REVIEW 6 minor 13 references
On non-monotonicity of logarithmic energy for random matrices
T0 review · 0 major / 6 minor · reviewed 2026-07-31 · grok-4.5
Pith's one-line read Dimensional monotonicity of penalised logarithmic energy fails for general Wigner and i.i.d. random matrices.
desk verdict Clean, explicit counterexamples that kill the unrestricted monotonicity conjectures of Chafaï–Dadoun–Youssef; the Wigner case is short and solid, the i.i.d. case rests on a thin but independently checkable rational inequality. 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 semicircle gap identity expressing FR(μ)−3/4 as the integral of the positive potential vsc plus a nonnegative Fourier discrepancy, together with collision-mass bounds that control how logarithmic energy of Gaussian-smoothed atomic spectral measures scales with log(1/ε).
What would settle it
Recompute the fourteen atoms and masses of the unregularised two-by-two Bernoulli spectral measure and the selected zero and outlier masses in dimension three; if the resulting L3−C2 is not strictly positive, the small-ε separation FC(ν3,ε)>FC(ν2,ε) collapses.
Extended reading notes
Core claim
The conjectural decrease of the quadratically penalised logarithmic energies FR and FC along mean empirical spectral distributions fails in general: there is a finite-energy real Wigner sequence with m2(ρn)=1 for which FR(ρ2)>FR(ρ1)=FR(σ)=3/4, and a concrete centred unit-variance Gaussian-regularised Bernoulli family for which FC(ν3,ε)>FC(ν2,ε) for all sufficiently small ε>0.
Load-bearing premise
The i.i.d. counterexample stands or falls on a strict numerical inequality between an exact two-dimensional collision mass and a partial three-dimensional collision mass obtained by enumerating atoms and selected eigenvalue branches.
Editorial extensions
If this is right
- Unrestricted dimensional monotonicity of FR and FC for mean ESDs does not hold for general Wigner or i.i.d. matrices.
- Any true monotonicity statement must impose further restrictions on the entry law or a large enough dimension threshold.
- Finite-energy counterexamples exist even when every mean ESD has a bounded density and all moments.
- Gaussian regularisation of discrete entry laws is enough to produce rigorous energy comparisons via collision masses.
Reading between the lines
- The same collision-mass comparison may extend to other sparse or lattice-supported entry laws, giving a systematic source of low-dimensional counterexamples.
- Whether monotonicity recovers for all entry laws after some N0, or only inside classical ensembles, is now the natural next quantitative question.
- The semicircle-gap identity may be reusable to certify energy increase whenever a positive mass leaks outside [−2,2].
Editorial analysis
A structured set of objections, weighed in public.
Referee Report
Summary. The manuscript constructs two counterexamples to the dimensional monotonicity questions for penalised logarithmic energies of mean empirical spectral distributions posed by Chafaï, Dadoun, and Youssef [5]. Theorem 1.1 takes a real symmetric Wigner sequence whose entries have the semicircle law itself, so ρ1 = σ; on the positive-probability event that all three entries of M2 lie in (3/2,2), the top eigenvalue exceeds 3/√2 > 2, and a gap identity (Proposition 2.2, FR(µ) − 3/4 = ∫vsc dµ + a nonnegative Fourier term) then yields FR(ρ2) ≥ FR(ρ1) + (p0³/2)(3/4 − log 2) > FR(ρ1). Theorem 1.2 regularises a Bernoulli law (rare atom probability p = 1/32) with a small circular Gaussian, and shows FC(ν3,ε) > FC(ν2,ε) for all small ε by separating limsup and liminf of FC/log(1/ε): an upper bound by the collision mass C2 of the exactly enumerated 14-atom measure ν0_2 (Lemmas 2.3, 4.2–4.3), and a lower bound L3 = a0² + a1² from stable zero/outlier eigenvalue branches of selected 3×3 configurations (Lemmas 2.4, 4.4–4.5, Prop. 4.6), with L3 − C2 given as an explicit positive rational.
