Fragility Edge: Thresholds Across Disciplines

• Fragility Edge is a threshold concept marking a system’s abrupt shift from routine perturbation handling to highly sensitive responses, defined by domain-specific metrics.
• It appears in seismic engineering, glass physics, topological media, finance, and operational models, using tools like lognormal curves, Monte Carlo simulations, and spectral analysis.
• Research highlights that accurately identifying the Fragility Edge often requires non-parametric or full-distribution approaches to capture its multifaceted nature.

Searching arXiv for papers using the term and related topics to ground the synthesis. arXiv search query: "Fragility Edge OR fragility edge" Across several contemporary literatures, “Fragility Edge” denotes a boundary at which a system becomes acutely sensitive to perturbation, but the object called the edge is domain-specific. In seismic engineering it is the knee of a fragility curve governed by the median intensity $\theta$ and dispersion $\beta$; in glass physics it marks crossovers between fragile, strong, and superstrong dynamics; in topological media it refers either to the instability of edge transport under disorder or to defect-based probes of fragile topology; in finance and network science it appears as a spectral, geometric, or combinatorial threshold; and in operational models it is a true critical point at which small disturbances trigger cascades (Mai et al., 2017, Ciarella et al., 2019, Mannaï et al., 2023, Bhandari, 4 Jul 2026, Moran et al., 2023, Limkumnerd, 30 May 2026).

1. Domain scope and semantic structure

The literature suggests that “Fragility Edge” is not a single standardized construct but a family of threshold concepts. In each case, the term links a control parameter to a sharp change in susceptibility: earthquake intensity in structural reliability, density or temperature in glass formation, disorder and topology in edge transport, dominant-factor concentration in markets, or schedule buffer in socio-technical systems. The edge may therefore be geometric, spectral, statistical, or literal, depending on the ontology of the system under study (Mai et al., 2017, Ciarella et al., 2019, Mannaï et al., 2023, Bhandari, 4 Jul 2026, Moran et al., 2023, Wu et al., 2024).

Domain	Edge variable or object	Interpretation
Seismic fragility	$\theta$, $\beta$, or the curve knee	Failure-transition location and steepness
Glass-forming matter	$\rho_c$, $\theta_A=T_A/T_g$, fragile–strong crossover	Boundary between dynamical regimes
Topological media	Edge channels, impurity bands, screw dislocations	Stability or probing of boundary/defect modes
Finance and networks	$\beta r(B)=1$, edge curvature, critical edge fraction	Threshold for bubbles or structural disintegration
Operational dynamics	$B_c$, $x_{\mathrm{crit}}$, or $a_{\mathrm{edge}}$	Critical buffer, precipice, or onset amplitude

A recurring theme is that the edge is rarely defined solely by the single “worst” direction. Several papers instead emphasize distributions, tails, or ensembles: non-parametric seismic curves rather than assumed lognormality, the full conditional distribution of an engineering demand parameter in PBEE, the breadth of pre-failure gain spectra, or the aggregate geometry of network edges rather than a single weakest link (Mai et al., 2017, Zhu et al., 2022, Limkumnerd, 30 May 2026, Samal et al., 2021).

2. Seismic fragility as a failure-transition edge

In seismic reliability, the fragility edge is the transition in the conditional failure probability

$\beta$0

where $\beta$1 is the maximum inter-storey drift ratio, $\beta$2 a prescribed drift threshold, and $\beta$3 an intensity measure such as PGA or $\beta$4. The classical lognormal representation,

$\beta$5

makes the edge explicit: $\beta$6 is the median intensity at $\beta$7 failure probability and $\beta$8 controls steepness, with smaller $\beta$9 corresponding to a steeper rise. The 2017 study compares this parametric form with two non-parametric procedures—binned Monte Carlo simulation and kernel density estimation—and shows that the accuracy of lognormal curves depends on the intensity measure, the failure criterion, and especially the parameter-estimation method (Mai et al., 2017).

