pith. sign in

arxiv: 2606.29675 · v1 · pith:5PIKWGU6new · submitted 2026-06-29 · 💻 cs.LG · cond-mat.dis-nn· stat.ML

I-BBS: Coordinate-Free Inference of Latent Sub-Manifolds Using Random Distance Matrix Theory

Igor Halperin This is my paper

Pith reviewed 2026-06-30 07:01 UTC · model grok-4.3

classification 💻 cs.LG cond-mat.dis-nnstat.ML
keywords latent sub-manifoldsdistance matrix spectrumeigenvalue multipletscoordinate-free inferencemanifold dimensionnoise robustnessrandom matrix theory
0
0 comments X

The pith

The dimension of a latent sub-manifold is recovered from the multiplicity of a top eigenvalue multiplet in its distance matrix.

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

This paper develops I-BBS to recover the dimension and geometry of a latent low-dimensional sub-manifold from the pairwise distance matrix of points in a high-dimensional ambient space, without needing coordinates or full access to that space. It models the mixing of manifold signal with off-manifold components using two classes of generative noise and shows that the eigenvalues reorganize collectively, so geometry is read from two surviving integer-stable signatures instead of individual eigenvalues. The multiplicity of the top non-Perron multiplet fixes the manifold dimension d, and a parameter-free law governs how those multiplet positions shrink as noise grows. These signatures prove more stable than the continuous spectral slope on synthetic spheres of dimensions one through three, and a blind test recovers both the manifold and the noise model from one matrix.

Core claim

Bogomolny, Bohigas and Schmit observed that the spectrum of the pairwise distance matrix on N points sampled from a smooth d-dimensional manifold encodes a signature of the underlying geometry. I-BBS recovers the latent geometry from the multiplicity of the top non-Perron multiplet, which fixes d, and a parameter-free law for the shrinkage of these multiplet positions as noise increases, even when the two generative noise classes cause collective reorganization of the eigenvalues.

What carries the argument

The top non-Perron multiplet in the eigenvalue spectrum of the noisy distance matrix, whose multiplicity fixes the manifold dimension d and whose positions follow a parameter-free shrinkage law under increasing noise.

If this is right

  • The integer signatures are far more stable under noise than the continuous spectral slope on synthetic spheres S1, S2 and S3.
  • A blind test recovers both the manifold and the noise model from a single distance matrix.
  • The method applies directly to neural-network representations and the dynamic training regime.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The approach could extend to any data source that supplies only pairwise distances, such as similarity graphs or partial observations.
  • Verification on non-spherical manifolds would test whether the multiplet signatures generalize beyond the synthetic spheres used.
  • The parameter-free shrinkage law might link to broader properties of random distance matrices independent of the specific noise models.

Load-bearing premise

The two generative noise classes sufficiently capture how real data mixes latent manifold signal with off-manifold components so that the claimed integer signatures remain identifiable and stable.

What would settle it

Finding that the multiplicity of the top non-Perron multiplet fails to match the known dimension d or changes with noise level on a synthetic sphere under either noise model would falsify the recovery claim.

Figures

Figures reproduced from arXiv: 2606.29675 by Igor Halperin.

