Pith. sign in

REVIEW 4 major objections 4 minor 25 references

Sparse Phase Retrieval with Redundant Dictionary via $\ell_q (0<q\le 1)$-Analysis Model

T0 review · 4 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 The paper proves null-space and strong-dictionary-RIP conditions under which dictionary-sparse signals are recovered exactly and stably from magnitude-only measurements by ℓ_q analysis for 0<q≤1.

desk verdict The S-DRIP stable-recovery extension is worth a look, but the real NSP theorem is false as stated; a simple counterexample kills Section 3. read the letter →

arxiv 2506.04576 v1 pith:OWTJIEME submitted 2025-06-05 cs.IT math.IT

Haiye Huo , Li Xiao This is my paper
classification cs.ITmath.IT MSC 41A2794A12
keywords sparsephaseretrievalredundantdictionaryℓ_q-analysismodelnullspacepropertystrongrestrictedisometryexactrecoverystablephaselessmeasurements
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

This paper asks when a signal that is sparse in a redundant dictionary can be recovered from magnitude-only measurements by minimizing $\|D^*x\|_q^q$ over the phaseless constraint. It gives two necessary-and-sufficient null-space conditions for exact recovery, one in the real case and one in the complex case, and a sufficient strong-dictionary-RIP condition for stable recovery in noise. The stable-recovery condition works for S-DRIP of order $tk$ with $0

What carries the argument

The argument runs on the strong dictionary restricted isometry property (S-DRIP), a two-sided bound on $\|A_I Dv\|_2^2$ for every row subset $I$ of size at least $m/2$ and every $k$-sparse coefficient vector $v$. The paper pairs this with the dictionary RIP (D-RIP), whose constant appears inside the stable-recovery bound, and passes from the phaseless problem to a linear one by selecting the larger subset of rows where the signs of $x_0$ and the candidate agree. The proof then uses two decomposition tools: a convex-combination representation of the residual on the off-support set as a mixture of $k$-sparse vectors, and a norm inequality that converts $\ell_q$ tail information into $\ell_2$ estimates. These tools carry the $q$-dependence through the constants $2^{2/q-2}$ and $2^{2/q-1}$ in the final bound.

What would settle it

A numerical search over small $m,n,N,k$ with random tight frames that finds a dictionary-$k$-sparse $x_0$ and a spurious solution $\hat{x}$ with $\|D^*\hat{x}\|_q\le\|D^*x_0\|_q$ but with the agreeing-sign set $\Lambda$ of size larger than $k$ would refute Theorem 3.1(b)$\Rightarrow$(a) as stated.

Watch

Extended reading notes

Core claim

On the paper's own terms, the central claim is that the $\ell_q$-analysis model (6) inherits the recovery guarantees of $\ell_1$-analysis while allowing $0<q\le 1$ and a weaker order requirement on the strong dictionary RIP. The exact-recovery claims are Theorems 3.1 and 3.2: a dictionary-$k$-sparse signal is the unique minimizer of $\|D^*x\|_q^q$ subject to $|Ax|=|Ax_0|$, up to a sign in the real case and up to a unit-modulus phase in the complex case, if and only if a null-space comparison inequality holds for the relevant partition of the measurements into sign classes. The stable-recovery claim is Theorem 4.1: if $A$ satisfies the S-DRIP of order $tk$ with constants $\theta_-,\theta_+\in(0,2)$ obeying $\max\big((3+2^{2/q-2})(1-\theta_-)/(2-\theta_-),(3+2^{2/q-2})(\theta_+-1)/\theta_+\big)<t<4/3$, then every solution $\hat{x}$ to (6) satisfies $\min\{\|\hat{x}-x_0\|_2,\|\hat{x}+x_0\|_2\}\le c_1\epsilon+c_2\,2^{2/q-1}\sigma_k(D^*x_0)_q/k^{1/q-1/2}$, with the constants $c_1,c_2$ explicit in the paper.

Load-bearing premise

The equivalence between exact recovery and the null-space condition depends on the unproved fact that the rows where a spurious solution agrees in sign form a set of at most $k$ indices.

