pith. sign in

arxiv: 2606.01179 · v1 · pith:C6OHNT5Wnew · submitted 2026-05-31 · 💻 cs.LG · cs.AI

Physics-Informed Deep Learning for Entropy Prediction in Heterogeneous Systems: Thermodynamic and Information-Theoretic Case Studies

Biswajeet Sahoo , Debadutta Patra This is my paper

Pith reviewed 2026-06-28 17:55 UTC · model grok-4.3

classification 💻 cs.LG cs.AI
keywords physics-informed neural networksentropy productionSecond Law of ThermodynamicsFokker-Planck equationSoftplus activationdata efficiencyRuppeiner geometry
0
0 comments X

The pith

A unified neural framework enforces the Second Law exactly while learning entropy from reactor ODEs and market PDEs.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper presents a Physics-Informed Deep Learning framework that embeds both differential-equation residuals and thermodynamic bounds inside one network architecture. It tests the approach on a continuous stirred-tank reactor governed by ODEs and on an inverse Fokker-Planck model that recovers latent drift and diffusion from financial time series. Softplus activations are placed on selected outputs to keep entropy production non-negative and diffusion positive. The resulting models produce zero Second-Law violations and retain more than 90 percent accuracy when trained on only 30 percent of the data. A subsequent Ruppeiner geometric analysis of the learned entropy surface locates thermodynamic phase instabilities.

Core claim

By placing Softplus constraints on network outputs, the PIDL architecture solves the CSTR ODE system while satisfying the Second Law at every point and solves the inverse Fokker-Planck PDE while guaranteeing positive diffusion coefficients and naturally producing Shannon entropy; three model variants confirm that the shared-encoder version achieves absolute thermodynamic admissibility and high data efficiency across both domains.

What carries the argument

Softplus-constrained outputs inside a shared-encoder network that jointly minimizes PDE residuals and enforces non-negativity of entropy production and diffusion.

If this is right

  • The same architecture can be applied to other systems whose governing equations must respect the Second Law without post-hoc correction.
  • Post-training Ruppeiner analysis of the entropy surface can locate instabilities even when the network was trained only on sparse data.
  • Quantitative risk models in finance gain a built-in guarantee that inferred diffusion remains positive.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same positivity constraint pattern could be transferred to other conservation laws such as mass or energy balance in process models.
  • Data-efficiency results suggest the method may be useful when measurements are expensive or limited in real-time control settings.

Load-bearing premise

Enforcing Softplus constraints on selected network outputs is enough to make the Second Law hold exactly for every learned solution in both the reactor and financial models.

What would settle it

A single predicted entropy-production rate that is negative anywhere in the CSTR domain or a negative diffusion coefficient anywhere in the financial model would show that the admissibility guarantee does not hold.

Figures

Figures reproduced from arXiv: 2606.01179 by Biswajeet Sahoo, Debadutta Patra.

Figure 1
Figure 1. Figure 1: Schematic overview of the PIDL framework. (a) The unified architecture featuring [PITH_FULL_IMAGE:figures/full_fig_p006_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Schematic of the jacketed continuous stirred-tank reactor (CSTR). Entropy generation [PITH_FULL_IMAGE:figures/full_fig_p009_2.png] view at source ↗
Figure 4
Figure 4. Figure 4: Predicted versus reference (RK4) CSTR state trajectories. (a) Molar concen￾tration CA(t) and (b) bulk fluid tempera￾ture T(t). The PIDL model successfully cap￾tures the initial transient and steady-state regimes, outperforming the unconstrained baseline within the ±2ε noise bounds [PITH_FULL_IMAGE:figures/full_fig_p013_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Predicted volumetric entropy generation rate [PITH_FULL_IMAGE:figures/full_fig_p013_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Predicted probability density function p(x, t) of S&P 500 log-returns. (a) Spatiotem￾poral contour map and (b) selected cross-sectional marginal distributions. The PIDL solutions match KDE estimates, accurately capturing leptokurtic transitions during financial crises. 14 [PITH_FULL_IMAGE:figures/full_fig_p014_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Temporal evolution of Shannon entropy H(t) (2000–2022). Entropy peaks correlate strongly with NBER recession periods and systemic market shocks (e.g., 2008 Lehman collapse), serving as a quantitative indicator of market disorder. 15 [PITH_FULL_IMAGE:figures/full_fig_p015_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Inferred drift and diffusion coefficients. (a) Mean-reverting drift function [PITH_FULL_IMAGE:figures/full_fig_p016_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Performance evaluation of different PIDL model variants. Variant III (shared-encoder [PITH_FULL_IMAGE:figures/full_fig_p017_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: t-SNE projection of the shared latent space [PITH_FULL_IMAGE:figures/full_fig_p018_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Data efficiency learning curves. (a) CSTR concentration MAPE and (b) Shannon [PITH_FULL_IMAGE:figures/full_fig_p019_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: Ruppeiner scalar curvature map R(R) (T, CA). Derived entirely from the PIDL￾learned entropy surface, regions of negative curvature R(R) < 0 successfully identify physical domains prone to thermodynamic instability without requiring explicit bifurcation training. 9 Discussion The results presented in Sections 5–8 collectively establish the PIDL framework as a technically sound and computationally efficient… view at source ↗
read the original abstract

