An Efficient Memory Gradient Method for Extreme M-Eigenvalues of Elastic type Tensors

Zhuolin Du, Xudong LiarXiv 2026 · arXiv

Abstract

M-eigenvalues of fourth order hierarchically symmetric tensors play a significant role in nonlinear elastic material analysis and quantum entanglement problems. This paper focuses on computing extreme M-eigenvalues for such tensors. To achieve this, we first reformulate the M-eigenvalue problem as a sequence of unconstrained optimization problems by introducing a shift parameter. Subsequently, we develop a memory gradient method specifically designed to approximate these extreme M-eigenvalues. Under this framework, we establish the global convergence of the proposed method. Finally, comprehensive numerical experiments demonstrate the efficacy and stability of our approach.

Introduction

Fourth-order hierarchically symmetric tensors, often referred to as "elastic-type" tensors, are fundamental in modeling material deformation under stress and analyzing quantum entanglement. A critical challenge in these domains is the computation of M-eigenvalues, which are essential for determining material stability and positive definiteness. As tensor dimensions increase, traditional computational methods often suffer from high costs or poor convergence. This paper introduces a novel Memory Gradient Method (MGM) that leverages historical gradient information to accelerate convergence and improve numerical stability compared to standard first-order methods.

Methodology

The proposed approach addresses the M-eigenvalue problem through two primary innovations:

1. Unconstrained Reformulation and Adaptive Shifting

The constrained M-eigenvalue problem is transformed into a sequence of unconstrained optimization problems. To ensure the algorithm converges to a meaningful, non-zero solution (the largest M-eigenvalue) even when eigenvalues are non-positive, an adaptive shift parameter ($t$) is introduced. This parameter effectively shifts the tensor's spectrum, preventing the algorithm from converging to trivial points.

2. Memory Gradient Method (MGM)

Unlike steepest descent methods that only use the current gradient, the MGM utilizes a weighted combination of the current gradient and $N$ previous search directions. This "memory" allows the algorithm to capture the curvature of the optimization landscape—similar to quasi-Newton methods like BFGS—but without the prohibitive memory and computational overhead of calculating Hessian matrices.

Algorithm 1: The proposed iterative framework for computing extreme M-eigenvalues.

To maintain numerical stability, the algorithm incorporates a rescaling step (ensuring $||x|| = ||y||$). This ensures that the mathematical sequences remain bounded, a prerequisite for proving global convergence as outlined in Algorithm 1.

Numerical Experiments

The efficacy of the MGM was evaluated across several benchmarks, including small-scale structured tensors, uniformly random tensors, and large-scale random tensors with dimensions up to $50 \times 60 \times 50 \times 60$.

Key Results

The MGM (variants MGM-1 and MGM-2) was compared against established benchmarks like SS-HOPM and SIPM:

  • Efficiency: MGM-1 consistently required fewer iterations and less CPU time. In specific benchmarks, MGM-1 converged in 5 iterations compared to 15 for SS-HOPM.
  • Scalability: The performance advantage of MGM became more pronounced as tensor dimensions increased, highlighting its suitability for high-dimensional problems.
  • Robustness: Experiments with multiple random starts showed high stability, with the method being relatively insensitive to initial conditions.

Conclusion

The Memory Gradient Method provides a robust and efficient framework for computing extreme M-eigenvalues of fourth-order tensors. While $N=3$ was identified as an empirical "sweet spot" for memory length, future research could explore theoretical methods for determining optimal $N$. Additionally, extending this framework to higher-order tensors (e.g., 6th or 8th order) remains a promising avenue for complex physics simulations.

Summary updated

Abstract

M-eigenvalues of fourth order hierarchically symmetric tensors play a significant role in nonlinear elastic material analysis and quantum entanglement problems. This paper focuses on computing extreme M-eigenvalues for such tensors. To achieve this, we first reformulate the M-eigenvalue problem as a sequence of unconstrained optimization problems by introducing a shift parameter. Subsequently, we develop a memory gradient method specifically designed to approximate these extreme M-eigenvalues. Under this framework, we establish the global convergence of the proposed method. Finally, comprehensive numerical experiments demonstrate the efficacy and stability of our approach.
BibTeX
@misc{du2026efficient,
  author = {Zhuolin Du and Xudong Li},
  title = {{An Efficient Memory Gradient Method for Extreme M-Eigenvalues of Elastic type Tensors}},
  year = {2026},
  eprint = {2602.01152},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2602.01152},
  howpublished = {arXiv},
}
Figure