🌀 nonlinear sciences

The information path functional approach for solution of a controllable stochastic problem

This paper proposes an information path functional approach to solve controllable stochastic problems by deriving optimal control functions that simultaneously address system identification and state consolidation, revealing that the accumulated information from process inner connections exceeds the sum of individual state entropies.

Original authors: Vladimir S. Lerner

Published 2026-06-03

📖 6 min read🧠 Deep dive

CC BY 3.0

Original authors: Vladimir S. Lerner

Original paper licensed under CC BY 3.0 (http://creativecommons.org/licenses/by/3.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

The Big Picture: Taming the Wild River

Imagine you are trying to navigate a wild, unpredictable river (the stochastic system). The water is choppy, the currents shift randomly, and you can't see exactly where the next wave will come from. This is how many real-world systems work: stock markets, weather patterns, or even biological processes. They are full of "noise" and uncertainty.

The author, Vladimir Lerner, proposes a new way to steer this river. Instead of just trying to predict the next wave, he suggests building a map (a dynamic model) that learns from the river as you go, and then using that map to steer the boat with perfect precision.

The core idea is a "dual strategy": you are simultaneously identifying the river's hidden rules and controlling the boat to follow the best possible path.

Key Concepts Explained

1. The "Entropy Functional" (The Measure of Chaos)

Think of Entropy as a measure of how confused or disordered the river is.

Standard View: Usually, we measure the confusion of the river by looking at individual snapshots (states) and adding them up. It's like counting how many ripples are in a pond at 1:00 PM, then 1:01 PM, then 1:02 PM.
Lerner's View: The author argues that the river has "hidden connections" between those ripples. If you cut the river into separate snapshots, you lose the story of how the water flows from one moment to the next.
The Innovation: He uses a special mathematical tool called an Entropy Functional (EF). This tool measures the entire journey of the water, not just the snapshots. It captures the "secret information" hidden in the flow between the moments. He claims this functional holds more information than the sum of all the individual parts.

2. The "Cut-Off" (The Impulse Control)

How do we extract this hidden information? Lerner suggests using Impulse Controls.

The Analogy: Imagine the river is flowing smoothly. Suddenly, you drop a giant, heavy anchor (an impulse) at a specific moment, stopping the flow for a split second, then immediately lifting it to let the water surge forward again.
The Result: This "cut-off" breaks the continuous flow. By doing this, the author shows that you can measure exactly how much information was "hidden" in the connection between the moments before and after the cut. It's like realizing that the silence between two musical notes holds as much meaning as the notes themselves.

3. The "Information Path Functional" (The Perfect Map)

Once we have this measure of hidden information, we want to find the best path.

The Goal: We want to turn the chaotic, random river into a smooth, predictable highway.
The Method: Lerner uses a "Variation Principle" (similar to how a light beam chooses the fastest path). He looks for a path where the "confusion" (entropy) is minimized, but in a very specific way: he finds the path that minimizes the worst-case confusion.
The Result: This creates an Information Path Functional (IPF). Think of this as a "perfect map" that the boat follows. This map isn't just a guess; it is a mathematical model that approximates the random river with the highest possible probability.

4. The "Punched" Points (The Checkpoints)

The journey isn't one long smooth line. It's a series of segments connected by special checkpoints called "Punched Points" (or Discrete Points).

The Analogy: Imagine driving a car on a foggy road. You can't see the whole road. Instead, you drive a short distance, then you hit a "checkpoint" where the fog clears for a second.
What Happens at the Checkpoint:
1. Stop & Measure: You pause (the "cut-off") to measure the road conditions (identify the model).
2. Update: You update your GPS (the dynamic model) with this new data.
3. Resume: You drive the next segment based on the updated map.
The Magic: At these checkpoints, the random "noise" of the river is converted into a solid "signal" (a dynamic model). The author calls this a Macrodynamic process. It turns the uncertain future into a certain plan.

5. The "Spiral" and the "Network"

As the boat moves, it doesn't just go in a straight line.

The Geometry: The path the boat takes looks like a spiral winding around a cone.
The Network: As the boat completes segments, it builds a structure called an Information Network (IN). Think of this as a tree growing. Every time the boat finishes a segment and hits a checkpoint, it adds a new "branch" to the tree.
The Code: This tree structure allows the system to "encode" the entire history of the river into a compact code. It's like compressing a whole movie into a single file size that still contains all the plot details.

How It Works in Practice (The "Dual Strategy")

The paper describes a system that does two things at once:

Identification: It watches the random process to figure out the rules (the "drift" and "diffusion" of the river).
Control: It applies the "impulse" (the anchor drop) to force the system into a state where those rules can be seen clearly.

By doing this, the system creates a feedback loop:

The random process gives data.
The system extracts the "hidden info" at the checkpoints.
The system updates its model.
The system steers the process toward the best possible outcome.

The Bottom Line

Lerner's paper claims that by using these "cut-off" moments and measuring the "hidden connections" between them, we can:

Predict random systems better than traditional methods.
Control them with high precision.
Encode the entire system's behavior into a simple, hierarchical code (the Information Network).

He argues that this approach is superior to standard methods because it doesn't just look at the "dots" (individual states); it understands the "lines" connecting them, revealing a hidden order in what looks like pure chaos.

Note: The paper focuses entirely on the mathematical theory, the derivation of these equations, and examples of how the math works (like the "Example 3" in the appendix). It does not discuss specific medical applications, future commercial products, or clinical uses, but rather presents a new mathematical framework for solving control problems in any system that behaves randomly.