Editorial extensions

If this is right

  • For $q=1$, Theorems 3.1 and 3.2 reduce to the known $\ell_1$-analysis NSP conditions, so the noiseless exact-recovery results contain the earlier characterization as a special case.
  • Theorem 4.1 supplies the first S-DRIP guarantee for the $\ell_q$-analysis model with order $tk$ and $0<t<4/3$, where previous work required $t\ge 2$.
  • Setting $\epsilon=0$ and taking $D^*x_0$ exactly $k$-sparse turns the stable bound into exact recovery: the S-DRIP condition then forces the minimizer to be $\pm x_0$.
  • With $D=I_n$ and $q=1$, Corollary 4.5 gives a strong-RIP stable-recovery bound for standard sparse phase retrieval via $\ell_1$ minimization in the range $0<t<4/3$.

Reading between the lines

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

  • The NSP equivalence is proved for noiseless exact recovery; nothing in the paper makes the same partition-based condition necessary for the noisy stable-recovery bound, so the two regimes should not be conflated.
  • Because the constants in Theorem 4.1 grow like $2^{2/q}$, the admissible range of $t$ narrows sharply as $q\to 0$; a natural check is whether a different technique could remove that dependence or whether the bound is essentially tight at small $q$.
  • The results describe the ideal minimizer of (6), not an algorithm that reaches it, so an implementable solver for $\ell_q$-analysis phase retrieval that attains the stated bound is the practical next step the authors leave open.
  • The sign-subset argument might extend to complex stable recovery through a stronger partitioned condition, but the paper explicitly leaves S-DRIP-based results in the real field, so the complex stable case remains open.
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

4 major / 4 minor

Summary. The paper studies sparse phase retrieval with a redundant tight frame D, using the ℓ_q-analysis model (0<q≤1). It claims two contributions: (i) null-space-property (NSP) conditions that are necessary and sufficient for exact recovery of dictionary-k-sparse signals in the noiseless real and complex cases (Theorems 3.1 and 3.2), and (ii) a strong dictionary RIP (S-DRIP) condition of order tk with 0<t<4/3 that guarantees stable recovery in the noisy real case (Theorem 4.1). The stable-recovery proof follows the Zhang-Li framework through an auxiliary lemma whose proof is placed in the appendix. The paper is theoretical and contains no numerical experiments.

Significance. If the results were correct, the stable-recovery range 0<t<4/3 would be a genuine extension of existing ℓ_q-analysis phase-retrieval guarantees, and the NSP characterization would be the first for the ℓ_q-analysis model. The paper is clearly organized and carefully builds on prior work by Gao and by Cao-Huang. However, the exact-recovery theorem is false as stated, and the stable-recovery lemma leaves several central algebraic identities unproved. These issues are load-bearing, so the significance of the contribution cannot be assessed as the manuscript stands.

