Quantum diffusion and delocalization in one-dimensional band matrices via the flow method

Kevin Yang; Sofiia Dubova

REVIEW 3 major objections 6 minor 2 cited by

Quantum diffusion and delocalization in one-dimensional band matrices via the flow method

T0 review · 3 major / 6 minor · reviewed 2026-08-11 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read This paper proves that bulk eigenvectors of one-dimensional Gaussian random band matrices are delocalized and that the resolvent exhibits quantum diffusion, once the bandwidth W exceeds N^{8/11} times a small power, improving the…

desk verdict A genuine threshold improvement via the flow method, but the proof rests on a sketched combinatorial lemma (Lemma 22) that needs a full proof before the main theorems can be trusted. read the letter →

arxiv 2412.15207 v1 pith:KE7K4KVN submitted 2024-12-19 math.PR math-phmath.MP

Sofiia Dubova , Kevin Yang This is my paper · ORCID

classification math.PRmath-phmath.MP MSC 15B5260B2082B44

keywords randombandmatricesquantumdiffusioneigenvectordelocalizationlocalsemicirclelawflowmethoddiagrammaticexpansionbandwidththresholdbulkspectrum

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 establishes a quantitative form of quantum diffusion for one-dimensional Gaussian random band matrices and derives bulk eigenvector delocalization from it. The main result is that, when the bandwidth satisfies W ≥ $N^{{8/11+υ}}$ and the spectral scale is η ≍ $W^{2}$$N^{{-2}}$, the smoothed absolute-square resolvent matrix T_xy is close to an explicit diffusion profile Θ_xy, with error $W^{{-7/4}}$$η^{{-3/2}}$. From this entrywise approximation the paper obtains the local law |G_xy − m(z)δ_xy|^2 ≺ $W^{{-1}}$$η^{{-1/2}}$, and then, via the delocalization criterion of [21], that at most a fraction ≈ √ε of bulk eigenvectors can be localized to any scale ℓ ≪ N. A sympathetic reader cares because this moves the rigorous delocalization threshold closer to the conjectured W ≫ $N^{{1/2}}$ transition and explains the transition through the spectral gap of a random-walk generator.

What carries the argument

The central object is the T-matrix T(z)_ab, a doubly smoothed version of |G_xy(z)|^2, compared with Θ(z), which is the resolvent of a random-walk generator on the discrete torus. The machinery is an SDE flow in which the entries of the band matrix evolve as Brownian motions and the spectral parameter travels from w_0 = −m(z)^{-1} to w_1 = z; the error E_t = T_t − Θ_t satisfies an SDE with martingale, drift, and quadratic terms. The quadratic term is controlled by a stopping-time argument, the martingale term by quadratic-variation estimates, and the drift term by Gaussian integration by parts expressed through diagrammatic expansions, namely a loop expansion and a regular-vertex expansion. Expanding the drift integrand twice yields main terms bounded deterministically and fluctuation terms bounded by even-moment estimates.

What would settle it

Write out the even-moment expansion of the fluctuation terms F_{i,s}(z) in Lemma 22 for p = 1 and p = 2, enumerate every diagrammatic term, and check each against the claimed bound $W^{{3δ_stop/10}}$$W^{{-1+2ε}}$|Im w_t|^{-1} · $W^{{-1}}$|Im w_t|^{-1/2}; one oversized term would invalidate Proposition 11 and hence Theorems 2 and 4.

Watch

Extended reading notes

Core claim

The central claim is a comparison between the T-matrix, defined by T(z)_ab = ∑_{x,y} $S^{{1/2}}$_{ax}|G_{xy}(z)|^2 $S^{{1/2}}$_{yb}, and the diffusion profile Θ(z) = |m(z)|^2S (1 − |m(z)|^2S)^{-1}, where S is the doubly stochastic variance matrix and m(z) is the Stieltjes transform of the semicircle law. Under |E| < 2, η ≍ $W^{2}$$N^{{-2}}$, and W ≥ $N^{{8/11+υ}}$, the paper proves max_{x,y}|T_xy − Θ_xy| ≺ $W^{{-7/4}}$$η^{{-3/2}}$, and hence |G_xy − m(z)δ_xy|^2 ≺ $W^{{-1}}$$η^{{-1/2}}$. The proof uses the flow method: the spectral parameter moves at constant speed in the upper half-plane while the matrix entries evolve as Brownian motions with variance profile S, and the error E_t = T_t − Θ_t is split into martingale, drift, and quadratic parts and controlled up to a stopping time.

