pith. sign in

arxiv: 2605.15985 · v1 · pith:QILNBVLSnew · submitted 2026-05-15 · 🪐 quant-ph · cond-mat.stat-mech· cs.NE· physics.bio-ph

Thermodynamic Networks: Harnessing Non-Equilibrium Steady States for Computation

Patryk Lipka-Bartosik , Gianmichele Blasi , Javier Lalueza Pu\'ertolas , G\'eraldine Haack , Mart\'i Perarnau-Llobet , Nicolas Brunner This is my paper

Pith reviewed 2026-05-20 19:36 UTC · model grok-4.3

classification 🪐 quant-ph cond-mat.stat-mechcs.NEphysics.bio-ph
keywords thermodynamic networksnon-equilibrium steady statesnegative differential conductanceuniversal function approximationquantum dot networksenzymatic networksphysics-based computationreservoir computing
0
0 comments X

The pith

Negative differential conductance lets thermodynamic networks approximate any function using non-equilibrium steady states.

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

The paper introduces thermodynamic networks made of finite-size reservoirs that exchange conserved quantities and relax into a non-equilibrium steady state encoding a computational solution. It shows that the presence of negative differential conductance turns these networks into universal function approximators, while networks without it can only produce monotonic functions. Training exploits the system's natural drive toward equilibrium rather than external optimization. The approach is illustrated with quantum dot networks and enzymatic reaction networks on tasks such as sine approximation and MNIST classification. This establishes a direct physical mechanism linking non-equilibrium dynamics to computational expressivity.

Core claim

Thermodynamic networks consisting of finite reservoirs exchanging conserved quantities reach a non-equilibrium steady state that solves chosen computational problems. Negative differential conductance is the property that lifts the restriction to monotonic functions and enables approximation of arbitrary continuous functions. Without NDC the steady-state mapping remains monotonic; with NDC the network becomes a universal approximator. The networks are trained by protocols that harness equilibration, and both quantum-dot and enzymatic platforms are shown to achieve high performance once engineered for NDC.

What carries the argument

Negative Differential Conductance (NDC), the non-monotonic current-voltage response that supplies the non-linearity required for universal approximation in the steady-state mapping of conserved quantities.

If this is right

  • Quantum dot and enzymatic networks can be engineered for NDC to perform both regression and classification tasks.
  • Computation occurs autonomously once the network relaxes to its steady state, with no ongoing external driving required.
  • Training reduces to choosing initial conditions or couplings that let natural equilibration reach the desired encoding.
  • The framework applies to any platform where reservoirs exchange conserved quantities and NDC can be realized.

Where Pith is reading between the lines

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

  • Similar NDC-based mechanisms could appear in biological signaling or metabolic networks, offering a physical route to complex computation without explicit programming.
  • Energy costs of computation might be analyzed directly from the thermodynamic cost of maintaining the required non-equilibrium steady state.
  • Small arrays of quantum dots could serve as an experimental testbed to measure how closely approximation error tracks the strength of engineered NDC.
  • The approach suggests a design principle for other physical reservoirs: add tunable negative conductance regions to expand the class of representable functions.

Load-bearing premise

Finite-size reservoirs exchanging conserved quantities can be arranged and tuned so that their non-equilibrium steady state directly encodes the solution once negative differential conductance is present.

What would settle it

Construct a thermodynamic network deliberately lacking NDC and show it still approximates a non-monotonic target such as the sine function to high accuracy after standard equilibration training.

Figures

Figures reproduced from arXiv: 2605.15985 by G\'eraldine Haack, Gianmichele Blasi, Javier Lalueza Pu\'ertolas, Mart\'i Perarnau-Llobet, Nicolas Brunner, Patryk Lipka-Bartosik.