That comparison is technically important because it separates two different sources of error: model form and estimator bias. In the three-storey steel frame analyzed with $\theta$0 synthetic motions, bMCS and KDE overlay closely across intensity measures and thresholds, including the knee region, whereas linear-regression cloud analysis can misplace $\theta$1 and distort $\theta$2 when residuals are non-normal or heteroscedastic. Structure-specific intensity measures such as $\theta$3 or $\theta$4 tend to correlate more strongly with drift than PGA, yielding steeper and better-localized edges at low-to-moderate drift limits. Bootstrap bands widen with increasing drift threshold and intensity, indicating growing epistemic uncertainty around the edge even when sample size is large (Mai et al., 2017).

Within PBEE, the 2022 SPCE framework generalizes the same problem from a single fragility curve to the full conditional distribution of the engineering demand parameter under a stochastic ground motion model. The method represents $\theta$5 with a stochastic polynomial chaos expansion over the SGMM parameters and a latent Gaussian variable, fits the coefficients by maximum likelihood with quadrature over the latent variable, and then computes fragility as an exceedance probability of the surrogate conditional distribution. The reported advantage is that SPCE estimates both the conditional PDF and the fragility functions more accurately than linear, kernel, or probit competitors in the case studies, particularly for multi-IM fragility surfaces and higher thresholds (Zhu et al., 2022). A common misconception in this literature is that fragility analysis is exhausted by fitting a lognormal CDF; these papers instead treat the edge as an empirical transition whose accurate localization may require non-parametric or distributional surrogates (Mai et al., 2017, Zhu et al., 2022).

3. Fragility crossovers in glass-forming matter

In glass physics, fragility describes how abruptly a material’s relaxation time or viscosity changes as the glass transition is approached. The standard metric is the Angell fragility index

$\theta$6

with low $\theta$7 indicating strong behavior and high $\theta$8 indicating fragile behavior. The vitrimer study identifies a density-controlled fragility edge: high-density systems with $\theta$9 are fragile, $\beta$0 is Arrhenius-like strong, and $\beta$1 is superstrong, with $\beta$2 essentially independent of density in the superstrong regime. The crossover density is reported as $\beta$3, and the mechanism is traced to the temperature sensitivity of the static structure factor $\beta$4, especially at its main peak $\beta$5; high-density vitrimers show pronounced growth of $\beta$6 on cooling, whereas low-density superstrong vitrimers show anomalously weak growth (Ciarella et al., 2019).

A closely related but temperature-based version appears in the Arrhenius crossover literature, where the edge is identified with $\beta$7, the reduced temperature at which transport departs from high-temperature Arrhenius behavior and becomes super-Arrhenius. That crossover occurs near $\beta$8 in metallic glass-formers, near $\beta$9 in fragile molecular liquids, and above $\rho_c$0 in strong network liquids. The same work reports an inverse correlation between $\rho_c$1 and kinetic fragility and notes that the high-temperature activation barrier is approximately $\rho_c$2 for metallic and molecular liquids, independent of fragility and $\rho_c$3 (Jaiswal et al., 2016). In the review of bulk metallic glasses, the corresponding transition is expressed as a fragile–strong crossover in VFT parameters: Zr-based alloys often show strong behavior near $\rho_c$4 with $\rho_c$5–$\rho_c$6, but higher-temperature melt data can be fit by a more fragile branch with $\rho_c$7–$\rho_c$8, as in Vitreloy 1 (Busch et al., 2014).