Load-bearing premise

The proof of the crucial drift estimate (Proposition 11) relies on a moment-expansion bound (Lemma 22) whose combinatorial control is only sketched, with the assertion that the verification is 'straightforward to verify'; if that graph-counting bound is not valid, the drift estimate and with it Theorems 2 and 4 do not follow from the written argument.

Editorial extensions

If this is right

For W ≥ N^{8/11+υ} and η ≍ W^2N^{-2}, the local law |G_xy − m(z)δ_xy|^2 ≺ W^{-1}η^{-1/2} holds in the bulk, with the stated error size.
The fraction of bulk eigenvectors localized to any scale ℓ ≪ N is at most √ε + O(N^{-c}), so no positive fraction of bulk eigenvectors is localized.
Quantum diffusion holds through the relaxation time: the T-matrix is close to the resolvent of a random-walk generator with diffusivity proportional to W^2, matching the Thouless-time heuristic.
The rigorous delocalization threshold for one-dimensional Gaussian band matrices drops from W ≫ N^{3/4} to W ≫ N^{8/11}.
The diffusion-profile comparison provides a natural explanation of the conjectured W ≫ N^{1/2} transition, since the random-walk spectral gap is of order W^2N^{-2}.

Reading between the lines

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

If the operator-norm estimate conjectured in Section 5 of the paper can be proved, the bound in Lemma 10 would improve by a factor |Im w_s|^{-1/2}, plausibly pushing the bandwidth threshold below N^{8/11} toward the conjectured N^{1/2} transition.
The same flow-plus-graphical-expansion route could be adapted to non-Gaussian band models by replacing the Gaussian integration-by-parts identities with cumulant expansions; the first test would be whether the resulting diagrammatic bounds remain of the same form.
A direct check of the argument would be to carry out the omitted combinatorial verification in Lemma 22 for the lowest even moments p = 1 and p = 2, enumerating every diagrammatic term to see whether the claimed fluctuation bound holds.
The delocalization result, if it extends to asymmetric variance profiles or to higher dimensions, would connect the band-matrix transition to the known high-dimensional delocalization results, where the required bandwidth is essentially W ≫ L^ε.

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.

Desk Editor's Note

read the letter

Here's my take on Dubova-Yang, arXiv 2412.15207. The paper does something real: it pushes the proven delocalization threshold for 1D Gaussian band matrices from W >> N^{3/4} down to W >> N^{8/11}, and it does so with the flow method, which is new for this model. The quantum diffusion result for the T-matrix is also new. The overall architecture—splitting E_t into martingale, drift, and nonlinear parts, with a stopping time bootstrap—is coherent and follows the EKY/YYY line. The diffusion-profile interpretation of Theta is a nice touch, because it makes the conjectured W >> N^{1/2} transition look natural rather than magical.

The soft spot is exactly where the reader flagged it. Proposition 11, the drift estimate that carries the whole paper, rests on Lemma 22. That lemma controls the fluctuation terms F_{i,s} and F_{0,s} via a 2p-th moment expansion, and the proof is not written out. The manuscript says 'It is straightforward to verify' the four graph invariants, then gives a word-level algorithm for the summation. There is no induction, no case analysis, no count of the graph families for general p. If any of those invariants fail—say a component is a cycle rather than a tree, or the edge count is off—the Chebyshev bound gains a polynomial factor that the W^{3*delta_stop/10} slack may not absorb, and Theorem 2 and 4 collapse. So the gap is load-bearing, though not obviously wrong: the invariants are plausible, and the algorithm resembles the standard machinery from [38,39,40]. It may well be fillable. But as it stands, the main theorem is conditional on an unchecked combinatorial claim.

