Universal non-Hermitian transport in disordered systems

Bo Li; Chuan Chen; Zhong Wang

REVIEW 2 major objections 4 minor 6 cited by

Universal non-Hermitian transport in disordered systems

T0 review · 2 major / 4 minor · reviewed 2026-08-12 · deepseek-v4-flash

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

Pith's one-line read Even when disorder localizes every eigenstate, a wave packet still spreads; the spreading distance follows universal scaling set by the imaginary spectrum's tail.

desk verdict New and largely convincing scaling theory for non-Hermitian disorder transport, but the weak-disorder Gaussian-tail universality is an intermediate regime, not the true long-time asymptote, for bounded disorder. read the letter →

arxiv 2411.19905 v2 pith:EYEO32LU submitted 2024-11-29 quant-ph cond-mat.dis-nncond-mat.stat-mechphysics.optics

Bo Li , Chuan Chen , Zhong Wang This is my paper · ORCID

classification quant-phcond-mat.dis-nncond-mat.stat-mechphysics.optics MSC 82B4481Q12

keywords non-HermitianAndersonlocalizationimaginarydisorderdynamicaldelocalizationimaginary-partdensityofstatesuniversalscalingsub-ballistictransportsubdiffusionLindbladmasterequation

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

Disorder that makes every eigenstate of a lattice exponentially localized normally freezes wave propagation; the paper's point is that this logic fails when the disorder is non-Hermitian, i.e. when the on-site potential is a random imaginary gain or loss. In that setting a wave packet can still spread, because localized eigenstates with larger imaginary energies amplify in time and come to dominate the wavefront. The paper derives a universal law for the spreading distance: it depends only on the tail of the imaginary-part density of states and on spatial dimension, not on microscopic details. Concretely, Gaussian imaginary-part density of states gives $|x_c|\sim t/(\ln t)^{1/2}$, a uniform one gives $|x_c|\sim t^{1/(d+1)}$, and a linear tail gives $|x_c|\sim t^{2/(d+2)}$. These results explain earlier experiments and predict new slower-than-ballistic and subdiffusive behaviors that have no Hermitian counterpart.

What carries the argument

The load-bearing object is the imaginary-part density of states, $\rho(\lambda)=\int d\varepsilon\,\rho(\varepsilon+i\lambda)$, the density of the imaginary parts of the complex eigenvalues; its tail, not its bulk, sets the late-time exponents. The optimization principle is the weight-factor competition $W=-|x|/\xi+\lambda_x t$ combined with the extreme-value condition $\int_{\lambda_{\max}}^\infty \rho\,d\lambda\sim |x|^{-d}$. For the weak-disorder limit the central limit theorem supplies a Gaussian $\rho$; for strong bounded disorder a short perturbative argument supplies a linear tail. A separate renormalization-group calculation, based on a Cole-Hopf transformation to a nonlinear equation, determines the $d=3$ delocalized-phase exponent $3/5$.

What would settle it

Measure or compute exactly the density of imaginary eigenvalues near the top of the spectrum for a 1D lattice with $V_x\in[-W,W]$ and $W\gg t_0$; if the edge density scales as $\Delta^p$ with $p\neq 1$, where $\Delta=W-\lambda$, then the predicted late-time exponent $t^{2/(d+2)}$ is wrong. A direct dynamical test is the predicted early-to-late crossover $|x_c|\sim t^{1/2}\to t^{2/3}$ in that same model, which finite-size-controlled simulations or an optical-lattice experiment could confirm or reject.

Watch

Extended reading notes

Core claim

The central claim is that in the eigenstate-localized regime of a non-Hermitian disordered lattice, the average spreading distance is controlled by a simple optimization. Writing the weight of a localized eigenstate centered at $x$ as $W(x,t)=-|x|/\xi+\lambda_x t$, where $\xi$ is the localization length and $\lambda_x=\mathrm{Im}\,E_x$, the wavefront is dominated by the eigenstate with the largest such weight. Extreme-value statistics over the $\sim |x|^d$ available centers gives $\int_{\lambda_{\max}(|x|)}^\infty \rho(\lambda)d\lambda \sim |x|^{-d}$, which ties the maximal imaginary part to distance through the tail of the imaginary-part density of states $\rho(\lambda)$. Solving $\partial W/\partial |x|=0$ then yields the universal scalings: $t/(\ln t)^{1/2}$ for Gaussian $\rho$, $t^{1/(d+1)}$ for uniform $\rho$, and $t^{2/(d+2)}$ for a linear tail, with late-time dynamics set by the tail. The paper validates the 1D and 2D scalings numerically, shows the weak-disorder limit is universally Gaussian by the central limit theorem, and maps Lindblad open systems with random losses onto the same problem.

