Bond Fracture Cascade Analysis of Defensive Networks in Elite Football: A Statistical Mechanics Approach

Bond Fracture Cascade Analysis of Defensive Networks in Elite Football: A Statistical Mechanics Approach | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Article Bond Fracture Cascade Analysis of Defensive Networks in Elite Football: A Statistical Mechanics Approach JIong zhang, Lin Zuo This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-9397572/v1 This work is licensed under a CC BY 4.0 License Status: Under Revision Version 1 posted 8 You are reading this latest preprint version Abstract Collective defensive organisation in football is a key determinant of match outcome, yet existing analytical approaches lack a principled physical framework for capturing defensive structural fragility at the moment of goal concession. Here we introduce a bond fracture cascade framework inspired by condensed-matter physics, in which spatial relationships between defending players are modelled as a network of interacting bonds governed by a Morse-type potential. Applying this framework to 190732 possessions from seven elite competitions—including the FIFA World Cup 2022, UEFA Euro 2020 and 2024, the Women's World Cup 2023, Women's Euro 2022 and 2025, and the Bundesliga 2023/24—we test twelve hypotheses linking bond-fracture metrics to goal outcomes. Goal-scoring possessions exhibit lower first-fracture bond strength (− 5.90 vs − 2.73 a.u.; p = 1.24 × 10⁻¹²⁵), larger avalanche size (3.77 vs 2.09; p = 2.01 × 10⁻¹¹⁶), and larger weakness gap (− 6.08 vs − 3.27 a.u.; p = 1.36 × 10⁻¹³⁵). Goal probability rises monotonically with defensive coordination number, from 1.89% to 5.26% across bond-count bins (r = 0.975, p = 0.0002). A Defensive Engagement Index (DEI) derived from the bond network shows a consistent directional difference between goal and non-goal possessions across all seven competitions (pooled p = 5.8 × 10⁻⁷). Logistic regression confirms cascade features retain independent predictive value beyond proximity to goal (distance to goal β = −0.459; first-fracture strength β = −0.126, p < 0.001). These results establish a physically principled, competition-agnostic signature of defensive structural collapse in football. Physical sciences/Mathematics and computing Physical sciences/Physics football analytics defensive network bond fracture cascade Morse potential statistical mechanics Defensive Engagement Index Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Figure 7 1. Introduction The emergence of collective defensive organisation in team sports represents one of the most compelling manifestations of complex adaptive systems in athletic competition. In football, defensive structure is not merely a tactical preference but a dynamical property of a coupled multi-agent system whose stability under perturbation ultimately determines competitive outcome. Yet despite the explosion of quantitative sports science over the past two decades 1,2 , the physical mechanisms by which defensive structures fail—and the moment at which they become vulnerable to penetration—remain poorly understood at the level of first principles. Traditional approaches to defensive analysis have relied on positional metrics such as team length, team width, and the area of the convex hull occupied by defending players 3,4 . While these measures capture aggregate spatial organisation, they are insensitive to local structural heterogeneity and cannot identify the specific interplayer connections whose disruption precipitates goal concession. Network-based approaches have offered partial remedies: passing networks reveal information flow patterns within teams 5,6,7 , and proximity networks quantify spatial coupling between players 8,9 . However, the dynamical question of how and in what order these coupling structures fail during a dangerous attack has not been addressed with a physically grounded model. The physics of fracture and cascade failure in disordered systems provides a productive conceptual bridge. In condensed matter, the sequential rupture of bonds in a network under stress—so-called bond percolation and avalanche phenomena 10,11 —has been used to explain failure in materials ranging from granular media 12 to biological tissues 13 and infrastructure networks 14,15 . The Morse potential, originally formulated to describe diatomic molecular vibrations 16 , has proven particularly versatile for modelling interactions with a characteristic equilibrium distance and asymmetric decay behaviour. Its anharmonic form captures both the attractive pull towards an equilibrium separation and the energetic penalty for excessive compression, properties that are physically analogous to the tactical imperative for defending players to maintain optimal inter-player spacing 17 . Statistical mechanics approaches to sports have gained traction in recent years. Entropy-based measures have been applied to quantify unpredictability in passing sequences 18,19 , while percolation theory has been used to study connectivity in team formations 20 . Agent-based models have simulated pressing dynamics 21 , and complexity science frameworks have characterised the phase-transition-like behaviour of team coordination 22,23 . Despite these advances, no prior work has applied bond-fracture cascade analysis—the sequential failure of specific structural connections—to model defensive collapse in football. The present study addresses this gap through three principal contributions: First, we introduce a bond-mechanics framework in which the pairwise spatial relationships between defending players are assigned Morse-potential-derived strengths, capturing the energetic cost of deviating from an equilibrium defensive spacing. Second, we define a suite of cascade fracture metrics—first-fracture strength, mean fracture strength, failure step count, avalanche size, and weakness gap—and test twelve hypotheses linking these metrics to goal outcomes across 190732 ball possessions from seven elite competitions. Third, we construct a Defensive Engagement Index (DEI) that aggregates bond network properties into a single scalar capable of characterising defensive structural integrity in a competition-agnostic manner. By grounding tactical football analysis in the physics of fracture mechanics, we aim to provide a principled, interpretable, and universally applicable framework for understanding when and why defensive structures fail—findings with direct implications for coaching, match preparation, and real-time performance monitoring. 2. Methods 2.1 Dataset Event-level tracking data with 360° spatial coordinates were obtained from the StatsBomb 360 open dataset for 190732 ball possessions across seven elite competitions: the FIFA World Cup 2022 (n = 38442 possessions), UEFA Euro 2020 (n = 32747), UEFA Euro 2024 (n = 33583), the FIFA Women's World Cup 2023 (n = 32155), UEFA Women's Euro 2022 (n = 14232), UEFA Women's Euro 2025 (n = 15668), and the German Bundesliga 2023/24 season (n = 23905). Of these, 4831 possessions (2.53%) concluded with a goal. Each possession record includes the Cartesian coordinates of all defending players at the moment of ball reception, the spatial origin of the possession, and competition metadata. 2.2 Bond Network Construction For each possession, a spatial bond network was constructed from the positions of the defending players. Two defenders i and j were connected by a bond if their Euclidean distance r ij did not exceed the maximum bonding distance D MAX = 15 m, a threshold chosen to capture tactically meaningful interplayer connections while excluding remote coupling. Each possession thus yielded a weighted undirected graph G = (V, E) where V is the set of defending players and E the set of active bonds. The mean coordination number across all possessions was 12.14 ± 8.74 bonds per possession (range 1–55). 2.3 Bond Strength: Morse Potential Bond strength was assigned using a modified Morse potential. For each bond of length r ij , the Gaussian-attenuated strength was defined as: $${S_{Gauss}}(r)={D_e} \cdot \exp \left[ { - \frac{{{{(r - {r_0})}^2}}}{{2{\sigma ^2}}}} \right]$$ 1 where r₀ = 10.0 m is the equilibrium bond length (set to the empirical mean inter-player distance), σ = 3.25 m is the standard deviation of observed bond lengths, and D e is a normalisation constant. This Gaussian formulation produces a symmetric bell-shaped strength profile: bonds near the equilibrium distance are maximally strong, while over- or under-extended bonds are progressively weaker. Negative bond strength values arise for bonds that are substantially compressed below or stretched beyond the equilibrium range, reflecting the repulsive component of the Morse analogue. 2.4 Attack Damage Model To capture the spatially heterogeneous pressure exerted by an attacking possession, a damage scalar D was computed for each possession: $$D={D_{goal}} \times {D_{density}}=\exp \left( { - \frac{{{d_{goal}}}}{{{\lambda _{goal}}}}} \right) \times \exp \left( { - \frac{{{n_{near}}}}{{{n_0}}}} \right)$$ 2 where d goal is the Euclidean distance from the possession origin to the goal centre (fixed at [105, 34] m in standard pitch coordinates), λ goal = 20 m is the spatial decay constant, n near is the number of defenders within 5 m of the ball carrier, and n₀ = 2.0. Bond strengths were subsequently modified by a spatially attenuated damage term that reduces the strength of bonds close to the point of attack, with a spatial decay constant λ spread = 5 m. 2.5 Cascade Metrics For each possession, bond strengths were sorted in ascending order and the following cascade fracture metrics were computed as shown in Table 1 .: Table 1 Cascade fracture metrics computed for each ball possession. Metric Definition First fracture strength Strength of the weakest bond (first to fail under cascade loading) Mean fracture strength Mean strength across the 30 weakest bonds Failure step Number of bond ruptures required for 20% of bonds to fail Avalanche size Maximum absolute strength jump between successive fracture events Weakness gap Difference: first fracture strength minus mean fracture strength; measures local heterogeneity Resilience index 1 − failure step / n_bonds; fraction of bonds surviving cascade (range 0–1) 2.6 Defensive Engagement Index (DEI) A composite Defensive Engagement Index was constructed by combining three normalised bond network properties via a sigmoid (logistic) function: $$DEI=\sigma \left( {0.35 \cdot \ln \frac{n}{{12}}+0.35 \cdot \ln \frac{{\bar {s}}}{{{s_0}}}+0.30 \cdot \ln \frac{d}{{40}}} \right)$$ 3 where n is the coordination number (number of active bonds), s̄ is the mean Gaussian bond strength, s₀ = 1.0 a.u. is the reference strength, d is the distance from the possession origin to the goal, and the reference values 12 and 40 m correspond to the empirical global means. The weights were chosen to reflect the relative tactical importance of structural density (0.35), bond quality (0.35), and spatial position (0.30). The sigmoid transformation maps DEI to the interval [0, 1], where higher values indicate greater defensive structural engagement. 