Other issues are minor by comparison. Lemma 10's upgrade from discrete to continuous time is sketched, though that sort of net argument is standard. The stopping time definition is a legitimate bootstrap, not a circularity. I don't see a free-parameter problem; the constants are not fitted to any data.

Who is this for? Specialists in random matrix theory, especially those working on band matrices or the flow method. A general math reader will not get much without a large investment. It deserves a serious referee: the result is important enough and the architecture is plausible enough that a specialist should spend the time to verify Lemma 22. My own verdict is conditional—I would not build on the theorem until Lemma 22 is proved in full. But I would send it to review, with the expectation of heavy revision.

Referee Report

3 major / 6 minor

Summary. The paper studies Gaussian random band matrices on the one-dimensional torus and proves quantum diffusion and eigenvector delocalization under the condition W ≫ N^{8/11}. The main result, Theorem 2, compares the T-matrix T_xy to a diffusion profile Θ_xy with error W^{-7/4} η^{-3/2} when η ≍ W^2 N^{-2}; Theorem 4 derives the local law |G_xy - m(z)δ_xy|^2 ≺ W^{-1} η^{-1/2}, and Corollary 5 gives complete delocalization of bulk eigenvectors. The proofs are organized around a flow method: an SDE for the resolvent and T-matrix, a stopping-time construction, and a decomposition of the error E_t into martingale (E^M), drift (E^D), and nonlinear (E^S) terms. The drift term is controlled by a graphical expansion culminating in Proposition 11, whose proof rests on the even-moment estimate in Lemma 22.

Significance. If the main theorems are correct, the paper gives a substantial improvement over the previous best threshold W ≫ N^{3/4} for delocalization and quantum diffusion in one-dimensional band matrices, and it introduces the flow method as a new tool in this setting. The statement of Theorem 2 is sharp in its explicit error rate, and the derivation of delocalization from the diffusion profile is conceptually clean and matches the conjectured threshold W ≫ N^{1/2} up to the technical gap. The paper also contains useful auxiliary estimates (Lemmas 23–25) that are stated carefully. The main reservation is that the central new estimate, Proposition 11, depends on a graph-counting lemma whose proof is only sketched; until that proof is completed, the main theorems are not fully established.

major comments (3)

[§4, proof of Lemma 22] Lemma 22 is the load-bearing estimate for the drift term: it bounds the fluctuation terms F_{i,s}(z) and F_{0,s}(z) that appear after two rounds of Gaussian integration by parts, and Proposition 11, Theorem 8, Theorem 2, and Theorem 4 all depend on it. The proof asserts, without a complete case analysis or induction, that each term in the moment expansion satisfies the four invariants: 4p double edges attached to distinct inner vertices, 2p−k tree components with respect to waved edges, 12p−2k oriented G-edges split into disjoint loops, and connectivity of every vertex to an a-vertex. The text says 'It is straightforward to verify' and then describes an algorithmic re-expansion in words. This is not a complete proof of a bound that is polynomial in W: a single missing factor of N or W from an extra cycle or an unsummed edge would break the Chebyshev bound and with it the estimate on E^{D,stop}_t. Please supply a complete proof—an induction on the expansion steps with all graph types enumerated, or a fully specified combinatorial argument that establishes these invariants and the resulting moment bound.
[§4, Lemma 22] The moment bound for M_{i,s,ab}^{2p} is obtained by a summation procedure that 'covers the resulting graph with 2p disjoint trees rooted at 2p a-vertices' and selects 4p−k summation edges. This step is not rigorous as written: it is not specified how the trees are chosen for an arbitrary graph satisfying the stated invariants, why the selected edges are always available, and how the constants depend on p and k. Since the final bound is required for all p in a Chebyshev estimate, the proof must show that the procedure applies uniformly, without hidden p-dependent combinatorial factors that could affect the N^{-D} rate.
[§4, Proposition 11] The proof of Proposition 11 combines Lemmas 18–22, but the only stochastic estimate among them is Lemma 22; Lemmas 20 and 21 are deterministic bounds for the main terms. Because Lemma 22 is not fully proved, the bootstrap behind Theorem 8 is incomplete. In particular, the stopping time τstop in (3.3)–(3.5) is shown to be self-propagating only if the drift term E^{D,stop}_t is controlled at the stated rate. Please clarify the logical dependence and provide the missing proof of Lemma 22 before the main theorems can be accepted.

