Abstract:We prove Central Limit Theorem for non-stationary random products of $SL(2, \mathbb{R})$ matrices, generalizing the classical results by Le Page and Tutubalin that were obtained in the case of iid random matrix products.
What problem does this paper attempt to address?
The problem that this paper attempts to solve is: **Prove the central limit theorem (CLT) for non - stationary random SL(2,R) matrix products**. Specifically, the authors hope to generalize the classical central limit theorem from products of independent and identically distributed (iid) random matrices to the non - stationary case.
### Problem Background
1. **Review of Classical Results**:
- For a sequence of independent and identically distributed (iid) random variables \(\{\xi_n\}\), if they have a finite expectation \(a\) and a finite variance \(\sigma^2\), then the strong law of large numbers (LLN) shows that:
\[
\frac{1}{n}\sum_{i = 1}^{n}\xi_i\to a\quad\text{(almost surely)}
\]
The central limit theorem (CLT) shows that:
\[
\frac{\sum_{i = 1}^{n}\xi_i - na}{\sqrt{n}\sigma}\xrightarrow{d}N(0,1)
\]
- For products of random matrices, Furstenberg and Kesten have already proven the law of large numbers in multiplicative form, and Furstenberg's theorem guarantees the positivity of the Lyapunov exponent.
2. **Existing Work**:
- For products of independent and identically distributed random matrices, Le Page, Tutubalin et al. have already proven the corresponding central limit theorem.
- Benoist and Quint further optimized the moment conditions, making these results more general.
### Research Objectives
- **Central Limit Theorem in the Non - stationary Case**: The main contribution of this paper is to prove the central limit theorem for non - stationary random SL(2,R) matrix products. Specifically, for a set of probability measures \(K\) satisfying certain conditions and the corresponding sequence of random matrices \(\{A_i\}\), the authors prove that:
\[
\frac{\log \|T_n\| - L_n}{\sqrt{\text{Var}(\log \|T_n\|)}}\xrightarrow{d}N(0,1)
\]
where \(T_n = A_nA_{n - 1}\cdots A_1\), and the convergence is uniform with respect to the choice of the sequence of measures.
### Main Results
- **Theorem 1.4**: Under certain conditions (for example, there exists some \(\gamma\geq9\) and a constant \(M > 0\) such that for all \(\mu\in K\), \(E_\mu(\log \|A\|)^\gamma < M\)), the random variable
\[
\frac{\log \|T_n\| - L_n}{\sqrt{\text{Var}(\log \|T_n\|)}}
\]
converges to the standard normal distribution \(N(0,1)\).
- **Variance Estimation**: There also exist constants \(C_1,C_2 > 0\) and an index \(n_0\) such that for all \(n\geq n_0\) and all \(\mu_1,\ldots,\mu_n\in K\)
\[
C_1n\leq\text{Var}(\log \|T_n\|)\leq C_2n
\]
### Significance
This result provides a theoretical basis for the central limit theorem in non - stationary random dynamical systems, further expanding the application scope of classical probability theory and random matrix theory.