2.7 Hypothesis Framework Twelve pre-specified hypotheses were tested (Mann-Whitney U test, two-sided, α = 0.001 after Bonferroni correction for 12 tests)as shown in Table 2 : Table 2 Pre-specified hypotheses and predicted directions of effect (goal vs non-goal possessions). H Metric Predicted direction (goal vs non-goal) H1 Weakest bond strength (Gauss) Lower H2 Weakest bond length Larger H3 First fracture strength Lower H4 Mean fracture strength Lower H5 Failure step Higher (longer cascade until collapse) H6 Avalanche size Larger H7 Attack damage Higher H8 Distance to goal Smaller H9 Resilience index Higher (longer sustained attack) H10 Coordination number Higher (deeper penetration) H11 DEI Lower H12 Weakness gap Larger (more local vulnerability) 2.8 Statistical Analysis Group differences were assessed using the two-sided Mann-Whitney U test. Effect sizes were quantified using the rank-biserial correlation r rb . Monotonic association with goal outcome was quantified using Spearman's rank correlation ρ. To address the spatial confound of distance to goal, logistic regression was performed with two nested models: Model A (cascade features only) and Model B (cascade features plus spatial controls: attack damage, distance to goal, and DEI). All cascade predictors were standardised (z-scored) prior to regression to enable comparison of coefficient magnitudes. The logistic regression was estimated via iteratively reweighted least squares (IRLS). All analyses were conducted in Python 3.11 using NumPy, SciPy, and pandas. 3. Results 3.1 Bond Strength Distributions (H3, H12) Figure 1 displays the distributions of first fracture strength and weakness gap for goal and non-goal possessions. Goal-scoring possessions exhibit substantially lower first fracture strength (mean: −5.90 a.u., median: −1.18 a.u.) compared to non-goal possessions (mean: −2.73 a.u., median: −0.41 a.u.; Mann-Whitney U, p = 1.24 × 10 − 125 ), confirming H3. The weakness gap—defined as the difference between first fracture and mean fracture strength—is similarly more negative in goal possessions (mean: −6.08 a.u.) than non-goal possessions (mean: −3.27 a.u.; p = 1.36 × 10 − 135 ; r rb = 0.21), confirming H12. The long negative tails of the goal-possession distributions indicate that a subset of goal-scoring situations involves extreme local structural vulnerability in the defending network, well beyond the typical range of non-goal possessions. 3.2 Cross-Competition Consistency (H3 Stratified) Figure 2 presents the distribution of first fracture strength stratified by competition. The systematic displacement of goal-possession boxes (orange) below non-goal boxes (blue) is consistent across all seven competitions, from the Bundesliga 2023/24 (p = 2.28 × 10 − 8 ) to the Women's World Cup 2023 (p = 1.53 × 10 − 29 ). The Women's Euro 2022 shows the largest absolute effect (goal mean: −9.60 vs non-goal mean: −3.51), potentially reflecting the tactical characteristics of the women's game including higher defensive line heights and wider inter-player spacings. The robustness of H3 across competitions spanning men's and women's football, club and international contexts, strongly supports the generalisability of the bond fracture framework. 3.3 Cascade Fracture Metrics (H3, H6, H9, H12) Figure 3 provides a four-panel comparison of the principal cascade metrics. In addition to the first fracture strength (H3, p = 1.24 × 10 − 125 ), goal possessions show larger avalanche size (H6: goal mean 3.77 vs 2.09; p = 2.01 × 10 − 116 ), higher resilience index (H9: goal mean 0.725 vs 0.703; p = 1.74 × 10 − 30 ), and more negative weakness gap (H12: goal mean − 6.08 vs − 3.27; p = 1.36 × 10 − 135 ). The directional findings for H9 (higher resilience in goal possessions) and H10 (higher coordination number) warrant interpretation: goal-scoring sequences tend to be longer and more sustained attacks, involving more bonds and more cascade steps before structural collapse. This is consistent with H5 (goal possessions show higher mean failure step count: 3.42 vs 2.84; p = 5.63 × 10 − 88 ), suggesting that successful attacks involve a process of gradual structural attrition—'slow burn' cascade accumulation—before the eventual catastrophic avalanche event that enables penetration. 3.4 Goal Probability and Coordination Number (H10) Figure 4 shows goal probability as a function of defensive coordination number, binned into seven intervals. There is a strong, monotonically increasing relationship: possessions that engage only 1–5 defensive bonds yield a goal probability of 1.89%, rising continuously to 5.26% for possessions with more than 30 bonds (linear trend: r = 0.975, p = 0.0002; slope = 0.61% per bin). Wilson 95% confidence intervals are shown for each bin; all bins contain at least 5,000 possessions. This finding, which initially appears counterintuitive—more bonds should imply better-covered defence—is resolved by recognising that higher coordination number in the attacking possession context reflects greater depth of penetration into the defensive structure. A possession that has generated 30 + defensive bonds has advanced far into the defensive organisation, precisely the condition under which structural fracture is most consequential. 3.5 Defensive Engagement Index by Competition (H11) Figure 5 presents DEI distributions stratified by competition, sorted by the magnitude of the goal vs non-goal difference. Goal possessions show lower DEI than non-goal possessions across all seven competitions in the expected direction. The effect reaches conventional significance in three competitions individually (Euro 2024: p = 2.6 × 10⁻⁴; Women's Euro 2022: p = 0.003; Women's World Cup 2023: p = 0.001) and in the pooled analysis (p = 5.8 × 10⁻⁷), partially confirming H11. The Women's Euro 2022 and Women's World Cup 2023 show the largest separation between goal and non-goal DEI distributions, while the Women's Euro 2025 shows the smallest effect (ns). The within-competition effect size is small (pooled rank-biserial r꜀ β = −0.042), and DEI is best interpreted as a composite structural index whose discriminative power is most robust in pooled analysis rather than within individual competitions; full per-competition results are reported in Supplementary Table S2. 3.6 Logistic Regression: Cascade Features Beyond Spatial Position Figure 6 presents the standardised logistic regression coefficients from Model B (cascade features plus spatial controls). Distance to goal is, as expected, the dominant predictor (β = −0.459, p < 10 − 300 ), confirming that spatial proximity to the target is a necessary condition for goal threat. However, after controlling for distance to goal, attack damage, and DEI, cascade features retain statistically significant independent predictive value: first fracture strength (β = −0.126, p = 4.66 × 10 − 15 ), resilience index (β = −0.083, p = 0.002), and attack damage (β = 0.052, p = 0.001). The significance of first fracture strength in Model B is particularly important: it demonstrates that the structural fragility of the defensive bond at the point of first failure carries predictive information about goal outcomes independent of how far up the pitch the possession originates, directly addressing the potential confound that cascade metrics merely proxy for spatial position. 3.7 Feature Correlation Structure Figure 7 displays the Spearman rank correlation matrix across all bond fracture and spatial features. Several structural patterns are notable. First fracture strength and weakness gap are nearly perfectly correlated (ρ = 0.97), confirming their mathematical relationship and justifying the removal of weakness gap from the regression model to avoid collinearity. Coordination number and failure step are extremely highly correlated (ρ = 0.98), reflecting the fact that longer possessions naturally involve both more bonds and more cascade steps. Attack damage and distance to goal are strongly inversely correlated (ρ = −0.85), as the damage function is defined primarily through spatial proximity to goal. Correlations with the goal outcome (rightmost column) are modest in absolute magnitude (range |ρ| = 0.03–0.07), consistent with the known difficulty of predicting individual goal events from single-possession metrics in a sport characterised by low scoring rates and high variance. 