major comments (4)
  1. [Theorem 3.1, Section 3.1] Theorem 3.1 is false as stated. Take D=[I_2,0] in R^{2×4}, which is a tight frame, and A in R^{4×2} with rows (1,1), (1,-1), (-1,1), (-1,-1). Let k=1 and fix any q in (0,1]. For every Lambda subset of [1:4] with |Lambda|≤1, the matrix A_{Lambda^c} has at least three rows that span R^2, so N(A_{Lambda^c})={0}; hence condition (b) holds vacuously. But x0=e1 and x_hat=e2 both lie in D R_1, satisfy |Ax0|=|Ax_hat|=(1,1,1,1), and have ||D^*x0||_q^q=||D^*x_hat||_q^q=1, while x_hat is not ±x0. Thus (a) fails. The gap is in the proof of (b)⇒(a): the constructed set Lambda={i : a_i^T(x0+x_hat)=0} is never shown to have cardinality at most k, and in this example |Lambda|=2. A correct theorem would need to quantify over all Lambda subset of [1:m], or impose additional hypotheses that force the sign-flip set to be small.
  2. [Appendix A, Eqs. (A.11), (A.12), (A.17), (A.18)] The stable-recovery proof of Lemma 4.2 depends on two algebraic identities and two inequalities that are asserted without proof. Equations (A.11) and (A.12) are introduced with the statement that the proofs are similar to (14) and (15) in [22], and (A.17)-(A.18) are stated directly with no derivation. These relations are the mechanism by which the D-RIP estimates are converted into the quadratic inequality for ||D^*_{S0}h||_2, and the dictionary terms involving D⊥ are new relative to [22]. Saying 'similar to [22]' is not a self-contained proof for such load-bearing steps; the authors should provide complete derivations or a detailed line-by-line transfer from [22].
  3. [Appendix A, final paragraph] The treatment of non-integer tk is not valid. The text says that if tk is not an integer, one sets t' = ceil(tk)/k and works with delta_{t'k}; however t' need not satisfy t' < 4/3. For example, with k=2 and t=1.3, t'=1.5. Thus the rounding argument does not preserve the hypothesis of Theorem 4.1. The theorem should either be restricted to integer tk or supplied with a correct rounding/monotonicity argument.
  4. [Theorem 3.2, proof of (a)⇒(b)] In the proof of (a)⇒(b), after Eq. (30), the assertion that x_hat is not in {c x0 : c in S} is unjustified and can fail under the stated relations. For instance, with p=2, d1=1, d2=-1, and phi1=i phi2, the definitions (29)-(30) give x0=(i-1)phi2 and x_hat=-(i+1)phi2, so x_hat=i x0 with |i|=1. In that case the constructed pair is a trivial phase ambiguity and does not contradict (a), yet condition (24) fails with equality. The proof therefore does not establish the implication as written; an additional argument ruling out this case, or a reformulation of condition (b), is needed.
minor comments (4)
  1. [Appendix A, Eq. (A.4)] The displayed line '... = k alpha^q = b·( (r/k) b alpha )^q' is garbled and does not parse; please rewrite the argument that leading to ||D^*_{S0^c}h||_∞ ≤ alpha.
  2. [Section 2, Definition 2.2] The S-DRIP definition quantifies over subsets I with |I|≥m/2, and the proof of Theorem 4.1 later applies it to a specific set S defined by a sign pattern. Please state explicitly that the min/max in (14) imply the two-sided estimate for every such subset.
  3. [Theorem 3.2] For p=1, condition (23) is vacuous, and the theorem should state what happens in that case; also the denominators d1-d_j require the distinctness of d_i, which is stated but should be made explicit in the condition.
  4. [Abstract and Section 5] There are minor language issues: 'where, in particularly' should be 'where, in particular', and 'for exact/stable recovery or dictionary-sparse signals' in Section 5 should be 'of dictionary-sparse signals'.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the recovery bounds are derived from stated S-DRIP/D-RIP assumptions and independent external lemmas, not from the conclusions being proved.

full rationale

The paper does not fit parameters, rename conclusions as assumptions, or rely on a load-bearing self-citation. The NSP theorems in Section 3 follow the proof pattern of Gao's earlier work with the q-norm replaced, while the stable recovery result in Theorem 4.1 is obtained by partitioning the row indices into sign-coherent sets and invoking Lemma 4.2, whose proof is given in the appendix. Lemma 4.2 itself uses only the stated D-RIP condition and external lemmas (Lemma 2.4 from Zhang and Li, Lemma 2.5 from Cai and Zhang), none of which assume the target recovery inequality. The S-DRIP constants are hypotheses, not fitted quantities, and equation (35) is genuinely a consequence of these hypotheses. The authors' self-citations [9] and [10] are contextual references on weighted sparse recovery and do not carry the derivation. The most serious issue found is a correctness gap in Theorem 3.1: the proof of (b) implies (a) constructs a sign-flip set Lambda without proving |Lambda| <= k, and a counterexample with D=[I2,0], A having rows (±1,±1), k=1 makes (b) vacuously true while the required uniqueness fails. This is a validity or statement-error problem, not circularity, so the circularity score remains zero. No circular step satisfies the requirement of a quoted textual reduction to the paper's own inputs.