minor comments (6)

[Title page] The affiliation contains a typo: 'Univeristy' should be 'University'.
[§4] The sentence 'The analysis of the drift term E D,stop t (z) requires requires expanding the integrands' contains a doubled word 'requires requires'.
[Notation] The text uses the unconventional notation /llbracket1, N/rrbracket for {1,...,N}; this is not defined in the introduction and appears without explanation. Please define it or use standard notation.
[§3.3] In the proof of Theorem 8, the event {τstop,2 ≠ 1} ∩ {τstop,1 = 1} is handled via Corollary 14, but the role of the assumption η ≍ W^2 N^{-2} in that Corollary is not explicit; please state where this assumption is used.
[§5] The conjectured improved bound (5.2) is presented as a conjecture without a proof; this is fine for a discussion section, but the word 'conjecture' should be used explicitly and the dependence on the main theorem should be clarified.
[Appendix A] The reference to [15] for the heat-kernel estimate (A.2) in Lemma 23 may need a specific equation or lemma number; the current citation points to a paper on a different process and relies on an 'inspection of its proof'.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the diffusion profile is an independent target and the stopping-time bootstrap is legitimate.

full rationale

I checked the derivation chain for reductions of the predicted quantities to the inputs. The target objects are genuinely independent: T_t is defined from the matrix resolvent, Θ_t is the explicit diffusion profile |m(z)|^2 S (1 - t|m(z)|^2 S)^{-1}, and no parameter in the final bounds is fitted to the data being predicted. The stopping times in (3.3)-(3.5) are phrased in terms of the desired smallness, but the paper proves, via Lemmas 9 and 10 and Proposition 11, that those thresholds are not crossed with overwhelming probability; this is a standard bootstrap, not a definitional identity. The drift analysis in Section 4 relies on Lemma 16, whose loop and regular-vertex expansions are stated as identities and cited to published independent work [38]; these identities do not assume the smallness of E_t or the local law. Corollary 5 invokes Proposition 7.1 of [21] as an external criterion, and the paper verifies its hypotheses from (2.2) rather than building the conclusion into the hypotheses. The proof of Lemma 22 contains an omitted combinatorial verification -- the text says 'It is straightforward to verify that each term in the expansions satisfies...' and then describes an algorithmic re-expansion in words. This is a completeness/correctness gap in a load-bearing graph-counting step, but it is not circularity: the claimed invariants (12p-2k oriented G-edges, 2p-k tree components, 4p inner vertices per double edge) would support a bound, not assume the theorem. No fitted parameter is renamed as a prediction, and no load-bearing step reduces to a self-citation chain by the authors of this paper. Therefore the appropriate circularity score is 0.

Assumptions & free parameters 2 free parameters · 5 assumptions · 0 invented entities

The central claim rests on no new physical entities or fitted constants. The free parameters δstop and ε are technical slack variables. The main assumptions are the banded variance profile with a square root of the same form, standard semicircle Stieltjes transform facts, an external heat kernel estimate, and structural identities from prior works by the same group.

free parameters (2)

δstop = small fixed positive constant
Appears in stopping time definitions (3.3)-(3.5); the estimates require it to be sufficiently small but independent of N, and it does not affect the asymptotic threshold.
ε in Lemma 24 = arbitrary fixed positive
A technical slack parameter in the off-diagonal heat kernel bound; final bounds hold for any fixed ε>0, so it is not fitted.