Load-bearing premise

The argument assumes the imaginary-part density of states has exactly the tail shapes it quotes: Gaussian all the way out for weak disorder and linear at the band edge for strong bounded disorder, so a different tail shape would change every quoted exponent.

Editorial extensions

If this is right

Even with every eigenstate exponentially localized, a non-Hermitian disordered lattice does not stop transport: the wave packet spreads indefinitely, with a distance that grows as a power of time or as $t/(\ln t)^{1/2}$.
The spreading exponent is fixed by the tail of the imaginary-part density of states and by the spatial dimension, so different disorder distributions give different universality classes of transport.
In the strong-disorder limit with a uniform bounded potential, 1D spreading should cross over from $t^{1/2}$ to $t^{2/3}$, and 2D from $t^{1/3}$ to $t^{1/2}$, as the linear edge tail takes over.
The same dynamics appears in Lindblad open systems with random local losses, because the two-point correlation matrix obeys the same non-Hermitian evolution.
In $d=3$, weak disorder flows to a renormalization-group fixed point giving $|x|\sim t^{3/5}$, an exponent independent of the disorder statistics.

Reading between the lines

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

Editorial extension: because the spreading exponent encodes the tail shape, time-resolved measurement of $|x_c|(t)$ could serve as a practical probe of the imaginary-part density of states near the spectral edge, without resolving individual eigenvalues.
Editorial extension: the predicted crossover from the uniform bulk scaling to the linear-tail scaling, for instance $t^{1/2}\to t^{2/3}$ in 1D, should occur at a time that grows with system size; locating this crossover numerically could discriminate the linear-tail mechanism from finite-size artifacts.
Editorial extension: the RG fixed point at $d=3$ suggests the same setup may host a disorder-driven delocalization transition; tuning disorder strength across it and measuring the $t^{3/5}$ exponent would test a prediction that lies outside the eigenstate-localized regime covered by the main optimization argument.

Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

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

Colleague —

Read this one. It is the first quantitative theory I know that turns the 'jumpy' non-Hermitian transport seen in photonic experiments into concrete scaling exponents. The extreme-value optimization is simple, and the resulting Table I — Gaussian ImDOS gives t/(ln t)^{1/2}, uniform gives t^{1/(d+1)}, linear gives t^{2/(d+2)} — is genuinely new. The strong-disorder 1D and 2D numerics support the uniform/linear predictions; the crossover from t^{1/2} to t^{2/3} in 1D is visible and the size-effect discussion is careful. The RG section is speculative but clearly marked, and the Lindblad mapping is a useful bridge to experiments. The ImDOS is measured in the same systems, but since the exponents are not fitted back to dynamics, I do not see a damaging circularity.

The soft spot is the weak-disorder Gaussian tail. For bounded V in [-W,W], every eigenvalue has Im E in [-W,W]; the ImDOS has compact support. The Gaussian form from the central limit theorem describes the bulk of the distribution, not the extreme tail that the optimization actually selects. An eigenstate averaging ξ sites has a power-law tail (W - λ)^{ξ-1} near W, and using that exact tail in Eqs. (2)-(3) gives |x| ~ t^{1/(1+d/ξ)}, not t/(ln t)^{1/2}. The Gaussian scaling survives only for ln|x| << ξ, i.e. as an intermediate regime. The SM admits the CLT argument fails near the band edge, but the compact-support cutoff is sharper and applies to band-center states too. This does not kill the strong-disorder results — the linear tail there is independently derived from rare-event perturbation theory — but it does falsify the paper's 'universal weak-disorder limit' claim as a strict t→∞ statement.

Minor points: no code, no error bars, and the 2D late-time linear-tail prediction is not really checked.

Who should read it: anyone working on non-Hermitian disordered lattices or disorder in dissipative photonics. It deserves a serious referee, but the referee should push for a clear distinction between intermediate and true asymptotic scaling, and for a numerical test at large ξ with bounded disorder.

Recommendation: send to peer review with major revision.

Referee Report

2 major / 4 minor

Summary. The paper develops an extreme-value optimization theory for wave-packet spreading in non-Hermitian disordered lattices with purely imaginary disorder. It argues that although eigenstates are exponentially localized, the imaginary parts of their energies provide temporal amplification, and the spreading distance is controlled by the eigenstate with the largest growth factor within a spatial volume. The resulting scaling laws are determined by the tail of the imaginary-part density of states (ImDOS): Gaussian ImDOS gives |x_c| ~ t/(ln t)^{1/2}, uniform ImDOS gives |x_c| ~ t^{1/(d+1)}, and linear ImDOS gives |x_c| ~ t^{2/(d+2)}. The theory is supported by 1D and 2D numerical simulations, and a dynamical renormalization group analysis is added to predict |x| ~ t^{3/5} for the 3D delocalized phase. A mapping to Lindblad master equations shows that the same scaling applies to open systems with random losses.

