Same worst case, different risk: how transient response shapes failure probability

arXiv:2606.00789 · cond-mat.stat-mech, nlin.CD · Submitted 2026-05-30 · Read on arXiv

Listen

Radio episode about this paper

Transcript

Introduction to the show: ident: Condensed Matter. The day in condensed matter: materials, phases and the measurements under them.

Kai: I'm Kai.

Mira: And I'm Mira, here with Lev.

Kai: We're looking at the paper: "Same worst case, different risk".

Mira: Here's the gist of the paper. Pre-failure response spectra predict finite-amplitude fragility; this work introduces a single pre-failure quantity whose distribution predicts the dangerous volume by unifying extremal constructions as the leading edge of one fragility curve The Core Fragility Problem Failure theories often identify a single leading route to failure: the most unstable mode, weakest link, minimum-action escape path, or optimal perturbation Yet finite-amplitude susceptibility depends not only on the nearest route but on how much of perturbation space lies near dangerous directions This distinction is framed as a fragility problem where the failure distance is the smallest amplitude that crosses a prescribed boundary and the fragility curve is the fraction of directions that fail below a given amplitude The authors show that this curve is predicted before any failure occurs by the tail of a single pre-failure quantity: the boundary-normalized fragility gain computed from the linearized response The Predictive Mechanism The exact nonlinear target is defined as the failure-distance distribution relative to a prescribed boundary, where for a declared perturbation ensemble µ, i.e., a probability measure over normalized directions u, the susceptibility object is Pfail(B) = Pu∼µ

b∗(u) ≤ B: The authors introduce the pre-failure predictor by approximating the response using linearized response and defining the boundary-directed linear response as lt(u) = DΦx0(t)

Rtu: The fragility gain is then motivated by g(u) = max

lt(u): + m(t), which represents the fraction of the remaining failure margin consumed per unit amplitude in direction u To leading order, b∗(u) ≃ 1/g(u) Validation through Geometry and Data The mechanism is demonstrated in a high-dimensional nonlinear non-normal network where the system with broader response-channel breadth has a larger nonlinear fragility curve, isolating breadth from the worst direction In this validation, two systems with matched strongest directional gain Gmax,dir are compared based on channel-mode breadth, where system B2 is broadened and A2 is concentrated The direct high-dimensional validation shows that the broader system has a larger nonlinear fragility curve even though the strongest directional gain is matched Scalar Limit and Robustness Testing In deterministic traffic breakdown, the framework extends to a scalar response amplitude where the gain tail collapses to a response spectrum and a margin field The scalar dangerous-volume estimate is Bth ≃ MR−1∞ exp(−Sresp/2) The deterministic traffic implementation confirms that response-path breadth lowers calibrated jam thresholds once the strongest response is matched, with residual margins screening but never reversing the effect This provides an independent scalar test of the mechanism in a distinct physical system The results establish a distributional view of finite-amplitude failure where response-spectrum breadth measures how many identifiable response channels can contribute comparable boundary-directed amplification beyond the worst direction The framework extends to applied systems through the per-path margin field Mp and certified bands of Eq.

Kai: (6) Beyond the nearest route to failure lies a measurable volume of nearly dangerous directions, and its breadth is predicted—before any failure occurs—by the tail of the pre-failure response spectrum The traffic calculation provides an independent scalar test of the same mechanism in a distinct physical system The residual margin, not ambient jam statistics, accounts for the remainder No external funding was received for this research Data and Code Availability The complete reproducibility package for this manuscript, including source code, configuration files, saved numerical outputs, plotting routines, and manuscript sources is available in the Zenodo archive cited in Ref.

13: The archive is structured by analysis component rather than by manuscript section The high-dimensional validation module contains the saved A2/B2 gain samples, nonlinear threshold measurements, fragility curves, summary files, and plotting routine used to generate Fig.

Kai: 1 of the main text The traffic module contains the frozen selected arrangements, response-path spectra, threshold protocol table, residual decomposition, and the plots shown in Figs.