Technical Summary: The Information Path Functional Approach for Controllable Stochastic Problems

Problem Statement
The paper addresses the challenge of optimal control for stochastic systems described by Ito's controllable stochastic differential equations. A central difficulty in this domain is the concurrent solution of two problems: the identification of the random process's dynamic model (specifically its controllable drift and diffusion functions) and the synthesis of an optimal control strategy. Traditional methods often treat identification and control as separate tasks or rely on standard Shannon entropy measures, which the author argues fail to capture the inner connections and dependencies between the states of a random process. The paper posits that a "complete solution" to this joint problem has remained unknown despite decades of research, necessitating a new approach that generalizes the universal nature of information models for specific applications.

Methodology
The proposed methodology is based on the Information Path Functional (IPF) approach, which draws an analogy to R.P. Feynman's path functional in quantum mechanics. The core of the approach involves the following steps:

Entropy Functional (EF) Formulation: The process is modeled using a controllable Ito equation. An entropy functional (EF) is defined based on the Radon-Nikodym derivative between the a priori (uncontrolled) and a posteriori (controlled) probability measures of the process. This EF is shown to be equivalent to the Kullback-Leibler divergence for continuous variables.
Variation Problem (VP) and Minimax Principle: The author formulates a variation problem to minimize the EF. This leads to a "minimax principle" where the optimal transformation of a random process into a dynamic process is achieved by minimizing the entropy of the random process while maximizing this minimum. This converts the uncertainty of the random process into the certainty of a dynamic information process.
Dynamic Approximation via IPF: The solution to the variation problem yields an Information Path Functional (IPF). The IPF serves as a dynamic analogy to the EF, defined on the trajectories of a sought "Information Dynamic Model" (IDM). The IPF approximates the random trajectories of the a priori process with maximal probability.
Impulse Control and "Cutting Off": A critical mechanism in this approach is the use of impulse controls (composed of step-wise up and down actions) to "cut off" the diffusion process at discrete moments. This action dissolves correlations between states at specific points (called "punched" discrete points or DPs). The paper demonstrates that the entropy functional measure of the uncut process contains more information than the sum of Shannon entropies of its separated states; the difference represents "hidden information" stored in the connections between states.
Constraint Imposition: The variation problem imposes dynamic constraints (via the Jacobi-Hamilton and Kolmogorov equations) at these discrete points. These constraints link the micro-level diffusion process with the macro-level dynamic model. The control acts to turn these constraints on and off, effectively selecting extremal segments from the random process to form a piece-wise optimal macrodynamic process.
Information Invariants: The solution generates "information invariants" derived from the eigenvalues of the system's operator. These invariants allow for the prediction of control actions and the encoding of the identified model.

Key Contributions and Results

Joint Identification and Control: The paper presents a unified framework where the optimal control action simultaneously identifies the dynamic model (matrix $A$ ) and controls the system. The control is synthesized as a feedback function of macrostates at the discrete points.
Hidden Information and Feller's Kernel: The methodology quantifies "hidden information" lost when a process is cut off. The paper claims that the functional information measure contains more information than the sum of Shannon entropies of separated states, providing a measure of Feller's kernel. It estimates that each cut-off point loses approximately 0.5 Nats (~0.772 bits) of information.
Dual Strategy: The optimal strategy is described as "dual," where the IPF optimum predicts current control actions not only based on a total path goal but also by renovating the values of the controllable drift and diffusion for subsequent actions. This allows for real-time identification and correction of the goal.
Hierarchical Information Network (IN): For multidimensional processes, the approach leads to the consolidation and aggregation of information trajectories into a hierarchical structure called an Information Network (IN). The process is modeled as a sequence of extremal segments forming spiral trajectories on conic surfaces, with nodes representing the consolidation of information.
Optimal Encoding: The paper connects the IPF approach to Shannon's information theory, proposing that the identified dynamic operator and control can be optimally encoded using the derived information invariants. This allows for the compression of random process information into invariant quantities, facilitating the construction of a shortest optimal program (linking to Algorithmic Information Theory).
Practical Implementation: The paper asserts that the procedure has been implemented in practice, specifically in electro-technological processes involving direct measuring of diffusion conductivity. It claims that this method expedites closed-loop dynamics and minimizes errors (reported as not exceeding 10%) compared to statistical methods of computing correlation functions.

Significance and Claims
The paper claims that the IPF approach offers a complete solution to the problem of optimal control and identification for stochastic systems, a problem the author states has remained unsolved by conventional methods. By utilizing the variation principle on the entropy functional, the approach transforms the inherent uncertainty of random processes into a deterministic dynamic model with maximal probability.

The significance lies in:

Theoretical Advancement: It extends information theory beyond static state probabilities to dynamic path functionals, capturing the "inner connections" of a process that traditional Shannon entropy misses.
Unified Framework: It provides a single mathematical structure (the IPF and its invariants) that simultaneously solves identification, optimal control, and information encoding.
Practical Utility: The author claims the method is applicable to a wide class of controllable information systems and has been successfully applied to real-world industrial processes, offering a robust alternative to standard statistical identification techniques.

The paper concludes that this approach reveals regularities within random processes that are otherwise obscured by uncertainty, effectively converting the "uncertainty of a random process into the certainty of a dynamic information process."