REVIEW 3 major objections 3 minor 5 cited by
Localization of One-Dimensional Random Band Matrices
T0 review · 3 major / 3 minor · reviewed 2026-08-05 · deepseek-v4-flash
Pith's one-line read This paper proves that, for a general class of one-dimensional random band matrices with bandwidth $W$, all eigenvectors are exponentially localized at the sharp scale $W^2$ whenever $W^2\ll n$, with probability tending to one.
desk verdict Plausible and potentially important localization theorem at the sharp W² scale, but the proof is invisible and the class compatibility with the cited delocalization results is unverified; deserves a serious referee. 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 central quantitative object is the exponential decay scale $W^2$ for eigenvector components of a random band matrix. The paper's argument is built around proving this decay uniformly for all bandwidths satisfying $W^2\ll n$; the sharp scale is what makes the result match the conjectured transition. The complementary delocalization inputs are two cited results that cover the regime $W^2\gg n$.
What would settle it
Find, for some sequence of $n\times n$ random band matrices with $W^2\ll n$ (e.g. $W=n^{0.4}$), an eigenvector whose amplitudes do not decay exponentially at scale $W^2$—for instance, a vector spread over $n$ sites with entries of size $n^{-1/2}$. A rigorous construction or a numerical observation of such delocalized eigenvectors for a non-vanishing fraction of the spectrum would contradict the theorem.
Extended reading notes
Core claim
On the paper's own terms, the central discovery is that the eigenvector localization length in one-dimensional random band matrices is of order $W^2$: for $W^2 \ll n$, with probability tending to one, each eigenvector is exponentially localized, meaning there is a localization center and the eigenvector amplitude decays exponentially over a length scale of order $W^2$. The sharpness of this scale means the decay rate cannot be improved to a longer scale. Combined with the delocalization results cited in the paper for $W^2\gg n$, this would establish the long-conjectured localization–delocalization transition for this class of random band matrices.
Load-bearing premise
The combined transition claim assumes that the class of random band matrices handled in this paper is exactly the same class for which the cited delocalization results were proved, and the abstract does not display either set of hypotheses.
Editorial extensions
If this is right
- For $W^2\ll n$, eigenvector localization at scale $W^2$ holds with probability tending to one, for all eigenvectors.
- The localization–delocalization transition for one-dimensional random band matrices is pinned at $W\sim \sqrt{n}$.
- The exponential decay scale $W^2$ is sharp, meaning the localization length cannot be polynomially larger.
- The result applies to a general class of band matrices, not only Gaussian or Bernoulli entries.
Reading between the lines
- If the compatibility assumption holds, the transition is sharp at $W\asymp \sqrt{n}$, which would rule out an intermediate 'critical' phase of extended but non-ergodic eigenvectors.
- A testable numerical signature: in the regime $W^2\ll n$, eigenvalue spacings should follow Poisson statistics, characteristic of localized states; the paper's eigenvector result would support that.
- The same proof strategy may extend to other one-dimensional disordered ensembles, such as block band matrices or random Schrödinger operators with finite-range off-diagonal disorder.
Editorial analysis
A structured set of objections, weighed in public.
Referee Report
Summary. The manuscript (arXiv:2508.05802) concerns n×n random band matrices of bandwidth W. Its central claim, stated in the abstract, is that when W^2 ≪ n, with high probability every eigenvector is localized and decays exponentially at the sharp scale W^2. The abstract further asserts that, taken together with delocalization results of Yau–Yin and Erdős–Riabov, this establishes the conjectured localization–delocalization transition for a large class of random band matrices. As submitted, only the abstract is available; no theorem statement, proof, precise hypotheses, or quantitative probability estimates are provided. My assessment is therefore necessarily limited to the claims as stated.
Significance. If the central theorem is correct, this is a major contribution to random matrix theory. Proving exponential localization at the predicted scale W^2 for a general class of band matrices, and combining it with the complementary delocalization regime, would resolve a long-standing conjecture and provide the sharp transition picture. The claimed scale is explicit and empirically falsifiable, and the reliance on independent delocalization theorems by Yau–Yin and Erdős–Riabov is appropriate and strengthens the overall narrative. However, the significance is conditional: it depends on the missing proof and on the unstated compatibility of the matrix class with the cited delocalization results.
major comments (3)
- [Abstract, first sentence] The central assertion, 'we prove that with high probability the eigenvectors ... are localized and decay exponentially at the sharp scale W^2', is stated without a precise theorem. The abstract gives no definition of the 'general class' of random band matrices, no assumptions on entry distributions or band shape, no probability estimate (rate of convergence, uniformity in W and n), and no specification of whether all eigenvectors or only bulk eigenvectors are covered. These are load-bearing details: without them the claim cannot be checked. This is not a minor omission; it is the core mathematical content of the paper.