4. Discussion The present study introduces and validates a bond fracture cascade framework for analysing defensive structural failure in football. Across 190732 possessions from seven elite competitions, we demonstrate that goal-scoring attacks are systematically associated with lower first-fracture bond strength, larger cascade avalanches, and greater local structural vulnerability in the defending network—findings that are consistent across men's and women's football and across both club and international competition contexts. The most practically significant finding is the monotonic relationship between defensive coordination number and goal probability (Fig. 4 , r = 0.975). Under the bond mechanics interpretation, coordination number in the goal-possession context reflects the depth to which an attacking sequence has penetrated the defensive structure before the possession concludes. Possessions that activate more than thirty defensive bonds achieve goal rates nearly three times those of possessions activating fewer than five bonds. This suggests a cascade threshold phenomenon: once a possession has sufficiently disrupted the peripheral layers of defensive bonding, the probability of finding and exploiting a catastrophic fracture path increases markedly. The counterintuitive directions of H9 (higher resilience in goal possessions) and H10 (higher coordination number) are mechanistically interpretable within the cascade framework. Goal-scoring sequences are not simply faster disruptions of defensive networks; they tend to be longer, more sustained attacks that accumulate damage across a greater number of bonds before the cascade becomes irreversible. This 'slow burn' characteristic—high failure step count, high coordination number, high resilience in the sense of surviving many partial fractures—is precisely the signature of a systemic, progressive defensive breakdown rather than a lucky individual error. From a coaching perspective, this implies that vulnerability to conceding goals is not always expressed in the immediate structural properties of the defensive line at a single moment, but may be better captured by the trajectory of structural deterioration over the course of an extended possession. The logistic regression results (Fig. 6 ) address a fundamental question of confounding: are cascade metrics merely proxies for spatial proximity to goal? The persistence of first fracture strength (β = −0.126, p = 4.66 × 10 − 15 ) and resilience index (β = −0.083, p = 0.002) as significant predictors after controlling for distance to goal and DEI demonstrates that structural bond fragility provides genuine incremental predictive information beyond what is captured by position alone. This is consistent with the observation that many high-xG (expected goals) situations do not result in goals, and vice versa: the physical structure of defensive organisation at the moment of ball receipt carries information that spatial proximity metrics do not fully encode. Several limitations warrant acknowledgement. The bond strength model depends on parameters (r₀, σ, D MAX ) calibrated to the empirical distributions of this dataset; future work should assess parameter sensitivity and explore alternative potential functions. The analysis treats each possession as an independent event, whereas in practice, defensive fatigue and in-game tactical adjustments create temporal dependencies across possessions within a match. The modest absolute magnitudes of Spearman correlations with goal outcome (|ρ| ≤ 0.07) reflect the irreducibly stochastic nature of goal events in football, and caution against interpreting the framework as a precise goal probability predictor rather than a structural vulnerability indicator. Finally, the negative values of first fracture strength arise from bonds compressed or stretched well beyond equilibrium; while physically meaningful within the Morse framework, their interpretation in coaching applications should be accompanied by appropriate normalisation. Future extensions of this framework might incorporate temporal dynamics—tracking how bond network properties evolve across successive events within a possession—and explore machine learning approaches that jointly model structural and positional features. Application to real-time tracking data could enable live defensive fragility scores that inform in-match tactical adjustments. 5. Conclusions We have introduced a bond fracture cascade framework for modelling defensive structural failure in football, grounded in the physics of Morse-potential bond networks and avalanche phenomena. Applied to 190732 possessions from seven elite competitions, the framework yields robust structural signatures of goal-scoring attacks across men's and women's, club and international contexts: eleven hypotheses (H3–H10, H12) are confirmed at p < 10⁻⁶ in pooled analysis, with H11 (DEI) partially supported (pooled p = 5.8 × 10⁻⁷; significant in 3/7 competitions individually). The Defensive Engagement Index offers a physically motivated composite measure of defensive structural integrity, most informative in pooled analysis. Logistic regression confirms that cascade features carry predictive information beyond spatial proximity to goal. These results establish a physically grounded, interpretable foundation for quantitative analysis of defensive collapse in football, with direct applications in tactical analysis and performance monitoring. Declarations Competing Interests The authors declare no competing interests. Funding This work was supported by the 2024 Guangdong Provincial Research Program on High-Quality Development of Youth Campus Football and School Physical Education (Grant No. 24SXZPT50). Author Contribution Jiongzhang: Conceptualisation, methodology, software, formal analysis, writing – original draft. Linzuo: Conceptualisation, supervision, writing – review & editing. Data Availability The raw event tracking data (StatsBomb 360) used in this study are publicly available via the StatsBomb Open Data GitHub repository (https://github.com/statsbomb/open-data). The analysis code (Python scripts 01–05) and the derived cascade feature datasets supporting the findings of this study are available from the corresponding author upon reasonable request. References Duarte, R., Araújo, D., Correia, V. & Davids, K. Sports teams as superorganisms: Implications of sociobiological models of behaviour for research and practice in team sports performance analysis. Sports Med. 42, 633–642 (2012). Gudmundsson, J. & Horton, M. Spatio-temporal analysis of team sports. ACM Comput. Surv. 50, 1–34 (2017). Moura, F. A. et al. Analysis of football game-related statistics using multivariate analysis. J. Hum. Kinet. 36, 183–194 (2013). Fernandez, J. & Bornn, L. Wide open spaces: A statistical technique for measuring space creation in professional soccer. MIT Sloan Sports Analytics Conference (2018). Pena, J. L. & Touchette, H. A network theory analysis of football strategies. arXiv:1206.6904 (2012). Cotta, C., Mora, A. M., Merelo, J. J. & Merelo-Molina, C. A network analysis of the 2010 FIFA World Cup champion team play. J. Syst. Sci. Complex. 26, 21–42 (2013). Buldú, J. M. et al. Defining a historic football team: Using Network Science to study Guardiola's F.C. Barcelona. Sci. Rep. 9, 13602 (2019). Duarte, R. et al. Basketball teams as dynamic networks: Implications for team training and practice. J. Sports Sci. Med. 15, 428–434 (2017). Rein, R. & Memmert, D. Big data and tactical analysis in elite soccer: future challenges and opportunities for sports science. SpringerPlus 5, 1410 (2016). Sornette, D. Critical Phenomena in Natural Sciences. Springer, Berlin (2006). Zapperi, S., Vespignani, A. & Stanley, H. E. Plasticity and avalanche behaviour in microfracturing phenomena. Nature 388, 658–660 (1997). Jaeger, H. M. & Nagel, S. R. Physics of the granular state. Science 255, 1523–1531 (1992). Alava, M. J., Nukala, P. K. V. V. & Zapperi, S. Statistical models of fracture. Adv. Phys. 55, 349–476 (2006). Buldyrev, S. V., Parshani, R., Paul, G., Stanley, H. E. & Havlin, S. Catastrophic cascade of failures in interdependent networks. Nature 464, 1025–1028 (2010). Gao, J., Buldyrev, S. V., Stanley, H. E. & Havlin, S. Networks formed from interdependent networks. Nat. Phys. 8, 40–48 (2012). Morse, P. M. Diatomic molecules according to the wave mechanics. II. Vibrational levels. Phys. Rev. 34, 57–64 (1929). Bartlett, R., Button, C., Robins, M., Dutt-Mazumder, A. & Kennedy, G. Analysing team coordination patterns from player movement trajectories in soccer. J. Sports Sci. 30, 1577–1587 (2012). Lago-Peñas, C. & Lago-Ballesteros, J. Game location and team quality effects on performance profiles in professional soccer. J. Sports Sci. Med. 10, 465–471 (2011). Gonçalves, B. et al. Entropy measures reveal collective tactical behaviours in association football. Int. J. Sports Physiol. Perform. 12, 1164–1172 (2017). Grund, T. U. Network structure and team performance: The case of English Premier League soccer teams. Soc. Networks 34, 682–690 (2012). Narizuka, T. & Yamamoto, K. Clustering algorithm for formations in football games. Sci. Rep. 9, 1–8 (2019). Duarte, R. et al. Competing with teammates and opponents in complex adaptive systems. J. Hum. Kinet. 48, 5–16 (2015). Vilar, L., Araújo, D., Davids, K. & Bar-Yam, Y. Science of winning soccer: emergent pattern-forming dynamics in association football. J. Syst. Sci. Complex. 