Entropy production governs irreversibility and uncertainty in both physical and information-theoretic systems. While Physics-Informed Neural Networks (PINNs) successfully solve differential equations, current architectures remain inherently domain-specific. The extraction of domain-invariant entropy representations across fundamentally different physical laws remains unexplored. This paper introduces a unified Physics-Informed Deep Learning (PIDL) framework that simultaneously enforces differential equation residuals and information-theoretic bounds within a single neural architecture. We demonstrate this framework via two canonical studies: (i) a thermodynamic continuous stirred-tank reactor (CSTR) model solving governing ODEs, where a Softplus constraint strictly enforces the Second Law of Thermodynamics; and (ii) an information-theoretic financial market model solving the inverse Fokker-Planck PDE to infer latent drift and diffusion coefficients, guaranteeing diffusion positivity via a Softplus constraint while naturally inducing Shannon entropy. Three model variants are evaluated: two domain-specific baselines and one shared-encoder architecture. The PIDL framework guarantees absolute thermodynamic admissibility with zero Second-Law violations and exhibits exceptional data efficiency, retaining >90% predictive accuracy using merely 30% of available training data. Furthermore, a post-hoc Ruppeiner Riemannian geometric analysis of the learned entropy surface successfully identifies thermodynamic phase instabilities. This methodology provides a robust, domain-agnostic architecture for physics-constrained entropy modeling, advancing applications in sustainable process design and quantitative financial risk assessment.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

3 major / 0 minor

Summary. The paper introduces a unified Physics-Informed Deep Learning (PIDL) framework that enforces both differential-equation residuals and information-theoretic bounds in a single neural architecture. It demonstrates the approach on two case studies: (i) a thermodynamic CSTR model whose governing ODEs are solved subject to a Softplus constraint asserted to enforce the Second Law exactly, and (ii) an inverse Fokker-Planck PDE for a financial market model in which Softplus is used to guarantee positive diffusion while inducing Shannon entropy. Three architectures (two domain-specific baselines and one shared-encoder) are compared; the manuscript claims zero Second-Law violations, retention of >90 % predictive accuracy with only 30 % of the training data, and successful post-hoc Ruppeiner geometric identification of thermodynamic instabilities.

Significance. If the enforcement mechanism can be shown to guarantee thermodynamic admissibility and the accuracy claims are substantiated with quantitative metrics and baselines, the work would offer a domain-agnostic architecture for entropy-constrained learning with clear relevance to sustainable process design and quantitative finance. The combination of physics residuals, positivity constraints, and subsequent Riemannian analysis is conceptually attractive, but the absence of supporting evidence in the abstract leaves the practical significance difficult to assess.