Figure 1
Figure 1. Figure 1: contrasts the two: on S 2 the delocalised slope follows the BBS prediction −d/(d − 1) with the SO(3) multiplet plateaus 1, 3, 5, . . . at the top, while on S 127 the spectrum is a single dominant, only weakly decaying band of the largest ∼ 128 eigenvalues (the ℓ = 1 multiplet, spread out in the MP regime) ending in a sharp drop, the delocalised window [2, √ N] not reaching past this first multiplet (√ N ≈ … view at source ↗
Figure 2
Figure 2. Figure 2: Empirical spectral density of the negative eigenvalues of the geodesic matrix (Perron [PITH_FULL_IMAGE:figures/full_fig_p009_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Angular-momentum-level shrinkage fℓ(ε) for degrees ℓ = 1, 3, 5: parameter-free prediction (19) (lines) and simulated multiplet shrinkage (markers) on the S 2/S124 setup. 0 1 2 3 4 5 latent degree r 10 21 10 18 10 15 10 12 10 9 10 6 10 3 10 0 |ars( )| (1,0) dimension cluster (a) sector map, = 0.32, D d = 125 s=0 s=1 s=2 s=3 s=4 0.1 0.2 0.3 0.4 0.5 0.6 0.00 0.05 0.10 0.15 0.20 0.25 0.30 0.35 0.40 m argin |a1… view at source ↗
Figure 4
Figure 4. Figure 4: Product-kernel spectrum of the RSM operator [PITH_FULL_IMAGE:figures/full_fig_p017_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: ESDs of the latent M(d) (blue), residual M(D−d) (red), and RSM ambient M(D) at ε ∈ {0.10, 0.22, 0.32, 0.45} (ηcos ≈ {1, 5, 10, 20}%, Eq. (21)) for a uniform reference (top-left) and beta-product concentrated single-particle densities at 40%, 35%, 30%, 25%, and 20% sphere coverage (sharper beta peaks, deeper ergodicity breaking; the per-sphere sharpness κlat, κres is shown in each panel). N = 1000, d = 3, D… view at source ↗
Figure 6
Figure 6. Figure 6: Multiplicity recovery P(hˆ 1 = h(1, d)) vs noise η for RSM and FSM on S 1 , S2 , S3 (N = 1000, D = 128, 20 realisations, detector threshold τ = 0.30). RSM recovers it up to η = 0.80; FSM up to η = 0.50, then fails when its inter-multiplet gap closes (Sec. 5.2). 5.2 Eigenvalue stability and the shrinkage law The multiplicity is an integer; the eigenvalues that carry it are not. We track the mean magnitude o… view at source ↗
Figure 7
Figure 7. Figure 7: Top: ℓ = 1 multiplet shrinkage |Λ (D) ℓ=1|/|Λ (d) ℓ=1| vs η for RSM and FSM on S 1 , S2 , S3 , with the Funk–Hecke prediction f1 (dashed); RSM lies on it, FSM shrinks more slowly. The green curve is the quasi-degenerate block reduction (Pymablock [43], Appendix E) of the bounded spliced mean drift; it tracks RSM and f1 across the full range on all three manifolds. Bottom: the log inter-multiplet gap with t… view at source ↗
Figure 8
Figure 8. Figure 8: Blind identification from a single M(D) , pooled over 0.05 ≤ η ≤ 0.50 and S 1–S 3 . (a) Manifold confusion from the gap-protected multiplicity (identity). (b) Blind ℓ = 2 component ratio |λˆ 2/λˆ 1| from the top cos M(D) eigenvectors: RSM at the parity floor, FSM rising with η, separated by a geometric-mean boundary (dashed). (c) The resulting noise-model confusion matrix. 5.4 Statistical validation: null … view at source ↗
Figure 9
Figure 9. Figure 9: Statistical validation of the multiplicity detector with [PITH_FULL_IMAGE:figures/full_fig_p024_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Anisotropic-sampling negative control on [PITH_FULL_IMAGE:figures/full_fig_p026_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: c). The same sensitivity makes the localised branch the sharpest fingerprint of the noise mechanism: the isotropic RSM and the anisotropic, ℓ = 2-injecting FSM fill the small-|Λ| region with visibly different shapes, FSM flattening it toward a Wigner-like plateau (see Fig. 11d). This is the residual random-matrix discrimination of Appendix C.5, read directly off the localised branch. 10 1 10 2 10 3 rank K… view at source ↗
Figure 12
Figure 12. Figure 12: RSM forward drift (left) and volatility (right) correction prefactors versus [PITH_FULL_IMAGE:figures/full_fig_p032_12.png] view at source ↗
Figure 13
Figure 13. Figure 13: Convergence of the angular-momentum Gegenbauer expansion of the RSM drift (one fixed [PITH_FULL_IMAGE:figures/full_fig_p035_13.png] view at source ↗
Figure 14
Figure 14. Figure 14: Empirical spectral densities of the negative eigenvalues of the geodesic distance matrix [PITH_FULL_IMAGE:figures/full_fig_p037_14.png] view at source ↗
Figure 15
Figure 15. Figure 15: Signal limit: the top eigenvalues |ΛK|/N of the stored matrices (N = 1000, markers with error bars over 20 realisations) against the operator spectrum eig(Tp¯) from an independent M = 4000 sample (lines), on S 2 (left) and S 124 (right). The four single-particle patterns are reproduced by their respective p¯; the per-angle pattern (sc3) shows the larger scatter expected of a quenched anisotropy. C.3 Nonun… view at source ↗
Figure 16
Figure 16. Figure 16: shows the S 2 rank plot: the operator counting tracks the simulated |ΛK| and bends from a steeper small-K slope toward the βBBS reference at larger K. The counting route (C.6) and the resolvent (see Sec. C.4) are not independent of each other, both built on the same operator spectrum; the independent check is this operator prediction against the direct diagonalisation, which it matches [PITH_FULL_IMAGE:f… view at source ↗
Figure 17
Figure 17. Figure 17: Closed-form bulk density from the deformed-MP self-consistent equation [PITH_FULL_IMAGE:figures/full_fig_p043_17.png] view at source ↗
Figure 18
Figure 18. Figure 18: Free multiplicative deconvolution of the operator spectrum (the inverse of the Silverstein [PITH_FULL_IMAGE:figures/full_fig_p044_18.png] view at source ↗
Figure 19
Figure 19. Figure 19: Residual-spectrum RMT consistency on S 2 (N = 1000, D − d = 125). Top row: oracle residual R = M(D) − M(d) . Bottom row: operational residual Rˆ = M(D) − Mˆ (d) from (C.16) with Klat = ⌊ √ N⌋. Columns are ε ∈ {0.10, 0.22, 0.32, 0.45}. The dashed curve is the exploratory semicircle reference (C.17) at the residual’s own entry variance; the annotations give the ratio of the latent eigenvalue |Λ (d) | to the… view at source ↗
Figure 20
Figure 20. Figure 20: Finite-N delocalised-slope bias |βˆ deloc − βBBS| versus N for the geodesic matrix on S 1 , S2 , S3 (uniform and product-Beta sampling, two rank windows). The black line is the determin￾istic operator-only bias from the ranked {N|aℓ |} (no noise): it accounts for most of the drift, the full random-matrix curves lying close to it. The curves are nearly parallel across the uniform and product-Beta sampling … view at source ↗
Figure 21
Figure 21. Figure 21: Delocalised-branch slope bias |∆β(N, d)| vs N on latent samples of S d−1 . Solid lines: power-law fits with a shallow exponent ≈ 0.2–0.3 for d = 2, 3, 4, tracking the deterministic operator￾window drift of Sec. D.1; the rate is shallower than the N −1/2 of a noise-only estimate. Applied to ambient data, the corrected slope matches the BBS asymptote at ε = 0 but drifts monotonically downward under noise: o… view at source ↗
Figure 22
Figure 22. Figure 22: Quasi-degenerate perturbation theory (Pymablock [43]) against the full simulation. (a) [PITH_FULL_IMAGE:figures/full_fig_p050_22.png] view at source ↗
read the original abstract