26, 73–84 (2013). Additional Declarations No competing interests reported. Supplementary Files supplementaryinformation.docx Cite Share Download PDF Status: Under Revision Version 1 posted Editorial decision: Revision requested 18 May, 2026 Reviews received at journal 12 May, 2026 Reviewers agreed at journal 10 May, 2026 Reviewers invited by journal 22 Apr, 2026 Editor invited by journal 20 Apr, 2026 Editor assigned by journal 16 Apr, 2026 Submission checks completed at journal 16 Apr, 2026 First submitted to journal 12 Apr, 2026 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-9397572","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":631112595,"identity":"6a7073c7-357f-493d-be72-e07962932cf5","order_by":0,"name":"JIong zhang","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA60lEQVRIiWNgGAWjYBACxmYwdQCImQ8wwNlEamFLIE4LA0IZjwFxWpjbmY89/PLnjrw5/5pv0rxtDHJ8NxIYPxfgdRhburEMzzPDnTPebgNpMZa8kcAsPQOvFh4zaQmJw4wbbpzddhuoJXHDjQQ2Zh6CWgwO22+4ceYZSEs9UVokPyQcTtxwvocNpCXBgLAWtjRphgOHkzfcYDP/OeechOHMMw+bpfFpMew/fEzyx5/DthvOH35s8KbMRp7vePLBz3i1NAADGqxAIoGBiYdBAmRzAx4NDAzyICU/QCz+A1DGKBgFo2AUjAI0AADxwlDFUb4acAAAAABJRU5ErkJggg==","orcid":"","institution":"Wuyi University","correspondingAuthor":true,"prefix":"","firstName":"JIong","middleName":"","lastName":"zhang","suffix":""},{"id":631112596,"identity":"d4c31ef0-04ae-4de4-8ba3-552ed0d47eca","order_by":1,"name":"Lin Zuo","email":"","orcid":"","institution":"Wuyi University","correspondingAuthor":false,"prefix":"","firstName":"Lin","middleName":"","lastName":"Zuo","suffix":""}],"badges":[],"createdAt":"2026-04-13 01:53:10","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-9397572/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-9397572/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":108205613,"identity":"8a3e8578-7358-467b-a864-31e81a8663c3","added_by":"auto","created_at":"2026-04-30 12:44:21","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":115861,"visible":true,"origin":"","legend":"\u003cp\u003eBond fracture strength distributions for goal (orange) and non-goal (blue) possessions. Left: first fracture strength. Right: weakness gap (first minus mean fracture strength). Violin plots show the full distribution; embedded boxes indicate the interquartile range and median. Mann-Whitney p-values are reported above each panel.\u003c/p\u003e","description":"","filename":"floatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-9397572/v1/10ee6639fae80704502966d0.png"},{"id":108205614,"identity":"07c613b8-b65b-43ba-a911-13c9c699387d","added_by":"auto","created_at":"2026-04-30 12:44:21","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":85743,"visible":true,"origin":"","legend":"\u003cp\u003eFirst fracture strength stratified by competition. Each competition panel shows box plots for goal (orange) and non-goal (blue) possessions. Significance markers above each panel indicate the result of a two-sided Mann-Whitney U test (*** p \u0026lt; 0.001). Box limits indicate the 25th and 75th percentiles; whiskers extend to 1.5 × IQR.\u003c/p\u003e","description":"","filename":"floatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-9397572/v1/4a226af3267e8c207b919e6a.png"},{"id":108205615,"identity":"0078a363-b91a-4e8a-8f87-4de9277ee114","added_by":"auto","created_at":"2026-04-30 12:44:21","extension":"jpeg","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":58577,"visible":true,"origin":"","legend":"\u003cp\u003eFour-panel comparison of cascade fracture metrics for goal (orange) and non-goal (blue) possessions. From left to right: first fracture strength (H3), avalanche size (H6), resilience index (H9), and weakness gap (H12). Violin plots with embedded box-and-whisker overlays. Mann-Whitney p-values are indicated above each panel.\u003c/p\u003e","description":"","filename":"floatimage3.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-9397572/v1/b0532baca57cffa4812cef98.jpeg"},{"id":108205619,"identity":"cd4d5843-9853-4b6e-b4f0-bbe38068570f","added_by":"auto","created_at":"2026-04-30 12:44:21","extension":"jpeg","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":55085,"visible":true,"origin":"","legend":"\u003cp\u003eGoal probability as a function of defensive coordination number (n_bonds), binned in intervals of five. Bars show observed goal rate; error bars indicate 95% Wilson score confidence intervals. Sample sizes (n) are annotated above each bar. The dashed line shows the linear trend (r = 0.975, p = 0.0002).\u003c/p\u003e","description":"","filename":"floatimage4.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-9397572/v1/fb620fd8127b3f46fa9329a5.jpeg"},{"id":108205616,"identity":"717f37a4-1ffb-4d26-b677-ecbefa743c3f","added_by":"auto","created_at":"2026-04-30 12:44:21","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":103286,"visible":true,"origin":"","legend":"\u003cp\u003eDefensive Engagement Index (DEI) by competition. Competitions are sorted by the magnitude of the goal vs non-goal difference (left: largest difference; right: smallest). Box plots show the interquartile range and median; whiskers extend to 1.5 × IQR. Significance markers indicate Mann-Whitney U test results (*** p \u0026lt; 0.001, ** p \u0026lt; 0.01, ns p ≥ 0.05). DEI reaches significance in three of seven competitions individually; full per-competition results are reported in Supplementary Table S2.\u003c/p\u003e","description":"","filename":"floatimage5.png","url":"https://assets-eu.researchsquare.com/files/rs-9397572/v1/e4f49763462d1b4aa1ee4ece.png"},{"id":108205617,"identity":"20fdb41f-afd1-4613-9dff-303d8a9c0077","added_by":"auto","created_at":"2026-04-30 12:44:21","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":128607,"visible":true,"origin":"","legend":"\u003cp\u003eStandardised logistic regression coefficients for predictors of goal possession outcome (Model B, n = 190,732). Bars show coefficients; error bars indicate 95% confidence intervals. Positive coefficients (orange) indicate features associated with increased goal probability; negative coefficients (blue) indicate protective features. Significance markers: *** p \u0026lt; 0.001, ns not significant.\u003c/p\u003e","description":"","filename":"floatimage6.png","url":"https://assets-eu.researchsquare.com/files/rs-9397572/v1/d99007e503b595629f95e738.png"},{"id":108491492,"identity":"9598dc5f-f81e-4c3f-9b5e-20accde342ba","added_by":"auto","created_at":"2026-05-05 09:54:10","extension":"jpeg","order_by":7,"title":"Figure 7","display":"","copyAsset":false,"role":"figure","size":124152,"visible":true,"origin":"","legend":"\u003cp\u003eSpearman rank correlation matrix for all bond fracture and spatial features, including the binary goal outcome. Colours indicate the sign and magnitude of ρ (red: positive; blue: negative). Cell annotations show ρ rounded to two decimal places.\u003c/p\u003e","description":"","filename":"floatimage7.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-9397572/v1/64ba078eb84ce1ec83ddfaf6.jpeg"},{"id":108494992,"identity":"71f98032-8427-4051-8cc4-7290239d09e5","added_by":"auto","created_at":"2026-05-05 10:08:25","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":881682,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-9397572/v1/4ebd1a0c-5a38-49b0-b992-538d594f9f14.pdf"},{"id":108205612,"identity":"b3737381-8235-4555-a695-17aa4a230c2a","added_by":"auto","created_at":"2026-04-30 12:44:20","extension":"docx","order_by":0,"title":"","display":"","copyAsset":false,"role":"supplement","size":20516,"visible":true,"origin":"","legend":"","description":"","filename":"supplementaryinformation.docx","url":"https://assets-eu.researchsquare.com/files/rs-9397572/v1/c9cfc77debbe67de2397b5ac.docx"}],"financialInterests":"No competing interests reported.","formattedTitle":"Bond Fracture Cascade Analysis of Defensive Networks in Elite Football: A Statistical Mechanics Approach","fulltext":[{"header":"1. Introduction","content":"\u003cp\u003eThe emergence of collective defensive organisation in team sports represents one of the most compelling manifestations of complex adaptive systems in athletic competition. In football, defensive structure is not merely a tactical preference but a dynamical property of a coupled multi-agent system whose stability under perturbation ultimately determines competitive outcome. Yet despite the explosion of quantitative sports science over the past two decades\u003csup\u003e1,2\u003c/sup\u003e, the physical mechanisms by which defensive structures fail\u0026mdash;and the moment at which they become vulnerable to penetration\u0026mdash;remain poorly understood at the level of first principles.\u003c/p\u003e \u003cp\u003eTraditional approaches to defensive analysis have relied on positional metrics such as team length, team width, and the area of the convex hull occupied by defending players\u003csup\u003e3,4\u003c/sup\u003e. While these measures capture aggregate spatial organisation, they are insensitive to local structural heterogeneity and cannot identify the specific interplayer connections whose disruption precipitates goal concession. Network-based approaches have offered partial remedies: passing networks reveal information flow patterns within teams\u003csup\u003e5,6,7\u003c/sup\u003e, and proximity networks quantify spatial coupling between players\u003csup\u003e8,9\u003c/sup\u003e. However, the dynamical question of how and in what order these coupling structures fail during a dangerous attack has not been addressed with a physically grounded model.\u003c/p\u003e \u003cp\u003eThe physics of fracture and cascade failure in disordered systems provides a productive conceptual bridge. In condensed matter, the sequential rupture of bonds in a network under stress\u0026mdash;so-called bond percolation and avalanche phenomena\u003csup\u003e10,11\u003c/sup\u003e\u0026mdash;has been used to explain failure in materials ranging from granular media\u003csup\u003e12\u003c/sup\u003e to biological tissues\u003csup\u003e13\u003c/sup\u003e and infrastructure networks\u003csup\u003e14,15\u003c/sup\u003e. The Morse potential, originally formulated to describe diatomic molecular vibrations\u003csup\u003e16\u003c/sup\u003e, has proven particularly versatile for modelling interactions with a characteristic equilibrium distance and asymmetric decay behaviour. Its anharmonic form captures both the attractive pull towards an equilibrium separation and the energetic penalty for excessive compression, properties that are physically analogous to the tactical imperative for defending players to maintain optimal inter-player spacing\u003csup\u003e17\u003c/sup\u003e.