major comments (3)
  1. [Abstract] Abstract: The central claim that the PIDL framework “guarantees absolute thermodynamic admissibility with zero Second-Law violations” is not supported by the stated architecture. The Softplus is described as acting on network outputs (concentrations or an auxiliary variable), yet entropy production σ in the CSTR ODE system is determined by the state trajectory and its time derivative; no equation is supplied showing that σ itself is the constrained non-negative quantity. Consequently the zero-violation guarantee does not logically follow from the given constraint.
  2. [Abstract] Abstract: The data-efficiency claim (“retaining >90 % predictive accuracy using merely 30 % of available training data”) is presented without any quantitative metric (e.g., relative L² error, MAE), baseline comparisons, error bars, or verification that the learned solutions satisfy the underlying ODE/PDE residuals. The same paragraph asserts “exceptional data efficiency” while supplying none of the standard diagnostics needed to evaluate it.
  3. [Abstract] Abstract: For the inverse Fokker-Planck financial model the manuscript states that Softplus “guarantees diffusion positivity,” but again supplies no explicit mapping from the constrained network output to the diffusion coefficient that appears in the entropy-production expression. Without this mapping the claim of exact thermodynamic/information-theoretic admissibility remains unsubstantiated.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the constructive comments on the abstract. We agree that the claims regarding thermodynamic admissibility and data efficiency require explicit supporting details within the abstract to be fully substantiated. We will revise the abstract in the resubmission to address these points directly while preserving conciseness. Point-by-point responses are provided below.

read point-by-point responses

  1. Referee: [Abstract] Abstract: The central claim that the PIDL framework “guarantees absolute thermodynamic admissibility with zero Second-Law violations” is not supported by the stated architecture. The Softplus is described as acting on network outputs (concentrations or an auxiliary variable), yet entropy production σ in the CSTR ODE system is determined by the state trajectory and its time derivative; no equation is supplied showing that σ itself is the constrained non-negative quantity. Consequently the zero-violation guarantee does not logically follow from the given constraint.

    Authors: We agree that the abstract does not supply the explicit mapping or equation. In the manuscript, an auxiliary network output is defined as the entropy production rate σ to which Softplus is applied, ensuring σ ≥ 0 by construction before the ODE residuals are enforced. We will revise the abstract to include a concise statement of this construction (e.g., “with Softplus applied directly to the entropy production rate”). revision: yes

  2. Referee: [Abstract] Abstract: The data-efficiency claim (“retaining >90 % predictive accuracy using merely 30 % of available training data”) is presented without any quantitative metric (e.g., relative L² error, MAE), baseline comparisons, error bars, or verification that the learned solutions satisfy the underlying ODE/PDE residuals. The same paragraph asserts “exceptional data efficiency” while supplying none of the standard diagnostics needed to evaluate it.

    Authors: The referee is correct that the abstract states the claim without accompanying quantitative diagnostics. The body of the manuscript reports relative L² errors, baseline comparisons, error bars, and residual norms. We will revise the abstract to incorporate a brief quantitative qualifier or reference to these results. revision: yes

  3. Referee: [Abstract] Abstract: For the inverse Fokker-Planck financial model the manuscript states that Softplus “guarantees diffusion positivity,” but again supplies no explicit mapping from the constrained network output to the diffusion coefficient that appears in the entropy-production expression. Without this mapping the claim of exact thermodynamic/information-theoretic admissibility remains unsubstantiated.

    Authors: We concur that the abstract omits the explicit mapping. The manuscript sets the diffusion coefficient equal to the Softplus of the relevant network output, which is then substituted into the entropy-production term. We will add a clarifying phrase to the abstract stating this mapping. revision: yes

Circularity Check

1 steps flagged

Softplus constraint on outputs makes 'absolute thermodynamic admissibility' and 'zero Second-Law violations' true by construction

specific steps
  1. self definitional [Abstract]
    "where a Softplus constraint strictly enforces the Second Law of Thermodynamics; ... guaranteeing diffusion positivity via a Softplus constraint while naturally inducing Shannon entropy. The PIDL framework guarantees absolute thermodynamic admissibility with zero Second-Law violations"

    The zero-violation guarantee is obtained by making the constrained quantity (entropy production or diffusion) the direct Softplus(NN) output; non-negativity therefore holds identically by the activation function, not as a derived property of the learned trajectory or PDE residual.

