Stochastic thermodynamics for classical non-Markov jump processes

This paper establishes a general theory of stochastic thermodynamics for classical non-Markov jump processes by introducing a "Fourier embedding" technique that converts memory-dependent dynamics into Markovian field systems, thereby enabling the derivation of the second law and time-reversal symmetry conditions for a broad class of history-dependent fluctuations.

The Gist

Imagine you are trying to predict the weather. In a standard, simple model (what physicists call a Markov process), you only look at the sky right now. If it's sunny, you predict tomorrow will be sunny. If it's raining, you predict rain. The past doesn't matter; only the present moment counts.

But in the real world, weather is non-Markovian. It has a "memory." If it rained all week, the ground is soaked, and even if the sun comes out today, the humidity will stay high tomorrow. The system remembers its history.

For decades, physicists have struggled to apply the laws of thermodynamics (the rules of energy and heat) to these "memory-filled" systems. It's like trying to use a map of a flat world to navigate a mountain range; the old tools just don't fit.

This paper by Kiyoshi Kanazawa and Andreas Dechant introduces a brilliant new tool to solve this problem. Here is the story of their discovery, explained simply.

The Problem: The "Black Box" of Memory

In the microscopic world (like tiny particles in water or biological cells), things don't just move randomly; they often move based on what happened a second ago, or ten seconds ago.

• The Issue: When a system has memory, the math gets messy. If you try to reverse time (like playing a video backward), the system might behave in ways that seem impossible or "causally broken" (like a ball rolling uphill because it remembers falling down earlier).
• The Consequence: Without a clear way to handle this, we couldn't calculate how much energy is wasted or how much entropy (disorder) is created in these complex, real-world systems.

The Solution: The "Fourier Embedding" Trick

The authors invented a mathematical magic trick called Fourier Embedding.

The Analogy: The Orchestra vs. The Soloist
Imagine a solo violinist (the system we care about) playing a song. But the violinist is being influenced by a massive, invisible orchestra (the memory) that is playing along.

• Old View: We only hear the violin. The music sounds weird and unpredictable because we can't see the orchestra. We can't write down the rules of the music.
• The New Trick: The authors say, "Let's make the orchestra visible!" They take the invisible history and translate it into a set of auxiliary variables (think of them as extra musicians in the orchestra).
• The Result: Suddenly, the violinist and the whole orchestra are playing together in a way that is predictable and follows standard rules. The "memory" is no longer a ghostly force; it's just a bunch of extra variables we can track.

They call this the Fourier Embedding because they use a specific mathematical language (Fourier transforms, which break waves into frequencies) to turn the "history" into these extra variables.

Why This Matters: The "Gauge Invariance"

Here is the most exciting part. In physics, sometimes the answer you get depends on how you look at the problem. If you change your measuring stick, the answer changes. This is bad for fundamental laws.

The authors proved that even though they added these "extra musicians" (the auxiliary variables) to solve the problem, the laws of thermodynamics remain the same.

• Whether you look at just the violinist or the whole orchestra, the amount of energy used and the heat generated is identical.
• They showed that the "Second Law of Thermodynamics" (which says disorder always increases) holds true, even for systems with strong memory.

Two New Models They Built

To prove their theory works, they built two new types of "memory machines":

1. The Forgetful Spin: Imagine a tiny magnet that flips up or down. In their model, the chance of it flipping depends on how many times it flipped in the past. It's like a coin that remembers if it landed on heads five times in a row and gets "tired" of flipping.
2. The Hiker with a Backpack: Imagine a hiker walking randomly. In their model, the hiker's next step depends on the entire path they've walked so far, not just where they are standing right now.

They showed that for both of these complex scenarios, you can still calculate the work done and the heat generated, and the Second Law is never broken.

The Big Picture

Before this paper, if you had a system with strong memory, you were stuck. You couldn't easily apply thermodynamics.

• Before: "This system is too complex; we can't define its heat or work."
• After: "We can translate this complex system into a simpler, memory-less one, solve it, and translate the answer back. The laws of physics still apply."

