Functional Dismantling of Network Relaxation through Slow-Branch Susceptibility

Kaiming Luo , Huiying Zhou

Authors on Pith no claims yet

Pith reviewed 2026-05-07 17:56 UTC · model grok-4.3

classification ⚛️ physics.soc-ph nlin.AO

keywords relaxationnetworksdismantlingfunctionalrobustnessspectralsusceptibilityasymmetric

0 comments

The pith

A modal susceptibility score derived from the real part of the tracked slow Laplacian branch quantifies first-order changes in relaxation rate under node deletion in directed networks.

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

Networks often model how things like signals or flows relax back to a steady state after a disturbance. In networks with one-way connections, this relaxation involves complex modes that standard undirected methods miss. The authors focus on the slowest decaying non-stationary mode, tracked by the real part of a selected nonzero eigenvalue of the network Laplacian. They treat node removal as a change in the operator's dimension and derive a modal susceptibility score that predicts how much this removal alters the slow decay rate, using the left and right eigenvectors of the mode. In the special case of symmetric connections, the score reduces to a weighted version of the well-known Fiedler vector approach. Tests on both artificial graphs and real networks indicate that this score highlights different critical nodes than degree, betweenness, or other common attack strategies. The framework aims to resolve ambiguities in choosing which mode matters for robustness when the network is not reciprocal.

Core claim

Node deletion is then treated as a dimension-changing compression of the operator, leading to a modal susceptibility (MS) score that estimates the first-order reduction of the branch-tracked relaxation rate from the biorthogonal support of the slow mode.

Load-bearing premise

That a unique slow collective mode can be reliably identified and tracked across node deletions, and that the first-order perturbation via biorthogonal projection accurately captures the change in the real part of the relevant Laplacian eigenvalue without higher-order effects dominating.

Figures

Figures reproduced from arXiv: 2604.24279 by Huiying Zhou, Kaiming Luo.

**Figure 1.** Figure 1: FIG. 1. Spectral interpretation of the slow relaxation scale in a directed weighted ring. (a) Minimal directed relaxation model view at source ↗

**Figure 2.** Figure 2: FIG. 2. Illustration of the modal-susceptibility mechanism view at source ↗

**Figure 3.** Figure 3: FIG. 3. Branch-aware attack performance in directed networks. Top row: monotone remaining branch-robustness ratio view at source ↗

**Figure 4.** Figure 4: FIG. 4. Branch-aware attack performance in the symmetric weighted-undirected limit. Top row: monotone remaining branch view at source ↗

**Figure 5.** Figure 5: FIG. 5. Finite-size scaling of the AUC gap between MS and view at source ↗

**Figure 6.** Figure 6: FIG. 6. Validation of modal susceptibility (MS) in directed view at source ↗

read the original abstract

Robustness of relaxation on asymmetric networks is not determined by connectivity alone, because the slow collective mode can be complex and may change its spectral identity under adaptive damage. We introduce a slow-branch susceptibility framework for functional dismantling of network relaxation. Starting from the projected relaxation dynamics, we show that the relevant robustness observable is the real part of the selected nonzero Laplacian branch, which controls the long-time decay of the nonstationary sector. Node deletion is then treated as a dimension-changing compression of the operator, leading to a modal susceptibility (MS) score that estimates the first-order reduction of the branch-tracked relaxation rate from the biorthogonal support of the slow mode. In the reciprocal limit, the same construction reduces to the weighted Fiedler sector, placing directed and weighted-undirected networks within a common spectral-response formulation. Tests on synthetic and real-world networks show that MS identifies vulnerability patterns that differ from standard centrality-based attacks and edge-level spectral proxies. These results resolve a modal-selection ambiguity in non-Hermitian robustness analysis and provide a spectral basis for functional dismantling in asymmetric networks.

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.

