Trajectory of Probabilities, Probability on Trajectories, and the Stochastic-Quantum Correspondence

This paper clarifies the conceptual distinction between "trajectories of probabilities" and "probabilities on trajectories" to resolve ambiguities in stochastic-quantum correspondence, establishing a rigorous framework that characterizes the non-unique relationship between probability dynamics and their implementing processes while disentangling concepts like linearity, Markovianity, and divisibility.

The Gist

The Big Picture: Two Ways to Look at a Movie

Imagine you are watching a movie about a coin flipping in the air. You want to describe how the coin behaves over time. The authors of this paper argue that there are two completely different ways to describe this movie, and confusing them has led to a lot of mistakes in physics (especially regarding Quantum Mechanics).

1. The "Trajectory of Probabilities" (The Weather Forecast):

   • What it is: This is like a weather report. It tells you, "At 1:00 PM, there is a 50% chance of rain. At 2:00 PM, there is a 70% chance."
   • The Focus: It only cares about the current state of the odds. It doesn't care how the weather got there or what happened in between. It's a map of probabilities evolving over time.
   • The Metaphor: Think of a GPS navigation screen showing your estimated time of arrival. It updates based on traffic, but it doesn't record the specific route you took, just the current prediction.

2. The "Probability on Trajectories" (The Reality of the Journey):

   • What it is: This is like a security camera recording every single flip of the coin. It records the entire history: "Heads, then Tails, then Heads." It assigns a probability to every possible story or history the coin could tell.
   • The Focus: It cares about the connections between moments. It asks: "If the coin was Heads at 1:00, what is the chance it is Tails at 2:00?"
   • The Metaphor: Think of a choose-your-own-adventure book. Every page turn is a specific path. This method assigns a probability to every single possible book you could write.

The Core Problem: Mixing Up the Map and the Journey

The paper says that many scientists have been making a category error. They are looking at the Weather Forecast (the changing odds) and assuming it must follow the rules of the Security Camera (the specific paths).

The "Coin Toss" Analogy:
Imagine a coin that has a secret, shifting weight inside it.

• The Forecast (Trajectory of Probabilities): We know the coin starts fair (50/50). Then, the weight shifts, making it 90% likely to be Heads. Then it shifts back.
• The Mistake: Scientists often assume that because the forecast changes linearly (smoothly), the underlying mechanism (the coin's path) must be a simple, memoryless process (like a standard Markov chain where the next flip only depends on the current one).

The Reality:
The authors show that you can have a very complex, messy, "memory-full" coin (where the past matters) that still produces a perfectly smooth, simple-looking forecast.

• Analogy: Imagine a chaotic dance floor (the trajectories). You can have a chaotic dance where people are constantly bumping into each other, remembering who they bumped, and changing steps based on the last 10 minutes. Yet, if you stand on a balcony and just count how many people are dancing in the center vs. the edge, the numbers might rise and fall in a perfectly smooth, predictable line.
• The Lesson: Just because the numbers (probabilities) look simple and linear, it doesn't mean the reality (the trajectories) is simple or memoryless.

Key Concepts Explained Simply

1. Linearity vs. Non-Linearity

• The Myth: Many physicists believe that probability evolution must be linear (like mixing paint: 50% red + 50% blue = 50% purple). They think this is a fundamental law of nature.
• The Truth: The paper shows that linearity is only guaranteed if you are looking at a statistical mixture (a big crowd of different coins, some biased one way, some another).
• The Metaphor:
  • Linear (Statistical Mixture): You have a bag of 100 coins. 50 are weighted to Heads, 50 to Tails. If you pick a coin at random, the odds are 50/50. If you mix them, the odds stay 50/50. This is linear.
  • Non-Linear (Single System): Now imagine a single coin that is fair, but its internal weight shifts based on a complex rule (like $r^2$). If you start with a 50/50 chance, the next step might jump to 25/75. The math doesn't add up linearly.
  • The Takeaway: Quantum mechanics is more like the single, shifting coin. You cannot assume it behaves like the bag of mixed coins.

2. Divisibility vs. Decomposability

• The Confusion: Scientists often ask, "Can we break the evolution of time into small, independent steps?"
• Decomposability (The Good News): Yes, usually. You can say, "The state at 2:00 is determined by the state at 1:00." The future depends on the present.
• Divisibility (The Bad News): This is a stricter rule. It says, "The rule that takes you from 1:00 to 2:00 is the same type of rule as the one from 2:00 to 3:00, just applied again."
• The Metaphor:
  • Decomposable: You can walk from New York to London, then London to Paris. You can break the trip into legs.
  • Divisible: You can walk from New York to London using a specific stride, and then from London to Paris using the exact same stride.
  • The Paper's Point: Quantum mechanics is decomposable (you can break it into steps) but not divisible (the rules for each step change in a way that doesn't fit a simple "multiply by a matrix" pattern).