- [Abstract, last sentence] The asserted 'localization–delocalization transition' depends on a compatibility premise that is not displayed. The cited delocalization results are for particular matrix models (Yau–Yin arXiv:2501.01718; Erdős–Riabov arXiv:2506.06441), but the abstract does not state that the 'general class' treated here coincides with the classes in those papers. If the entry distribution, dependence structure, or bandwidth growth conditions differ, the concatenation of localization for W^2 ≪ n and delocalization for W^2 ≫ n does not establish a transition. Furthermore, a genuine transition statement normally requires control near the critical window W ~ n^{1/2}; the two regimes W^2 ≪ n and W^2 ≫ n leave that window uncovered. The abstract needs to either state the exact transition statement or soften the claim.
- [Abstract, 'sharp scale W^2'] The phrase 'sharp scale' is not defined. It presumably means that the localization length is of order W^2, but one needs a quantitative formulation: e.g., sup-norm decay of eigenvector entries at rate exp(-c|x-y|/W^2) with explicit constants, or a matching lower bound. Without this definition, the claim is ambiguous and the comparison to the conjectured transition cannot be evaluated. This is a technical but load-bearing clarification.
minor comments (3)
- [Abstract, notation] The asymptotic condition 'W^2 ≪ n' should be quantified as W = W(n) with W^2/n → 0. The current notation is informal.
- [Abstract, terminology] The terms 'localized' and 'exponentially decay' are used without definitions. A precise definition in terms of eigenvector entry distributions or localization centers is needed.
- [Abstract, references] The references to Yau–Yin and Erdős–Riabov appear relevant, but the abstract should indicate whether the cited results hold in the same W-regime and for the same matrix class.
Circularity Check
No circularity found: abstract shows independent derivation and complementary external citations.
full rationale
No full text is available, only the abstract. The abstract makes a mathematical claim (localization at scale W^2 for W^2 ≪ n) and cites two delocalization papers by other authors as complementary results, not as inputs to the proof. There are no fitted parameters, no self-citations, no equations, and no definitional identifications visible. The combined transition statement depends on the external delocalization theorems being correct and on the matrix classes matching, but that is a condition for the combined claim, not a circular reduction. A citation of external complementary results does not constitute circularity. Under the hard rules, circularity can only be claimed when a specific reduction is exhibited; none can be exhibited from the abstract alone. Therefore the honest finding is no significant circularity, score 0.
Assumptions & free parameters
assumptions (3)
- domain assumption The 'general class' of n×n random band matrices satisfies unspecified technical hypotheses (entry distribution, independence, symmetry) required by the proof.
- domain assumption The delocalization theorems of Yau and Yin (arXiv:2501.01718) and Erdős and Riabov (arXiv:2506.06441) are correct and apply to the same matrix class with complementary regime W² ≫ n.
- standard math Standard probabilistic and spectral tools (e.g., resolvent or Wegner-type estimates) used in the proof are valid for the matrix model.
Cite this review
Pith. "Pith review of Localization of One-Dimensional Random Band Matrices." pith.science (2026). https://pith.science/paper/EXMIPI6I
@misc{pith2026250805802,
author = {Pith},
title = {Pith review of: Localization of One-Dimensional Random Band Matrices},
year = {2026},
howpublished = {\url{https://pith.science/paper/EXMIPI6I}},
note = {Machine review of arXiv:2508.05802}
}
abstract
We consider a general class of $n\times n$ random band matrices with bandwidth $W$. When $W^2\ll n$, we prove that with high probability the eigenvectors of such matrices are localized and decay exponentially at the sharp scale $W^2$. Combined with the delocalization results of Yau and Yin [arXiv:2501.01718] and of Erd\H{o}s and Riabov [arXiv:2506.06441], this establishes the conjectured localization-delocalization transition for a large class of random band matrices.
Forward citations
Cited by 5 Pith papers
-
On a Rosenzweig-Porter-type model
Provides uniform local laws and localization analysis for the general Rosenzweig-Porter model H = H0 + λW, generalizing previous results on deformed Wigner matrices.
-
Localization Lengths of Power-Law Random Band Matrices
The paper establishes rigorous lower bounds on eigenvector localization lengths for power-law random band matrices in four regimes of the decay exponent α, verifying a physical conjecture via new resolvent techniques.
-
On a Rosenzweig-Porter-type model
Uniform single- and two-resolvent local laws for arbitrary deformations of Wigner matrices yield ETH and localization profiles for all λ and H0.
-
Gradual eigenvector ergodization in coupled Ginibre matrices
Explicit large-N asymptotic formula for gradual eigenvector ergodization in two coupled Ginibre matrices, plus vanishing of eigenvalue density at the origin beyond critical scaled coupling |tilde c|=1.
-
Characteristic polynomials of non-Hermitian random band matrices near the threshold
The second correlation function of characteristic polynomials for non-Hermitian random band matrices is studied asymptotically in the critical regime W proportional to sqrt(N) as N and W tend to infinity.
Reviewed August 5, 2026 · model on record in the stance chip above.
Discussion (0). Sign in to comment.