Desk Editor's Note private letter to a colleague

The paper gives a biorthogonal modal susceptibility score to rank node deletions by their effect on relaxation rate in directed networks, but the first-order estimate for finite deletions needs direct checks against recomputed spectra.

read the letter

The paper's main new piece is a modal susceptibility score built from the left and right eigenvectors of the tracked slow branch of the non-Hermitian Laplacian. Node deletion is framed as a dimension-reducing compression, and the score estimates the resulting drop in the real part of that eigenvalue. In the reciprocal limit the construction collapses to a weighted Fiedler sector, which keeps the directed and undirected cases inside one spectral picture. The tests on synthetic graphs and real networks show that ranking nodes by this score produces attack patterns different from standard centrality or edge-spectral proxies, so it does pick up something about the dynamics rather than just connectivity. That is the useful part: it gives a concrete way to think about functional robustness when the slow mode is complex and can shift identity under damage. The soft spot is the first-order biorthogonal formula itself. Deletion is a finite change, not an infinitesimal perturbation, so the actual shift in the relaxation rate can include higher-order terms or a swap in which mode is slowest after the matrix shrinks. The abstract reports that the score differs from other measures on the test cases, but it does not show a side-by-side comparison of the predicted reduction versus the eigenvalue recomputed on the damaged network. Without that check it is hard to know how far the linear term dominates in practice. This work is aimed at people already using spectral tools on asymmetric networks, such as infrastructure models with one-way flows or ecological systems with directed interactions. A reader who wants a new observable for how damage slows collective decay would find something to try here. I would send it for peer review. The construction is original and the problem is real, even if the validation of the approximation needs to be tightened before the claims are fully convincing.

Axiom & Free-Parameter Ledger

0 free parameters · 3 axioms · 1 invented entities

The framework rests on standard spectral properties of (possibly non-symmetric) graph Laplacians and first-order perturbation theory for eigenvalue changes under rank-one or low-rank updates induced by node deletion. No free parameters are explicitly introduced in the abstract description.

axioms (3)

domain assumption Relaxation dynamics on the network are governed by the (possibly non-Hermitian) Laplacian operator.
Standard modeling assumption for linear diffusion or consensus processes on graphs.
domain assumption The long-time decay of the nonstationary sector is controlled by the real part of a selected nonzero eigenvalue.
Follows from spectral theory of linear operators; the paper selects the slow branch as the relevant one.
standard math Biorthogonal left and right eigenvectors exist and can be used to compute first-order perturbations.
Standard result from linear algebra for non-symmetric matrices.

invented entities (1)

Modal susceptibility (MS) score no independent evidence
purpose: Quantifies the first-order impact of node deletion on the tracked slow relaxation rate.
Newly defined quantity constructed from the biorthogonal support of the slow mode.

pith-pipeline@v0.9.0 · 5487 in / 1696 out tokens · 115773 ms · 2026-05-07T17:56:40.170787+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

31 extracted references

[1]

M. E. J. Newman,Networks: An Introduction(Oxford University Press, Oxford, 2010)

2010
[2]

Albert and A.-L

R. Albert and A.-L. Barab´ asi, Rev. Mod. Phys.74, 47 (2002)

2002
[3]

Arenas, A

A. Arenas, A. D´ ıaz-Guilera, J. Kurths, Y. Moreno, and C. Zhou, Phys. Rep.469, 93 (2008)

2008
[4]

J. G. Restrepo, E. Ott, and B. R. Hunt, Phys. Rev. E 71, 036151 (2005)

2005
[5]

Chung, Annals of Combinatorics9, 1 (2005)

F. Chung, Annals of Combinatorics9, 1 (2005)

2005
[6]

Battiston, E

F. Battiston, E. Amico, A. Barrat, G. Bianconi, G. Fer- raz de Arruda, B. Franceschiello, I. Iacopini, S. K´ efi, V. Latora, Y. Moreno, M. M. Murray, T. P. Peixoto, F. Vaccarino, and G. Petri, Nature Physics17, 1093 (2021)