Polymer nanocomposites supply a further operationalization. In the nanoparticle-doped polymer melt study, the fragility edge is not a universal threshold in density or temperature but a tunable boundary in the temperature dependence of cooperative motion. Attractive nanoparticle–polymer interactions increase $\rho_c$9, increase $\theta_A=T_A/T_g$0, increase fragility, and increase average string length $\theta_A=T_A/T_g$1; non-attractive interactions produce the opposite effect. Within the Adam–Gibbs interpretation used there,

$\theta_A=T_A/T_g$2

and the derivative term $\theta_A=T_A/T_g$3 is reported to be $\theta_A=T_A/T_g$4–$\theta_A=T_A/T_g$5 times larger than $\theta_A=T_A/T_g$6, implying that fragility is governed primarily by the temperature dependence of cooperativity rather than its absolute magnitude (Starr et al., 2011). Taken together, these papers replace a purely phenomenological strong/fragile dichotomy with a boundary problem: the edge is where a microstructural control parameter causes a qualitative change in the form of slowdown (Ciarella et al., 2019, Jaiswal et al., 2016, Busch et al., 2014, Starr et al., 2011).

4. Literal edge-state fragility and fragile topology

In topological and mesoscopic transport, “fragility edge” often refers to the actual instability of edge modes. The 2023 analysis of the modified Haldane model directly overturns the idea that antichiral edge states are immune to disorder. Because the model is gapless and lacks a protecting bulk invariant, onsite Anderson disorder breaks the chiral symmetry of the SSH-like one-dimensional reduction and opens scattering channels from edge modes into counterpropagating bulk modes. Quantitatively, the disorder-averaged winding number at $\theta_A=T_A/T_g$7 is strongly suppressed already for $\theta_A=T_A/T_g$8, and transfer-matrix scaling shows that the normalized localization length decreases with strip width across the spectrum at strong disorder, indicating localization of both edge and bulk channels (Mannaï et al., 2023). A related correction appears in the quantum spin Hall literature: non-local resistance in a four-terminal QSH bar is not generically more robust than local resistance. Even a single disordered voltage probe reduces the ideal $\theta_A=T_A/T_g$9 from $\beta r(B)=1$0 to $\beta r(B)=1$1 when $\beta r(B)=1$2, and dephasing plus spin-flip scattering makes the non-local signal highly configuration-dependent (Mani et al., 2016).

The 2026 theory of gapless helical edge and hinge states introduces “topological fragility” in a different but related sense. In a one-dimensional helical channel, spin–momentum locking yields a perfect current-induced spin polarization, and random spin–orbit coupling of Elliott–Yafet type then turns a modest external magnetic field into a robust elastic backscattering channel without gap opening. The resulting bilinear magnetoelectric resistance is odd in both $\beta r(B)=1$3 and $\beta r(B)=1$4 and is predicted to be parametrically larger in 1D than in 2D because the exchange splitting scales directly with current, $\beta r(B)=1$5 (Gorini et al., 25 May 2026). This usage preserves the language of topological protection but stresses that nonlinear transport plus random SOC can create a fragility channel even when linear-response conductance remains ideal at $\beta r(B)=1$6.

Photonic Chern insulators show a further refinement. Weak disorder can disrupt unidirectional chiral transport even when the global Chern number remains intact, because randomly placed nonmagnetized impurities create spatially extended defect states near the single-impurity resonance frequency $\beta r(B)=1$7, and weak coupling among them forms a narrow, topologically trivial impurity band inside the gap. Near $\beta r(B)=1$8, reciprocal “necklace state” channels couple to the chiral edge states and break one-way propagation, despite preservation of local Chern number away from the impurity-band window (Shi et al., 8 May 2026). By contrast, in fragile topology proper the absence of gapless edge states is not a sign of triviality. The phononic-crystal experiment on screw dislocations shows that fragile bands with filling anomaly can support one-dimensional gapless bound modes along a screw dislocation even though ordinary edges are gapped. The dislocation implements a twisted boundary condition,

$\beta r(B)=1$9

and the resulting spectral flow across the gap constitutes a bulk–defect, rather than bulk–edge, correspondence for fragile topology (Wu et al., 2024). A central misconception is thus domain-dependent: in some systems edge transport is less protected than previously thought, whereas in fragile topology the lack of boundary modes does not exhaust the topology (Mannaï et al., 2023, Shi et al., 8 May 2026, Wu et al., 2024).