Bogomolny, Bohigas and Schmit (BBS) found that the spectrum of the pairwise distance matrix on N points sampled from a smooth d-dimensional manifold encodes a signature of the underlying geometry. We develop I-BBS (Inference-BBS), a coordinate-free method that identifies a low-dimensional latent sub-manifold embedded in a high-dimensional ambient distance matrix alone, without accessing an ambient high-dimensional vector space. It therefore applies even when that space is only partly observable or undefined. We model the ambient embedding by two classes of generative noise, model-based and model-free. The noise mixes the latent signal with off-manifold components, so the eigenvalues reorganise collectively and the latent geometry cannot be read off eigenvalue by eigenvalue. We recover it instead from two integer-stable signatures that survive the noise: the multiplicity of the top non-Perron multiplet, which fixes $d$, and a parameter-free law for how the multiplet positions shrink as the noise grows. On synthetic spheres $S^1$, $S^2$ and $S^3$ these integer signatures are far more stable under noise than the continuous spectral slope, and a blind test recovers both the manifold and the noise model from a single distance matrix. Applications to neural-network representations and to the dynamic training regime are developed in two companion papers.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 0 minor

Summary. The manuscript introduces I-BBS, a coordinate-free extension of BBS random distance matrix theory for inferring the dimension d and geometry of a latent d-dimensional sub-manifold embedded in an ambient distance matrix. It models the embedding via two generative noise classes (model-based and model-free), under which eigenvalues reorganize collectively; geometry is recovered from two integer-stable signatures—the multiplicity of the top non-Perron multiplet (fixing d) and a parameter-free shrinkage law for multiplet positions with increasing noise. Synthetic tests on spheres S¹–S³ demonstrate greater stability than continuous spectral slope, and a blind test recovers both manifold and noise model from a single distance matrix. Applications to neural-network representations are noted for companion papers.

