REVIEW 3 major objections 3 minor 24 references
Reconstruction of missing information in diffraction patterns and holograms by iterative phase retrieval
T0 review · 3 major / 3 minor · reviewed 2026-08-14 · deepseek-v4-flash
Pith's one-line read The paper shows that an object can be reconstructed from a diffraction pattern or hologram even when most measured intensity samples are missing, provided the experiment is oversampled and the missing pixels are randomly scattered.
desk verdict A useful counting bound for missing-pixel tolerance in CDI and holography, but the paper sells a necessary condition as sufficient and only the symmetrized pipeline supports the headline example. read the letter →
The pith
A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.
The reading
What carries the argument
The load-bearing device is a counting inequality: the number of measured intensity values must exceed the number of unknown object pixels. Written with the linear oversampling ratio $\sigma = N/N_0$, it becomes the missing-pixel thresholds $f < 1 - 2/\sigma^2$ for diffraction patterns and $f < 1 - 1/\sigma^2$ for holograms. The algorithm that carries the recovery is the hybrid input-output (HIO) iterative phase retrieval routine with a tight object support; at each iteration the missing Fourier or hologram amplitudes are replaced by the current estimate, so the support constraint effectively fills gaps in the data.
What would settle it
Take an object different from the one used here and add Poisson noise to its diffraction pattern, delete a random fraction $f$ just below $1 - 2/\sigma^2$ at $\sigma = 8$, and run the same HIO protocol with a loose support. If reconstructions fail or stagnate at error levels well above the no-missing baseline, the counting bound is necessary but not sufficient.
Extended reading notes
Core claim
The paper's central claim is that the information content of an oversampled diffraction measurement is far larger than the object itself, and that excess information can be spent on missing pixels. Counting equations against unknowns gives $f < 1 - 2/\sigma^2$ for a real-valued object in coherent diffraction imaging and $f < 1 - 1/\sigma^2$ for an amplitude object in holography. Simulations with the hybrid input-output algorithm, replacing each missing intensity by the current iterate, confirm the counting picture: at high oversampling ratios the object is recovered even when most pixels are missing, and symmetrization of the centrosymmetric diffraction pattern improves convergence and extends recovery toward the theoretical limit. The missing values themselves are reconstructed along with the object, so no separate inpainting step is needed.
Load-bearing premise
The paper assumes that satisfying the counting inequality is enough for the iterative algorithm to converge to the true object; this is supported by noise-free simulations of one real-valued test object with known support, not by a proof.
Editorial extensions
If this is right
- Detectors with random dead or saturated pixels can tolerate high pixel-loss fractions when the measurement is oversampled; no separate inpainting step is needed.
- At linear oversampling ratio 8, retaining only about 5% of the diffraction-pattern intensities can be enough to recover both the object and the missing values for a real-valued object.
- In-line holography inherits the same resilience, with the reference-wave extent playing the role of oversampling and a threshold of $f < 1 - 1/\sigma^2$.
- Centrally missing values, such as those hidden by a beamstop or a saturated central spot, are a different problem: even 1% missing in the center defeats recovery, so such gaps require other strategies.
Reading between the lines
- In experiments with noise and imperfect support, the safe missing fraction will probably sit below the counting bound; a calibration curve of $f$ versus reconstruction error would be a direct test.
- The same counting logic could be applied to complex-valued or three-dimensional objects, giving thresholds of the same form; whether iterative phase retrieval reaches them is an open question.
- The center-missing failure is a clue that low spatial frequencies carry the object's background; combining phase retrieval with extrapolation or a prior on low frequencies might rescue beamstop data.
- If these thresholds hold under noise, they suggest a detector-design trade-off: oversample more in exchange for accepting more defective pixels.
Signed reviews
Editorial analysis
A structured set of objections, weighed in public.
Referee Report
Summary. This paper studies the recovery of missing intensity pixels in coherent diffraction imaging (CDI) and in-line holography using iterative phase retrieval. For a real-valued object, the author derives from a simple counting argument that the fraction f of missing pixels should not exceed 1 - 2/σ² for a diffraction pattern and 1 - 1/σ² for a hologram, where σ is the linear oversampling ratio. Simulations on a single real-valued amplitude object (the "man" image) with known tight support and no noise show that random missing pixels can be recovered up to high f when the diffraction pattern is centro-symmetrised, that missing pixels concentrated in the centre ruin reconstruction even at small f, and that holograms tolerate somewhat larger missing fractions. The paper concludes that even 5% of measured intensity values at σ = 8 suffice for CDI and 6% at σ = 4 for holography.