Paper summary: Kai: To recap what we've seen so far, this paper introduces a distributional view of finite amplitude failure where response spectrum breadth measures how many identifiable response channels can contribute comparable boundary directed amplification beyond the worst direction. They claim that this breadth is predicted before any failure occurs by the tail of the pre-failure response spectrum.

Mira: The thesis is that instead of focusing on one single leading route to failure, which some theories do, you need to consider how much perturbation space lies near those dangerous directions when considering finite amplitude susceptibility. They define the fragility curve based on this view: it's the fraction of directions that fail below a given amplitude.

Lev: From a real hardware standpoint, what’s interesting is how they motivate the fragility gain g of u by taking the maximum of that boundary directed linear response and adding in some running margin term m of t. They say this represents the fraction of remaining failure margin consumed per unit amplitude in direction u, and to leading order, b star of u is related to one over g(u).

Kai: So the key innovation isn't just the per-direction sensitivity itself, but how they normalize it by that running margin m of t—which converts a response slope into a directional failure distance—and then using the distribution of this gain g of u over the entire perturbation ensemble instead of just its highest value.

Mira: That means their central predictive object is equation four, which states that at an amplitude B, the dangerous volume is precisely the tail of that boundary normalized gain distribution. This moves us from finding a single threshold to predicting a whole volume based on this distribution.

Lev: The paper does show this mechanism in a high-dimensional nonlinear non normal network, comparing two systems A2 and B2 that have matched strongest directional gain G max dir of seven point five zero zero but different channel mode breadth Schmin values of one point three five four for the broader system B2 compared to the concentrated system A2.

Kai: That comparison is what really gets me; they show that even when you match that dominant gain, if you broaden the response channels, like in system B2, you get a larger nonlinear fragility curve because of that increased breadth.

Conclusion: Mira: So looking at the title "Same worst case, different risk: how transient response shapes failure probability", it really boils down to how the transient nature of the response dictates the risk profile when you move away from just looking at one specific failure path.

Lev: The implication for us is that we need to consider this breadth—this volume of nearly dangerous directions beyond the nearest route—because it’s predicted before any failure occurs by that tail of the pre-failure response spectrum.

Kai: It suggests that when we model complex systems, like traffic or circuits, we can't just focus on the single most unstable mode; we have to account for how many other nearby pathways contribute to the finite amplitude failure probability.

Mira: In simple terms, this work establishes that response spectrum breadth is a pre-failure coordinate for finite amplitude fragility beyond the strongest path. It means there’s a measurable volume of nearly dangerous directions that we can predict before anything actually breaks down.

Lev: And they do test this concept in a different physical system too, using deterministic traffic calculations to provide an independent scalar test of the same mechanism through their response spectrum breadth findings.

Surachate Limkumnerd

Department of Physics, Faculty of Science, Chulalongkorn University

cond-mat.stat-mech, nlin.CD

Submitted: 2026-05-30

Updated: 2026-11-02

Comments: 17 pages, 7 figures

License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/

Importance score: 91/100

The gist: Pre-failure response spectra predict finite-amplitude fragility; this work introduces a single pre-failure quantity whose distribution predicts the dangerous volume by unifying extremal constructions…

Key concepts

Fragility Problem
This problem distinguishes between finding the single most unstable route to failure and understanding how much of the entire perturbation space near dangerous directions contributes to susceptibility. The fragility curve is defined by the smallest amplitude needed to cross a boundary, and its shape is predicted by the tail of a single pre-failure quantity.
Pre-failure Predictor
This is the quantity that predicts failure before it happens. It uses linearized response approximations to define a 'fragility gain' g(u) for each direction u, representing how much failure margin is consumed per unit amplitude in that direction. The distribution of this gain across directions predicts the dangerous volume.
Response-Spectrum Breadth
This is the key new quantity. It measures the number of identifiable response channels that can contribute comparable boundary-directed amplification beyond the worst possible direction. This breadth emerges as a pre-failure coordinate for finite-amplitude fragility, indicating failure risk in directions beyond the single most unstable path.

