Skip to content
DraconiticResearch
Search Sources API
203,948 studies31 sourcesUpdated 15 min ago
PreprintCC BY 4.0arXiv

Gaussian Critical-Threshold Instability in Real Phase Retrieval

Abstract

We establish the natural scale of instability in real Gaussian phase retrieval at the critical injectivity threshold. Let A be a (2M−1)× M matrix with independent standard Gaussian entries and let KM=binom2M−1M. For every wM→∞ with log wM=o(M), we prove that ℙ{(wM√MKM)−1≤ω(A)≤ wM/(√MKM)}→1, where ω(A) is the Balan--Wang stability parameter. Consequently, −M−1logω(A)→log4 in probability. For full-spark matrices at this threshold, ω(A) equals both the optimal lower Lipschitz constant of x↦|Ax| and the minimum least singular value over all square row submatrices. The upper bound follows from a second-moment analysis of overlapping minors. A weighted Gaussian inverse-tail asymptotic and inverse-Wishart concentration yield asymptotic independence for central overlaps; rectangular hard-edge bounds control the remaining overlaps. The matching lower bound follows from a union bound and a square Gaussian hard-edge estimate.

Related

  1. Quantitative non-Gaussian and Gaussian approximation for exponential random graphs in the single-phase critical regime
    Yi-Kun Zhao
    arXivPreprintarXiv licenceFull textProbabilityarXiv
  2. Boolean threshold functions, neuron capacity, and memory retrieval
    Xinyuan Xie
    arXivPreprintarXiv licenceFull textCombinatorics & discrete mathematicsProbabilityarXiv
  3. Instability of Gaussian elimination is exponentially rare (proof of partial result)
    Lloyd N. Trefethen
    arXivPreprintCC BY 4.0Full textNumerical analysis & scientific computingarXiv
  4. The injectivity threshold of a Gaussian ReLU layer
    Xiaohui Xie
    arXivPreprintarXiv licenceFull textProbabilityarXiv
  5. Threshold Scattering for the Three-Dimensional Focusing Energy-Critical Nonlinear Schrödinger Equation
    Pang-hung Chung, Dan Han
    arXivPreprintarXiv licenceFull textAnalysis & PDEsarXiv
BibTeX
@misc{haggblom2026gaussian,
  title = {{Gaussian Critical-Threshold Instability in Real Phase Retrieval}},
  author = {Christian E. Häggblom},
  howpublished = {arXiv},
  year = {2026},
  month = sep,
  doi = {10.48550/arxiv.2609.28198},
  eprint = {2609.28198},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2609.28198},
  note = {CC BY 4.0},
}
RIS
TY  - UNPB
TI  - Gaussian Critical-Threshold Instability in Real Phase Retrieval
AU  - Christian E. Häggblom
PY  - 2026
DA  - 2026/09/23
PB  - arXiv
DO  - 10.48550/arxiv.2609.28198
UR  - https://arxiv.org/abs/2609.28198
AB  - We establish the natural scale of instability in real Gaussian phase retrieval at the critical injectivity threshold. Let A be a (2M−1)× M matrix with independent standard Gaussian entries and let K_M=binom2M−1M. For every w_M→∞ with log w_M=o(M), we prove that ℙ{(w_M√MK_M)^−1≤ω(A)≤ w_M/(√MK_M)}→1, where ω(A) is the Balan--Wang stability parameter. Consequently, −M^−1logω(A)→log4 in probability. For full-spark matrices at this threshold, ω(A) equals both the optimal lower Lipschitz constant of x↦|Ax| and the minimum least singular value over all square row submatrices. The upper bound follows from a second-moment analysis of overlapping minors. A weighted Gaussian inverse-tail asymptotic and inverse-Wishart concentration yield asymptotic independence for central overlaps; rectangular hard-edge bounds control the remaining overlaps. The matching lower bound follows from a union bound and a square Gaussian hard-edge estimate.
ER  -