\u003c/p\u003e \u003cp\u003eStatistical mechanics approaches to sports have gained traction in recent years. Entropy-based measures have been applied to quantify unpredictability in passing sequences\u003csup\u003e18,19\u003c/sup\u003e, while percolation theory has been used to study connectivity in team formations\u003csup\u003e20\u003c/sup\u003e. Agent-based models have simulated pressing dynamics\u003csup\u003e21\u003c/sup\u003e, and complexity science frameworks have characterised the phase-transition-like behaviour of team coordination\u003csup\u003e22,23\u003c/sup\u003e. Despite these advances, no prior work has applied bond-fracture cascade analysis\u0026mdash;the sequential failure of specific structural connections\u0026mdash;to model defensive collapse in football.\u003c/p\u003e \u003cp\u003eThe present study addresses this gap through three principal contributions:\u003c/p\u003e \u003cp\u003eFirst, we introduce a bond-mechanics framework in which the pairwise spatial relationships between defending players are assigned Morse-potential-derived strengths, capturing the energetic cost of deviating from an equilibrium defensive spacing. Second, we define a suite of cascade fracture metrics\u0026mdash;first-fracture strength, mean fracture strength, failure step count, avalanche size, and weakness gap\u0026mdash;and test twelve hypotheses linking these metrics to goal outcomes across 190732 ball possessions from seven elite competitions. Third, we construct a Defensive Engagement Index (DEI) that aggregates bond network properties into a single scalar capable of characterising defensive structural integrity in a competition-agnostic manner.\u003c/p\u003e \u003cp\u003eBy grounding tactical football analysis in the physics of fracture mechanics, we aim to provide a principled, interpretable, and universally applicable framework for understanding when and why defensive structures fail\u0026mdash;findings with direct implications for coaching, match preparation, and real-time performance monitoring.\u003c/p\u003e"},{"header":"2. Methods","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003e2.1 Dataset\u003c/h2\u003e \u003cp\u003eEvent-level tracking data with 360\u0026deg; spatial coordinates were obtained from the StatsBomb 360 open dataset for 190732 ball possessions across seven elite competitions: the FIFA World Cup 2022 (n\u0026thinsp;=\u0026thinsp;38442 possessions), UEFA Euro 2020 (n\u0026thinsp;=\u0026thinsp;32747), UEFA Euro 2024 (n\u0026thinsp;=\u0026thinsp;33583), the FIFA Women's World Cup 2023 (n\u0026thinsp;=\u0026thinsp;32155), UEFA Women's Euro 2022 (n\u0026thinsp;=\u0026thinsp;14232), UEFA Women's Euro 2025 (n\u0026thinsp;=\u0026thinsp;15668), and the German Bundesliga 2023/24 season (n\u0026thinsp;=\u0026thinsp;23905). Of these, 4831 possessions (2.53%) concluded with a goal. Each possession record includes the Cartesian coordinates of all defending players at the moment of ball reception, the spatial origin of the possession, and competition metadata.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec4\" class=\"Section2\"\u003e \u003ch2\u003e2.2 Bond Network Construction\u003c/h2\u003e \u003cp\u003eFor each possession, a spatial bond network was constructed from the positions of the defending players. Two defenders \u003cem\u003ei\u003c/em\u003e and \u003cem\u003ej\u003c/em\u003e were connected by a bond if their Euclidean distance r\u003csup\u003eij\u003c/sup\u003e did not exceed the maximum bonding distance D\u003csup\u003eMAX\u003c/sup\u003e = 15 m, a threshold chosen to capture tactically meaningful interplayer connections while excluding remote coupling. Each possession thus yielded a weighted undirected graph G = (V, E) where V is the set of defending players and E the set of active bonds. The mean coordination number across all possessions was 12.14\u0026thinsp;\u0026plusmn;\u0026thinsp;8.74 bonds per possession (range 1\u0026ndash;55).\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec5\" class=\"Section2\"\u003e \u003ch2\u003e2.3 Bond Strength: Morse Potential\u003c/h2\u003e \u003cp\u003eBond strength was assigned using a modified Morse potential. For each bond of length r\u003csup\u003eij\u003c/sup\u003e, the Gaussian-attenuated strength was defined as:\u003cdiv id=\"Equ1\" class=\"Equation\"\u003e\u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equ1\" name=\"EquationSource\"\u003e\n$${S_{Gauss}}(r)={D_e} \\cdot \\exp \\left[ { - \\frac{{{{(r - {r_0})}^2}}}{{2{\\sigma ^2}}}} \\right]$$\u003c/div\u003e\u003cdiv class=\"EquationNumber\"\u003e1\u003c/div\u003e\u003c/div\u003e\u003c/p\u003e \u003cp\u003ewhere r₀ = 10.0 m is the equilibrium bond length (set to the empirical mean inter-player distance), σ\u0026thinsp;=\u0026thinsp;3.25 m is the standard deviation of observed bond lengths, and D\u003csup\u003ee\u003c/sup\u003e is a normalisation constant. This Gaussian formulation produces a symmetric bell-shaped strength profile: bonds near the equilibrium distance are maximally strong, while over- or under-extended bonds are progressively weaker. Negative bond strength values arise for bonds that are substantially compressed below or stretched beyond the equilibrium range, reflecting the repulsive component of the Morse analogue.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec6\" class=\"Section2\"\u003e \u003ch2\u003e2.4 Attack Damage Model\u003c/h2\u003e \u003cp\u003eTo capture the spatially heterogeneous pressure exerted by an attacking possession, a damage scalar D was computed for each possession:\u003cdiv id=\"Equ2\" class=\"Equation\"\u003e\u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equ2\" name=\"EquationSource\"\u003e\n$$D={D_{goal}} \\times {D_{density}}=\\exp \\left( { - \\frac{{{d_{goal}}}}{{{\\lambda _{goal}}}}} \\right) \\times \\exp \\left( { - \\frac{{{n_{near}}}}{{{n_0}}}} \\right)$$\u003c/div\u003e\u003cdiv class=\"EquationNumber\"\u003e2\u003c/div\u003e\u003c/div\u003e\u003c/p\u003e \u003cp\u003ewhere d\u003csup\u003egoal\u003c/sup\u003e is the Euclidean distance from the possession origin to the goal centre (fixed at [105, 34] m in standard pitch coordinates), λ\u003csup\u003egoal\u003c/sup\u003e\u0026thinsp;=\u0026thinsp;20 m is the spatial decay constant, n\u003csup\u003enear\u003c/sup\u003e is the number of defenders within 5 m of the ball carrier, and n₀ = 2.0. Bond strengths were subsequently modified by a spatially attenuated damage term that reduces the strength of bonds close to the point of attack, with a spatial decay constant λ\u003csup\u003espread\u003c/sup\u003e\u0026thinsp;=\u0026thinsp;5 m.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec7\" class=\"Section2\"\u003e \u003ch2\u003e2.5 Cascade Metrics\u003c/h2\u003e \u003cp\u003eFor each possession, bond strengths were sorted in ascending order and the following cascade fracture metrics were computed as shown in Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e.:\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eCascade fracture metrics computed for each ball possession.\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"2\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMetric\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eDefinition\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eFirst fracture strength\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eStrength of the weakest bond (first to fail under cascade loading)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMean fracture strength\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eMean strength across the 30 weakest bonds\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eFailure step\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eNumber of bond ruptures required for 20% of bonds to fail\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAvalanche size\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eMaximum absolute strength jump between successive fracture events\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eWeakness gap\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eDifference: first fracture strength minus mean fracture strength; measures local heterogeneity\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eResilience index\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e1\u0026thinsp;\u0026minus;\u0026thinsp;failure step / n_bonds; fraction of bonds surviving cascade (range 0\u0026ndash;1)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec8\" class=\"Section2\"\u003e \u003ch2\u003e2.6 Defensive Engagement Index (DEI)\u003c/h2\u003e \u003cp\u003eA composite Defensive Engagement Index was constructed by combining three normalised bond network properties via a sigmoid (logistic) function:\u003cdiv id=\"Equ3\" class=\"Equation\"\u003e\u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equ3\" name=\"EquationSource\"\u003e\n$$DEI=\\sigma \\left( {0.35 \\cdot \\ln \\frac{n}{{12}}+0.35 \\cdot \\ln \\frac{{\\bar {s}}}{{{s_0}}}+0.30 \\cdot \\ln \\frac{d}{{40}}} \\right)$$\u003c/div\u003e\u003cdiv class=\"EquationNumber\"\u003e3\u003c/div\u003e\u003c/div\u003e\u003c/p\u003e \u003cp\u003ewhere n is the coordination number (number of active bonds), s̄ is the mean Gaussian bond strength, s₀ = 1.0 a.u. is the reference strength, d is the distance from the possession origin to the goal, and the reference values 12 and 40 m correspond to the empirical global means. The weights were chosen to reflect the relative tactical importance of structural density (0.35), bond quality (0.35), and spatial position (0.30). The sigmoid transformation maps DEI to the interval [0, 1], where higher values indicate greater defensive structural engagement.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec9\" class=\"Section2\"\u003e \u003ch2\u003e2.7 Hypothesis Framework\u003c/h2\u003e \u003cp\u003eTwelve pre-specified hypotheses were tested (Mann-Whitney U test, two-sided, α\u0026thinsp;=\u0026thinsp;0.001 after Bonferroni correction for 12 tests)as shown in Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e:\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab2\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003ePre-specified hypotheses and predicted directions of effect (goal vs non-goal possessions).