Significance. If correct, this work would establish a novel and potentially universal class of dynamical scaling in non-Hermitian disordered systems, connecting the tails of the non-Hermitian density of states to transport exponents and explaining experimental observations of dynamical delocalization despite spectral localization. The optimization framework is physically transparent and the numerical support for the uniform and linear ImDOS cases (Figs. 2 and 3) is reasonably convincing. The paper also makes a falsifiable prediction for 3D weak-disorder dynamics. However, the central claim that the Gaussian ImDOS tail dictates the long-time limit is not the true asymptotic for bounded disorder, which is a load-bearing issue that requires revision.

major comments (2)

[Universal scaling under weak disorder; SM Sec. II B] The claim that the Gaussian ImDOS tail determines the long-time scaling is undermined by the compact support of bounded disorder. For V_x in [-W,W], every eigenvalue satisfies Im(E) in [-W,W], so the ImDOS is exactly zero beyond W. The Gaussian form in SM Eq. (19) is a central-limit statement about typical eigenstates, but Eq. (2) of the main text requires the extreme-value tail near the upper edge. For an eigenstate effectively averaging xi bounded variables, P(Im(E) > W - Delta) behaves as Delta^xi, which gives lambda_max ~ W - C r^{-d/xi} and hence |x_c| ~ t^{1/(1+d/xi)} in the strict t -> infinity limit. The t/(ln t)^{1/2} law can hold only in the intermediate regime ln t << xi, not as the true asymptotic. The paper should state this crossover explicitly and either reframe the Gaussian result as an intermediate asymptotic or justify why the bounded-support tail is irrelevant in the thermodynamic limit.
[Renormalization group analysis; Eq. (46)] The one-loop dynamical RG flow of Eq. (46) is used to predict a 3D delocalization transition and the scaling |x| ~ t^{3/5}. This claim is not supported by numerical simulation or an independent non-perturbative argument, and the derivation follows the Hermitian KPZ/directed-polymer literature without addressing the complex-valued nature of the field Phi in the non-Hermitian problem. The fixed point g*_2 = 1/(2K_d) at d=3 is presented as an exact statement, but the one-loop beta function may not be reliable for the original lattice model with purely imaginary disorder. This result should be explicitly labeled as a speculative extension, or accompanied by numerical validation.

minor comments (4)

[Throughout] The notation is difficult to parse because of OCR artifacts (e.g., "ïxð" for the delta function, "À" for the localization length, "Ã" for the standard deviation). A careful proofreading and a notation table would improve readability.
[Fig. 2(a)] The crossover from t^{1/2} to t^{2/3} is inferred from overlapping curves for different system sizes, but no quantitative comparison with the predicted crossover scale is provided. A finite-size scaling analysis of the crossover time would strengthen the claim.
[SM Sec. II A] The derivation of the linear tail in the strong-disorder limit is heuristic and relies on perturbative arguments; the statement following Eq. (13) that the coefficient in front of Delta is not explicitly given makes the derivation incomplete. The numerical agreement with the predicted linear-tail scaling, however, partially compensates for this.
[Liouvillian dynamics] The mapping to the Lindblad master equation is clearly presented, but the claim that the same scaling applies to purely dissipative systems (SM Sec. IV B) is stated without a derivation of the effective Hamiltonian H and its ImDOS; adding a brief derivation would make the mapping more transparent.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the transport scaling is derived from the independently specified ImDOS tail, not fitted to the transport observables.

full rationale

The derivation chain is: Hamiltonian (Eq. 1) to eigenstate localization and complex eigenvalues, then the imaginary-part density of states (ImDOS), then the extreme-value estimate of the maximal growth rate via Eq. (2), then optimization of the weight factor via Eq. (3), and finally the space-time exponents. Each link connects a distinct physical quantity: the ImDOS is a spectral input, while the spreading distance is a dynamical output. The numerical comparisons in Figs. 2-3 test the resulting exponents against independently computed wave-packet dynamics; the ImDOS histograms are not tuned to reproduce the dynamics, and no transport-fitting parameter is renamed as a prediction. The paper's own caveats, including the breakdown of the central-limit argument near the band edge and the bounded-support cutoff of the Gaussian tail for weak bounded disorder, are limitations on the validity of the assumed tail shape and therefore affect correctness of the asymptotic claim, but they do not close a logical loop: a different tail input would give different exponents, which is exactly the advertised input-output structure. The linear-tail and Gaussian-tail derivations in the Supplemental Material are perturbative arguments, not assumptions imported from the authors' prior work, and the RG calculation follows standard methods of Medina et al. via an explicit flow equation. The Liouvillian section is a derivation mapping the master equation to the same non-Hermitian Hamiltonian problem. References to [51] are the paper's own supplementary calculations, which provide detail rather than external self-citation of an unverified premise. No step satisfies the required criterion of exhibiting a specific reduction of the prediction to its own input, so the analysis is self-contained rather than circular.