Terminology

Summary

Pre-failure response spectra predict finite-amplitude fragility; this work introduces a single pre-failure quantity whose distribution predicts the dangerous volume by unifying extremal constructions as the leading edge of one fragility curve

The Core Fragility Problem

Failure theories often identify a single leading route to failure: the most unstable mode, weakest link, minimum-action escape path, or optimal perturbation Yet finite-amplitude susceptibility depends not only on the nearest route but on how much of perturbation space lies near dangerous directions This distinction is framed as a fragility problem where the failure distance is the smallest amplitude that crosses a prescribed boundary and the fragility curve is the fraction of directions that fail below a given amplitude The authors show that this curve is predicted before any failure occurs by the tail of a single pre-failure quantity: the boundary-normalized fragility gain computed from the linearized response

The Predictive Mechanism

The exact nonlinear target is defined as the failure-distance distribution relative to a prescribed boundary, where for a declared perturbation ensemble µ, i.e., a probability measure over normalized directions u, the susceptibility object is Pfail(B) = Pu∼µ [b∗(u) ≤ B] The authors introduce the pre-failure predictor by approximating the response using linearized response and defining the boundary-directed linear response as lt(u) = DΦx0(t)[Rtu] The fragility gain is then motivated by g(u) = max [lt(u)]+ m(t), which represents the fraction of the remaining failure margin consumed per unit amplitude in direction u To leading order, b∗(u) ≃ 1/g(u)

Validation through Geometry and Data

The mechanism is demonstrated in a high-dimensional nonlinear non-normal network where the system with broader response-channel breadth has a larger nonlinear fragility curve, isolating breadth from the worst direction In this validation, two systems with matched strongest directional gain Gmax,dir are compared based on channel-mode breadth, where system B2 is broadened and A2 is concentrated The direct high-dimensional validation shows that the broader system has a larger nonlinear fragility curve even though the strongest directional gain is matched

Scalar Limit and Robustness Testing

In deterministic traffic breakdown, the framework extends to a scalar response amplitude where the gain tail collapses to a response spectrum and a margin field The scalar dangerous-volume estimate is Bth ≃ MR−1∞ exp(−Sresp/2) The deterministic traffic implementation confirms that response-path breadth lowers calibrated jam thresholds once the strongest response is matched, with residual margins screening but never reversing the effect This provides an independent scalar test of the mechanism in a distinct physical system

Conclusion and Significance

The results establish a distributional view of finite-amplitude failure where response-spectrum breadth measures how many identifiable response channels can contribute comparable boundary-directed amplification beyond the worst direction The framework extends to applied systems through the per-path margin field Mp and certified bands of Eq. (6) Beyond the nearest route to failure lies a measurable volume of nearly dangerous directions, and its breadth is predicted—before any failure occurs—by the tail of the pre-failure response spectrum The traffic calculation provides an independent scalar test of the same mechanism in a distinct physical system The residual margin, not ambient jam statistics, accounts for the remainder No external funding was received for this research

Data and Code Availability

The complete reproducibility package for this manuscript, including source code, configuration files, saved numerical outputs, plotting routines, and manuscript sources is available in the Zenodo archive cited in Ref. [13] The archive is structured by analysis component rather than by manuscript section The high-dimensional validation module contains the saved A2/B2 gain samples, nonlinear threshold measurements, fragility curves, summary files, and plotting routine used to generate Fig. p. 5 p. 5 1 of the main text The traffic module contains the frozen selected arrangements, response-path spectra, threshold protocol table, residual decomposition, and the plots shown in Figs. p. 7 p. 7 S2 and S3 The deterministic traffic results use frozen selected arrangements and target sites Threshold measurements are performed only after selection, and the protocol-level statistics are computed from the saved threshold table Auxiliary exploratory inputs are retained for traceability but are not the authority for the reported scalar-limit traffic results The author thanks Prof. Dr. Pranut Potiyaraj for encouragement and for urging completion of this project