full rationale

The paper's headline guarantee of zero Second-Law violations is achieved by directly constraining the relevant network output (entropy production or diffusion coefficient) with Softplus, so non-negativity holds by the choice of activation rather than emerging from the ODE/PDE solution or independent verification. The data-efficiency claim (>90% accuracy on 30% data) is an empirical fit result on the training distribution. No external benchmarks or parameter-free derivations are invoked to support the admissibility claim beyond the architectural constraint itself.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 0 invented entities

Review performed on abstract only; ledger entries are therefore limited to elements explicitly named in the abstract. The central claims rest on the assumption that Softplus activations enforce physical bounds exactly and that a shared encoder extracts invariant entropy features.

free parameters (1)
  • neural network parameters
    Weights and biases fitted during training to minimize residuals plus constraints; standard in deep learning but not enumerated.
axioms (1)
  • domain assumption Softplus activation strictly enforces the Second Law of Thermodynamics and diffusion positivity
    Invoked to guarantee thermodynamic admissibility and positive diffusion in both case studies.

pith-pipeline@v0.9.1-grok · 5781 in / 1451 out tokens · 25617 ms · 2026-06-28T17:55:17.827164+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

50 extracted references · 2 canonical work pages · 1 internal anchor

  1. [1]

    Clausius, R. (1865). Ueber verschiedene f¨ ur die Anwendung bequeme Formen der Haupt- gleichungen der mechanischen W¨ armetheorie.Annalen der Physik, 125(7), 353–400

  2. [2]

    (1967).Introduction to Thermodynamics of Irreversible Processes, 3rd ed

    Prigogine, I. (1967).Introduction to Thermodynamics of Irreversible Processes, 3rd ed. Interscience Publishers, New York

    1967

  3. [3]

    R., & Mazur, P

    de Groot, S. R., & Mazur, P. (1984).Non-Equilibrium Thermodynamics. Dover Publica- tions, New York

    1984

  4. [4]

    Shannon, C. E. (1948). A mathematical theory of communication.Bell System Technical Journal, 27(3), 379–423

    1948

  5. [5]

    M., & Thomas, J

    Cover, T. M., & Thomas, J. A. (2006).Elements of Information Theory, 2nd ed. John Wiley & Sons, Hoboken, NJ

    2006

  6. [6]

    (2016).Advanced Engineering Thermodynamics, 4th ed

    Bejan, A. (2016).Advanced Engineering Thermodynamics, 4th ed. John Wiley & Sons, Hoboken, NJ

    2016

  7. [7]

    Callen, H. B. (1985).Thermodynamics and an Introduction to Thermostatistics, 2nd ed. John Wiley & Sons, New York

    1985

  8. [8]

    N., & Stanley, H

    Mantegna, R. N., & Stanley, H. E. (1999).An Introduction to Econophysics: Correlations and Complexity in Finance. Cambridge University Press, Cambridge

    1999

  9. [9]

    (2004).Financial Modelling with Jump Processes

    Cont, R., & Tankov, P. (2004).Financial Modelling with Jump Processes. Chapman & Hall/CRC, Boca Raton, FL

    2004

  10. [10]

    Raissi, M., Perdikaris, P., & Karniadakis, G. E. (2019). Physics-informed neural networks: A deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations.Journal of Computational Physics, 378, 686–707

    2019

  11. [11]

    D., & Karniadakis, G

    Jagtap, A. D., & Karniadakis, G. E. (2020). Extended physics-informed neural networks (XPINNs): A generalized space-time domain decomposition based deep learning framework for nonlinear partial differential equations.Communications in Computational Physics, 28(5), 2002–2041

    2020

  12. [12]

    Lu, L., Meng, X., Mao, Z., & Karniadakis, G. E. (2021). DeepXDE: A deep learning library for solving differential equations.SIAM Review, 63(1), 208–228. 23

    2021

  13. [13]

    Wang, S., Sankaran, S., & Perdikaris, P. (2022). Respecting causality is all you need for training physics-informed neural networks.arXiv, 2203.07404

    work page arXiv 2022

  14. [14]