Figure 1
Figure 1. Figure 1: FIG. 1 [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: FIG. 2 [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: (b), and the corresponding sign reversal of the dif￾ferential conductance shown in [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: FIG. 4 [PITH_FULL_IMAGE:figures/full_fig_p008_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: FIG. 5 [PITH_FULL_IMAGE:figures/full_fig_p009_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: FIG. 6 [PITH_FULL_IMAGE:figures/full_fig_p010_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: FIG. 7 [PITH_FULL_IMAGE:figures/full_fig_p011_7.png] view at source ↗
read the original abstract

We introduce thermodynamic networks, a general framework for autonomous, physics-based computation using non-equilibrium steady states. These networks are modeled as a collection of finite-size reservoirs that exchange conserved quantities--such as electric charge or molecular number--while relaxing to a non-equilibrium steady state, which encodes the solution of a computational problem. We identify Negative Differential Conductance (NDC) as the critical physical property governing the computational expressivity of the thermodynamic network. While networks lacking NDC are restricted to computing monotonic functions, the presence of NDC enables universal function approximation. For the training of the network, we use protocols that take advantage of the natural tendency of the system to equilibrate. We illustrate the versatility of our approach via two different platforms: quantum dot networks and enzymatic reaction networks. Both systems can be engineered to have NDC, enabling high performance in standard benchmarks, including sine function approximation and MNIST digit classification. Overall, our work establishes a rigorous link between non-equilibrium steady states and computational expressivity.

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 / 2 minor

Summary. The manuscript introduces thermodynamic networks as collections of finite-size reservoirs exchanging conserved quantities (e.g., charge or molecular number) that relax to a non-equilibrium steady state (NESS) encoding the solution to a computational problem. The central claim is that Negative Differential Conductance (NDC) is the key property enabling universal function approximation, while its absence restricts the networks to monotonic functions only. Training exploits natural relaxation dynamics rather than explicit fitting, and the framework is illustrated on quantum-dot and enzymatic-reaction platforms for sine approximation and MNIST classification.

Significance. If the central claims are rigorously established, the work would provide a concrete link between non-equilibrium thermodynamics and computational expressivity, potentially opening routes to autonomous, low-energy physical computers. The emphasis on NDC as the expressivity switch and the use of equilibration for training are distinctive strengths; the two concrete platforms supply falsifiable benchmarks that could be reproduced.

major comments (3)
  1. [Abstract / central claim on NDC] The universality claim (NDC enables approximation of arbitrary continuous functions via NESS observables) is load-bearing yet unsupported by any derivation or density argument in the provided text. No explicit map from inputs to steady-state currents/potentials is shown to be dense in C[0,1], nor is a reference to a relevant theorem supplied.
  2. [Discussion of NDC and steady-state encoding] The assumption that engineering NDC yields a unique, asymptotically stable NESS is not justified. Standard circuit theory and non-equilibrium thermodynamics show that NDC regions frequently produce multiple fixed points or limit cycles; the manuscript supplies neither Lyapunov conditions nor numerical evidence confirming uniqueness once NDC is present.
  3. [Training protocols] The training protocol is described only at the level of 'natural relaxation.' It is unclear how the target function is encoded in the choice of reservoir parameters or boundary conditions without introducing hidden fitting degrees of freedom that undermine the claimed parameter-free character.
minor comments (2)
  1. Notation for conserved quantities, reservoir indices, and the precise definition of 'thermodynamic network' should be introduced with an explicit diagram or set of equations early in the manuscript.
  2. Performance numbers for sine approximation and MNIST (error metrics, comparison baselines) are mentioned but not tabulated; a small results table would improve clarity.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the careful reading and constructive criticism of our manuscript. The comments identify important points that require clarification and strengthening. We respond to each major comment below and indicate the revisions we will make.

read point-by-point responses

  1. Referee: [Abstract / central claim on NDC] The universality claim (NDC enables approximation of arbitrary continuous functions via NESS observables) is load-bearing yet unsupported by any derivation or density argument in the provided text. No explicit map from inputs to steady-state currents/potentials is shown to be dense in C[0,1], nor is a reference to a relevant theorem supplied.