\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"3\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eH\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eMetric\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003ePredicted direction (goal vs non-goal)\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eH1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eWeakest bond strength (Gauss)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eLower\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eH2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eWeakest bond length\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eLarger\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eH3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eFirst fracture strength\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eLower\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eH4\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eMean fracture strength\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eLower\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eH5\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eFailure step\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eHigher (longer cascade until collapse)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eH6\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eAvalanche size\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eLarger\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eH7\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eAttack damage\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eHigher\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eH8\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eDistance to goal\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eSmaller\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eH9\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eResilience index\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eHigher (longer sustained attack)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eH10\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eCoordination number\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eHigher (deeper penetration)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eH11\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eDEI\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eLower\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eH12\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eWeakness gap\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eLarger (more local vulnerability)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec10\" class=\"Section2\"\u003e \u003ch2\u003e2.8 Statistical Analysis\u003c/h2\u003e \u003cp\u003eGroup differences were assessed using the two-sided Mann-Whitney U test. Effect sizes were quantified using the rank-biserial correlation r\u003csup\u003erb\u003c/sup\u003e. Monotonic association with goal outcome was quantified using Spearman's rank correlation ρ. To address the spatial confound of distance to goal, logistic regression was performed with two nested models: Model A (cascade features only) and Model B (cascade features plus spatial controls: attack damage, distance to goal, and DEI). All cascade predictors were standardised (z-scored) prior to regression to enable comparison of coefficient magnitudes. The logistic regression was estimated via iteratively reweighted least squares (IRLS). All analyses were conducted in Python 3.11 using NumPy, SciPy, and pandas.\u003c/p\u003e \u003c/div\u003e"},{"header":"3. Results","content":"\u003cdiv id=\"Sec12\" class=\"Section2\"\u003e \u003ch2\u003e3.1 Bond Strength Distributions (H3, H12)\u003c/h2\u003e \u003cp\u003eFigure \u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e displays the distributions of first fracture strength and weakness gap for goal and non-goal possessions. Goal-scoring possessions exhibit substantially lower first fracture strength (mean: \u0026minus;5.90 a.u., median: \u0026minus;1.18 a.u.) compared to non-goal possessions (mean: \u0026minus;2.73 a.u., median: \u0026minus;0.41 a.u.; Mann-Whitney U, p\u0026thinsp;=\u0026thinsp;1.24 \u0026times; 10\u003csup\u003e\u0026minus;\u0026thinsp;125\u003c/sup\u003e), confirming H3. The weakness gap\u0026mdash;defined as the difference between first fracture and mean fracture strength\u0026mdash;is similarly more negative in goal possessions (mean: \u0026minus;6.08 a.u.) than non-goal possessions (mean: \u0026minus;3.27 a.u.; p\u0026thinsp;=\u0026thinsp;1.36 \u0026times; 10\u003csup\u003e\u0026minus;\u0026thinsp;135\u003c/sup\u003e; r\u003csup\u003erb\u003c/sup\u003e = 0.21), confirming H12. The long negative tails of the goal-possession distributions indicate that a subset of goal-scoring situations involves extreme local structural vulnerability in the defending network, well beyond the typical range of non-goal possessions.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec13\" class=\"Section2\"\u003e \u003ch2\u003e3.2 Cross-Competition Consistency (H3 Stratified)\u003c/h2\u003e \u003cp\u003eFigure \u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e presents the distribution of first fracture strength stratified by competition. The systematic displacement of goal-possession boxes (orange) below non-goal boxes (blue) is consistent across all seven competitions, from the Bundesliga 2023/24 (p\u0026thinsp;=\u0026thinsp;2.28 \u0026times; 10\u003csup\u003e\u0026minus;\u0026thinsp;8\u003c/sup\u003e) to the Women's World Cup 2023 (p\u0026thinsp;=\u0026thinsp;1.53 \u0026times; 10\u003csup\u003e\u0026minus;\u0026thinsp;29\u003c/sup\u003e). The Women's Euro 2022 shows the largest absolute effect (goal mean: \u0026minus;9.60 vs non-goal mean: \u0026minus;3.51), potentially reflecting the tactical characteristics of the women's game including higher defensive line heights and wider inter-player spacings. The robustness of H3 across competitions spanning men's and women's football, club and international contexts, strongly supports the generalisability of the bond fracture framework.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec14\" class=\"Section2\"\u003e \u003ch2\u003e3.3 Cascade Fracture Metrics (H3, H6, H9, H12)\u003c/h2\u003e \u003cp\u003eFigure \u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e provides a four-panel comparison of the principal cascade metrics. In addition to the first fracture strength (H3, p\u0026thinsp;=\u0026thinsp;1.24 \u0026times; 10\u003csup\u003e\u0026minus;\u0026thinsp;125\u003c/sup\u003e), goal possessions show larger avalanche size (H6: goal mean 3.77 vs 2.09; p\u0026thinsp;=\u0026thinsp;2.01 \u0026times; 10\u003csup\u003e\u0026minus;\u0026thinsp;116\u003c/sup\u003e), higher resilience index (H9: goal mean 0.725 vs 0.703; p\u0026thinsp;=\u0026thinsp;1.74 \u0026times; 10\u003csup\u003e\u0026minus;\u0026thinsp;30\u003c/sup\u003e), and more negative weakness gap (H12: goal mean\u0026thinsp;\u0026minus;\u0026thinsp;6.08 vs\u0026thinsp;\u0026minus;\u0026thinsp;3.27; p\u0026thinsp;=\u0026thinsp;1.36 \u0026times; 10\u003csup\u003e\u0026minus;\u0026thinsp;135\u003c/sup\u003e). The directional findings for H9 (higher resilience in goal possessions) and H10 (higher coordination number) warrant interpretation: goal-scoring sequences tend to be longer and more sustained attacks, involving more bonds and more cascade steps before structural collapse. This is consistent with H5 (goal possessions show higher mean failure step count: 3.42 vs 2.84; p\u0026thinsp;=\u0026thinsp;5.63 \u0026times; 10\u003csup\u003e\u0026minus;\u0026thinsp;88\u003c/sup\u003e), suggesting that successful attacks involve a process of gradual structural attrition\u0026mdash;'slow burn' cascade accumulation\u0026mdash;before the eventual catastrophic avalanche event that enables penetration.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec15\" class=\"Section2\"\u003e \u003ch2\u003e3.4 Goal Probability and Coordination Number (H10)\u003c/h2\u003e \u003cp\u003eFigure \u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e shows goal probability as a function of defensive coordination number, binned into seven intervals. There is a strong, monotonically increasing relationship: possessions that engage only 1\u0026ndash;5 defensive bonds yield a goal probability of 1.89%, rising continuously to 5.26% for possessions with more than 30 bonds (linear trend: r\u0026thinsp;=\u0026thinsp;0.975, p\u0026thinsp;=\u0026thinsp;0.0002; slope\u0026thinsp;=\u0026thinsp;0.61% per bin). Wilson 95% confidence intervals are shown for each bin; all bins contain at least 5,000 possessions. This finding, which initially appears counterintuitive\u0026mdash;more bonds should imply better-covered defence\u0026mdash;is resolved by recognising that higher coordination number in the attacking possession context reflects greater depth of penetration into the defensive structure. A possession that has generated 30\u0026thinsp;+\u0026thinsp;defensive bonds has advanced far into the defensive organisation, precisely the condition under which structural fracture is most consequential.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec16\" class=\"Section2\"\u003e \u003ch2\u003e3.5 Defensive Engagement Index by Competition (H11)\u003c/h2\u003e \u003cp\u003eFigure \u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003e presents DEI distributions stratified by competition, sorted by the magnitude of the goal vs non-goal difference. Goal possessions show lower DEI than non-goal possessions across all seven competitions in the expected direction. The effect reaches conventional significance in three competitions individually (Euro 2024: p\u0026thinsp;=\u0026thinsp;2.6 \u0026times; 10⁻⁴; Women's Euro 2022: p\u0026thinsp;=\u0026thinsp;0.003; Women's World Cup 2023: p\u0026thinsp;=\u0026thinsp;0.001) and in the pooled analysis (p\u0026thinsp;=\u0026thinsp;5.8 \u0026times; 10⁻⁷), partially confirming H11. The Women's Euro 2022 and Women's World Cup 2023 show the largest separation between goal and non-goal DEI distributions, while the Women's Euro 2025 shows the smallest effect (ns). The within-competition effect size is small (pooled rank-biserial r꜀\u003csub\u003eβ\u003c/sub\u003e = \u0026minus;0.042), and DEI is best interpreted as a composite structural index whose discriminative power is most robust in pooled analysis rather than within individual competitions; full per-competition results are reported in Supplementary Table S2.