    E., Kevrekidis, I

    Karniadakis, G. E., Kevrekidis, I. G., Lu, L., Perdikaris, P., Wang, S., & Yang, L. (2021). Physics-informed machine learning.Nature Reviews Physics, 3(6), 422–440

    2021

  15. [15]

    G., Pearlmutter, B

    Baydin, A. G., Pearlmutter, B. A., Radul, A. A., & Siskind, J. M. (2018). Automatic differ- entiation in machine learning: A survey.Journal of Machine Learning Research, 18(153), 1–43

    2018

  16. [16]

    D., & Karniadakis, G

    Mao, Z., Jagtap, A. D., & Karniadakis, G. E. (2020). Physics-informed neural networks for high-speed flows.Computer Methods in Applied Mechanics and Engineering, 360, 112789

    2020

  17. [17]

    Haghighat, E., Raissi, M., Moure, A., Gomez, H., & Juanes, R. (2021). A physics-informed deep learning framework for inversion and surrogate modeling in solid mechanics.Computer Methods in Applied Mechanics and Engineering, 379, 113741

    2021

  18. [18]

    He, Q., & Tartakovsky, A. M. (2021). Physics-informed neural network method for for- ward and backward advection-dispersion equations.Water Resources Research, 57(7), e2020WR029479

    2021

  19. [19]

    Onsager, L. (1931). Reciprocal relations in irreversible processes I.Physical Review, 37(4), 405–426

    1931

  20. [20]

    (2014).Modern Thermodynamics: From Heat Engines to Dissipative Structures, 2nd ed

    Kondepudi, D., & Prigogine, I. (2014).Modern Thermodynamics: From Heat Engines to Dissipative Structures, 2nd ed. John Wiley & Sons, Chichester

    2014

  21. [21]

    S., & Salamon, P

    Andresen, B., Berry, R. S., & Salamon, P. (1984). Thermodynamics in finite time.Physics Today, 37(9), 62–70

    1984

  22. [22]

    Y., Wan Alwi, S

    Liew, P. Y., Wan Alwi, S. R., Klemeˇ s, J. J., Varbanov, P. S., & Manan, Z. A. (2013). Total site heat integration with seasonal energy availability.Chemical Engineering Transactions, 35, 19–24

    2013

  23. [23]

    (1989).The Fokker–Planck Equation: Methods of Solution and Applications, 2nd ed

    Risken, H. (1989).The Fokker–Planck Equation: Methods of Solution and Applications, 2nd ed. Springer-Verlag, Berlin

    1989

  24. [24]

    (2003).Stochastic Differential Equations: An Introduction with Applications, 6th ed

    Øksendal, B. (2003).Stochastic Differential Equations: An Introduction with Applications, 6th ed. Springer, Berlin

    2003

  25. [25]

    Black, F., & Scholes, M. (1973). The pricing of options and corporate liabilities.Journal of Political Economy, 81(3), 637–654

    1973

  26. [26]

    Cont, R. (2001). Empirical properties of asset returns: Stylized facts and statistical issues. Quantitative Finance, 1(2), 223–236

    2001

  27. [27]

    Caruana, R. (1997). Multitask learning.Machine Learning, 28(1), 41–75

    1997

  28. [28]

    Ruder, S. (2017). An overview of multi-task learning in deep neural networks.arXiv, 1706.05098

    work page internal anchor Pith review Pith/arXiv arXiv 2017

  29. [29]

    J., & Yang, Q

    Pan, S. J., & Yang, Q. (2010). A survey on transfer learning.IEEE Transactions on Knowl- edge and Data Engineering, 22(10), 1345–1359

    2010

  30. [30]

    D., & Karniadakis, G

    Perdikaris, P., Raissi, M., Damianou, A., Lawrence, N. D., & Karniadakis, G. E. (2017). Nonlinear information fusion algorithms for data-efficient multi-fidelity modelling.Proceed- ings of the Royal Society A, 473(2198), 20160751. 24

    2017

  31. [31]

    Goswami, S., Anitescu, C., Chakraborty, S., & Rabczuk, T. (2020). Transfer learning enhanced physics informed neural network for phase-field modeling of fracture.Theoretical and Applied Fracture Mechanics, 106, 102447