    Authors: We agree that a fully rigorous density argument establishing that NDC-equipped networks can approximate arbitrary continuous functions is not developed in the current text. The manuscript instead demonstrates the claim through explicit constructions: the quantum-dot and enzymatic examples show that NDC permits non-monotonic steady-state responses (e.g., sine approximation) while its absence restricts the networks to monotonic maps. To address the referee’s concern, we will add a new subsection that sketches a density argument. Specifically, we will show that the steady-state current map, when NDC is present, can realize a family of non-monotonic basis functions whose linear combinations are dense in C[0,1] by appealing to the Stone–Weierstrass theorem after a suitable change of variables induced by the NDC characteristic. We will also cite relevant results from non-equilibrium circuit theory that support the existence of such maps. This addition will be placed after the general framework section. revision: partial

  2. Referee: [Discussion of NDC and steady-state encoding] The assumption that engineering NDC yields a unique, asymptotically stable NESS is not justified. Standard circuit theory and non-equilibrium thermodynamics show that NDC regions frequently produce multiple fixed points or limit cycles; the manuscript supplies neither Lyapunov conditions nor numerical evidence confirming uniqueness once NDC is present.

    Authors: The referee correctly notes that NDC can destabilize fixed points in general dynamical systems. In the manuscript we implicitly assume that device parameters are chosen so that a unique, globally attractive NESS exists; this assumption is verified numerically for the specific quantum-dot and enzymatic networks we simulate. We acknowledge that a general Lyapunov or contraction-mapping argument is absent. We will therefore revise the stability discussion to include (i) a brief statement of the conditions under which the network equations remain contractive even with NDC (drawing on standard results for monotone systems with bounded negative slopes), and (ii) additional numerical diagnostics (phase portraits and eigenvalue spectra of the Jacobian at the operating point) confirming uniqueness and asymptotic stability for the parameter regimes used in the sine and MNIST examples. These additions will appear in the platform-specific sections. revision: yes

  3. Referee: [Training protocols] The training protocol is described only at the level of 'natural relaxation.' It is unclear how the target function is encoded in the choice of reservoir parameters or boundary conditions without introducing hidden fitting degrees of freedom that undermine the claimed parameter-free character.

    Authors: We clarify that the term “parameter-free” refers to the absence of iterative numerical optimization (gradient descent, back-propagation, etc.). The physical parameters—reservoir volumes, baseline conductances, and fixed boundary chemical potentials—are chosen once, on the basis of the problem statement, and then held constant while the system relaxes. The target function is encoded by mapping the input variable to a boundary condition (e.g., input voltage or substrate concentration) and reading the desired output from a steady-state observable (current or product flux). Because the mapping is fixed by the physical topology and the NDC characteristic, no additional tunable weights are introduced after the initial design. We will expand the training-protocol paragraph to make this encoding explicit, including a step-by-step description of how the sine and MNIST tasks are mapped onto boundary conditions, and we will add a short paragraph contrasting this approach with conventional machine-learning training to emphasize the lack of hidden fitting degrees of freedom. revision: partial

Circularity Check

0 steps flagged

No significant circularity; derivation self-contained

full rationale

The paper introduces thermodynamic networks as a framework where non-equilibrium steady states encode computations, with NDC identified as the key property separating monotonic from universal approximation capabilities. Training exploits natural relaxation to equilibrium rather than explicit parameter fitting to target functions. No load-bearing steps reduce claims to self-definitions, fitted inputs renamed as predictions, or self-citation chains; the link between NDC and expressivity is presented via physical modeling and concrete platforms (quantum dots, enzymatic networks) with benchmarks, remaining independent of the target results by construction.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 1 invented entities

The framework rests on standard non-equilibrium thermodynamics plus the new postulate that steady-state flows can be made to represent arbitrary functions via NDC; no explicit free parameters are named in the abstract.

axioms (1)
  • domain assumption Finite-size reservoirs exchanging conserved quantities relax to a non-equilibrium steady state whose flows encode computational outputs.
    Invoked in the opening definition of thermodynamic networks.
invented entities (1)
  • thermodynamic network no independent evidence
    purpose: General framework for autonomous physics-based computation
    Newly introduced collection of reservoirs whose steady state solves problems.

pith-pipeline@v0.9.0 · 5744 in / 1201 out tokens · 48330 ms · 2026-05-20T19:36:24.897445+00:00 · methodology

discussion (0)

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

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

112 extracted references · 112 canonical work pages · 1 internal anchor

  1. [1]

