Equivalent CM Models

TL;DR

This paper explores the relationships between different Gaussian conditionally Markov models, providing methods to convert between models and establish equivalences, which enhances modeling flexibility in stochastic sequence analysis.

Contribution

It introduces a unified approach to derive algebraically equivalent models among various Gaussian CM, reciprocal, and Markov sequences, extending beyond previous restrictions on transition matrices.

Findings

Unified method for model conversion and equivalence

Extension to models with singular transition matrices

Backward Markov model derived from forward model

Abstract

The conditionally Markov (CM) sequence contains different classes, including Markov, reciprocal, and so-called $CM_L$ and $CM_F$ (two CM classes defined in our previous work). Markov sequences are special reciprocal sequences, and reciprocal sequences are special $CM_L$ and $CM_F$ sequences. Each class has its own forward and backward dynamic models. The evolution of a CM sequence can be described by different models. For a given problem, a model in a specific form is desired or needed, or one model can be easier to apply and better than another. Therefore, it is important to study the relationship between different models and to obtain one model from another. This paper studies this topic for models of nonsingular Gaussian (NG) $CM_L$, $CM_F$, reciprocal, and Markov sequences. Two models are \textit{probabilistically equivalent (PE)} if their stochastic sequences have the same distribution, and are \textit{algebraically equivalent (AE)} if their stochastic sequences are path-wise identical. A unified approach is presented to obtain an AE forward/backward $CM_L$/$CM_F$/reciprocal/Markov model from another such model. As a special case, a backward Markov model AE to a forward Markov model is obtained. While existing results are restricted to models with nonsingular state transition matrices, our approach is not. In addition, a simple approach is presented for studying and determining Markov models whose sequences share the same reciprocal/$CM_L$ model.