Significance. If the claimed integer signatures prove robust and the two noise classes are representative, the method would provide a genuinely coordinate-free route to manifold dimension and geometry recovery from distance data alone, even when the ambient vector space is only partially observable. This would be a notable contribution to manifold learning and random-matrix applications in high-dimensional data analysis.

major comments (2)
  1. [Abstract] The central claim that the two generative noise classes (model-based and model-free) sufficiently capture real-data mixing of latent manifold signal with off-manifold components, so that the integer signatures remain identifiable, is asserted in the abstract but not shown to be exhaustive. No demonstration is supplied that the multiplet multiplicity and shrinkage law survive other plausible perturbations (e.g., low-rank structured noise or manifold-dependent noise) that could split or shift the multiplet and render d unrecoverable.
  2. [Abstract] The abstract describes the signatures and synthetic sphere tests but supplies no derivations, error analysis, or quantitative results; support for the parameter-free law and noise-model recovery therefore cannot be assessed.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their thoughtful review and constructive comments on our manuscript. We respond to each major comment below, indicating where revisions will be incorporated.

read point-by-point responses

  1. Referee: [Abstract] The central claim that the two generative noise classes (model-based and model-free) sufficiently capture real-data mixing of latent manifold signal with off-manifold components, so that the integer signatures remain identifiable, is asserted in the abstract but not shown to be exhaustive. No demonstration is supplied that the multiplet multiplicity and shrinkage law survive other plausible perturbations (e.g., low-rank structured noise or manifold-dependent noise) that could split or shift the multiplet and render d unrecoverable.

    Authors: We appreciate the referee drawing attention to the scope of our noise modeling. The manuscript introduces the model-based and model-free classes as two representative generative mechanisms under which the eigenvalues reorganize collectively, allowing recovery via the integer signatures; it does not assert that these classes are exhaustive or that they capture every possible real-data perturbation. The synthetic sphere experiments demonstrate stability of the multiplicity and shrinkage law specifically under the proposed noise classes. We agree that additional perturbations (such as low-rank structured noise or manifold-dependent noise) are not tested and could potentially affect identifiability. In revision we will update the abstract and add a dedicated limitations paragraph in the discussion to explicitly scope the claims to the two modeled noise classes and note that robustness to other perturbation types remains an open question for future work. revision: yes

  2. Referee: [Abstract] The abstract describes the signatures and synthetic sphere tests but supplies no derivations, error analysis, or quantitative results; support for the parameter-free law and noise-model recovery therefore cannot be assessed.

    Authors: The abstract is written as a high-level summary of the method, signatures, and key experimental outcomes. The derivations of the integer signatures and parameter-free shrinkage law, together with the error analysis, quantitative stability comparisons against spectral slope, and details of the blind noise-model recovery test, are all contained in the main body of the manuscript (theoretical sections and experimental results). Because the abstract's role is to provide an overview rather than technical detail, we do not plan to expand it with derivations or quantitative tables. revision: no

Circularity Check

0 steps flagged

No significant circularity; derivation self-contained against external benchmarks

full rationale

The paper extends the established BBS spectral signature by introducing two explicit generative noise classes (model-based and model-free) to model off-manifold mixing, then identifies the multiplicity of the top non-Perron multiplet (fixing d) and a claimed parameter-free shrinkage law as the surviving integer signatures. No equations or fitting procedures are exhibited in the abstract or description that would reduce the shrinkage law to a fitted parameter renamed as prediction, nor is any self-citation load-bearing, uniqueness theorem imported from the authors, or ansatz smuggled via prior work. Validation occurs on synthetic spheres S^1–S^3 with blind recovery tests, which are independent of the target result rather than tautological. The central claim therefore rests on the modeling assumptions and empirical stability rather than reducing by construction to its inputs.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

