Maximum Likelihood Estimation based on Random Subspace EDA: Application to Extrasolar Planet Detection

Bin Liu, Ke-Jia Chen

Introduction

Detecting extrasolar planets using the Radial Velocity (RV) method presents a significant computational challenge. Astronomers search for subtle "wobbles" in a star's velocity caused by the gravitational influence of orbiting planets. Mathematically, this involves fitting a highly nonlinear Keplerian model to noisy time-series data. As the number of planets in a system increases, the search space expands exponentially—a phenomenon known as the "curse of dimensionality." Standard evolutionary search methods often struggle with the resulting high-dimensional, multimodal likelihood surfaces, frequently becoming trapped in local optima.

Methodology: Random Subspace EDA (RS-EDA)

To address these challenges, researchers have introduced the Random Subspace Estimation of Distribution Algorithm (RS-EDA). While traditional EDAs build probabilistic models of promising solutions to guide the search, they often fail in high-dimensional spaces because the amount of data required to model all dimensions simultaneously becomes prohibitive.

RS-EDA overcomes this by decomposing the complex search space into smaller, manageable subspaces. The process follows four key stages:

  1. Correlation Estimation: The algorithm analyzes a pool of promising solutions to identify variables that exhibit strong correlations (e.g., orbital period and eccentricity).
  2. Random Subspace Partitioning: Parameters are partitioned into smaller, overlapping subspaces. Highly correlated variables are grouped together to preserve the underlying physical structure of the model.
  3. Subspace EDA: Optimization is performed within these lower-dimensional subspaces. This reduction in dimensionality makes model-fitting significantly more efficient and less prone to sample starvation.
  4. Global Update: Local improvements from the subspaces are integrated back into the global best estimate, and the cycle repeats.
Flowchart of the RS-EDA algorithm showing the iterative process of correlation estimation, subspace partitioning, and local optimization.

Experimental Evaluation

The algorithm was evaluated using both synthetic benchmarks and real-world astronomical data:

  • Synthetic Benchmark: The Rotated Schaffers F7 (RSF7) function, a known "torture test" for optimizers due to its multimodal and nonseparable nature.
  • Real-World Data: Radial velocity measurements for the star HD73526, which is known to host a multi-planet system.

Key Results and Discussion

The experiments highlighted the scalability and robustness of RS-EDA compared to standard methods like Particle Swarm Optimization (PSO), Covariance Matrix Adaptation Evolution Strategy (CMA-ES), and traditional EDA.

  • High-Dimensional Performance: In 20-dimensional tests, RS-EDA significantly outperformed its competitors. While traditional EDA often diverged and CMA-ES/PSO struggled with the complex landscape in higher dimensions, RS-EDA maintained convergence toward the global optimum.
  • Exoplanet Verification: When applied to the HD73526 system, RS-EDA successfully fitted both 1-planet and 2-planet models. Using the Bayesian Information Criterion (BIC), the algorithm confirmed that a 2-planet model provided a statistically superior fit, aligning with established astronomical findings.
  • Convergence Characteristics: As shown in the performance plots, RS-EDA demonstrates a robust ability to improve solution quality over time, eventually surpassing standard EDA which tends to plateau prematurely on complex surfaces.
Convergence comparison of RS-EDA against EDA, PSO, and CMA-ES on a benchmark optimization task.

Conclusion

RS-EDA provides a powerful framework for solving high-dimensional maximum likelihood estimation problems in astronomy. By using correlation as a "compass" to guide subspace partitioning, the algorithm effectively navigates peaky likelihood surfaces that stymie traditional methods. While it may be slower than specialized local searchers in low-dimensional problems due to the partitioning overhead, its scalability makes it a promising tool for detecting complex multi-planet systems. Future work will explore more sophisticated subspace construction techniques and test the algorithm on even higher-dimensional "Grand Tack" style planetary models.

Summary updated

Abstract

This paper addresses maximum likelihood (ML) estimation based model fitting in the context of extrasolar planet detection. This problem is featured by the following properties: 1) the candidate models under consideration are highly nonlinear; 2) the likelihood surface has a huge number of peaks; 3) the parameter space ranges in size from a few to dozens of dimensions. These properties make the ML search a very challenging problem, as it lacks any analytical or gradient based searching solution to explore the parameter space. A population based searching method, called estimation of distribution algorithm (EDA), is adopted to explore the model parameter space starting from a batch of random locations. EDA is featured by its ability to reveal and utilize problem structures. This property is desirable for characterizing the detections. However, it is well recognized that EDAs can not scale well to large scale problems, as it consists of iterative random sampling and model fitting procedures, which results in the well-known dilemma curse of dimensionality. A novel mechanism to perform EDAs in interactive random subspaces spanned by correlated variables is proposed and the hope is to alleviate the curse of dimensionality for EDAs by performing the operations of sampling and model fitting in lower dimensional subspaces. The effectiveness of the proposed algorithm is verified via both benchmark numerical studies and real data analysis.
BibTeX
@inproceedings{liu2017maximum,
  author = {Bin Liu and Ke-Jia Chen},
  title = {{Maximum Likelihood Estimation based on Random Subspace EDA: Application to Extrasolar Planet Detection}},
  booktitle = {SEAL},
  year = {2017},
  doi = {10.1007/978-3-319-68759-9\_1},
}
Figure