Summary

Think of this paper as a universal translator. It takes the confusing, history-dependent language of complex real-world systems and translates it into the simple, clean language of standard physics. This allows scientists to finally study the thermodynamics of things like biological cells, financial markets, or complex materials, where the past always influences the future.

They didn't just fix a math problem; they opened the door to understanding how energy flows in the messy, memory-filled world we actually live in.

Technical Summary

1. Problem Statement

Stochastic thermodynamics provides a framework for analyzing energy and entropy in small, noisy systems. However, its foundational results (First and Second Laws) rely heavily on the Markov assumption (memoryless dynamics), where transition rates depend only on the current state.

• The Gap: Real-world systems (biological, chemical, neural, and physical) often exhibit strong memory effects (non-Markovianity), where fluctuations depend explicitly on the system's entire history.
• The Challenge: Extending stochastic thermodynamics to general non-Markov jump processes has been elusive.
  • Standard mathematical tools (like the Master Equation) fail for history-dependent intensities.
  • Time-reversal symmetry becomes ill-defined or acausal in non-Markov systems, making it difficult to define heat, work, and entropy production consistently with thermodynamic laws.
  • Previous approaches (e.g., Generalized Langevin Equations, Semi-Markov processes) are limited to specific classes (linear/Gaussian or weak memory) and do not cover general non-Gaussian, history-dependent jump processes.

2. Methodology: Fourier Embedding

The authors introduce a novel mathematical technique called Fourier Embedding to convert a non-Markov jump process into a high-dimensional Markovian field dynamics.

• The Core Idea: Instead of treating the history $X_t = \{x_{t-\tau}\}_{\tau \ge 0}$ as a functional argument, the authors map the velocity history $v_{t-\tau}$ into an auxiliary field of Fourier modes.
• The Transformation:
  They define auxiliary variables $z_t(s)$ and $w_t(s)$ for wave numbers $s > 0$ via the Fourier transform of the velocity:
  $$z_t(s) = \int_0^\infty d\tau \, v_{t-\tau} \sin(s\tau), \quad w_t(s) = \frac{1}{s} \frac{\partial z_t(s)}{\partial t}$$
  The extended state is $\Gamma_t = (x_t, \{z_t(s), w_t(s)\}_{s>0})$.
• Resulting Dynamics: The original non-Markov jump process is equivalent to a Markovian system of Stochastic Partial Differential Equations (SPDEs) for $\Gamma_t$:
  $$\frac{dx_t}{dt} = \xi^{CP}_{\lambda_a[y|\Gamma_t]}, \quad \frac{\partial z_t(s)}{\partial t} = s w_t(s), \quad \frac{\partial w_t(s)}{\partial t} = -s z_t(s) + \xi^{CP}_{\lambda_a[y|\Gamma_t]}$$
  Here, the jump term acts simultaneously on the target system $x_t$ and all auxiliary modes $w_t(s)$.
• Field Master Equation (fME): The probability density functional $P_t[\Gamma]$ evolves according to a Field Master Equation:
  $$\frac{\partial P_t[\Gamma]}{\partial t} = (\mathcal{L}_A + \mathcal{L}_J) P_t[\Gamma]$$
  where $\mathcal{L}_A$ is the advective Liouvillian (describing the harmonic oscillators) and $\mathcal{L}_J$ is the jump Liouvillian.

3. Key Assumptions and Conditions

To derive thermodynamic laws, the authors establish three critical assumptions for the Fourier-embeddable class:

1. Assumption 0 (Stationarity): The fME admits a stationary probability density functional (PDF).
2. Assumption 1 (Time-Reversal Symmetry): The system satisfies a detailed balance condition for the target-system intensity $\lambda_a[y|\Gamma]$:
   $$\frac{1}{\beta} \ln \frac{\lambda_{a^*}[\epsilon(x-x')|\Gamma'^*]}{\lambda_a[x'-x|\Gamma]} = E_a[\Gamma'] - E_a[\Gamma]$$
   where $E_a$ is the total energy functional, $\beta$ is the inverse temperature, and $*$ denotes time-reversal operations.
3. Assumption 2 (Intensity Invariance): The total jump intensity is invariant under time reversal: $\lambda_{tot;a}[\Gamma] = \lambda_{tot;a^*}[\Gamma^*]$.
4. Assumption 3 (Exclusive Controllability): The controllable part of the energy depends only on the target system $x$ and external parameter $a$, independent of the auxiliary fields: $E_{ctrl}(x, a)$.