assumptions (5)

domain assumption The variance profile matrix S admits a matrix square root S^{1/2} of the same banded form with a compactly supported symmetric density.
Used to define the T-matrix and in averaging estimates; the authors call it 'entirely technical' and expect it can be removed (Section 2, after (2.1)).
standard math Bulk Stieltjes transform properties: for |E|<2, |Im m(z)| ≍ 1 and 1 - |m(z)|^2 ≍ η.
Used throughout to change variables σ=Im w_s and to bound Θ entries (Lemma 23, Lemma 3.5/6.2 of [21]/[22]).
standard math Heat kernel estimate (A.2) for the band random walk, derived by adapting (A.12) from Dembo-Tsai [15] to general W.
Used to prove the pointwise bounds in Lemma 23; the authors provide a sketch but do not reproduce the full W-dependence derivation (Appendix A, proof of Lemma 23).
standard math The loop expansion and regular vertex expansion identities of Lemma 16 are valid for all s∈[0,1] by extension of Lemma 3.5 and Lemma 3.14 of [38].
These identities are the backbone of the drift expansion in Section 4; they are quoted from prior work.
standard math Proposition 7.1 of [21] applies to the present model, giving the delocalization bound from the local law.
Used to prove Corollary 5; the paper checks the hypotheses but relies on that external criterion.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Quantum diffusion and delocalization in one-dimensional band matrices via the flow method." pith.science (2026). https://pith.science/paper/KE7K4KVN

@misc{pith2026241215207,
  author       = {Pith},
  title        = {Pith review of: Quantum diffusion and delocalization in one-dimensional band matrices via the flow method},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/KE7K4KVN}},
  note         = {Machine review of arXiv:2412.15207}
}

abstract

We study a class of Gaussian random band matrices of dimension $N \times N$ and band-width $W$. We show that delocalization holds for bulk eigenvectors and that quantum diffusion holds for the resolvent, all under the assumption that $W \gg N^{8/11}$. Our analysis is based on a flow method, and a refinement of it may lead to an improvement on the condition $W \gg N^{8/11}$.

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 2 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

The Zigzag Strategy for Random Band Matrices
math.PR 2025-06 accept novelty 8.0 of 10

For random band matrices with width W >> sqrt(N), the paper proves complete delocalization, Wigner-Dyson statistics, and quantum unique ergodicity for general distributions and variance profiles.
Delocalization of One-Dimensional Random Band Matrices
math.PR 2025-01 conditional novelty 8.0 of 10

For one-dimensional block band matrices with W > N^{1/2+c}, the paper proves the local semicircle law, eigenvector delocalization, quantum unique ergodicity, and GUE universality.

Reference graph

Works this paper leans on

40 extracted references · 40 canonical work pages · cited by 2 Pith papers

[1]

Aizenman and S

M. Aizenman and S. Molchanov. Localization at large diso rder and at extreme energies: an elementary derivation. Communications in Mathematical Physics , 157:245–278, 1993

work page 1993
[2]

Aizenman and S

M. Aizenman and S. Warzel. Extended states in a lifshitz t ail regime for random Schr¨ odinger operators on trees. Physical Review Letters, 106:136804, 2011

work page 2011
[3]

Aizenman and S

M. Aizenman and S. Warzel. Resonant delocalization for r andom Schr¨ odinger operators on tree graphs. Journal of the European Mathematical Society , 15:1167–1222, 2013

work page 2013
[4]

P . W . Anderson. Absence of diffusion in certain random la ttices. Physical Review Letters , 109:1492– 1505, 1958

work page 1958
[5]

Z. Bao, L. Erd˝ os, H.-T. Y au, and J. Yin. Delocalization f or a class of random block band matrices. Probability Theory and Related Fields , 167:673–776, 2017. 33

work page 2017
[6]

Bogachev, S

L. Bogachev, S. Molchanov, and L. A. Pastur. On the densit y of states of random band matrices. Matematicheskie Zametki (English: Mathematical Notes) , 157 (English: 50):41–42 (English: 1232– 1242), 1991

work page 1991
[7]

Bourgade

P . Bourgade. Random band matrices. Proceedings of the International Congress of Mathematicia ns, pages 2759–2783, 2018

work page 2018
[8]

Bourgade, F

P . Bourgade, F. Y ang, H.-T. Y au, and J. Yin. Random band ma trices in the delocalized phase, II: Generalized resolvent estimates. Journal of Statistical Physics , 174:1189–1221, 2019

work page 2019

Show all 40 references

[9]