Supplemental Material Details

The Supplemental Material gives numerical and reproducibility details behind three empirical components of the main text: (i) a high-dimensional constructed validation geometry (Sec. S2), (ii) a minimal two-dimensional nonnormal check (Sec. p. 4 p. 4 S3), and (iii) a deterministic traffic scalar response-breadth test (Secs. p. 4 p. 4 S4 and S7) The three components play different roles The high-dimensional A2/B2 pair is the direct validation of the central prediction Plinfail(B) = P[g(u) > 1/B] against nonlinear thresholds, comparing a concentrated and a broadened gain geometry at matched Gmax,dir The two-dimensional non-normal system is now retained only as a transparent low-dimensional check of the same gain-tail predictor The traffic calculation is a scalar response-amplitude compression of the theory, not a full gain-tail validation The final part summarizes the Zenodo reproducibility archive and the organization of the deposited numerical evidence (Sec. p. 4 p. 4 S8) The high-dimensional validation module contains the saved A2/B2 gain samples, nonlinear threshold measurements, fragility curves, summary files, and plotting routine used to generate Fig. p. 5 1 of the main text The minimal two-dimensional module contains the corresponding low-dimensional gain and threshold data used for Fig. p. 6 S1 The traffic module contains the frozen selected arrangements, response-path spectra, threshold protocol table, residual decomposition, and the plots shown in Figs. p. 7 S2 and S3 The deterministic traffic results use frozen selected arrangements and target sites Threshold measurements are performed only after selection, and the protocol-level statistics are computed from the saved threshold table The provenance audit records which frozen evidence package supplies the manuscript traffic claims Auxiliary exploratory inputs are retained for traceability but are not the authority for the reported scalar-limit traffic results The complete reproducibility package for this manuscript, including source code, configuration files, saved numerical outputs, plotting routines, and manuscript sources is available in a public Zenodo reproducibility archive cited in Ref. p. 3 [13] The archive is structured by analysis component rather than by manuscript section

High-Dimensional Validation Geometry

The direct validation uses a state x ∈ R12 with weakly nonlinear dynamics x˙ = Ax + βx◦2, where x◦2 is the elementwise square and A is stable and non-normal The safe state is x0 = 0 and the failure diagnostic is Φ(x) = x1 with boundary Φc = 1 The boundary-directed linear response for a unit direction u is lt(u) = eT1exp(At)u, and the pre-failure fragility gain is g(u) = max t[lt(u)]+ Two constructed geometries are compared: a concentrated case A2 and a broadened case B2 They are matched in the dominant boundary-directed gain, Gmax,dir = 7.500, and their sampled gain maxima agree within 3. p. 4 p. 427% The broader system has much larger channel-mode breadth, Schmin = 1. p. 2 p. 2354 and ncheff = 4. p. 2 p. 277 compared to the concentrated system

Non-Normal Check Validation

The minimal two-dimensional non-normal validation uses a system x˙1 = −x1 + Kx2 + βx21, x˙2 = −λx2 with safe state x0 = (0, 0) and failure boundary x1 = Φc The linearized system at the safe state is x˙ = Ax, A = -1K0−λ The pre-failure fragility gain is therefore g(θ) = max 0≤t≤T[lt(θ)]+ Φc The nonlinear failure distance b∗(θ) is measured independently of Eq. (S4) by deterministic fixed-grid fourth-order Runge–Kutta time stepping over 0 ≤ t ≤ T The validation metrics reported are ρs(g,1/b∗) = 0. p. 69979, r(g,1/b∗) = 0. p. 69993, RMSE = 0. p. 60227