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

The paper introduces no fitted parameters or invented entities. Its results rest on the Parseval tight frame assumption, the S-DRIP/D-RIP conditions, and two cited norm/decomposition lemmas. The main unstated structural premise is that analysis sparsity of D^*x0 is compatible with the synthesis definition of D H_N^k.

assumptions (5)
  • domain assumption D is a Parseval tight frame with DD^*=I_n and ||D^*f||_2=||f||_2.
    Stated in Section 2, equations (8)-(9), and used throughout to relate D-RIP bounds on ADz to norms of z via (13).
  • domain assumption The measurement matrix A satisfies S-DRIP of order tk with constants θ_-, θ_+ in (0,2) satisfying the max inequality and the derived δ_tk bound.
    Theorem 4.1 assumes this property; it is the main sufficient condition, not derived.
  • standard math Lemma 2.4 (convex decomposition of vectors with bounded ℓ_q norm and ℓ_∞ norm) applies to the truncated analysis vectors D^*_{S^c}h.
    Taken from [23, Lemma 2.2] and used in Appendix A to decompose the tail; the application requires checking the norm conditions, which the paper does only partially.
  • standard math Lemma 2.5 (norm inequality for tail sums) from [2, Lemma 5.3].
    Used in (A.15) to bound ||D^*_{S^c}h||_2 in terms of the top k entries.
  • domain assumption The D-RIP monotonicity δ_k ≤ δ_tk and the relationship between S-DRIP of A and D-RIP of A_S.
    Used in the proof of Theorem 4.1 and in appendix inequalities (A.8)-(A.9).

how reviews work

0 comments
Cite this review

Pith. "Pith review of Sparse Phase Retrieval with Redundant Dictionary via $\ell_q (0<q\le 1)$-Analysis Model." pith.science (2026). https://pith.science/paper/OWTJIEME

@misc{pith2026250604576,
  author       = {Pith},
  title        = {Pith review of: Sparse Phase Retrieval with Redundant Dictionary via $\ell_q (0<q\le 1)$-Analysis Model},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/OWTJIEME}},
  note         = {Machine review of arXiv:2506.04576}
}
abstract

Sparse phase retrieval with redundant dictionary is to reconstruct the signals of interest that are (nearly) sparse in a redundant dictionary or frame from the phaseless measurements via the optimization models. Gao [7] presented conditions on the measurement matrix, called null space property (NSP) and strong dictionary restricted isometry property (S-DRIP), for exact and stable recovery of dictionary-$k$-sparse signals via the $\ell_1$-analysis model for sparse phase retrieval with redundant dictionary, respectively, where, in particularly, the S-DRIP of order $tk$ with $t>1$ was derived. In this paper, motivated by many advantages of the $\ell_q$ minimization with $0<q\leq1$, e.g., reduction of the number of measurements required, we generalize these two conditions to the $\ell_q$-analysis model. Specifically, we first present two NSP variants for exact recovery of dictionary-$k$-sparse signals via the $\ell_q$-analysis model in the noiseless scenario. Moreover, we investigate the S-DRIP of order $tk$ with $0<t<\frac{4}{3}$ for stable recovery of dictionary-$k$-sparse signals via the $\ell_q$-analysis model in the noisy scenario, which will complement the existing result of the S-DRIP of order $tk$ with $t\geq2$ obtained in [4].

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

25 extracted references · 25 canonical work pages

  1. [22]

    Zhang and S

    R. Zhang and S. Li. A proof of conjecture on restricted isometry property constants δtk(0< t <4 3 ).IEEE Trans. Inf. Theory, 64(3): 1699–1705, 2018

    work page 2018

  2. [1]

    J. F. Cai, Y. Jiao, X. Lu, et al. Sample-efficient sparse phase retrieval via stochastic alternating minimization.IEEE Trans. Signal Process.,70: 4951–4966, 2022

    work page 2022

  3. [2]

    Cai and A

    T. Cai and A. Zhang, Sharp RIP bound for sparse signal and low-rank matrix recovery, Appl. Comput. Harmon. Anal., vol. 35, no. 1, pp. 74–93, 2013

    work page 2013

  4. [3]

    E. J. Cand` es, Y. C. Eldar, D. Needell. Compressed sensing with coherent and redun- dant dictionaries.Appl. Comput. Harmon. Anal., 31(1): 59–73, 2011

    work page 2011

  5. [4]

    M. Cao, W. Huang. Sparse phase retrieval viaℓ p(0< p≤1) minimization.Int. J. Wavelets Multi. Inf. Process., 20(1): 2150034, 2022

    work page 2022

  6. [5]

