Eigenvector Geometry as a Universal Amplifier of Heavy-Tailed Fluctuations in Random Multiplicative Systems

Virgile Troude, Didier Sornette · 2025

Read Eigenvector Geometry as a Universal Amplifier of Heavy-Tailed Fluctuations in Random Multiplicative Systems by Virgile Troude, Didier Sornette (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2510-21755.txt.

Introduction

Virgile Troude and Didier Sornette find a new engine for the heavy tails and power laws that haunt physics, biology, and social systems. The classic explanation, Kesten processes, multiplies a quantity by random factors and adds noise, producing power laws when growth briefly turns supercritical and excursions run away before fading.

In several dimensions, they uncover a different lever: the eigenvectors of random non-normal matrices lean on each other instead of standing perpendicular, and that leaning amplifies fluctuations at every step. The amplification effectively raises the growth exponent and fattens the tail, and it strengthens with dimension until it dominates the classic mechanism. Polymer chains stretching in turbulent flow, where velocity gradients intermittently yank molecules long, illustrate the effect in a real system.

Catalog record

CREATORVirgile Troude, Didier Sornette
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE34 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2510-21755

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Eigenvector Geometry as a Universal Amplifier of Heavy-Tailed Fluctuations in Random Multiplicative Systems Virgile Troude, Didier Sornette Abstract Heavy-tailed fluctuations and power-law distributions pervade physics, biology, and the social sciences, with numerous mechanisms proposed for their emergence. Kesten processes, which are multiplicative stochastic recursions with additive noise or reinjection, provide a canonical explanation, where power-law tails arise from transient supercritical excursions as eigenvalues intermittently cross the stability boundary. Here we uncover a distinct and more general mechanism in multidimensional systems: non-normal eigenvector amplification . In random non-normal matrices, the non-orthogonality of eigenvectors, quantified by the condition number κ \kappa , induces transient growth that increases the effective Lyapunov exponent γ ≃ γ 0 + ⟨ ln ⁡ κ ⟩ \gamma\simeq\gamma_{0}+\langle\ln\kappa\rangle and lowers the tail exponent α ≃ − 2 γ / σ κ 2 \alpha\simeq-2\gamma/\sigma_{\kappa}^{2} , where σ κ 2 \sigma_{\kappa}^{2} is the variance of ln ⁡ κ \ln\kappa . As the system dimension N N grows, κ \kappa typically increases proportionally, making non-normal amplification the dominant source of scale-free behavior. We illustrate this mechanism in two representative systems: (i) polymer stretching in turbulent flows, where intermittent extensions arise from eigenvector amplification of velocity gradients and (ii) financial return distributions, where extending one-dimensional GARCH/Kesten processes to a multidimensional setting yields a collective origin for heavy-tailed market fluctuations and explains their near-universal exponents across assets. Heavy-tailed distributions are a hallmark of complex systems, arising in physics, finance, economics, biology, and many other domains. A canonical mechanism for their emergence is provided by the class of stochastic recursions studied by Kesten Kesten 1973 , where the interplay between multiplicative growth and additive noise naturally produces stationary distributions with power-law tails. Such Kesten processes have since become a cornerstone in the theoretical understanding of scale-free phenomena, with broad applications ranging from physics de Calan et al. 1985 ; Takayasu et al. 1997 ; Blank and Solomon 2000 , biology Statman et al. 2014 , to financial markets Lux and Sornette 2002 ; Lux 2008 ; Gabaix 2009 ; Derman et al. 2010 , and to models of wealth and income distribution Sornette 1998 ; Benhabib et al. 2016 . The classical understanding attributes the emergence of power laws in Kesten-type processes to spectral supercriticality , i.e., episodes in which eigenvalues of the random multiplicative operator transiently cross the unit circle. In this view, power law tails reflect rare bursts of exponential growth balanced by global stability on average. In this paper, we uncover a complementary and more general mechanism in multidimensional processes: non-normal eigenvector amplification . When random matrices are non-normal (not unitarily diagonalizable), transient growth can occur even when all eigenvalues lie strictly within the unit circle. This effect, arising from the non-orthogonality of eigenvectors, provides a distinct and generic route to heavy-tailed stationary distributions. Our analysis shows that non-normality not only lowers the tail exponent, thereby producing fatter tails, but can also shift the effective Lyapunov exponent, destabilizing the system in a way fundamentally different from spectral criticality. The power law mechanism rooted in the generic geometry of eigenvectors in high-dimensional random systems provides a unified explanation for heavy-tailed distributions in systems governed by multiplicative interactions, from turbulent flows to wealth and income dynamics. By identifying non-normal amplification