Bourgade, H.-T

P . Bourgade, H.-T. Y au, and J. Yin. Random band matrices in the delocalized phase, I: Quantum unique ergodicity and universality. Communications on Pure and Applied Mathematics , 73:1526–1596, 2020

2020
[10]

Bourgain and C

J. Bourgain and C. Kenig. On localization in the continu ous anderson-bernoulli model in higher dimen- sion. Inventiones Mathematicae, 161:389–426, 2005

2005
[11]

Carmona, A

R. Carmona, A. Klein, and F. Martinelli. Anderson local ization for Bernoulli and other singular poten- tials. Communications in Mathematical Physics , 108:41–66, 1987

1987
[12]

Casati, L

G. Casati, L. Molinari, and F. Izrailev. Scaling proper ties of band random matrices. Physical Review Letters, 64:1851–1854, 1990

1990
[13]

Cipolloni, R

G. Cipolloni, R. Peled, J. Schenker, and J. Shapiro. Dyn amical localization for random band matrices up to W ≪ N 1/4. Communications in Mathematical Physics , 405, 2024

2024
[14]

Damanik, R

D. Damanik, R. Sims, and G. Stolz. Localization for one- dimensional, continuum, Bernoulli-Anderson models. Duke Mathematical Journal, 114:59–100, 2002

2002
[15]

Dembo and L.-C

A. Dembo and L.-C. Tsai. Weakly asymmetric non-simple e xclusion process and the Kar- dar–Parisi–Zhang equation. Communications in Mathematical Physics , 341:219–261, 2016

2016
[16]

Ding and C

J. Ding and C. Smart. Localization near the edge for the A nderson Bernoulli model on the two dimen- sional lattice. Inventiones Mathematicae, 219:467–506, 2020

2020
[17]

Edwards and D

J. Edwards and D. Thouless. Numerical studies of locali zation in disordered systems. Journal of Physics C: Solid State Physics , 5:807–820, 1972

1972
[18]

K. Efetov. Supersymmetry in disorder and chaos . Cambridge University Press, 2017

2017
[19]

Erd˝ os and A

L. Erd˝ os and A. Knowles. Quantum diffusion and delocal ization for band matrices with general distri- bution. Annales de l’Institut Henri Poincare , 12:1227–1319, 2011

2011
[20]

Erd˝ os and A

L. Erd˝ os and A. Knowles. Quantum diffusion and eigenfunction delocalization in a random band matrix model. Communications in Mathematical Physics , 303:509–554, 2011

2011
[21]

Erd˝ os, A

L. Erd˝ os, A. Knowles, H.-T. Y au, and J. Yin. Delocaliza tion and diffusion proﬁle for random band matrices. Communications in Mathematical Physics , 323:367–416, 2013

2013
[22]

Erd˝ os and H.-T

L. Erd˝ os and H.-T. Y au. A dynamical approach to random m atrix theory. Courant Lecture Notes in Mathematics, 28:367–416, 2017

2017
[23]

Fr¨ olich, F

J. Fr¨ olich, F. Martinelli, E. Sccopola, and T. Spencer. Constructive proof of localization in the Anderson tight binding model. Communications in Mathematical Physics , 101:21–46, 1985

1985
[24]

Fr¨ olich and T

J. Fr¨ olich and T. Spencer. Absence of diffusion in the a nderson tight binding model for large disorder or low energy. Communications in Mathematical Physics , 83:151–184, 1988. 34

1988
[25]

Germinet and A

F. Germinet and A. Klein. A comprehensive proof of local ization for continuous Anderson models with singular random potentials. Journal of the European Mathematical Society , 15:53–143, 2013

2013
[26]

Li and L

L. Li and L. Zhang. Anderson-Bernoulli localization on the 3D lattice and discrete unique continuation principle. Duke Mathematical Journal, 171:327–415, 2022

2022
[27]

Peled, J

R. Peled, J. Schenker, M. Shamis, and S. Sodin. On the Weg ner orbital model. International Mathe- matics Research Notices, 2019:1030–1058, 2017

2019
[28]

Schenker

J. Schenker. Eigenvector localization for random band matrices with power law band width. Communi- cations in Mathematical Physics , 290:1065–1097, 2009

