Spectral norm bounds for block Markov chain random matrices

Jaron Sanders,Albert Senen-Cerda

DOI: https://doi.org/10.48550/arXiv.2111.06201

2021-11-11

Abstract:This paper quantifies the asymptotic order of the largest singular value of a centered random matrix built from the path of a Block Markov Chain (BMC). In a BMC there are $n$ labeled states, each state is associated to one of $K$ clusters, and the probability of a jump depends only on the clusters of the origin and destination. Given a path $X_0, X_1, \ldots, X_{T_n}$ started from equilibrium, we construct a random matrix $\hat{N}$ that records the number of transitions between each pair of states. We prove that if $\omega(n) = T_n = o(n^2)$, then $\| \hat{N} - \mathbb{E}[\hat{N}] \| = \Omega_{\mathbb{P}}(\sqrt{T_n/n})$. We also prove that if $T_n = \Omega(n \ln{n})$, then $\| \hat{N} - \mathbb{E}[\hat{N}] \| = O_{\mathbb{P}}(\sqrt{T_n/n})$ as $n \to \infty$; and if $T_n = \omega(n)$, a sparser regime, then $\| \hat{N}_\Gamma - \mathbb{E}[\hat{N}] \| = O_{\mathbb{P}}(\sqrt{T_n/n})$. Here, $\hat{N}_{\Gamma}$ is a regularization that zeroes out entries corresponding to jumps to and from most-often visited states. Together this establishes that the order is $\Theta_{\mathbb{P}}(\sqrt{T_n/n})$ for BMCs.

Probability

What problem does this paper attempt to address?

The problem that this paper attempts to solve is to quantify the asymptotic order of the largest singular value of the central random matrix constructed from the Block Markov Chain (BMC) path. Specifically, the paper focuses on constructing a random matrix $\hat{N}$ that records the number of transitions between each pair of states, given a BMC path $X_0, X_1, \ldots, X_{T_n}$ of length $T_n$ starting from the equilibrium state. The main objective of the paper is to prove the asymptotic behavior of the spectral norm $\|\hat{N} - E[\hat{N}]\|$ of the matrix $\hat{N} - E[\hat{N}]$ when the relationship between $T_n$ and $n$ satisfies different conditions.

Several key results are presented in the paper:
1. **Lower bound in the sparse case**: If $T_n = o(n^2)$, then $\sigma_1(\hat{N} - E[\hat{N}]) = \Omega_P(\sqrt{T_n/n})$.
2. **Upper bound in the dense case**: If $T_n = \Omega(n \ln n)$, then $\sigma_1(\hat{N} - E[\hat{N}]) = O_P(\sqrt{T_n/n})$.
3. **Upper bound in the sparser case**: If $T_n = \omega(n)$, then by regularizing the matrix $\hat{N}_\Gamma$ (i.e., the rows and columns corresponding to the states with the highest zero - out visit frequency), we can obtain $\sigma_1(\hat{N}_\Gamma - E[\hat{N}]) = O_P(\sqrt{T_n/n})$.

These results together indicate that in BMC, the order of the largest singular value is $\Theta_P(\sqrt{T_n/n})$.

### Key formulas
- **Lower bound in the sparse case**:
\[
\sigma_1(\hat{N} - E[\hat{N}]) = \Omega_P\left(\sqrt{\frac{T_n}{n}}\right)
\]

- **Upper bound in the dense case**:
\[
\sigma_1(\hat{N} - E[\hat{N}]) = O_P\left(\sqrt{\frac{T_n}{n}}\right)
\]

- **Upper bound in the sparser case**:
\[
\sigma_1(\hat{N}_\Gamma - E[\hat{N}]) = O_P\left(\sqrt{\frac{T_n}{n}}\right)
\]

### Main contributions of the paper
1. **Quantify the asymptotic behavior of the largest singular value of the random matrix constructed from the BMC path**, which is of great significance in the community detection problem.
2. **Provide the upper and lower bounds of the matrix spectral norm in the sparse and dense cases**, filling the gaps in existing research.
3. **Introduce a regularization technique** to control the spectral norm of the matrix by the states with the highest zero - out visit frequency, so that a tight upper bound can also be obtained in the sparse case.

These results not only deepen the understanding of the random matrix constructed from the BMC path but also provide theoretical support for the design and analysis of community detection algorithms.