as a fundamental source of heavy tails, our work extends the theoretical scope of Kesten processes and clarifies their relevance to the statistics of complex systems. Our work builds on a broad literature that first recognized the dynamical significance of non-normal operators. Pioneering studies in hydrodynamic stability and atmospheric dynamics by Farrell and Ioannou Farrell 1988 ; Farrell 1989 ; Farrell and Ioannou 1996a ; Farrell and Ioannou 1996b ; Farrell and Ioannou 2003 ; Farrell and Ioannou 2007 and in fluid mechanics by Trefethen and collaborators Trefethen et al. 1993 ; Embree and Trefethen 2005 showed that non-normality can induce large transient amplifications even when all eigenvalues indicate linear stability. Related ideas were later extended to other fields, such as subcritical magnetic dynamos Fedotov 2003 . Here we generalize this perspective beyond hydrodynamics and linear response, showing that non-normal eigenvector amplification constitutes a universal route to heavy-tailed fluctuations in high-dimensional stochastic systems. We consider the N -dimensional Kesten process \mathbf{x}_{t+1}=\mathbf{A}_{t}\mathbf{x}_{t}+\boldsymbol{\eta}_{t}, (1) where (\mathbf{A}_{t}) is an i.i.d. sequence of random matrices and \boldsymbol{\eta}_{t} is an additive noise term. Their long-term behavior is controlled by \pi_{t}=\|\boldsymbol{\Pi}_{t}\|\quad{\rm with}\quad\boldsymbol{\Pi}_{t}=\prod_{s=1}^{t}\mathbf{A}_{s}~, (2) where \Pi_{t} denotes the product of matrices up to time t and \pi_{t} is its L_{2} -norm. The stability of the process is characterized by the Lyapunov exponent \gamma , defined as \gamma:=\lim_{t\to\infty}\frac{1}{t}\,\mathbb{E}\!\left[\ln\pi_{t}\right]. (3) A negative exponent ( \gamma<0 ) implies asymptotic stability, with convergence to a stationary distribution, whereas a positive exponent ( \gamma>0 ) indicates instability, with the system diverging and growing exponentially in time. The tail exponent \alpha quantifies the heaviness of the tail of the stationary distribution, namely \mathbb{P}[\mathbf{n}\cdot\mathbf{x}_{t}>x_{n}]\sim x_{n}^{-\alpha},\quad x_{n}\to\infty~, (4) where \mathbf{n}\cdot\mathbf{x}_{t} denotes any projection (the same asymptotics hold for the L_{2} -norm). Consider the function \phi(\alpha):=\lim_{t\to\infty}\frac{1}{t}\,\ln\mathbb{E}\!\left[\pi_{t}^{\alpha}\right]~, (5) which is convex, satisfies \phi(0)=0 , and has derivative \phi^{\prime}(0)=\gamma . For \gamma<0 , the convexity of \phi(\alpha) ensures that there exists a unique \alpha>0 solving \phi(\alpha)=0 , which determines the tail exponent of the stationary distribution Furstenberg and Kesten 1960 ; Kesten 1973 . For \gamma\geq 0 , no positive solution exists and the process fails to admit a stationary heavy-tailed regime. Normal Kesten processes: spectral criticality . When the random matrices \mathbf{A}_{t} are normal (i.e. unitarily diagonalizable), they can be written as \mathbf{A}_{t}=\mathbf{U}_{t}\boldsymbol{\Lambda}_{t}\mathbf{U}_{t}^{\dagger} with \mathbf{U}_{t} unitary and \boldsymbol{\Lambda}_{t} diagonal. Then the product norm has an upper bound given by \pi_{t}\leq\prod_{s=1}^{t}\|\mathbf{A}_{s}\|=\prod_{s=1}^{t}~\rho_{s} , since, for normal matrices, the L_{2} -norm \|\mathbf{A}_{t}\| of \mathbf{A}_{t} is given by its spectral radius \rho_{t} . Using the monotonicity of the logarithm and the convexity of \phi(\alpha) ( 5 ), we obtain the following bounds for the Lyapunov and tail exponents \gamma\leq\ln\rho,\qquad\alpha\geq-\frac{2\ln\rho}{\sigma_{\rho}^{2}}~~{\rm for}~\ln\rho<0~, (6) where \ln\rho and \sigma_{\rho}^{2} are respectively the expected value and variance of the logarithm of the spectral radius. The upper bound for \gamma and the lower bound for \alpha are reached for the one-dimension case for which the Kesten process ( 1 ) reduces to x_{t+1}=\rho_{t}x_{t}+\eta_{t} . When the multiplicative noise is lognormally distributed, \ln\rho_{t}\sim\mathcal{N}(\ln\rho,\sigma_{\rho}^{2}) , expression ( 3 ) gives \gamma=\ln\rho and the solution of the equation \phi(\alpha)=0 is exactly \alpha=-\frac{2\ln\rho}{\sigma_{\rho}^{2}}~,~~~{\rm for}~\ln\rho<0~. (7) Hence, the one-dimensional Kesten process represents the worst-case scenario of the multidimensional Kesten process with normal matrices since the bounds ( 6 ) are exactly attained. Intuitively, the power law tail originates from the fact that sequences of successive \rho_{t}>1 , which generate transient exponential growth, have an exponentially small probability to occur as a function of their durations, so that the stationary distribution reflects a balance between rare but strong amplification bursts and global stability enforced by \ln\rho<0 Sornette 2006 . In other words, the heavy tails arise from eigenvalues stochastically crossing the unit circle, placing the system for brief periods in a spectrally supercritical state. Additional details and general proofs are presented in the companion paper Troude 2025 . Non-normal Kesten processes: eigenvector amplification . When the random matrices \mathbf{A}_{t} are non-normal , i.e. not unitarily diagonalizable, an additional amplification mechanism appears. Writing \mathbf{A}_{t}=\mathbf{P}_{t}\boldsymbol{\Lambda}_{t}\mathbf{P}_{t}^{-1} , where \boldsymbol{\Lambda}_{t}

Read the complete text