Significance. If the claimed bounds were established as sufficient conditions, they would provide simple quantitative guidance for experiments with dead pixels, beamstops, or saturated regions, which is practically valuable. The counting derivation is transparent and parameter-free, and the paper makes a useful, well-illustrated observation that missing low-frequency information in the centre of a diffraction pattern is far more damaging than uniformly distributed missing pixels. The error-metric comparison in Section 2.3 is also a useful practical note. However, the paper does not prove that the counting condition is sufficient; its own unsymmetrised simulations at σ = 8 fail at f = 0.8–0.9, well below the claimed limit of 0.969. The headline success at f = 0.95–0.97 relies on a symmetrisation step that changes the effective missing fraction, and all simulations are noise-free, use one object type, and assume a known tight support. The central claim therefore outruns the evidence, though the underlying phenomenon is real and the paper can be revised to state the necessary-condition status accurately.
major comments (3)
- [Section 2.1, Eqs. (2)–(4)] There is an algebraic inconsistency between Eq. (2) and Eq. (4). From Eq. (2), N²/2 > N0², and with σ = N/N0, one obtains σ > √2, not σ > 2. If the intended statement is the standard, more conservative linear oversampling condition σ > 2, then Eq. (2) needs to be changed (for example, to N²/4 > N0² if one counts independent equations after accounting for both the real-valuedness and the centrosymmetry of the intensity). As written, the derivation of Eq. (7) inherits this inconsistency, and the factor of 2 inside Eq. (7) deserves a clear justification.
- [Section 2.2.1 and Abstract; Eq. (7)] The claim that f < 1 - 2/σ² is the maximum allowable missing fraction (Abstract, Section 2.1, Conclusions) is not supported by the paper's own unsymmetrised results. For σ = 8, Eq. (7) allows f < 0.969, yet the unsymmetrised protocol of Section 2.2.1 fails at f = 0.8 and f = 0.9, with errors of 9.97×10⁻³ and 1.11×10⁻³ in Table 2 and stagnation in Fig. 3u. For σ = 4, the unsymmetrised protocol fails at f = 0.6 (Table 1), although Eq. (7) gives f < 0.875. The successes at f = 0.95–0.97 (Fig. 5, Table 4) are obtained only after the symmetrisation step in Section 2.2.2, which the paper itself notes changes the effective number of missing pixels. The counting argument in Eq. (6) is a necessary condition for the existence of a solution, not a proof that iterative phase retrieval will succeed; the manuscript should either rephrase the headline claim as a necessary condition, or explicitly and prominently qualify that the sufficiency assertion holds only for the symmetrised, noise-free, known-support examples actually simulated.
- [Section 3.1 (Eq. 14) and Section 3.2] The holography bound f < 1 - 1/σ² in Eq. (14) is likewise derived solely from a counting argument. The supporting simulation is a single noise-free in-line hologram of a real-valued amplitude object with a known tight support and a fixed reconstruction pipeline. The text says that reconstructions are "identical" up to f = 0.9 but only "still resemble" the original at f = 0.95, which is above the claimed limit of 0.938; this is not a demonstration that the bound is tight or sufficient. As with the CDI case, the paper should state clearly that Eq. (14) is a necessary condition, and that the empirical evidence for its sufficiency is limited to one idealised example. The current wording, including the Abstract's "should not exceed", overstates the strength of the result.
minor comments (3)
- [Section 2.3] In the paragraph on error metrics, the sentence "the reconstructions with the lowest error as defined by Eq. 7" should read "Eq. 8", since Eq. 7 is the missing-fraction bound, not an error metric.
- [Section 5 (Conclusions)] There is a typo in the sentence "a single inverse FT delivers a poorer reconstruction that the that obtained by iterative phase retrieval"; it should be "than that obtained".
- [Section 2.2.1, Fig. 2 caption] The figure caption lists subplot labels (a)–(t) for f = 0 to f = 0.9 but the text and Table 1 omit f = 0.6 and f = 0.7 in the caption listing; please check the caption for consistency.