3. The Quantum Connection (The "Interference" Problem)

This is the most important part for understanding Quantum Mechanics.

• The Quantum Weirdness: In quantum mechanics, particles can be in a "superposition" (like a coin spinning that is both Heads and Tails). When you measure it, the probabilities interfere with each other (like waves crashing).
• The Conflict: If you try to describe quantum mechanics using the "bag of mixed coins" (statistical mixture) logic, you get the wrong answer. The math of quantum interference is non-linear.
• The Paper's Verdict: You cannot force quantum mechanics into a "classical probability" box.
  • If you try to model a quantum particle as a classical coin with a hidden path, you fail.
  • The "interference" (the wavy, non-linear behavior) is the smoking gun that proves quantum probabilities are not just a statistical mix of hidden classical states.

Summary: What Should We Take Away?

1. Don't confuse the map with the territory. A smooth, linear graph of probabilities does not prove the underlying reality is simple or memoryless.
2. Quantum mechanics is special. It cannot be explained by simply saying "we just don't know the hidden details yet" (statistical mixing). The non-linear nature of quantum probability evolution is a fundamental feature, not a bug.
3. Stop forcing square pegs into round holes. Many recent theories try to force quantum mechanics to look like classical probability theory (by assuming linearity and divisibility). This paper argues that these theories are mathematically flawed because they ignore the difference between a "trajectory of probabilities" and a "probability on trajectories."

In a nutshell: The universe (at the quantum level) is not a simple game of dice where the odds just shift smoothly. It's a complex, interconnected dance where the history of the dance matters, and the rules of the dance are fundamentally different from the rules of a simple coin toss. Trying to describe the dance using the rules of the coin toss leads to confusion.

Technical Summary

1. Problem Statement

The paper addresses a fundamental conceptual ambiguity in the probabilistic description of physical systems, particularly within the context of stochastic–quantum correspondence and open quantum systems. The core problem is the conflation of two distinct probabilistic frameworks:

1. Trajectory of Probabilities: A description where the state of the system is a time-evolving probability vector $\vec{p}(t)$. This describes how the distribution of configurations changes over time.
2. Probability on Trajectories: A description where the system is modeled by a stochastic process (a probability measure $\mu$ over the space of all possible temporal trajectories). This assigns probabilities to specific histories (conjunctions of events at different times).

The authors argue that recent literature (e.g., Barandes [2025], Gillespie, Vacchini, Rivas) has failed to distinguish these two, leading to erroneous conclusions regarding:

• The linearity of probability dynamics (the assumption that probability evolution must be linear).
• The nature of transition matrices (confusing dynamical maps with conditional probability matrices).
• The concepts of divisibility and Markovianity (treating them as equivalent or assuming they imply linearity).
• The derivation of quantum mechanics from classical stochastic principles.

2. Methodology

The authors employ a rigorous axiomatic approach based on measure theory and probability theory to disentangle these concepts.

• Formal Definitions: They define Probability Dynamics ($P$) as a map $P: T \times S_N \to S_N$ (where $S_N$ is the probability simplex) and Canonical Stochastic Processes as probability measures $\mu$ on the space of trajectories $\Omega = C^T$.
• Implementation: They introduce the concept of implementation, where a stochastic process family $\mathcal{M}$ (a map from initial conditions $\vec{p}_0$ to measures $\mu_{\vec{p}_0}$) implements a probability dynamics $P$ if the marginal distributions of $\mu_{\vec{p}_0}$ match the trajectory $P(t, \vec{p}_0)$.
• Counter-Examples and Proofs: The paper constructs explicit counter-examples (e.g., coin toss models with hidden weights) to disprove common assumptions. It uses Kolmogorov's Extension Theorem to prove existence results regarding Markovian and non-Markovian implementations.
• Statistical Dynamics: The authors analyze specific physical scenarios (Deterministic, System-Ancilla, and Stochastic Statistical Dynamics) where probabilities represent ensemble frequencies to determine under what conditions linearity arises.
• Quantum Analysis: They apply their framework to quantum mechanics, analyzing Born rule probabilities, unitary evolution, and the role of interference.

3. Key Contributions and Concepts

A. Distinction Between Dynamics and Processes

The paper establishes that a single probability vector trajectory $\vec{p}(t)$ contains no information about multi-time joint probabilities or conditional probabilities. Conversely, a stochastic process contains this information but does not inherently define a functional law for how $\vec{p}(t)$ evolves from different initial conditions unless a family of processes is specified.

B. Linearity and the Law of Total Probability

The authors debunk the argument that the Law of Total Probability implies the linearity of probability dynamics.

