A Fast Monte Carlo algorithm for evaluating matrix functions with application in complex networks

Nicolas L. Guidotti,Juan A. Acebrón,José Monteiro

DOI: https://doi.org/10.1007/s10915-024-02500-w

2024-09-21

Abstract:We propose a novel stochastic algorithm that randomly samples entire rows and columns of the matrix as a way to approximate an arbitrary matrix function using the power series expansion. This contrasts with existing Monte Carlo methods, which only work with one entry at a time, resulting in a significantly better convergence rate than the original approach. To assess the applicability of our method, we compute the subgraph centrality and total communicability of several large networks. In all benchmarks analyzed so far, the performance of our method was significantly superior to the competition, being able to scale up to 64 CPU cores with remarkable efficiency.

Data Structures and Algorithms

What problem does this paper attempt to address?

### What problem does this paper attempt to solve?

This paper proposes a new stochastic algorithm for approximating arbitrary matrix functions through power series expansion. Compared to existing Monte Carlo methods, this algorithm has significant advantages in convergence speed by randomly sampling rows and columns of the entire matrix. Specifically, the algorithm addresses the following problems:

1. **Node centrality measures in large-scale networks**:
- Traditional Monte Carlo methods can only handle one matrix element at a time, resulting in slow convergence. The new algorithm significantly improves convergence speed by randomly sampling entire rows and columns.
- The algorithm can effectively compute important network metrics such as subgraph centrality and total communicability.

2. **Limitations of existing methods**:
- Direct methods (such as expm and Schur-Parlett algorithms) have high computational complexity and are only suitable for small-scale matrices.
- Methods based on Gaussian quadrature rules are prone to numerical instability for large-scale sparse matrices.
- Krylov methods have slow convergence or stagnation for matrices with large condition numbers.

3. **Improved algorithm performance**:
- The new algorithm performs excellently in multiple benchmark tests, particularly in estimating subgraph centrality and total communicability of large networks, outperforming other methods.
- The algorithm can scale to 64 CPU cores and is highly efficient.

In summary, this paper aims to propose an efficient and fast Monte Carlo algorithm to solve the problem of matrix function computation in large-scale network analysis, particularly excelling in the computation of subgraph centrality and total communicability.