    Chen and Y

    W. Chen and Y. Li. Stable recovery of signals with the high order D-RIP condition. Acta Math. Sci., 36(6): 1721–1730, 2016

    work page 2016

  7. [6]

    B. Gao, Y. Wang, and Z. Xu. Stable signal recovery from phaseless measurements.J. Fourier Anal. Appl., 22(4): 787–808, 2016

    work page 2016

  8. [7]

    B. Gao. Theℓ 1-analysis with redundant dictionary in phase retrieval.J. Fourier Anal. Appl., 23: 1097–1117, 2017

    work page 2017

Show all 25 references
  1. [8]

    M. A. Herman and T. Strohmer. High-resolution radar via compressed sensing.IEEE Trans. Signal Process., 57(6): 2275–2284, 2009

    2009

  2. [9]

    H. Huo, W. Sun, L. Xiao. New conditions on stable recovery of weighted sparse signals via weightedℓ 1 minimization.Circ. Syst. Signal Process., 37(7): 2866–2883, 2018

    2018

  3. [10]

    H. Huo. Stable recovery of weighted sparse signals from phaseless measurements via weightedℓ 1 minimization.Math. Meth. Appl. Sci., 45(9): 4929–4937, 2022

    2022

  4. [11]

    Lai and J

    M. Lai and J. Wang. An unconstrainedℓ q minimization with 0< q≤1 for sparse solution of underdetermined linear systems.SIAM J. Optim., 21(1): 82–101, 2011

    2011

  5. [12]

    A. I. Lvovsky and M. G. Raymer. Continuous-variable optical quantum-state tomog- raphy.Rev. Mod. Phys., 81(1): 299–332, 2009

    2009

  6. [13]

    R. P. Millane. Phase retrieval in crystallography and optics.J. Opt. Soc. Am. A, 7: 394–411, 1990

    1990

  7. [14]

    Shechtman, Y

    Y. Shechtman, Y. C. Eldar, O. Cohen, et al. Phase retrieval with application to optical imaging: a contemporary overview.IEEE Signal Proc. Mag., 32(3): 87–109, 2015

    2015

  8. [15]

    Q. Sun. Recovery of sparsest signals viaℓ q-minimization.Appl. Comput. Harmon. Anal., 32(3): 329–341, 2010. 20

    2010

  9. [16]

    Voroninski and Z

    V. Voroninski and Z. Xu. A strong restricted isometry property, with an application to phaseless compressed sensing.Appl. Comput. Harmon. Anal., 40(2): 386–395, 2016

    2016

  10. [17]

    A. Walther. The question of phase retrieval in optics.J. Mod. Optic., 10(1): 41–49, 1963

    1963

  11. [18]

    A. Wan. Uniform RIP conditions for recovery of sparse signals byℓ p (0< p≤1) minimization.IEEE Trans. Signal Process., 68: 5379–5394, 2020

    2020

  12. [19]

    Wang and Z

    Y. Wang and Z. Xu. Phase retrieval for sparse signals.Appl. Comput. Harmon. Anal., 37(3): 531–544, 2014

    2014

  13. [20]

    L. -J. Xie. Improved RIC bounds in terms ofδ 2s for hard thresholding-based algo- rithms.IEEE Signal Process. Lett., 30: 21–25, 2023

    2023

  14. [21]

    G. You, Z. H. Huang, and Y. Wang. A theoretical perspective of solving phaseless compressive sensing via its nonconvex relaxation.Inform. Sci., 415: 254–268, 2017

    2017

  15. [23]

    Zhang and S

    R. Zhang and S. Li. Optimal RIP bounds for sparse signals recovery viaℓ p minimiza- tion.Appl. Comput. Harmon. Anal.,47(3): 566-584, 2019

    2019

  16. [24]

    Zhang, Y

    D. Zhang, Y. Sun, F. Zhang, et al. Phase retrieval for signals with block sparsity using BOMP: Algorithms and recovery guarantees.Digit. Signal Process., 129: 103656, 2022

    2022

  17. [25]

    Zhou and J

    Z. Zhou and J. Yu. Phaseless compressive sensing using partial support information. Optim. Lett., 14: 1961–1973, 2020. 21

    1961

Pith tools

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