Circularity Check
No significant circularity: the missing-pixel bounds are derived by a parameter-free counting argument and tested against independent simulations; self-citations supply reconstruction algorithms, not the central result.
full rationale
The paper's central result, Eq. (7) f < 1 - 2/σ² and Eq. (14) f < 1 - 1/σ², is obtained by an explicit counting argument: for a real-valued object, the number of independent measured intensity samples after missing pixels, (1-f)N²/2, must exceed the number of unknown object pixels N0², which algebraically gives the stated bound in terms of the linear oversampling ratio σ = N/N0. No parameter is fitted to the target quantity, and the bound is not defined in terms of the simulation outcomes. The numerical simulations in Sections 2.2 and 3.2 are independent checks rather than inputs used to derive the inequality. The author's prior work [12,18,19] is cited for the iterative reconstruction algorithms and for background on oversampling failure, but those citations do not supply the bound; the same protocols could be replaced by standard HIO and angular-spectrum propagation. A genuine limitation is that the paper treats the counting condition as sufficient as well as necessary, and its own unsymmetrized σ=8 runs fail at f=0.8 even though Eq. (7) allows f < 0.969; successful high-f recovery is reported only after symmetrization (Section 2.2.2). This is an evidentiary/sufficiency gap and a correctness risk, not a circularity: the claim may be under-supported, but it is not true by construction. Therefore no circular step is exhibited.
Assumptions & free parameters
free parameters (4)
- HIO feedback parameter =
0.9
- Number of iterations =
2000
- Number of reconstruction trials =
100 (10 best averaged)
- Smoothing kernel schedule =
3x3 kernel, every 20 iterations
assumptions (6)
- domain assumption The system of equations can in principle have a solution if the number of equations exceeds the number of unknowns.
- standard math The diffraction pattern of a real-valued object is centro-symmetric, providing N^2/2 independent measurements.
- domain assumption Measured intensities are noise-free.
- domain assumption Missing pixels are randomly distributed over the diffraction pattern.
- domain assumption The object support is known and tight (128x128 pixels).
- domain assumption In holography, the object is an amplitude object and the reconstruction uses the convolution model and twin-image elimination.
Cite this review
Pith. "Pith review of Reconstruction of missing information in diffraction patterns and holograms by iterative phase retrieval." pith.science (2026). https://pith.science/paper/JPITZ5QR
@misc{pith2026190810205,
author = {Pith},
title = {Pith review of: Reconstruction of missing information in diffraction patterns and holograms by iterative phase retrieval},
year = {2026},
howpublished = {\url{https://pith.science/paper/JPITZ5QR}},
note = {Machine review of arXiv:1908.10205}
}
read the original abstract
It is demonstrated that an object distribution can be successfully retrieved from its diffraction pattern or hologram, even if some of the measured intensity samples are missing. The maximum allowable number of missing values depends on the linear oversampling ratio s, where the higher the value of s, the more intensity samples can be missing. For a real-valued object, the ratio of missing pixels to the total number of pixels should not exceed (1 - 2/s^2) or (1 - 1/s^2) in the acquired diffraction pattern or hologram, respectively. For example, even 5% of the measured intensity values at an oversampling ratio of s = 8 are sufficient to simultaneously retrieve the object distribution and the missing intensity values. It is important that the missing intensity values should not be concentrated in the centre, but should be randomly distributed over the acquired diffraction pattern.
Figures
Figures from the paper (1 more)
Reference graph
Works this paper leans on
-
[1]
Kotelnikov, On the transmission capacity of "ether" and wire in electrocommunications, Proc
V.A. Kotelnikov, On the transmission capacity of "ether" and wire in electrocommunications, Proc. 1st All-Union Conf. Technological Reconstruction of the Commun. Sector and Low-Current Eng., (1933) 1–19
work page 1933
-
[2]
C.E. Shannon, Communication in the presence of noise, Proceedings Institute of Radio Engineers, 37 (1949) 10–21
work page 1949
-
[3]
J.W. Miao, P. Charalambous, J. Kirz, D. Sayre, Extending the methodology of X-ray crystallography to allow imaging of micrometre-sized non-crystalline specimens, Nature, 400 (1999) 342–344
work page 1999
-
[4]
Gabor, A new microscopic principle, Nature, 161 (1948) 777–778
D. Gabor, A new microscopic principle, Nature, 161 (1948) 777–778
work page 1948
-
[5]
Gabor, Microscopy by reconstructed wave-fronts, Proc
D. Gabor, Microscopy by reconstructed wave-fronts, Proc. R. Soc. A, 197 (1949) 454–487
work page 1949
-
[6]
T. Latychevskaia, Y. Chushkin, F. Zontone, H.-W. Fink, Imaging outside the box: Resolution enhancement in X-ray coherent diffraction imaging by extrapolation of diffraction patterns, Appl. Phys. Lett., 107 (2015) 183102
work page 2015
-
[7]
J.W. Miao, Y. Nishino, Y. Kohmura, B. Johnson, C.Y. Song, S.H. Risbud, T. Ishikawa, Quantitative image reconstruction of GaN quantum dots from oversampled diffraction intensities alone, Phys. Rev. Lett., 95 (2005) 085503
work page 2005
-
[8]
D. Shapiro, P. Thibault, T. Beetz, V. Elser, M. Howells, C. Jacobsen, J. Kirz, E. Lima, H. Miao, A.M. Neiman, D. Sayre, Biological imaging by soft X-ray diffraction microscopy, Proc. Natl. Acad. Sci. USA, 102 (2005) 15343–15346
work page 2005
Show all 24 references
-
[9]
Nishino, Y
Y. Nishino, Y. Takahashi, N. Imamoto, T. Ishikawa, K. Maeshima, Three-dimensional visualization of a human chromosome using coherent X-ray diffraction, Phys. Rev. Lett., 102 (2009) 018101
2009
-
[10]
Seibert, T
M.M. Seibert, T. Ekeberg, F.R.N.C. Maia, M. Svenda, J. Andreasson, O. Jonsson, D. Odic, B. Iwan, A. Rocker, D. Westphal, M. Hantke, D.P. DePonte, A. Barty, J. Schulz, L. Gumprecht, N. Coppola, A. Aquila, M. Liang, T.A. White, A. Martin, C. Caleman, S. Stern, C. Abergel, V. Sel...