Significance. The paper settles, in the negative, Questions 1.1 and 1.2 of Chafaï–Dadoun–Youssef (RMTA 2024) for unrestricted entry laws — a conjecture backed in [5] by exact Gaussian computations and extensive numerics. The strengths are concreteness and verifiability: Theorem 1.1 is essentially parameter-free (the only quantity is p0 = σ((3/2,2))), rests on an exact semicircle gap identity (Prop. 2.2) of independent interest, and the strictness reduces to the elementary log 2 < 3/4. Theorem 1.2 is fully explicit: p = 1/32, exact collision mass of ν0_2 from a 14-atom enumeration, and a final comparison in exact rational arithmetic with no fitted constants. I spot-verified the load-bearing arithmetic (characteristic polynomials of Table 1, Col formula, C2 ≈ 0.392705, a0 ≈ 0.578388, L3 ≈ 0.397276, gap ≈ +0.0046) and it reproduces. The work leaves open, and clearly frames, the interesting question of monotonicity under restrictions or above a dimension threshold.
minor comments (6)
- [§4.1, Lemma 4.1] Lemma 4.1, proof: the covariance display 'E[G0(Gii−Tn)] = 1/√n − 1/√n = 0' only reads correctly with a complex conjugate on the second factor, i.e. E[G0 \overline{(Gii−Tn)}]. For proper complex Gaussians one needs vanishing of both the covariance and the pseudo-covariance; both do vanish here and the conclusion (independence of G0 and Rn) is correct and standard, but as printed the two 1/√n terms arise only in the conjugated product, so the display should be corrected and a half-sentence of justification added.
- [§4.3, Proposition 4.6] Proposition 4.6, proof: the displayed value '|c|√3 = √3/31' should be |c|√3 = √3/√31 = √(3/31) ≈ 0.311. Only positivity is used, so the argument is unaffected.
- [§4.2, Table 1] Table 1: the column header '√62 spec(A)' is confusing on first reading; the entries listed are the eigenvalues of √62A (equivalently, the atoms of ν0_2 scaled by √62). A one-line caption stating this, and noting that each empirical eigenvalue carries weight 1/2 (which explains the masses in Table 2), would make the enumeration easier to audit.
- [§2.2, Proposition 2.2] Proposition 2.2: the gap identity (8) is the equilibrium-condition form of the classical Fourier representation of the logarithmic kernel; a brief remark situating it relative to known relative-energy identities (e.g. in the free-entropy/large-deviation literature [1, 8]) would help readers, even if the self-contained proof is retained.
- [§4.4, Eq. (37)] The margin in (37) is L3−C2 ≈ 4.6×10⁻³, about 1.2% of C2; since the entire Theorem 1.2 funnels through this sign, a short appendix or ancillary file documenting the rational arithmetic (the expansion of (27), the 14-atom mass table, and the final subtraction) would materially assist referees and readers.
- [General] There are scattered typographical artifacts, presumably from source formatting: 'generalβ-ensembles' and 'For ann×nmatrixA' in §1, and run-together text around equations (e.g. 'Sincerε→0', 'log(1/sε)+1/4' line-break). A careful proofread of spacing around inline mathematics is advised.
Circularity Check
No circularity: explicit counterexample constructions with independently derived energy identities and enumerated collision masses.
full rationale
The paper disproves unrestricted dimensional monotonicity by constructing two explicit families (semicircle-entry Wigner matrices; Gaussian-regularised Bernoulli i.i.d. matrices) and verifying FR(ρ2)>FR(ρ1) and FC(ν3,ε)>FC(ν2,ε) from first principles. The gap identity (Prop. 2.2) is derived via Stieltjes inversion and Fourier representation of the log kernel, not fitted to the target inequality. Collision-mass limsup/liminf bounds (Lemmas 2.3–2.4, 4.2, Prop. 4.6) follow from Gaussian convolution estimates and Rouché/Riesz eigenvalue stability; C2 and L3 are obtained by finite configuration enumeration and rational arithmetic, then compared. Choosing p=1/32 so that L3>C2 is a legitimate existence parameter, not a fitted input renamed as a prediction. Citations are to classical RMT/potential theory and to the conjecture paper [5]; none is a self-citation load-bearing uniqueness claim. The derivation chain does not reduce any claimed inequality to its own definitional inputs.
Assumptions & free parameters
free parameters (1)
- Bernoulli rare-entry probability p =
1/32
assumptions (5)
- standard math Logarithmic energy E(μ) is well-defined in (−∞,+∞] for probability measures with finite second moment, and finite when μ has a bounded density (Lemma 2.1).
- standard math Semicircle and circular laws are the unique minimisers of FR and FC with the classical values m2(σ)=1, E(σ)=1/4, FR(σ)=3/4 and analogues for ω (displayed in (2)).
- domain assumption Wigner and i.i.d. matrix normalisations: independent upper-triangular (resp. all) entries, mean zero, variance one, Hermitian symmetry when required.
- standard math Roots of monic polynomials depend continuously on coefficients as unordered multisets; used to pass from regularised to unregularised mean spectral measures as ε↓0.
- standard math Operator-norm tail of a fixed-size complex Ginibre matrix allows P(∥G3∥op≤√log(1/ε))→1, so spectral clusters survive with high probability under ε-perturbation.