Assumptions & free parameters 0 free parameters · 4 assumptions · 0 invented entities

The central scaling predictions are parameter-free given the ImDOS form, but they inherit the paper's heuristic assumptions about the ImDOS tail and the single-eigenstate dominance. The 3D RG prediction adds a one-loop truncation assumption.

assumptions (4)

standard math Extreme value statistics: the largest growth factor λ_max within a volume of n=|x|^d sites satisfies P(λ>λ_max) ~ 1/n.
Used to derive Eq. (2) in the main text and SM Eq. (1); follows from Gumbel order statistics.
domain assumption Eigenstates of H with imaginary disorder are exponentially localized with a well-defined localization length ξ, so that the weight factor W = -|x|/ξ + λ_x t and the dominance of a single eigenstate hold.
Invoked in the main text before Eq. (2); supported by prior numerical and theoretical work (Refs. [33,42]) but not re-derived here.
domain assumption The tail of the imaginary-part density of states takes the assumed forms: Gaussian (weak disorder, by central limit theorem) and linear near the spectral edge (strong bounded disorder).
Load-bearing for Table I; derived heuristically in SM Sec. II A and II B, with the paper itself noting the CLT argument fails near band edges.
ad hoc to paper The one-loop dynamical RG flow (Eq. 46) determines the exact long-time scaling in 3D, giving |x| ~ t^{3/5}.
The paper presents z=5/3 at the d=3 fixed point without higher-loop or non-perturbative verification, and no numerical test is provided.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Universal non-Hermitian transport in disordered systems." pith.science (2026). https://pith.science/paper/EYEO32LU

@misc{pith2026241119905,
  author       = {Pith},
  title        = {Pith review of: Universal non-Hermitian transport in disordered systems},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/EYEO32LU}},
  note         = {Machine review of arXiv:2411.19905}
}

read the original abstract

In disordered Hermitian systems, localization of energy eigenstates prohibits wave propagation. In non-Hermitian systems, however, wave propagation is possible even when the eigenstates of Hamiltonian are exponentially localized by disorders. We find in this regime that non-Hermitian wave propagation exhibits novel universal scaling behaviors without Hermitian counterpart. Furthermore, our theory demonstrates how the tail of imaginary-part density of states dictates wave propagation in the long-time limit. Specifically, for the three typical classes, namely the Gaussian, the uniform, and the linear imaginary-part density of states, we obtain logarithmically suppressed sub-ballistic transport, and two types of subdiffusion with exponents that depend only on spatial dimensions, respectively. Our work highlights the fundamental differences between Hermitian and non-Hermitian Anderson localization, and uncovers unique universality in non-Hermitian wave propagation.

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 6 Pith papers

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

Lyapunov formulation of band theory for disordered non-Hermitian systems
cond-mat.dis-nn 2025-07 conditional novelty 7.0 of 10

A Lyapunov-exponent formulation gives exact spectral densities and a topological skin-Anderson transition criterion for disordered non-Hermitian 1D lattices.
Non-Hermitian delocalization in 1D via emergent compactness
cond-mat.dis-nn 2024-12 conditional novelty 7.0 of 10

Balanced pairs of imaginary on-site potentials can be arranged so the transfer matrix becomes SU(2)-like, producing real-energy delocalized states with an exact mobility edge.
Lifshitz-like Metastability and Optimal Dephasing in Dissipative Bosonic Lattices
quant-ph 2025-08 unverdicted novelty 6.0 of 10

In coupled bosonic lattices with non-uniform loss, an optimal intermediate dephasing rate speeds up equilibration, while stronger dephasing slows it by protecting quasi-dark modes.
Anisotropic Anderson localization in higher-dimensional nonreciprocal lattices
cond-mat.dis-nn 2025-07 conditional novelty 6.0 of 10

A 2D nonreciprocal Hatano-Nelson model hosts hybrid eigenstates with skin localization along one axis and Anderson localization along the other, yielding an ALM-HM-ALM reentrant transition.
Dephasing-induced jumps in non-Hermitian disordered lattices
cond-mat.dis-nn 2024-12 conditional novelty 6.0 of 10

Rapid dephasing localizes the eigenmodes of the incoherent propagator in weakly disordered non-Hermitian lattices, causing abrupt jumps between distant sites.
Universal Spreading Dynamics in Quasiperiodic Non-Hermitian Systems
cond-mat.dis-nn 2024-12 conditional novelty 6.0 of 10

In a quasiperiodic non-Hermitian lattice, wave packets spread diffusively with exponent 1/2 in the delocalized phase and subdiffusively with exponent 1/3 in the localized phase, in contrast to ballistic and halted Her...

Pith tools

Reviewed August 12, 2026 · model on record in the stance chip above.