Review is abstract-only; no explicit free parameters, invented entities, or additional axioms beyond the foundational BBS spectral observation are stated.

axioms (1)
  • domain assumption The spectrum of the pairwise distance matrix on N points sampled from a smooth d-dimensional manifold encodes a signature of the underlying geometry.
    Invoked as the starting point for developing I-BBS inference under noise.

pith-pipeline@v0.9.1-grok · 5771 in / 1177 out tokens · 17832 ms · 2026-06-30T07:01:44.708622+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

54 extracted references · 8 canonical work pages · 6 internal anchors

  1. [1]

    Spectral properties of distance matrices

    E. Bogomolny, O. Bohigas, and C. Schmit. Spectral properties of distance matrices.Journal of Physics A: Mathematical and General, 36:3595–3616, 2003. arXiv:nlin/0301044

    work page internal anchor Pith review Pith/arXiv arXiv 2003

  2. [2]

    Distance matrices and isometric embeddings

    E. Bogomolny, O. Bohigas, and C. Schmit. Distance matrices and isometric embeddings. arXiv:0710.2063, 2007

    work page internal anchor Pith review Pith/arXiv arXiv 2063

  3. [3]

    Halperin.Learning as Observable Matrix Dynamics: Diffusive Relaxations versus Phase Transitions.2026

    I. Halperin.Learning as Observable Matrix Dynamics: Diffusive Relaxations versus Phase Transitions.2026

    2026

  4. [4]

    Halperin.Grokking as Bagel Formation in Activation Space: Spectral Evidence for a Phase Transition.2026

    I. Halperin.Grokking as Bagel Formation in Activation Space: Spectral Evidence for a Phase Transition.2026

    2026

  5. [5]

    Bun, J.-P

    J. Bun, J.-P. Bouchaud, and M. Potters. Cleaning large correlation matrices: tools from Random Matrix Theory.Physics Reports, 666:1–109, 2017

    2017

  6. [6]

    Levina and P

    E. Levina and P. J. Bickel. Maximum likelihood estimation of intrinsic dimension.Advances in Neural Information Processing Systems (NeurIPS), 2004

    2004

  7. [7]

    Facco, M

    E. Facco, M. d’Errico, A. Rodriguez, and A. Laio. Estimating the intrinsic dimension of datasets by a minimal neighborhood information.Scientific Reports, 7(1):12140, 2017

    2017

  8. [8]

    Chazal and B

    F. Chazal and B. Michel. An introduction to Topological Data Analysis: fundamental and practical aspects for data scientists.Frontiers in Artificial Intelligence, 4:667963, 2021

    2021

  9. [9]

    Otter, M

    N. Otter, M. A. Porter, U. Tillmann, P. Grindrod, and H. A. Harrington. A roadmap for the computation of persistent homology.EPJ Data Science, 6(1):17, 2017

    2017

  10. [10]

    R. R. Coifman and S. Lafon. Diffusion maps.Applied and Computational Harmonic Analysis, 21(1):5–30, 2006

    2006

  11. [11]

    El Karoui.The spectrum of kernel random matrices

    N. El Karoui.The spectrum of kernel random matrices. Annals of Statistics38(1):1–50, 2010

    2010

  12. [12]

    On Euclidean random matrices in high dimension

    C. Bordenave.On Euclidean random matrices in high dimension.arXiv:1209.5888, 2012

    work page internal anchor Pith review Pith/arXiv arXiv 2012

  13. [13]

    Couillet and Z

    R. Couillet and Z. Liao.Random Matrix Methods for Machine Learning. Cambridge University Press, 2022

    2022

  14. [14]

    S. Lele. Euclidean Distance Matrix Analysis (EDMA): Estimation of Mean Form and Mean Form Difference.Mathematical Geology, 25(5):573–602, 1993

    1993

  15. [15]

    M´ ezard, G

    M. M´ ezard, G. Parisi, and A. Zee. Spectra of Euclidean random matrices.Nuclear Physics B, 559(3):689–710, 1999

    1999

  16. [16]

    Largest eigenvalue and top eigenvector statistics of large Euclidean random matrices