Cite this review
Pith. "Pith review of On non-monotonicity of logarithmic energy for random matrices." pith.science (2026). https://pith.science/paper/GNSNSLTX
@misc{pith2026260724170,
author = {Pith},
title = {Pith review of: On non-monotonicity of logarithmic energy for random matrices},
year = {2026},
howpublished = {\url{https://pith.science/paper/GNSNSLTX}},
note = {Machine review of arXiv:2607.24170}
}
read the original abstract
We construct a finite-energy Wigner counterexample and a concrete one-parameter family of Gaussian-regularised Bernoulli entry laws yielding counterexamples to the conjectural dimensional monotonicity of the quadratically penalised logarithmic energy for mean empirical spectral distributions by Chafa\"i, Dadoun, and Youssef.
Reference graph
Works this paper leans on
-
[5]
Monotonicity of the logarithmic energy for random matrices.Random Matrices: Theory and Applications 13, 2 (2024), 2450008
Chafa¨i, D., Dadoun, B.,andYoussef, P . Monotonicity of the logarithmic energy for random matrices.Random Matrices: Theory and Applications 13, 2 (2024), 2450008
2024
-
[1]
W., Guionnet, A.,andZeitouni, O.An Introduction to Random Matrices, vol
Anderson, G. W., Guionnet, A.,andZeitouni, O.An Introduction to Random Matrices, vol. 118 ofCambridge Studies in Advanced Mathematics. Cambridge University Press, Cambridge, 2010
2010
-
[2]
M., Barthe, F.,andNaor, A
Artstein, S., Ball, K. M., Barthe, F.,andNaor, A. Solution of Shannon’s problem on the monotonicity of entropy.Journal of the American Mathematical Society 17, 4 (2004), 975–982
2004
-
[3]
Large deviations for Wigner’s law and Voiculescu’s non-commutative entropy.Probability Theory and Related Fields 108, 4 (1997), 517–542
Ben-Arous, G.,andGuionnet, A. Large deviations for Wigner’s law and Voiculescu’s non-commutative entropy.Probability Theory and Related Fields 108, 4 (1997), 517–542. 21 Non-monotonicity of log-energy
1997
-
[4]
From Boltzmann to random matrices and beyond.Annales de la Faculté des Sciences de Toulouse
Chafa¨i, D. From Boltzmann to random matrices and beyond.Annales de la Faculté des Sciences de Toulouse. Mathématiques 24, 4 (2015), 641–689
2015
-
[6]
Maximal correlation and monotonicity of free entropy and of Stein discrepancy.Electronic Communications in Probability 26(2021), Paper No
Dadoun, B.,andYoussef, P . Maximal correlation and monotonicity of free entropy and of Stein discrepancy.Electronic Communications in Probability 26(2021), Paper No. 24, 10
2021
-
[7]
J.Log-Gases and Random Matrices, vol
Forrester, P . J.Log-Gases and Random Matrices, vol. 34 ofLondon Mathematical Society Monographs Series. Princeton University Press, Princeton, NJ, 2010
2010
-
[8]
77 of Mathematical Surveys and Monographs
Hiai, F.,andPetz, D.The Semicircle Law, Free Random Variables and Entropy, vol. 77 of Mathematical Surveys and Monographs. American Mathematical Society, Providence, RI, 2000
2000
Show all 13 references
-
[9]
Madiman, M.,andBarron, A. R. Generalized entropy power inequalities and monotonicity properties of information.IEEE Transactions on Information Theory 53, 7 (2007), 2317–2329
2007
-
[10]
B.,andTotik, V .Logarithmic Potentials with External Fields, vol
Saff, E. B.,andTotik, V .Logarithmic Potentials with External Fields, vol. 316 of Grundlehren der Mathematischen Wissenschaften. Springer-Verlag, Berlin, 1997
1997
-
[11]
A free analogue of Shannon’s problem on monotonicity of entropy
Shlyakhtenko, D. A free analogue of Shannon’s problem on monotonicity of entropy. Advances in Mathematics 208, 2 (2007), 824–833
2007
-
[12]
H-theorem and beyond: Boltzmann’s entropy in today’s mathematics
Villani, C. H-theorem and beyond: Boltzmann’s entropy in today’s mathematics. In Boltzmann’s Legacy. European Mathematical Society, Zürich, 2008, pp. 129–143
2008
-
[13]
Aspects of free probability
Voiculescu, D. Aspects of free probability. InXIVth International Congress on Mathematical Physics. World Scientific, Hackensack, NJ, 2005, pp. 145–157. 22
2005
Reviewed July 31, 2026 · model on record in the stance chip above.
Discussion (0). Continue with ORCID to comment.