systems VIII, 121855X (2022)

    2022

  2. [2]

    HARMONI at ELT: full scale prototype of the laser guide star wavefront sensor,

    P. Jouve et al., "HARMONI at ELT: full scale prototype of the laser guide star wavefront sensor," Proc. SPIE 12185, Adaptive Optics Systems VIII (2022)

    2022

  3. [3]

    Nicolle, Analyse de front d’onde pour les optiques adaptatives de nouvelle génération, PhD thesis, Université Paris XI (2006)

    M. Nicolle, Analyse de front d’onde pour les optiques adaptatives de nouvelle génération, PhD thesis, Université Paris XI (2006)

    2006

  4. [4]

    Comparison of centroid computation algorithms in a Shack -Hartmann sensor,

    S. Thomas, T. Fusco, A. Tokovinin et al., "Comparison of centroid computation algorithms in a Shack -Hartmann sensor," Mon. Not. R. Astron. Soc. 371, 323–336 (2006)

    2006

  5. [5]

    Results from the HARMONI Laser Guide Star wavefront sensor prototype,

    P. Rouquette et al., "Results from the HARMONI Laser Guide Star wavefront sensor prototype," Proc. AO4ELT7 (2023)

    2023

  6. [6]

    Performance of a CMOS sensor for laser guide star wavefront sensing,

    Z. Ke, F. P. Bustos, J. Atwood et al., "Performance of a CMOS sensor for laser guide star wavefront sensing," J. Astron. Telesc. Instrum. Syst. 8 (2022)

    2022

  7. [7]

    A story of errors and bias: the optimization of the LGS WFS for HARMONI,

    T. Fusco, B. Neichel, C. Correia et al., "A story of errors and bias: the optimization of the LGS WFS for HARMONI," Proc. AO4ELT6 (2019)

    2019

Pith tools

Reviewed August 1, 2026 · model on record in the stance chip above.