    P. Casaburi and P. Vivo.Largest eigenvalue and top eigenvector statistics of large Euclidean random matrices.arXiv:2604.26852, 2026

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  17. [17]

    Casaburi and P

    P. Casaburi and P. Vivo.Replica approach to extreme eigenvalues of Euclidean random matrices. Journal of Physics A: Mathematical and Theoretical, 2026

    2026

  18. [18]

    El Karoui

    N. El Karoui. Spectrum estimation for large dimensional covariance matrices using random matrix theory.The Annals of Statistics, 36(6):2757–2790, 2008. 51

    2008

  19. [19]

    Chen and R

    H. Chen and R. Ma.Statistical Inference for Manifold Similarity and Alignability across Noisy High-Dimensional Datasets.arXiv:2511.21074, 2025

    work page arXiv 2025

  20. [20]

    Rahimi and B

    A. Rahimi and B. Recht. Random features for large-scale kernel machines.Advances in Neural Information Processing Systems (NIPS), 2007

    2007

  21. [21]

    F. Liu, X. Huang, Y. Chen, and J. A. K. Suykens.Random Features for Kernel Approximation: A Survey on Algorithms, Theory, and Beyond. IEEE Transactions on Pattern Analysis and Machine Intelligence, 44(10):7128–7148, 2022 (arXiv:2004.11154, 2021)

    work page arXiv 2022

  22. [22]

    D. Paul. Asymptotics of sample eigenstructure for a large dimensional spiked covariance model. Statistica Sinica, 17(4):1617–1642, 2007

    2007

  23. [23]

    Donoho and M

    D. Donoho and M. Gavish. Minimax risk of matrix denoising by singular value thresholding. The Annals of Statistics, 42(6):2413–2440, 2014

    2014

  24. [24]

    Y. Yan, Y. Chen, and J. Fan. Inference for heteroskedastic PCA with missing data.The Annals of Statistics, 52(2):729–756, 2024

    2024

  25. [25]

    Ding and R

    X. Ding and R. Ma. Learning low-dimensional nonlinear structures from high-dimensional noisy data: an integral operator approach.The Annals of Statistics, 51(4):1744–1769, 2023

    2023

  26. [26]

    K. V. Mardia and P. E. Jupp.Directional Statistics. John Wiley & Sons, 2nd edition, 2000

    2000

  27. [27]

    Wu and N

    H.-T. Wu and N. Wu. Think globally, fit locally under the manifold setup: asymptotic analysis of locally linear embedding.The Annals of Statistics, 46(6B):3805–3837, 2018

    2018

  28. [28]

    J. Li. Asymptotic normality of interpoint distances for high-dimensional data with applications to the two-sample problem.Biometrika, 105(3):529–546, 2018

    2018

  29. [29]

    W. Li, Q. Wang, and J. Yao. Eigenvalue distribution of a high-dimensional distance covariance matrix with application.Statistica Sinica, 33(1):149–168, 2023

    2023

  30. [30]

    Meil˘ a and H

    M. Meil˘ a and H. Zhang. Manifold learning: what, how, and why.Annual Review of Statistics and Its Application, 11:393–417, 2024

    2024

  31. [31]

    Ding and R

    X. Ding and R. Ma. Kernel spectral joint embeddings for high-dimensional noisy datasets using duo-landmark integral operators.Journal of the American Statistical Association, 1–28, 2025

    2025

  32. [32]

    Smale and D.-X

    S. Smale and D.-X. Zhou. Geometry on probability spaces.Constructive Approximation, 30(3):311–323, 2009

    2009

  33. [33]

    Erd´ elyi, W

    A. Erd´ elyi, W. Magnus, F. Oberhettinger, and F. G. Tricomi.Higher Transcendental Functions, Vol. II(Bateman Manuscript Project). McGraw-Hill, 1953

    1953

  34. [34]

    Marsaglia and I

    G. Marsaglia and I. Olkin. Generating correlation matrices.SIAM Journal on Scientific and Statistical Computing, 5(2):470–475, 1984

    1984

  35. [35]

    I. J. Schoenberg. Positive definite functions on spheres.Duke Math. J.9(1942), 96–108

    1942

  36. [36]

    J. Baik, G. Ben Arous, and S. P´ ech´ e. Phase transition of the largest eigenvalue for nonnull complex sample covariance matrices.The Annals of Probability, 33(5):1643–1697, 2005