2021
[7]

Gallo, R

L. Gallo, R. Muolo, L. V. Gambuzza, V. Latora, M. Frasca, and T. Carletti, Communications Physics5, 263 (2022)

2022
[8]

Zhang, M

Y. Zhang, M. Lucas, and F. Battiston, Nature Commu- nications14, 1605 (2023)

2023
[9]

Nazerian, J

A. Nazerian, J. D. Hart, M. Lodi, and F. Sorrentino, Nature Communications15, 9003 (2024)

2024
[10]

Forrow, F

A. Forrow, F. G. Woodhouse, and J. Dunkel, Physical Review X8, 041043 (2018)

2018
[11]

J. C. Bronski and L. DeVille, SIAM Journal on Applied Mathematics74, 83 (2014)

2014
[12]

E. M. Shahrivar, M. Pirani, and S. Sundaram, Automat- ica83, 234 (2017)

2017
[13]

Grassia, M

M. Grassia, M. De Domenico, and G. Mangioni, Nature Communications12, 5190 (2021)

2021
[14]

Artime and M

O. Artime and M. De Domenico, Nature Communica- tions12, 2478 (2021)

2021
[15]

Fiedler, Czechoslovak Mathematical Journal23, 298 (1973)

M. Fiedler, Czechoslovak Mathematical Journal23, 298 (1973)

1973
[16]

Merris, Linear Algebra Appl.197–198, 143 (1994)

R. Merris, Linear Algebra Appl.197–198, 143 (1994)

1994
[17]

Mohar, Graph Theory, Combinatorics, and Applica- tions2, 871 (1991)

B. Mohar, Graph Theory, Combinatorics, and Applica- tions2, 871 (1991)

1991
[18]

Albert, H

R. Albert, H. Jeong, and A.-L. Barab´ asi, Nature406, 378 (2000)

2000
[19]

Cohen, K

R. Cohen, K. Erez, D. ben Avraham, and S. Havlin, Phys. Rev. Lett.86, 3682 (2001)

2001
[20]

Holme, B

P. Holme, B. J. Kim, C. N. Yoon, and S. K. Han, Phys. Rev. E65, 056109 (2002)

2002
[21]

Crucitti, V

P. Crucitti, V. Latora, and M. Marchiori, Phys. Rev. E 69, 045104 (2004)

2004
[22]

C. M. Schneider, A. A. Moreira, J. S. Andrade, S. Havlin, and H. J. Herrmann, Proceedings of the National Academy of Sciences108, 3838 (2011)

2011
[23]

J. Gao, S. V. Buldyrev, H. E. Stanley, and S. Havlin, Nature Physics8, 40 (2012)

2012
[24]

S. V. e. a. Buldyrev, Nature464, 1025 (2010)

2010
[25]

Luo, Chaos36, 023137 (2026)

K. Luo, Chaos36, 023137 (2026)

2026
[26]

Jiang, J.-a

S. Jiang, J.-a. Lu, J. Zhou, and Q. Dai, Phys. Rev. E 109, 054301 (2024)

2024
[27]

J. Zhou, Y. Zhang, J.-a. Lu, and G. Chen, IEEE Trans- actions on Systems, Man, and Cybernetics: Systems54, 2757 (2024)

2024
[28]

A. Hu, J. Zhou, S. Jiang, J. Hu, and J.-a. Lu, Inter- national Journal of Bifurcation and Chaos35, 2550076 (2025)

2025
[29]

Jiang, J

S. Jiang, J. Zhou, M. Small, J.-a. Lu, and Y. Zhang, Phys. Rev. Lett.130, 187402 (2023)

2023
[30]

X. Wen, J. Zhou, S. Jiang, Q. Zhou, and J.-a. Lu, Chaos, Solitons & Fractals201, 117389 (2025)

2025
[31]

R. A. Rossi and N. K. Ahmed, inAAAI(2015)

2015