    2020

  32. [32]

    Fogler, H. S. (2016).Elements of Chemical Reaction Engineering, 5th ed. Pearson, Upper Saddle River, NJ

    2016

  33. [33]

    Luyben, W. L. (1990).Process Modeling, Simulation, and Control for Chemical Engineers, 2nd ed. McGraw-Hill, New York

    1990

  34. [34]

    Cuomo, S., Cola, V. S. di, Giampaolo, F., Rozza, G., Raissi, M., & Piccialli, F. (2022). Scientific machine learning through physics-informed neural networks: Where we are and what’s next.Journal of Scientific Computing, 92(3), 88

    2022

  35. [35]

    Heston, S. L. (1993). A closed-form solution for options with stochastic volatility with applications to bond and currency options.Review of Financial Studies, 6(2), 327–343

    1993

  36. [36]

    Merton, R. C. (1976). Option pricing when underlying stock returns are discontinuous. Journal of Financial Economics, 3(1–2), 125–144

    1976

  37. [37]

    Bengio, Y., Courville, A., & Vincent, P. (2013). Representation learning: A review and new perspectives.IEEE Transactions on Pattern Analysis and Machine Intelligence, 35(8), 1798–1828

    2013

  38. [38]

    Liu, X.-Y., & Wang, J.-X. (2021). Physics-informed Dyna-style model-based deep rein- forcement learning for dynamic control.Proceedings of the Royal Society A: Mathematical, Physical and Engineering Sciences, 477, 20210618

    2021

  39. [39]

    Ruppeiner, G. (1995). Riemannian geometry in thermodynamic fluctuation theory.Reviews of Modern Physics, 67(3), 605–659

    1995

  40. [40]

    Ruppeiner, G. (2008). Thermodynamic curvature and phase transitions in Kerr–Newman black holes.Physical Review D, 78(2), 024016

    2008

  41. [41]

    (2016).Deep Learning

    Goodfellow, I., Bengio, Y., & Courville, A. (2016).Deep Learning. MIT Press, Cambridge, MA

    2016

  42. [42]

    Dugas, C., Bengio, Y., B´ elisle, F., Nadeau, C., & Garcia, R. (2000). Incorporating second- order functional knowledge for better option pricing.Advances in Neural Information Pro- cessing Systems, 13, 472–478

    2000

  43. [43]

    Yu, T., Kumar, S., Gupta, A., Levine, S., Hausman, K., & Finn, C. (2020). Gradient surgery for multi-task learning.Advances in Neural Information Processing Systems, 33, 5824–5836

    2020

  44. [44]

    P., & Ba, J

    Kingma, D. P., & Ba, J. L. (2015). Adam: A method for stochastic optimization. In Proceedings of the 3rd International Conference on Learning Representations (ICLR), San Diego, CA

    2015

  45. [45]

    Glorot, X., & Bengio, Y. (2010). Understanding the difficulty of training deep feedfor- ward neural networks. InProceedings of the 13th International Conference on Artificial Intelligence and Statistics (AISTATS), vol. 9, 249–256

    2010

  46. [46]

    Silverman, B. W. (1986).Density Estimation for Statistics and Data Analysis. Chapman & Hall, London

    1986

  47. [47]

    Van der Maaten, L., & Hinton, G. (2008). Visualizing data using t-SNE.Journal of Machine Learning Research, 9(86), 2579–2605. 25

    2008

  48. [48]

    Kovachki, N., Li, Z., Liu, B., Azizzadenesheli, K., Bhattacharya, K., Stuart, A., & Anand- kumar, A. (2023). Neural operator: Learning maps between function spaces with applica- tions to PDEs.Journal of Machine Learning Research, 24(89), 1–97

    2023

  49. [49]

    England, J. L. (2015). Dissipative adaptation in driven self-assembly.Nature Nanotechnol- ogy, 10(11), 919–923

    2015

  50. [50]

    Dewar, R. C. (2003). Information theory explanation of the fluctuation theorem, maximum entropy production and self-organized criticality in non-equilibrium stationary states.Jour- nal of Physics A: Mathematical and General, 36(3), 631–641. 26

    2003