5. Spectral, geometric, and combinatorial edges in networks and markets

Financial applications formalize the fragility edge as a spectral threshold. The 2026 operator-spectral theory defines a recovered self-adjoint dependence operator $B_c$0 from contemporaneous comovement and a discount factor $B_c$1, then identifies the edge by

$B_c$2

where $B_c$3 is the spectral radius. At this point the pricing resolvent $B_c$4 loses convergence, the norm $B_c$5 diverges, and a rational bubble becomes admissible if $B_c$6 is an eigenvalue of $B_c$7. In the empirical study of eighteen global equity indices from 2004 to 2024, the dominant factor strengthens in every documented crisis, the participation ratio falls from about $B_c$8 in calm periods to $B_c$9 in crisis, and the readings are explicitly described as coincident with crises rather than forecasts (Bhandari, 4 Jul 2026).

A different network literature works directly at the level of graph edges. In the threshold-network analysis of 69 global indices, four edge-based Ricci curvatures are used as fragility monitors. Forman–Ricci curvature is particularly associated with vulnerable, bridge-like connections; average FR decreases sharply during crisis windows, while average OR, MR, and HR increase as the network becomes denser, less modular, and more homogeneous (Samal et al., 2021). A later analytical study of stock-market Ollivier–Ricci curvature refines this result by computing

$x_{\mathrm{crit}}$0

on MST-plus-threshold graphs, with hop distance as ground metric and neighbor-normalized correlation weights as masses. There, average market curvature functions as an ex-post crash hallmark: it spikes during major crises and is most reliable when the rolling window $x_{\mathrm{crit}}$1 is large and the threshold $x_{\mathrm{crit}}$2 is calibrated, typically in the range $x_{\mathrm{crit}}$3 (García et al., 2024). The edge here is neither a mode nor a boundary state but a geometric property of pairwise links.

Graph-theoretic fragility can also be purely combinatorial. In the edge-removal framework of 2022, one fixes a target fraction $x_{\mathrm{crit}}$4 for the largest connected component and defines the critical fraction of removed edges as $x_{\mathrm{crit}}$5, where $x_{\mathrm{crit}}$6 is the minimal number of deletions needed to make the largest component at most $x_{\mathrm{crit}}$7. Relative fragility is then normalized against the complete graph by $x_{\mathrm{crit}}$8. The paper shows that graphs generally disintegrate faster than greedy targeted attack would anticipate and identifies complete equitable bipartite graphs as asymptotically robust under this measure (Fish et al., 2022). A still more local graph-based use appears in chess, where the interaction graph of legal attack and defense relations defines a fragility score

$x_{\mathrm{crit}}$9

with $a_{\mathrm{edge}}$0 indicating whether a piece is attacked and $a_{\mathrm{edge}}$1 its directed betweenness centrality. In a large corpus, fragility peaks around move $a_{\mathrm{edge}}$2, pawns account for approximately $a_{\mathrm{edge}}$3 of key pieces and knights for approximately $a_{\mathrm{edge}}$4, and the aligned average curves show an approximately universal build-up and decay around the peak (Barthelemy, 2024). These applications differ markedly in ontology, but all make the edge an empirically identifiable threshold in graph structure or graph-derived dynamics.

6. Operational criticality, supply-chain precipices, and pre-failure onset

In socio-technical systems, the fragility edge is formulated as a genuine critical point of timeliness. With delays propagated by

$a_{\mathrm{edge}}$5

the buffer $a_{\mathrm{edge}}$6 controls whether delays truncate, persist, or avalanche. The papers identify a critical buffer $a_{\mathrm{edge}}$7 such that $a_{\mathrm{edge}}$8 is robust, $a_{\mathrm{edge}}$9 allows accumulating delays and possible systemic crisis, and $\beta$00 produces avalanches of all sizes. A branching-style approximation expresses the boundary as $\beta$01, with $\beta$02 the spectral radius of the dependency operator and $\beta$03 (Moran et al., 2023). The edge is therefore an operational design threshold created by pressure toward efficiency, reduced redundancy, and tighter schedules.

