A convex optimization approach to the Lyapunov exponents

Christoph Kawan

2023-10-02

Abstract:The aim of this paper is to shed more light on some recent ideas about Lyapunov exponents and clarify the formal structures behind these ideas. In particular, we show that the vector of (averaged) Lyapunov exponents of a smooth measure-preserving dynamical system can be regarded as the solution of a vector-valued optimization problem on a space of Riemannian metrics. This result was first formulated and proved by Jairo Bochi in the language of linear cocycles and their conjugacies. We go a step further and prove that the optimization problem is geodesically convex with respect to the $L^2$-metric. Moreover, we derive some consequences of this fact.

Dynamical Systems

What problem does this paper attempt to address?

This paper aims to deeply explore some of the latest views on Lyapunov exponents and clarify the formal structures behind these views. Specifically, the author shows that the (average) Lyapunov exponent vector of a smooth measure - preserving dynamical system can be regarded as the solution to a vector - valued optimization problem defined on a Riemannian metric space. This result was initially proposed and proved by Jairo Bochi when studying linear homomorphisms and their conjugates. This paper further proves that this optimization problem is geodesically convex with respect to the $L^2$ metric and derives some consequences of this fact.

### Problems the paper attempts to solve

1. **Understanding of Lyapunov exponents**:
- The paper attempts to gain a deeper understanding of the concept of Lyapunov exponents and their role in dynamical systems. Lyapunov exponents are important indicators of the stability of dynamical systems and are widely used in the study of chaos theory and other complex systems.

2. **Geometric structure of the optimization problem**:
- By introducing a vector - valued optimization problem on a Riemannian metric space, the author links Lyapunov exponents to optimization theory. Specifically, they prove that the Lyapunov exponent vector can be regarded as the solution to an optimization problem that is geodesically convex with respect to the $L^2$ metric.

3. **Properties of the optimization problem**:
- The paper explores the properties of this optimization problem, including its convexity, existence, and computational methods. The author points out that although this optimization problem has certain advantages, there are also some challenges, such as the possible non - existence of a minimum, the infinite - dimensional nature of the problem, and non - smoothness.

### Main contributions

1. **Optimization representation of Lyapunov exponents**:
- The author proves that the Lyapunov exponent vector can be regarded as the solution to a vector - valued optimization problem, which provides a new perspective for understanding and computing Lyapunov exponents.

2. **Geodesic convexity**:
- The author further proves that this optimization problem is geodesically convex with respect to the $L^2$ metric, which provides a theoretical basis for the properties of the optimization problem.

3. **Geometric background of the optimization problem**:
- The paper discusses in detail the geometric structure of the Riemannian metric space, especially the properties with respect to the $L^2$ metric, providing background knowledge for further research on the optimization problem.

### Formula summary

- **Definition of Lyapunov exponents**:
\[
\lambda(x, v) := \lim_{n \to \infty} \frac{1}{n} \log \| d f^n_x v \|
\]
where $\| \cdot \|$ represents the norm induced by the given Riemannian metric.

- **Optimization representation of Lyapunov vectors**:
\[
\vec{\lambda}(f) = \inf_{g \in M_0} \int \vec{\sigma}_g(d f_x) d\mu(x)
\]
where $\vec{\sigma}_g(d f_x)$ represents the singular value vector of $d f_x$, and $M_0$ is the space of continuous Riemannian metrics.

- **Geodesic convexity**:
\[
\vec{S}_{f,\mu}(g) = \int \vec{\sigma}_g(d f_x) d\mu(x)
\]
The author proves that $\vec{S}_{f,\mu}$ is geodesically convex with respect to the $L^2$ metric.

### Conclusion

This paper provides new tools and methods for the study of Lyapunov exponents by introducing the perspective of optimization problems. These results not only deepen the understanding of Lyapunov exponents but also lay the foundation for further research in related fields.