    2005

  37. [37]

    Benaych-Georges and R

    F. Benaych-Georges and R. R. Nadakuditi. The eigenvalues and eigenvectors of finite, low rank perturbations of large random matrices.Advances in Mathematics, 227(1):494–521, 2011. 52

    2011

  38. [38]

    Pastur and V

    L. Pastur and V. Vasilchuk. On the law of addition of random matrices.Communications in Mathematical Physics, 214:249–286, 2000

    2000

  39. [39]

    Zee.Law of addition in random matrix theory

    A. Zee.Law of addition in random matrix theory. Nuclear Physics B, 474(3):726–744, 1996

    1996

  40. [40]

    Euclidean random matrices: solved and open problems

    G. Parisi.Euclidean random matrices: solved and open problems.InApplications of Random Matrices in Physics, NATO Science Series II, vol. 221, Springer, 2006 (arXiv:cond-mat/0512004)

    work page internal anchor Pith review Pith/arXiv arXiv 2006

  41. [41]

    L. D. Landau and E. M. Lifshitz.Quantum Mechanics: Non-Relativistic Theory(Course of Theoretical Physics, Vol. 3). Pergamon Press, 3rd edition, 1977 (§38–39)

    1977

  42. [42]

    Hose and U

    G. Hose and U. Kaldor. Quasidegenerate perturbation theory.The Journal of Physical Chemistry, 86(12):2133–2140, 1982

    1982

  43. [43]

    Araya Day, S

    I. Araya Day, S. Miles, H. K. Kerstens, D. Varjas, and A. R. Akhmerov. Pymablock: An algorithm and a package for quasi-degenerate perturbation theory.SciPost Physics Codebases, 50, 2025

    2025

  44. [44]

    L¨ owdin

    P.-O. L¨ owdin. Studies in perturbation theory. IV. Solution of eigenvalue problem by projection operator formalism.Journal of Mathematical Physics, 3(5):969–982, 1962

    1962

  45. [45]

    J. R. Schrieffer and P. A. Wolff. Relation between the Anderson and Kondo Hamiltonians. Physical Review, 149(2):491–492, 1966

    1966

  46. [46]

    A. J. Smola, Z. L. ´Ov´ ari, and R. C. Williamson. Regularization with dot-product kernels. In Advances in Neural Information Processing Systems 13 (NIPS 2000), pages 308–314. MIT Press, 2001

    2000

  47. [47]

    Davis and W

    C. Davis and W. M. Kahan. The rotation of eigenvectors by a perturbation. III.SIAM Journal on Numerical Analysis, 7(1):1–46, 1970

    1970

  48. [48]

    Helgason.Groups and Geometric Analysis: Integral Geometry, Invariant Differential Opera- tors, and Spherical Functions.Academic Press, 1984

    S. Helgason.Groups and Geometric Analysis: Integral Geometry, Invariant Differential Opera- tors, and Spherical Functions.Academic Press, 1984

    1984

  49. [49]

    Azevedo and V

    D. Azevedo and V. S. Barbosa. Covering numbers of isotropic reproducing kernels on compact two-point homogeneous spaces.Mathematische Nachrichten, 291(1):1–15, 2018

    2018

  50. [50]

    Euclidean random matrices and their applications in physics

    A. Goetschy and S. E. Skipetrov. Euclidean random matrices and their applications in physics. arXiv:1303.2880, 2013

    work page internal anchor Pith review Pith/arXiv arXiv 2013

  51. [51]

    V. N. Stepanov. The method of spherical harmonics for integral transforms on a sphere. Mathematical Structures and Modeling, 2(42):36–48, 2017

    2017

  52. [52]

    J. A. Mingo and R. Speicher.Free Probability and Random Matrices. Fields Institute Mono- graphs, Springer, 2017

    2017

  53. [53]

    Z. D. Bai and J. W. Silverstein.Spectral Analysis of Large Dimensional Random Matrices, 2nd ed. Springer, 2010

    2010

  54. [54]

    G. W. Anderson, A. Guionnet, and O. Zeitouni.An Introduction to Random Matrices. Cambridge University Press, 2010. 53

    2010