Traffic Model and Response Paths

The traffic calculation uses a deterministic heterogeneous car-following model on a periodic ring The response output is the absolute headway, normalized by the uniform headway scale A response path is one finite entry Gi,j,k ≠ 0 of this tensor The dominant response scale is Aresp = max pGp = log R∞ The collective response scale and response-path entropy are Fresp = 1/2logXp exp(2Gp), Sresp = 2(Fresp − Aresp) For the representative pair, the effective path count reported is neff = 1/P p π2p The selected low- and high-Sresp sets are matched in Aresp and separated in response-path breadth

Deterministic Threshold Protocols

The deterministic threshold calculation is performed only after the low- and high-Sresp arrangement sets and target vehicles have been identified The twelve calibrated protocols are localized velocity-headway perturbations of packet width wpkt ∈ 3, 4, 5, 7, 9, 11 The protocol-level high-minus-low shift is computed from the mean threshold in the high-Sresp group minus the mean threshold in the low-Sresp group The scalar constant-margin estimate used in the main text is Bth ≃ MR−1∞ e −Sresp/2 At fixed Aresp = log R∞, this gives ∆ log10 Bth = −∆Sresp/(2 ln 10) = −0.0748 The observed shifts realize 21%–42% of this ideal response-breadth value and remain negative for all twelve calibrated perturbation protocols The residual range is [0. p. 804364538, 0.05878246], so the protocol-dependent margin opposes part of the ideal responsebreadth lowering but does not reverse its sign

Traffic Statistics Robustness

The twelve paired high-minus-low protocol shifts in log10 Bth are - 0.02544963, -0.02691636, -0.02976646, - 0.02968214, - 0.02597187, - 0.01880431, - 0.02820472, - 0.02723169, - 0.03115869, - 0.03027241, - 0.02420196, and -

Final Summary

The mean of the twelve paired high-minus-low protocol shifts is −0.02614 All twelve shifts are negative The exact one-sided sign-test p-value is p = 2−12 = 2. p. 344 × 10−4, and the paired bootstrap 95% confidence interval for the mean shift is [−0. p. 302842, −0. p. 302345] The corresponding margin residuals are positive throughout, indicating partial cancellation of the response-breadth lowering The framework extends to applied systems through the per-path margin field Mp and certified bands of Eq. p. 3 (6) Beyond the nearest route to failure lies a measurable volume of nearly dangerous directions, and its breadth is predicted—before any failure occurs—by the tail of the pre-failure response spectrum The authors show that response-spectrum breadth emerges as a pre-failure coordinate for finite-amplitude fragility beyond the strongest path The residual margin, not ambient jam statistics, accounts for the remainder No external funding was received for this research The complete reproducibility package is available in a public Zenodo reproducibility archive cited in Ref. p. 3 [13]

--- Page 1 ---

Pre-failure response spectra predict finite-amplitude fragility; this work introduces a single pre-failure quantity whose distribution predicts the dangerous volume by unifying extremal constructions as the leading edge of one fragility curve

The results establish a distributional view of finite-amplitude failure where response-spectrum breadth measures how many identifiable response channels can contribute comparable boundary-directed amplification beyond the worst direction The framework extends to applied systems through the per-path margin field Mp and certified bands of Eq. (6) Beyond the nearest route to failure lies a measurable volume of nearly dangerous directions, and its breadth is predicted—before any failure occurs—by the tail of the pre-failure response spectrum The authors show that response-spectrum breadth emerges as a pre-failure coordinate for finite-amplitude fragility beyond the strongest path

The complete reproducibility package for this manuscript, including source code, configuration files, saved numerical outputs, plotting routines, and manuscript sources is available in a public Zenodo reproducibility archive cited in Ref. [13] The archive is structured by analysis component rather than by manuscript section The high-dimensional validation module contains the saved A2/B2 gain samples, nonlinear threshold measurements, fragility curves, summary files, and plotting routine used to generate Fig. 1 of the main text The traffic module contains the frozen selected arrangements, response-path spectra, threshold protocol table, residual decomposition, and the plots shown in Figs. S2 and S3 The deterministic traffic results use frozen selected arrangements and target sites Threshold measurements are performed only after selection, and the protocol-level statistics are computed from the saved threshold table Auxiliary exploratory inputs are retained for traceability but are not the authority for the reported scalar-limit traffic results The author thanks Prof. Dr. Pranut Potiyaraj for encouragement and for urging completion of this project