4. Key Contributions and Results

A. Formulation of Thermodynamic Laws

Under the above assumptions, the authors derive the First and Second Laws for the non-Markov system:

• First Law: $dE = dW + dQ$.
  • Work ($dW$) is defined via parameter changes.
  • Heat ($dQ$) is defined via the log-ratio of forward and backward transition intensities, consistent with the energy change.
• Second Law: The total entropy production rate (EPR) is non-negative:
  $$\dot{\sigma}_{tot} = \dot{\sigma}_{sys} + \dot{\sigma}_{bath} \ge 0$$
  Crucially, for equilibrium-to-equilibrium transitions, the cumulative entropy production is gauge invariant (independent of the choice of embedding variables):
  $$\sigma_{tot} = \beta (\langle \Delta W \rangle - \Delta F) \ge 0$$
  This proves that the Second Law holds regardless of the specific Markov embedding used, provided the target system satisfies the thermodynamic consistency conditions.

B. Uniqueness and Gauge Invariance

The paper addresses the non-uniqueness of Markov embeddings (e.g., Fourier vs. Laplace).

• Fourier vs. Laplace: The authors show that while the Laplace embedding is mathematically valid for reproducing the dynamics, it leads to an auxiliary field that is intrinsically out of equilibrium (non-zero housekeeping entropy).
• Decomposition: They demonstrate that the total entropy production in the Laplace frame ($\dot{\Sigma}_{tot}$) can be decomposed into housekeeping ($\dot{\Sigma}_{hk}$) and excess ($\dot{\Sigma}_{ex}$) parts. Remarkably, the excess entropy in the Laplace frame equals the total entropy in the Fourier frame ($\dot{\Sigma}_{ex} = \dot{\sigma}_{tot}$).
• Significance: The Fourier embedding is superior for thermodynamics because it allows the entire extended system (target + field) to be in thermal equilibrium, making the definition of time-reversal symmetry and entropy production natural and consistent.

C. Novel Non-Markov Models

The authors construct two specific models satisfying all assumptions, which were previously unformulable in stochastic thermodynamics:

1. History-Dependent Two-Level System: A classical spin ($x = \pm 1/2$) in a magnetic field where the flipping rate depends on the history of previous flips via a memory kernel.
2. History-Dependent Random Walk: A particle moving in a potential where the step size distribution and transition rates depend on the velocity history.
   • Both models utilize a Quadratic Non-Controllable Energy (QnCE) framework, where the auxiliary field energy is quadratic in $z$ and $w$, ensuring the existence of a stationary canonical distribution.

5. Significance and Impact

• Theoretical Breakthrough: This work establishes the first general framework for stochastic thermodynamics applicable to strong-memory, non-Gaussian jump processes. It resolves the long-standing issue of how to define time-reversal symmetry and entropy production for systems with explicit history dependence.
• Experimental Relevance: The framework provides a "guiding principle" for modeling realistic experimental setups (e.g., active matter, biological motors, quantum dots) where memory effects are dominant. It suggests that experimental models must satisfy time-reversal symmetry conditions (Assumptions 1-2) to be thermodynamically consistent.
• Machine Learning Integration: The authors propose that existing machine learning techniques for point processes can be adapted to incorporate these thermodynamic constraints, enabling "physics-informed" data-driven modeling of history-dependent fluctuations.
• Resolution of Ambiguity: By proving the gauge invariance of the Second Law (specifically the excess entropy), the paper clarifies that thermodynamic laws are robust against the mathematical choice of embedding, provided the physical system satisfies the underlying symmetry conditions.

In summary, Kanazawa and Dechant have successfully lifted the Markov constraint in stochastic thermodynamics, providing a rigorous mathematical foundation (via Fourier embedding) to analyze energy and entropy in complex, memory-laden systems.