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec17\" class=\"Section2\"\u003e \u003ch2\u003e3.6 Logistic Regression: Cascade Features Beyond Spatial Position\u003c/h2\u003e \u003cp\u003eFigure \u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003e presents the standardised logistic regression coefficients from Model B (cascade features plus spatial controls). Distance to goal is, as expected, the dominant predictor (β = \u0026minus;0.459, p\u0026thinsp;\u0026lt;\u0026thinsp;10\u003csup\u003e\u0026minus;\u0026thinsp;300\u003c/sup\u003e), confirming that spatial proximity to the target is a necessary condition for goal threat. However, after controlling for distance to goal, attack damage, and DEI, cascade features retain statistically significant independent predictive value: first fracture strength (β = \u0026minus;0.126, p\u0026thinsp;=\u0026thinsp;4.66 \u0026times; 10\u003csup\u003e\u0026minus;\u0026thinsp;15\u003c/sup\u003e), resilience index (β = \u0026minus;0.083, p\u0026thinsp;=\u0026thinsp;0.002), and attack damage (β\u0026thinsp;=\u0026thinsp;0.052, p\u0026thinsp;=\u0026thinsp;0.001). The significance of first fracture strength in Model B is particularly important: it demonstrates that the structural fragility of the defensive bond at the point of first failure carries predictive information about goal outcomes independent of how far up the pitch the possession originates, directly addressing the potential confound that cascade metrics merely proxy for spatial position.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec18\" class=\"Section2\"\u003e \u003ch2\u003e3.7 Feature Correlation Structure\u003c/h2\u003e \u003cp\u003eFigure \u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003e displays the Spearman rank correlation matrix across all bond fracture and spatial features. Several structural patterns are notable. First fracture strength and weakness gap are nearly perfectly correlated (ρ\u0026thinsp;=\u0026thinsp;0.97), confirming their mathematical relationship and justifying the removal of weakness gap from the regression model to avoid collinearity. Coordination number and failure step are extremely highly correlated (ρ\u0026thinsp;=\u0026thinsp;0.98), reflecting the fact that longer possessions naturally involve both more bonds and more cascade steps. Attack damage and distance to goal are strongly inversely correlated (ρ = \u0026minus;0.85), as the damage function is defined primarily through spatial proximity to goal. Correlations with the goal outcome (rightmost column) are modest in absolute magnitude (range |ρ| = 0.03\u0026ndash;0.07), consistent with the known difficulty of predicting individual goal events from single-possession metrics in a sport characterised by low scoring rates and high variance.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e"},{"header":"4. Discussion","content":"\u003cp\u003eThe present study introduces and validates a bond fracture cascade framework for analysing defensive structural failure in football. Across 190732 possessions from seven elite competitions, we demonstrate that goal-scoring attacks are systematically associated with lower first-fracture bond strength, larger cascade avalanches, and greater local structural vulnerability in the defending network\u0026mdash;findings that are consistent across men's and women's football and across both club and international competition contexts.\u003c/p\u003e \u003cp\u003eThe most practically significant finding is the monotonic relationship between defensive coordination number and goal probability (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e, r\u0026thinsp;=\u0026thinsp;0.975). Under the bond mechanics interpretation, coordination number in the goal-possession context reflects the depth to which an attacking sequence has penetrated the defensive structure before the possession concludes. Possessions that activate more than thirty defensive bonds achieve goal rates nearly three times those of possessions activating fewer than five bonds. This suggests a cascade threshold phenomenon: once a possession has sufficiently disrupted the peripheral layers of defensive bonding, the probability of finding and exploiting a catastrophic fracture path increases markedly.\u003c/p\u003e \u003cp\u003eThe counterintuitive directions of H9 (higher resilience in goal possessions) and H10 (higher coordination number) are mechanistically interpretable within the cascade framework. Goal-scoring sequences are not simply faster disruptions of defensive networks; they tend to be longer, more sustained attacks that accumulate damage across a greater number of bonds before the cascade becomes irreversible. This 'slow burn' characteristic\u0026mdash;high failure step count, high coordination number, high resilience in the sense of surviving many partial fractures\u0026mdash;is precisely the signature of a systemic, progressive defensive breakdown rather than a lucky individual error. From a coaching perspective, this implies that vulnerability to conceding goals is not always expressed in the immediate structural properties of the defensive line at a single moment, but may be better captured by the trajectory of structural deterioration over the course of an extended possession.\u003c/p\u003e \u003cp\u003eThe logistic regression results (Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003e) address a fundamental question of confounding: are cascade metrics merely proxies for spatial proximity to goal? The persistence of first fracture strength (β = \u0026minus;0.126, p\u0026thinsp;=\u0026thinsp;4.66 \u0026times; 10\u003csup\u003e\u0026minus;\u0026thinsp;15\u003c/sup\u003e) and resilience index (β = \u0026minus;0.083, p\u0026thinsp;=\u0026thinsp;0.002) as significant predictors after controlling for distance to goal and DEI demonstrates that structural bond fragility provides genuine incremental predictive information beyond what is captured by position alone. This is consistent with the observation that many high-xG (expected goals) situations do not result in goals, and vice versa: the physical structure of defensive organisation at the moment of ball receipt carries information that spatial proximity metrics do not fully encode.\u003c/p\u003e \u003cp\u003eSeveral limitations warrant acknowledgement. The bond strength model depends on parameters (r₀, σ, D\u003csup\u003eMAX\u003c/sup\u003e) calibrated to the empirical distributions of this dataset; future work should assess parameter sensitivity and explore alternative potential functions. The analysis treats each possession as an independent event, whereas in practice, defensive fatigue and in-game tactical adjustments create temporal dependencies across possessions within a match. The modest absolute magnitudes of Spearman correlations with goal outcome (|ρ| \u0026le; 0.07) reflect the irreducibly stochastic nature of goal events in football, and caution against interpreting the framework as a precise goal probability predictor rather than a structural vulnerability indicator. Finally, the negative values of first fracture strength arise from bonds compressed or stretched well beyond equilibrium; while physically meaningful within the Morse framework, their interpretation in coaching applications should be accompanied by appropriate normalisation.\u003c/p\u003e \u003cp\u003eFuture extensions of this framework might incorporate temporal dynamics\u0026mdash;tracking how bond network properties evolve across successive events within a possession\u0026mdash;and explore machine learning approaches that jointly model structural and positional features. Application to real-time tracking data could enable live defensive fragility scores that inform in-match tactical adjustments.\u003c/p\u003e"},{"header":"5. Conclusions","content":"\u003cp\u003eWe have introduced a bond fracture cascade framework for modelling defensive structural failure in football, grounded in the physics of Morse-potential bond networks and avalanche phenomena. Applied to 190732 possessions from seven elite competitions, the framework yields robust structural signatures of goal-scoring attacks across men's and women's, club and international contexts: eleven hypotheses (H3\u0026ndash;H10, H12) are confirmed at p\u0026thinsp;\u0026lt;\u0026thinsp;10⁻⁶ in pooled analysis, with H11 (DEI) partially supported (pooled p\u0026thinsp;=\u0026thinsp;5.8 \u0026times; 10⁻⁷; significant in 3/7 competitions individually). The Defensive Engagement Index offers a physically motivated composite measure of defensive structural integrity, most informative in pooled analysis. Logistic regression confirms that cascade features carry predictive information beyond spatial proximity to goal. These results establish a physically grounded, interpretable foundation for quantitative analysis of defensive collapse in football, with direct applications in tactical analysis and performance monitoring.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e \u003ch2\u003eCompeting Interests\u003c/h2\u003e \u003cp\u003eThe authors declare no competing interests.\u003c/p\u003e \u003c/p\u003e\u003ch2\u003eFunding\u003c/h2\u003e \u003cp\u003eThis work was supported by the 2024 Guangdong Provincial Research Program on High-Quality Development of Youth Campus Football and School Physical Education (Grant No. 24SXZPT50).\u003c/p\u003e\u003ch2\u003eAuthor Contribution\u003c/h2\u003e\u003cp\u003eJiongzhang: Conceptualisation, methodology, software, formal analysis, writing \u0026ndash; original draft. Linzuo: Conceptualisation, supervision, writing \u0026ndash; review \u0026amp; editing.