The Supplemental Material gives numerical and reproducibility details behind three empirical components of the main text: (i) a high-dimensional constructed validation geometry (Sec. S2), (ii) a minimal two-dimensional nonnormal check (Sec. S3), and (iii) a deterministic traffic scalar response-breadth test (Secs. S4 and S7) The three components play different roles The high-dimensional A2/B2 pair is the direct validation of the central prediction Plinfail(B) = P[g(u) > 1/B] against nonlinear thresholds, comparing a concentrated and a broadened gain geometry at matched Gmax,dir The two-dimensional non-normal system is now retained only as a transparent low-dimensional check of the same gain-tail predictor The traffic calculation is a scalar response-amplitude compression of the theory, not a full gain-tail validation The final part summarizes the Zenodo reproducibility archive and the organization of the deposited numerical evidence (Sec. S8)

The direct validation uses a state x ∈ R12 with weakly nonlinear dynamics x˙ = Ax + βx◦2, where x◦2 is the elementwise square and A is stable and non-normal The safe state is x0 = 0 and the failure diagnostic is Φ(x) = x1 with boundary Φc = 1 The boundary-directed linear response for a unit direction u is lt(u) = eT1exp(At)u, and the pre-failure fragility gain is g(u) = max t[lt(u)]+ Two constructed geometries are compared: a concentrated case A2 and a broadened case B2 They are matched in the dominant boundary-directed gain, Gmax,dir = 7.500, and their sampled gain maxima agree within 3.27% The broader system has much larger channel-mode breadth, Schmin = 1.354 and ncheff = 4.

Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.

Improvements for AI systems

  1. Predict finite-amplitude fragility using response spectrum breadth instead of just leading-path theory. The improved system can calculate a fragility curve by analyzing how many neardangerous pathways coexist beyond the strongest direction, allowing for a more accurate assessment of susceptibility that depends on the volume of nearly dangerous perturbations.

  2. Calculate the exact finite-amplitude fragility curve using the tail of a single pre-failure quantity. The system can use the boundary-normalized fragility gain computed from the linearized response to predict Pfail(B) rather than relying solely on measuring b∗(u) across many directions.

  3. Distinguish between threshold questions and fragility questions in instability theories. The improved AI can quantify the difference by using basin stability and survivability methods to measure volumes, thereby accurately determine if a system's behavior is controlled by a single route or a broader volume of dangerous directions.

  4. Implement an efficient pre-failure predictor based on margin consumption rather than full nonlinear trials. The system uses the boundary-directed linear response, namely the first variation of the failure diagnostic along Rtu and calculates the fragility gain g(u) = max 0≤t≤T [lt(u)]+ m(t) as a leading-order predictor where b∗(u) ≃ 1/g(u).

  5. Provide certified error envelopes for linear predictions. The system can calculate bounds using the amplitude-dependent gains g(B− (u)) and g(B+) to state that "P g(B− (u)) > 1/B ≤ Pfail(B) ≤ P g(B+) > 1/B," providing a formal error envelope for practical computation.

  6. Identify the relevant response spectrum breadth as a pre-failure coordinate for finite-amplitude fragility. The system can measure response-spectrum breadth by analyzing channel min-entropy and effective channel count to determine how concentrated or broad the dangerous response architecture is, which is crucial when the strongest direction is matched but fragility varies.

  7. Apply the framework to real systems like traffic models for scalar prediction. The system can use response-path spectrum and a scalar margin approximation to predict jam thresholds in deterministic environments, showing that response-path breadth lowers deterministic thresholds after selection and target sites are frozen.

Related papers