Adversarial Thermodynamics

Basic Information

• Paper ID: 2510.08298
• Title: Adversarial Thermodynamics
• Authors: Maite Arcos, Philippe Faist, Takahiro Sagawa, Jonathan Oppenheim
• Classification: quant-ph (Quantum Physics), cond-mat.stat-mech (Statistical Mechanics)
• Publication Date: October 9, 2025 (arXiv preprint)
• Paper Link: https://arxiv.org/abs/2510.08298

Abstract

In thermodynamics, an agent's ability to extract work is fundamentally constrained by its environment. Traditional frameworks struggle to capture strategic decision-making under uncertainty—particularly how an agent's risk tolerance determines the trade-off between extractable work and success probability in finite-scale experiments. This paper develops a non-equilibrium thermodynamics framework based on adversarial resource theory, modeling work extraction as an adversarial game between an agent and environmental constraints. From this perspective, we reformulate the Szilard engine as a game isomorphic to Kelly gambling—an information-theoretic model of optimal betting under uncertainty, but employing thermodynamic utility functions. Extending the framework to the finite-scale regime, we apply risk-return trade-offs to find an interpretation of Rényi divergence as extractable work given a failure probability constraint. By incorporating risk sensitivity through utility functions, we demonstrate that the guaranteed work amount rational agents are willing to accept (rather than risk-dependent protocols) is given by Rényi divergence. This provides a unified picture of thermodynamics and gambling, highlighting how generalized free energies emerge from adversarial settings.

Research Background and Motivation

Problem Background

1. Limitations of Traditional Thermodynamics: Traditional thermodynamic frameworks primarily apply to large systems in equilibrium states, relying on ensemble averaging. However, in small-scale, non-equilibrium systems in nanotechnology and biophysics, fluctuations dominate, and deterministic quantities like free energy must be replaced by probabilistic, protocol-dependent concepts.
2. Inadequacies of Existing Approaches:
   • Stochastic Thermodynamics: While embracing the inherent stochasticity of small-scale, non-equilibrium systems, it lacks a complete operational prescription
   • Resource Theory Approaches: Reformulate the second law as state transformation constraints, but fail to provide a complete description of how an agent's strategic choices directly determine the trade-off between work extraction and success probability
3. Core Challenge: How to connect an agent's risk tolerance to the risk-return trade-off in work extraction within a single finite-scale experiment.

Research Motivation

This paper aims to bridge this gap through the lens of expected utility theory and decision theory, treating work extraction as a decision-theoretic problem where optimal strategies are determined by the agent's sensitivity to fluctuations.

Core Contributions

1. Establishing an Adversarial Thermodynamics Framework: Based on adversarial resource theory, work extraction is modeled as an adversarial game between an agent and environmental constraints.
2. Discovering the Isomorphism Between the Szilard Engine and Kelly Gambling: Demonstrates that the adversarial Szilard engine is mathematically isomorphic to the Kelly betting problem, but with different utility function classes.
3. Identifying Relevant Utility Functions in Thermodynamics: Determines that Constant Absolute Risk Aversion (CARA) utility functions are the relevant risk aversion class in thermodynamics, distinct from Constant Relative Risk Aversion (CRRA) in gambling.
4. Providing an Operational Interpretation of Rényi Divergence: Proves that all Rényi divergences possess operational interpretations for work extraction, extending previous results limited to D₀ and D∞.
5. Unifying Stochastic and Resource Theory Perspectives: Through decision-theoretic principles, unifies the fluctuation sensitivity of stochastic thermodynamics with the generalized free energies of resource theory within a single framework.

Methodology Details

Task Definition

Adversarial Szilard Engine Setup:

• Participants: Bob (sets initial constraints), Alice (optimizes work extraction), Charlie (referee, executes randomness)
• Input: Empty box of volume V, binary probability distribution P_X(x)
• Output: Extracted work W
• Constraints: Isothermal process, finite-scale effects

Model Architecture

1. Basic Game Structure

Bob places partition → Charlie randomly places molecule → Alice chooses final partition → Work extraction

2. Work Extraction Formula

For single-round extraction, work is:

• When x=0 (left): w₀ = k_BT ln(Q^A/Q^B)
• When x=1 (right): w₁ = k_BT ln((1-Q^A)/(1-Q^B))

For n-round average work extraction:

W = n(D(P_X||Q^B_X) - D(P_X||Q^A_X))k_BT  (1)

3. Utility Function Framework

Employs CARA utility function:

u_r(w_x) = (1/r)(1 - exp(-rw_x))  (2)

where r is the risk parameter:

• r > 0: risk aversion
• r = 0: risk neutrality
• r < 0: risk seeking

4. Optimal Strategy

Derived through expected utility maximization:

Q^{A,r}_X(x) = P_X(x)^{1/(1+r)} Q^B_X(x)^{r/(1+r)} / Z  (7)

Technical Innovations

1. Physical Basis for Utility Function Selection: Identifies that the additive nature of thermodynamic systems requires CARA utility functions, rather than CRRA functions used in financial scenarios.
2. Mathematical Formulation of Risk-Return Trade-off: Transforms finite-scale work extraction into a "guessing type" decision-theoretic problem.
3. Thermodynamic Interpretation of Certainty Equivalence: Proves that certainty equivalence exactly equals Rényi divergence:

W_CE = D_{1/(1+r)}(P_X||Q^B_X)k_BT  (9)

Experimental Setup

Theoretical Verification Framework

This is primarily a theoretical work, verified through:

1. Mathematical Consistency Checks: Verification that classical results are recovered as r→0
2. Limiting Case Analysis: Examination of extreme risk aversion (r→∞) and risk-seeking (r→-∞) behavior
3. Comparison with Known Results: Comparison with original Szilard results and Kelly gambling theory

Evaluation Metrics

• Expected work extraction EW
• Certainty equivalence W_CE
• Success probability constraints
• Rényi divergence D_α

Experimental Results

Main Results

1. Expected Work Extraction

For risk aversion level r, expected work extraction is:

E[W] = (αD(P_X||Q^B_X) + (1-α)D_α(P_X||Q^B_X))k_BT  (8)

where α = 1/(1+r)

2. Certainty Equivalence

W_CE = D_{1/(1+r)}(P_X||Q^B_X)k_BT  (9)

3. Finite-Scale Work Bounds

In the finite-scale regime, work extraction bounds are:

W_n ≥ nD_μ(P_X||Q^B_X)k_BT + (μ/(1-μ))ln ε  (17)

Theoretical Findings

1. Risk-Neutral Correspondence: When r=0, the optimal strategy Q^A_X = P_X, exactly corresponding to non-equilibrium thermodynamic free energy.
2. Monotonicity Verification: Certainty equivalence monotonically decreases with increasing risk aversion, consistent with economic intuition.
3. Rationality Conditions: For risk-seeking behavior (r<-1), first-order stochastic dominance conditions are never violated, ensuring rational choice.

Related Work

Main Research Directions

1. Connections Between Stochastic Thermodynamics and Gambling: Works 11-16 establish links between stochastic thermodynamics and gambling strategies
2. Resource Theory Approaches: 4-8 develop resource-theoretic formulations of thermodynamics
3. Application of Expected Utility Theory in Thermodynamics: 16 applies expected utility theory to thermodynamic process evaluation

Advantages of This Work

• Provides deeper analogies rather than simple concept transfer
• Formally models work extraction as an adversarial game
• Reveals the essential role of decision theory in thermodynamics

Conclusions and Discussion

Main Conclusions

1. Finite-scale work extraction can be understood within a resource-theoretic framework based on adversarial gambling
2. Relevant risk aversion in thermodynamics is described by CARA utility functions
3. The coincidence of certainty equivalence with Rényi divergence provides an operational principle basis for generalized second laws
4. The introduction of risk aversion causes fluctuation sensitivity and generalized free energies to emerge from a single decision-theoretic principle

Limitations

1. Idealized Assumptions: Assumes Alice knows the prior distribution, which may not hold in practical applications
2. Binary Systems: Analysis primarily focuses on binary Szilard engines; while extensions to general cases exist, specific analysis is limited
3. Experimental Verification: Lacks actual experimental validation; primarily theoretical construction

Future Directions

1. Explore scenarios where Alice does not know the correct prior distribution
2. Study more complex multi-stage engine systems
3. Extend the framework to quantum thermodynamics
4. Explore potential connections with black hole thermodynamics

In-Depth Evaluation

Strengths

1. Strong Theoretical Innovation: First systematic unification of decision theory and thermodynamics, providing a novel theoretical perspective
2. Mathematical Rigor: Rigorous derivations, clear formula presentation, with detailed mathematical proofs in appendices
3. Interdisciplinary Integration: Successfully integrates concepts from thermodynamics, information theory, economics, and decision theory
4. Unifying Nature: Provides a unified understanding framework for stochastic thermodynamics and resource theory

Weaknesses

1. Limited Practical Applicability: The theoretical framework is quite abstract, with considerable distance to practical applications
2. Insufficient Verification: Lacks numerical simulations or experimental validation to support theoretical predictions
3. Complexity: Cross-disciplinary concepts may be difficult for non-specialist readers to understand

Impact

1. Academic Value: Provides new theoretical tools and perspectives for non-equilibrium thermodynamics
2. Inspirational Significance: May inspire more interdisciplinary research directions
3. Methodological Contribution: Adversarial game methods may apply to other physical problems

Applicable Scenarios

1. Theoretical analysis of small-scale thermodynamic systems
2. Information thermodynamics research
3. Resource-theoretic analysis of quantum thermodynamics
4. Modeling energy conversion processes in biological systems

References

The paper cites 32 important references spanning multiple fields including stochastic thermodynamics, resource theory, information theory, and economics, providing a solid theoretical foundation for interdisciplinary research.

Overall Assessment: This is a theoretically innovative interdisciplinary paper that successfully unifies thermodynamics, information theory, and economic theory within an adversarial game framework. While highly theoretical in nature, it provides a novel perspective for understanding finite-scale thermodynamic systems and possesses significant academic value and inspirational merit.