• The Fallacy: Arguments often claim $\vec{p}(t) = M(t) \vec{p}(0)$ implies linearity because $M(t)$ is a matrix.
• The Correction: The matrix $M_{\vec{p}_0}(t)$ (containing conditional probabilities) generally depends on the initial condition $\vec{p}_0$. Therefore, the map $\vec{p}_0 \mapsto M_{\vec{p}_0}(t)\vec{p}_0$ is nonlinear in general. Linearity only holds if the conditional probabilities are independent of the initial state (a "transition-constant" family), which is a specific physical assumption, not a mathematical necessity.

C. Decomposability vs. Divisibility

The paper clarifies the confusion between stepwise evolution concepts:

• Decomposability: A property of the probability dynamics $P$. It means the future state $\vec{p}(t)$ is uniquely determined by the intermediate state $\vec{p}(t')$. This is a general notion of "no memory" for the dynamics.
• Divisibility: A stronger property requiring that the stepwise evolution map can be represented by a stochastic matrix (i.e., it must be linear).
• Result: Every divisible dynamics is decomposable, but a decomposable dynamics is not necessarily divisible (even if linear). Divisibility fails if the intermediate state space is restricted (the range of $P_{t'}$ is a subset of the simplex) and the map cannot be extended to the whole simplex as a stochastic matrix.

D. Markovianity vs. Dynamics Properties

The authors prove that Markovianity is a property of the implementing stochastic process, not the probability dynamics itself.

• Proposition 9: Every probability dynamics admits a Markovian implementation (by constructing a process with independent time steps).
• Proposition 9: A non-degenerate probability dynamics also admits a non-Markovian implementation.
• Conclusion: There is no necessary link between the linearity/decomposability of a dynamics and the Markovianity of its implementation.

E. Statistical Dynamics

The paper identifies Statistical Dynamics as the specific physical context where linearity is justified. This occurs when probability vectors represent the frequency distributions of an ensemble of independent systems evolving under a deterministic or stochastic law.

• In this context, linearity arises from the statistical mixing of independent trajectories.
• However, this interpretation fails for quantum mechanics, where probabilities do not represent frequencies of pre-existing hidden states (due to no-go theorems like Bell and Kochen-Specker).

4. Key Results

1. Non-Uniqueness of Implementation: A single probability vector trajectory can be implemented by infinitely many stochastic processes (both Markovian and non-Markovian).
2. Generic Nonlinearity: Probability dynamics are generically nonlinear. Linearity is a special case derived from specific physical assumptions (statistical mixing), not a mathematical consequence of probability theory.
3. Decomposability $\neq$ Divisibility: The authors provide counter-examples showing that a linear probability dynamics can be decomposable (stepwise deterministic in terms of state) but not divisible (cannot be represented by a composition of stochastic matrices).
4. Quantum Probability Evolution is Nonlinear:
   • The evolution of Born probabilities $\vec{p}(t)$ is generally nonlinear due to quantum interference (cross terms in $|\langle i | U(t) | \psi \rangle|^2$).
   • While the evolution of the density matrix is linear, the map from initial probabilities to final probabilities is not, because the map from probabilities to density matrices is not linear (superposition $\neq$ mixture).
   • Quantum probability evolution is decomposable (the state at $t'$ determines the future) but not divisible in the stochastic sense (it cannot be modeled by a classical stochastic process with a transition matrix).

5. Significance

• Correction of Literature: The paper systematically corrects widespread misconceptions in the literature regarding the "stochastic–quantum correspondence." It demonstrates that attempts to derive quantum linearity from classical probabilistic assumptions (like the Law of Total Probability or statistical mixing) are flawed because they ignore the distinction between dynamics and processes and the specific ontological status of quantum probabilities.
• Clarification of Quantum Interference: The authors reframe quantum interference not as a failure of "divisibility" in a stochastic sense (as Barandes suggests), but as a manifestation of the nonlinearity of probability evolution arising from the superposition principle.
• Foundational Rigor: By separating the mathematical structures of "trajectories of probabilities" and "probabilities on trajectories," the paper provides a clearer framework for analyzing open quantum systems, decoherence, and the limits of classical stochastic models in describing quantum phenomena.
• Implications for Reconstruction Programs: The results challenge "General Probabilistic Theories" approaches that attempt to reconstruct quantum mechanics solely from assumptions about the linearity of probability evolution, showing that such linearity requires specific (and often quantum-incompatible) physical interpretations.

In summary, the paper argues that linearity is not a fundamental feature of probability evolution but a feature of specific statistical ensembles. Quantum mechanics, which does not fit the statistical ensemble interpretation of probabilities, exhibits generic nonlinear probability evolution, rendering standard stochastic–quantum correspondences based on linear transition matrices fundamentally inadequate.