Supply-network theory recasts the same logic in production terms. There the functional probability of a depth-$\beta$04 firm obeys

$\beta$05

with $\beta$06 essential inputs and $\beta$07 suppliers per input. For $\beta$08 and $\beta$09, there is a mechanical precipice at $\beta$10: below it, deep-limit reliability collapses to zero; above it, reliability jumps to a positive branch. In equilibrium, firms with intermediate productivity bunch at this precipice, so that small deteriorations in relationship strength can cause large output losses even though such fragility is inefficient from a planner’s perspective (Elliott et al., 2020). The “edge” here is not merely descriptive; it is an endogenous equilibrium object generated by underinvestment in robustness.

A more abstract dynamical version appears in the 2026 theory of finite-amplitude fragility. For perturbation directions $\beta$11 and amplitudes $\beta$12, the exact fragility curve is

$\beta$13

where $\beta$14 is the smallest amplitude causing failure in direction $\beta$15. The key result is that this curve is predicted, before failure occurs, by the tail of the boundary-normalized fragility gain

$\beta$16

through the leading-order relation $\beta$17 and hence $\beta$18. In that framework, a fragility edge can be operationalized either by the amplitude at which $\beta$19 has maximal slope or by the amplitude at which a fixed fraction of directions have become dangerous (Limkumnerd, 30 May 2026). This suggests a general distinction between the strongest route to failure and the breadth of near-dangerous directions, a distinction that also reappears in finance, seismic fragility, and network curvature.

7. Comparative synthesis

A plausible implication of this corpus is that “Fragility Edge” names a common structural problem rather than a common variable. First, the edge is always a boundary between routine perturbation handling and a regime of amplified sensitivity. Second, the boundary is usually encoded by a low-dimensional observable—$\beta$20 and $\beta$21, $\beta$22, $\beta$23, $\beta$24, $\beta$25, $\beta$26, $\beta$27, or a graph-level curvature statistic—even when the underlying mechanism is high-dimensional (Mai et al., 2017, Ciarella et al., 2019, Bhandari, 4 Jul 2026, Moran et al., 2023, Limkumnerd, 30 May 2026). Third, several papers show that single-parameter or single-route summaries can be misleading: LR cloud analysis can distort seismic knees, antichiral edge channels are not protected merely because they co-propagate, intact Chern numbers do not preclude impurity-mediated failure of unidirectionality, and the strongest directional gain alone does not determine finite-amplitude fragility (Mai et al., 2017, Mannaï et al., 2023, Shi et al., 8 May 2026, Limkumnerd, 30 May 2026).

The comparative literature also clarifies what the edge is not. It is not always predictive: the operator-spectral readings in markets are coincident with crises, and Ollivier–Ricci curvature is described as an ex-post crash hallmark rather than a forecasting device (Bhandari, 4 Jul 2026, García et al., 2024). It is not always visible at ordinary boundaries: fragile topology may have gapped edges yet still exhibit dislocation-bound spectral flow (Wu et al., 2024). Nor is it necessarily equivalent to a single “weakest link”: some frameworks instead treat fragility as a tail property of distributions over edges, modes, or perturbation directions (Samal et al., 2021, Limkumnerd, 30 May 2026).

In encyclopedia terms, the most defensible unified definition is therefore relational. “Fragility Edge” designates the threshold object—curve knee, crossover, spectral boundary, critical buffer, geometric edge statistic, or defect-enabled boundary condition—through which a system’s latent susceptibility becomes manifest. The specific mathematics varies across fields, but the recurrent content is a sharply organized transition from robust response to disproportionate consequence.