2011
-
[11]
J.M. Zuo, I. Vartanyants, M. Gao, R. Zhang, L.A. Nagahara, Atomic resolution imaging of a carbon nanotube from diffraction intensities, Science, 300 (2003) 1419–1421
2003
-
[12]
Latychevskaia, Iterative phase retrieval in coherent diffractive imaging: practical issues, Appl
T. Latychevskaia, Iterative phase retrieval in coherent diffractive imaging: practical issues, Appl. Optics, 57 (2018) 7187–7197
2018
-
[13]
J.W. Miao, D. Sayre, H.N. Chapman, Phase retrieval from the magnitude of the Fourier transforms of nonperiodic objects, J. Opt. Soc. Am. A, 15 (1998) 1662–1669
1998
-
[14]
Sayre, Some implications of a theorem due to Shannon, Acta Crystall., 5 (1952) 843–843
D. Sayre, Some implications of a theorem due to Shannon, Acta Crystall., 5 (1952) 843–843
1952
-
[15]
Fienup, Phase retrieval algorithms – a comparison, Appl
J.R. Fienup, Phase retrieval algorithms – a comparison, Appl. Optics, 21 (1982) 2758–2769
1982
-
[16]
J.W. Miao, D. Sayre, On possible extensions of X-ray crystallography through diffraction-pattern oversampling, Acta Crystallogr. Sect. A, 56 (2000) 596–605
2000
-
[17]
Miao, K.O
J.W. Miao, K.O. Hodgson, D. Sayre, An approach to three-dimensional structures of biomolecules by using single-molecule diffraction images, Proc. Natl. Acad. Sci. USA, 98 (2001) 6641– 6645
2001
-
[18]
Latychevskaia, H.-W
T. Latychevskaia, H.-W. Fink, Practical algorithms for simulation and reconstruction of digital in- line holograms, Appl. Optics, 54 (2015) 2424–2434
2015
-
[19]
Latychevskaia, H.-W
T. Latychevskaia, H.-W. Fink, Solution to the twin image problem in holography, Phys. Rev. Lett., 98 (2007) 233901
2007
-
[20]
Baraniuk, Compressive sensing, IEEE signal processing magazine, 24 (2007) 118–124
R.G. Baraniuk, Compressive sensing, IEEE signal processing magazine, 24 (2007) 118–124
2007
-
[21]
Brady, K
D.J. Brady, K. Choi, D.L. Marks, R. Horisaki, S. Lim, Compressive Holography, Opt. Express, 17 (2009) 13040–13049
2009
-
[22]
Candes, J
E.J. Candes, J. Romberg, T. Tao, Robust uncertainty principles: Exact signal reconstruction from highly incomplete frequency information, IEEE Trans. Inf. Theory, 52 (2006) 489–509
2006
-
[23]
Ratcliffe, Some aspects of diffraction theory and their application to the ionosphere, Rep
J.A. Ratcliffe, Some aspects of diffraction theory and their application to the ionosphere, Rep. Prog. Phys., 19 (1956) 188–267
1956
-
[24]
Goodman, Introduction to Fourier optics, 3 ed., Roberts & Company Publishers, 2004
J.W. Goodman, Introduction to Fourier optics, 3 ed., Roberts & Company Publishers, 2004
2004
Reviewed August 14, 2026 · model on record in the stance chip above.
Discussion (0). Continue with ORCID to comment.