2009
[29]

Shcherbina and T

M. Shcherbina and T. Shcherbina. Universality for 1d random band matrices. Communications in Mathematical Physics, 385:667–716, 2019

2019
[30]

Shcherbina

T. Shcherbina. Universality of the local regime for the block band matrices with a ﬁnite number of blocks. Journal of Statistical Physics , 155:466–499, 2014

2014
[31]

S. Sodin. The spectral edge of some random band matrices . Annals of Mathematics , 172:2223–2251, 2010

2010
[32]

T. Spencer. Random banded and sparse matrices. Oxford handbook of random matrix theory , pages 471–488, 2015

2015
[33]

Thouless

D. Thouless. Maximum metallic resistance in thin wires . Physical Review Letters , pages 1167–1169, 1977

1977
[34]

von Soosten and S

P . von Soosten and S. Warzel. Non-ergodic delocalizati on in the rosenzweig–porter model. Letters in Mathematical Physics, 109:905–922, 2019

2019
[35]

von Soosten and S

P . von Soosten and S. Warzel. Random characteristics fo r Wigner matrices. Electronic Journal of Probability, 24:1–12, 2019

2019
[36]

E. Wigner. Distribution of neutron resonance level spa cing. Columbia University Report CU175 , 1957

1957
[37]

C. Xu, F. Y ang, H.-T. Y au, and J. Yin. Bulk universality a nd quantum unique ergodicity for random band matrices in high dimensions. Annals of Probability, 52:765–837, 2024

2024
[38]

Y ang, H.-T

F. Y ang, H.-T. Y au, and J. Yin. Delocalization and quant um diffusion of random band matrices in high dimensions I: Self-energy renormalization. arXiv:2103.12048, 2021

2021 arXiv
[39]

Y ang, H.-T

F. Y ang, H.-T. Y au, and J. Yin. Delocalization and quant um diffusion of random band matrices in high dimensions II: T-expansion. Communications in Mathematical Physics , 396:527–622, 2022

2022
[40]

Y ang and J

F. Y ang and J. Yin. Random band matrices in the delocaliz ed phase, III: averaging ﬂuctuations. Proba- bility Theory and Related Fields , 179:451–540, 2021. 35

2021

Pith tools

Reviewed August 11, 2026 · model on record in the stance chip above.

[1] [1]

Aizenman and S

M. Aizenman and S. Molchanov. Localization at large diso rder and at extreme energies: an elementary derivation. Communications in Mathematical Physics , 157:245–278, 1993

work page 1993

[2] [2]

Aizenman and S

M. Aizenman and S. Warzel. Extended states in a lifshitz t ail regime for random Schr¨ odinger operators on trees. Physical Review Letters, 106:136804, 2011

work page 2011

[3] [3]

Aizenman and S

M. Aizenman and S. Warzel. Resonant delocalization for r andom Schr¨ odinger operators on tree graphs. Journal of the European Mathematical Society , 15:1167–1222, 2013

work page 2013

[4] [4]

P . W . Anderson. Absence of diffusion in certain random la ttices. Physical Review Letters , 109:1492– 1505, 1958

work page 1958

[5] [5]

Z. Bao, L. Erd˝ os, H.-T. Y au, and J. Yin. Delocalization f or a class of random block band matrices. Probability Theory and Related Fields , 167:673–776, 2017. 33

work page 2017

[6] [6]

Bogachev, S

L. Bogachev, S. Molchanov, and L. A. Pastur. On the densit y of states of random band matrices. Matematicheskie Zametki (English: Mathematical Notes) , 157 (English: 50):41–42 (English: 1232– 1242), 1991

work page 1991

[7] [7]

Bourgade

P . Bourgade. Random band matrices. Proceedings of the International Congress of Mathematicia ns, pages 2759–2783, 2018

work page 2018

[8] [8]

Bourgade, F

P . Bourgade, F. Y ang, H.-T. Y au, and J. Yin. Random band ma trices in the delocalized phase, II: Generalized resolvent estimates. Journal of Statistical Physics , 174:1189–1221, 2019

work page 2019