Abstract:An open stochastic system à la Willems is a system affected two qualitatively different kinds of uncertainty -- one is probabilistic fluctuation, and the other one is nondeterminism caused by lack of information. We give a formalization of open stochastic systems in the language of category theory. A new construction, which we term copartiality, is needed to model the propagating lack of information (which corresponds to varying sigma-algebras).
What problem does this paper attempt to address?
The problem that this paper attempts to solve is: how to use category theory to formally describe and analyze open stochastic systems that contain different kinds of uncertainties (probabilistic uncertainty, nondeterministic uncertainty and lack of information). Specifically, the author focuses on open linear systems that combine probability theory and non - deterministic models, especially physical systems with Gaussian noise and the problem of insufficient prior information in Bayesian statistics.
### Detailed Explanation:
1. **Different Types of Uncertainty**:
- **Probabilistic Uncertainty**: We don't know the exact value of a certain quantity, but we know its statistical distribution. For example, the normal distribution \( N(\mu,\sigma^{2}) \).
- **Nondeterministic Uncertainty**: The real quantity can be any feasible value, but we don't have more statistical information. This kind of uncertainty can be modeled by equality constraints of natural laws.
- **Lack of Information**: Due to incomplete measurement, some quantities can only be partially obtained. Mathematically, this lack of information can be formalized by the equivalence relation \( \Phi\subseteq X\times X \).
2. **Application of Category Theory**:
- The author introduces a new construction - copartiality, which is used to simulate the lack of information in the information propagation process (corresponding to the changing σ - algebra).
- Through the language of category theory, the author defines extended Gaussian distributions, which combine Gaussian probability and nondeterminism and are consistent with Willems' concept of Gaussian linear systems.
3. **Specific Examples**:
- **Resistor with Noise**: In an actual system, a resistor is not only constrained by Ohm's law \( V = RI \), but also affected by thermal noise, which can be modeled by \( V = RI+\epsilon \), where \( \epsilon\sim N(0,\sigma^{2}) \) is a Gaussian random variable.
- **Uninformative Prior**: In statistical inference, when there is a complete lack of prior information about the variable \( X \), how to model this situation. The author proposes the concept of extended Gaussian distribution to avoid using unnormalized measures.
4. **Compositionality**:
- Compositionality studies how to build complex systems from simple parts. Willems proposed the Behavioral Approach, which understands complex systems through "tearing", "zooming" and "linking".
- The author uses monoidal category theory to express this compositionality, so that different types of uncertainty can be dealt with in the same framework.
5. **Contributions**:
- Defines a Markov category GaussEx of extended Gaussian maps, which combines Gaussian probability and nondeterminism.
- Introduces copartiality to formalize the lack of information.
- Proves that the composition in category theory is consistent with Willems' interconnection concept (Theorem 5.6).
Through these works, the author provides a unified framework for dealing with open systems that contain multiple uncertainties, especially those systems that have both probabilistic noise and relational constraints.