\u003c/p\u003e\u003ch2\u003eData Availability\u003c/h2\u003e\u003cp\u003eThe raw event tracking data (StatsBomb 360) used in this study are publicly available via the StatsBomb Open Data GitHub repository (https://github.com/statsbomb/open-data). The analysis code (Python scripts 01\u0026ndash;05) and the derived cascade feature datasets supporting the findings of this study are available from the corresponding author upon reasonable request.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eDuarte, R., Ara\u0026uacute;jo, D., Correia, V. \u0026amp; Davids, K. Sports teams as superorganisms: Implications of sociobiological models of behaviour for research and practice in team sports performance analysis. Sports Med. 42, 633\u0026ndash;642 (2012).\u003c/li\u003e\n\u003cli\u003eGudmundsson, J. \u0026amp; Horton, M. Spatio-temporal analysis of team sports. ACM Comput. Surv. 50, 1\u0026ndash;34 (2017).\u003c/li\u003e\n\u003cli\u003eMoura, F. A. et al. Analysis of football game-related statistics using multivariate analysis. J. Hum. Kinet. 36, 183\u0026ndash;194 (2013).\u003c/li\u003e\n\u003cli\u003eFernandez, J. \u0026amp; Bornn, L. Wide open spaces: A statistical technique for measuring space creation in professional soccer. MIT Sloan Sports Analytics Conference (2018).\u003c/li\u003e\n\u003cli\u003ePena, J. L. \u0026amp; Touchette, H. A network theory analysis of football strategies. arXiv:1206.6904 (2012).\u003c/li\u003e\n\u003cli\u003eCotta, C., Mora, A. M., Merelo, J. J. \u0026amp; Merelo-Molina, C. A network analysis of the 2010 FIFA World Cup champion team play. J. Syst. Sci. Complex. 26, 21\u0026ndash;42 (2013).\u003c/li\u003e\n\u003cli\u003eBuld\u0026uacute;, J. M. et al. Defining a historic football team: Using Network Science to study Guardiola\u0026apos;s F.C. Barcelona. Sci. Rep. 9, 13602 (2019).\u003c/li\u003e\n\u003cli\u003eDuarte, R. et al. Basketball teams as dynamic networks: Implications for team training and practice. J. Sports Sci. Med. 15, 428\u0026ndash;434 (2017).\u003c/li\u003e\n\u003cli\u003eRein, R. \u0026amp; Memmert, D. Big data and tactical analysis in elite soccer: future challenges and opportunities for sports science. SpringerPlus 5, 1410 (2016).\u003c/li\u003e\n\u003cli\u003eSornette, D. Critical Phenomena in Natural Sciences. Springer, Berlin (2006).\u003c/li\u003e\n\u003cli\u003eZapperi, S., Vespignani, A. \u0026amp; Stanley, H. E. Plasticity and avalanche behaviour in microfracturing phenomena. Nature 388, 658\u0026ndash;660 (1997).\u003c/li\u003e\n\u003cli\u003eJaeger, H. M. \u0026amp; Nagel, S. R. Physics of the granular state. Science 255, 1523\u0026ndash;1531 (1992).\u003c/li\u003e\n\u003cli\u003eAlava, M. J., Nukala, P. K. V. V. \u0026amp; Zapperi, S. Statistical models of fracture. Adv. Phys. 55, 349\u0026ndash;476 (2006).\u003c/li\u003e\n\u003cli\u003eBuldyrev, S. V., Parshani, R., Paul, G., Stanley, H. E. \u0026amp; Havlin, S. Catastrophic cascade of failures in interdependent networks. Nature 464, 1025\u0026ndash;1028 (2010).\u003c/li\u003e\n\u003cli\u003eGao, J., Buldyrev, S. V., Stanley, H. E. \u0026amp; Havlin, S. Networks formed from interdependent networks. Nat. Phys. 8, 40\u0026ndash;48 (2012).\u003c/li\u003e\n\u003cli\u003eMorse, P. M. Diatomic molecules according to the wave mechanics. II. Vibrational levels. Phys. Rev. 34, 57\u0026ndash;64 (1929).\u003c/li\u003e\n\u003cli\u003eBartlett, R., Button, C., Robins, M., Dutt-Mazumder, A. \u0026amp; Kennedy, G. Analysing team coordination patterns from player movement trajectories in soccer. J. Sports Sci. 30, 1577\u0026ndash;1587 (2012).\u003c/li\u003e\n\u003cli\u003eLago-Pe\u0026ntilde;as, C. \u0026amp; Lago-Ballesteros, J. Game location and team quality effects on performance profiles in professional soccer. J. Sports Sci. Med. 10, 465\u0026ndash;471 (2011).\u003c/li\u003e\n\u003cli\u003eGon\u0026ccedil;alves, B. et al. Entropy measures reveal collective tactical behaviours in association football. Int. J. Sports Physiol. Perform. 12, 1164\u0026ndash;1172 (2017).\u003c/li\u003e\n\u003cli\u003eGrund, T. U. Network structure and team performance: The case of English Premier League soccer teams. Soc. Networks 34, 682\u0026ndash;690 (2012).\u003c/li\u003e\n\u003cli\u003eNarizuka, T. \u0026amp; Yamamoto, K. Clustering algorithm for formations in football games. Sci. Rep. 9, 1\u0026ndash;8 (2019).\u003c/li\u003e\n\u003cli\u003eDuarte, R. et al. Competing with teammates and opponents in complex adaptive systems. J. Hum. Kinet. 48, 5\u0026ndash;16 (2015).\u003c/li\u003e\n\u003cli\u003eVilar, L., Ara\u0026uacute;jo, D., Davids, K. \u0026amp; Bar-Yam, Y. Science of winning soccer: emergent pattern-forming dynamics in association football. J. Syst. Sci. Complex. 26, 73\u0026ndash;84 (2013).\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"scientific-reports","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"scirep","sideBox":"Learn more about [Scientific Reports](http://www.nature.com/srep/)","snPcode":"","submissionUrl":"","title":"Scientific Reports","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Scientific Reports","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"football analytics, defensive network, bond fracture cascade, Morse potential, statistical mechanics, Defensive Engagement Index","lastPublishedDoi":"10.21203/rs.3.rs-9397572/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-9397572/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eCollective defensive organisation in football is a key determinant of match outcome, yet existing analytical approaches lack a principled physical framework for capturing defensive structural fragility at the moment of goal concession. Here we introduce a bond fracture cascade framework inspired by condensed-matter physics, in which spatial relationships between defending players are modelled as a network of interacting bonds governed by a Morse-type potential. Applying this framework to 190732 possessions from seven elite competitions\u0026mdash;including the FIFA World Cup 2022, UEFA Euro 2020 and 2024, the Women's World Cup 2023, Women's Euro 2022 and 2025, and the Bundesliga 2023/24\u0026mdash;we test twelve hypotheses linking bond-fracture metrics to goal outcomes. Goal-scoring possessions exhibit lower first-fracture bond strength (\u0026minus;\u0026thinsp;5.90 vs\u0026thinsp;\u0026minus;\u0026thinsp;2.73 a.u.; p\u0026thinsp;=\u0026thinsp;1.24 \u0026times; 10⁻\u0026sup1;\u0026sup2;⁵), larger avalanche size (3.77 vs 2.09; p\u0026thinsp;=\u0026thinsp;2.01 \u0026times; 10⁻\u0026sup1;\u0026sup1;⁶), and larger weakness gap (\u0026minus;\u0026thinsp;6.08 vs\u0026thinsp;\u0026minus;\u0026thinsp;3.27 a.u.; p\u0026thinsp;=\u0026thinsp;1.36 \u0026times; 10⁻\u0026sup1;\u0026sup3;⁵). Goal probability rises monotonically with defensive coordination number, from 1.89% to 5.26% across bond-count bins (r\u0026thinsp;=\u0026thinsp;0.975, p\u0026thinsp;=\u0026thinsp;0.0002). A Defensive Engagement Index (DEI) derived from the bond network shows a consistent directional difference between goal and non-goal possessions across all seven competitions (pooled p\u0026thinsp;=\u0026thinsp;5.8 \u0026times; 10⁻⁷). Logistic regression confirms cascade features retain independent predictive value beyond proximity to goal (distance to goal β = \u0026minus;0.459; first-fracture strength β = \u0026minus;0.126, p\u0026thinsp;\u0026lt;\u0026thinsp;0.001). These results establish a physically principled, competition-agnostic signature of defensive structural collapse in football.\u003c/p\u003e","manuscriptTitle":"Bond Fracture Cascade Analysis of Defensive Networks in Elite Football: A Statistical Mechanics Approach","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2026-04-30 12:44:16","doi":"10.21203/rs.3.rs-9397572/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Revision requested","date":"2026-05-18T12:02:01+00:00","index":"","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2026-05-12T12:16:31+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"138638684935286469874105086894835196199","date":"2026-05-10T10:33:09+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2026-04-22T09:06:23+00:00","index":"","fulltext":""},{"type":"editorInvited","content":"","date":"2026-04-20T18:15:06+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2026-04-16T04:47:44+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2026-04-16T04:47:02+00:00","index":"","fulltext":""},{"type":"submitted","content":"Scientific Reports","date":"2026-04-13T01:37:36+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"scientific-reports","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"scirep","sideBox":"Learn more about [Scientific Reports](http://www.nature.com/srep/)","snPcode":"","submissionUrl":"","title":"Scientific Reports","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Scientific Reports","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"25d1407c-d885-4a29-99b0-9dbcc9d3fea4","owner":[],"postedDate":"April 30th, 2026","published":true,"recentEditorialEvents":[{"type":"decision","content":"Revision requested","date":"2026-05-18T12:02:01+00:00","index":"","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2026-05-12T12:16:31+00:00","index":69,"fulltext":""},{"type":"reviewerAgreed","content":"138638684935286469874105086894835196199","date":"2026-05-10T10:33:09+00:00","index":64,"fulltext":""}],"rejectedJournal":[],"revision":"","amendment":"","status":"in-revision","subjectAreas":[{"id":67173128,"name":"Physical sciences/Mathematics and computing"},{"id":67173129,"name":"Physical sciences/Physics"}],"tags":[],"updatedAt":"2026-05-18T12:10:25+00:00","versionOfRecord":[],"versionCreatedAt":"2026-04-30 12:44:16","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-9397572","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-9397572","identity":"rs-9397572","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}