Eigenvalues and Eigenvectors of Tau Matrices with Applications to Markov Processes and Economics

Sven-Erik Ekström, Carlo Garoni, Adam Jozefiak, Jesse PerlaarXiv 2020 · arXiv

Introduction

In many scientific fields, systems are represented using tridiagonal matrices—matrices where only the main diagonal and the two adjacent diagonals have non-zero entries. A specific subset of these, known as the $\tau$ algebra, is well-understood. However, real-world applications often involve "boundary effects" (like reflecting walls in a random walk or specific policy constraints in economics) that modify the very first and last entries of these matrices.

These modifications lead to the $\tau_{\epsilon, \phi}$ algebra. While specific instances of these matrices had been solved, a general, unified spectral theory was missing. Specifically, researchers lacked a clear way to handle "outliers"—eigenvalues that fall outside the standard range. These outliers are not just mathematical curiosities; in Markov processes, they represent the "spectral gap" that determines how fast a system reaches equilibrium and what its final steady state looks like.

The structure of the generalized tridiagonal matrix $T_{n,\epsilon,\phi}$ showing boundary modifications $\epsilon$ and $\phi$.

Methodology

The authors employ a mix of classical linear algebra and the theory of linear difference equations to provide a comprehensive spectral analysis. The eigenvalue problem $Tv = \lambda v$ is treated as a second-order recurrence relation, allowing for the derivation of characteristic equations based on trigonometric functions for standard eigenvalues and hyperbolic functions for outliers.

Key methodological components include:

  • Asymptotic Analysis: Deriving where outliers "land" as the matrix size $n$ grows toward infinity, providing simple formulas for complex matrix behavior.
  • Kronecker Products: Extending 1D models to multidimensional ones (e.g., multi-asset portfolios) using tensor products.
  • Numerical Validation: Theoretical results were validated using the Julia programming language with 1024-bit high-precision arithmetic to ensure asymptotic estimates hold in practice.
Numerical validation comparing theoretical asymptotic predictions against computed outlier values for increasing matrix size $n$.

Key Insights and Results

The paper demonstrates that if the boundary parameters $|\epsilon|$ or $|\phi|$ are greater than 1, the matrix produces outliers that dominate long-term system behavior, such as the stabilization rate of wealth inequality. A significant finding is the $\epsilon\phi = 1$ case, where the eigendecomposition simplifies, making it ideal for modeling reflected diffusion processes.

Major Findings:

  • Unified Spectral Theory: A complete set of equations for eigenvalues and eigenvectors across five distinct cases covering the entire real spectrum.
  • Steady-State Formulas: Explicit formulas for stationary distributions in multidimensional reflected diffusion processes.
  • Convergence Rates: Derivation of the "second eigenvalue" ($\nu_1$) as a precise measure of convergence speed.
  • Economic Sensitivities: Analytical "shortcuts" for computing derivatives of wealth mean and variance with respect to drift ($\mu$) and volatility ($\sigma^2$), facilitating comparative statics in economic modeling.

Limitations and Future Work

Current models assume constant drift and variance. To model "top-heavy" income inequality (Pareto tails), future research would need to incorporate state-dependent coefficients. Additionally, while this work focuses on symmetric matrices, extending these results to non-symmetric generators of general Markov chains remains an open challenge. The authors also suggest that the $\tau_{\epsilon, \phi}$ algebra appears in isogeometric analysis and finite element methods, representing a fruitful area for future spectral research.

Summary updated

BibTeX
@misc{ekstrom2020eigenvalues,
  author = {Sven-Erik Ekström and Carlo Garoni and Adam Jozefiak and Jesse Perla},
  title = {{Eigenvalues and Eigenvectors of Tau Matrices with Applications to Markov Processes and Economics}},
  year = {2020},
  eprint = {2008.10554},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2008.10554},
}
Figure