    Physical setup The network consists of electronic reservoirs (the nodes) connected by single-level quantum dots (the edges), as shown Fig. 2. A reservoirαis characterized by a chemical potentialµ α and temperatureT α. Each quantum dot sitting on the edge connecting two reser- voirs(α, β)has a single discrete energy levelϵαβ through which electrons tunnel ...

    work page

  2. [2]

    Transport through a quantum dot The steady-state current through a quantum dot con- necting reservoirsαandβcan be expressed as the weak- coupling limit of coherent transport through a single res- onant level, Iαβ =γ αβ fα(ϵαβ)−f β(ϵαβ) ,(15) whereγ αβ is the tunnel coupling and fℓ(E) = 1 1 +e (E−µℓ)/kB Tℓ (16) is the Fermi–Dirac distribution of reservoirℓ...

    work page

  3. [3]

    NDC mechanism (electrostatic shift) Under certain conditions, the energy level of a quan- tum dot can be shifted by its electrostatic environment, in particular by the potentialsµα andµ β of the two reser- voirs connected to the dot. Here we model this depen- dence as a linearelectrostatic shift, namely ϵαβ =a αβ +b αβ (µα −µ β),(17) wherea αβ is the bare...

    work page

  4. [4]

    Breaking cooperativity Cooperativity requiresG αβ ≥0for all edges and all operating points (see Sec. IID). Using Eq. (18) it then follows that this is true if and only if Tβ ˜fα Tα ˜fβ ≤ 1 +b αβ bαβ .(19) The right-hand side specifies thecooperativity threshold, namely the maximum asymmetry between source and drain that ensures cooperative behaviour. Nota...

    work page

  5. [5]

    For that we consider a network with architecture 2-2-1, i.e

    A minimal example: XOR problem We now test the mechanism of NDC on the simplest non-monotone Boolean function: exclusive-or (XOR). For that we consider a network with architecture 2-2-1, i.e. two input reservoirs(x 1, x2), two hidden reservoirs (h1, h2), and one output(y), fully connected by quantum dots with parameters{w αβ, aαβ, bαβ}. We then train the ...

    work page

  6. [6]

    We now test whether the same prin- ciple can be scaled up to realistic problems

    Scaling to image classification The XOR example demonstrates the mechanism of folding due to NDC. We now test whether the same prin- ciple can be scaled up to realistic problems. For that we consider the MNIST handwritten digit classification benchmark [42]: Classifying28×28grayscale images into ten digit categories (0–9). Each input pixel is encoded as t...

    work page

  7. [7]

    There we show the performance of thermodynamic networks operating in different transport regimes as a function of the number of neurons in the hidden layer. Although the high dimensionality of the parameter space leads to challenging training scenarios where the network can converge to multistable solutions, the final networks achieve robust performance o...

    work page

  8. [8]

    reservoirs that contain molecules of different chemical species with varying concentrationsc α

    Physical setup The nodes of the network are pools of molecular sub- strates, i.e. reservoirs that contain molecules of different chemical species with varying concentrationsc α. The transport between two nodes is mediated by an enzyme that catalyzes the conversion between compoundαinto another compoundβ. Each such reaction defines an edge of the network, ...

    work page

  9. [9]

    These parameters satisfy the analog of lo- cal detailed balance, known as Haldane relation [47], Vf Kβ/(VrKα) =K eq, whereK eq is the equilibrium con- stant of the reaction

    Enzymatic transport A simple and well-studied model of reversible enzyme kinetics is the Michaelis–Menten mechanism [47], which expresses the reaction flux fromαtoβas Iαβ(cα,cβ) = 1 1 + cα Kα + cβ Kβ Vf Kα cα − Vr Kβ cβ ,(22) whereV f /r are the tunable parameters that character- ize the maximal forward and reverse rates, andK α/β, are the Michaelis const...

    work page

  10. [10]

    This is a generic phenomenon made possi- ble bysubstrate inhibition[26]

    NDC mechanism (substrate inhibition) Toaccessnon-monotonecomputation, weneedachem- ical analogue of NDC, namely a mechanism that makes the reaction flux respond non-monotonically to substrate concentration. This is a generic phenomenon made possi- ble bysubstrate inhibition[26]. More specifically, at high concentrations, excess substrate molecules bind to...

    work page

  11. [11]

    Wenowturntoregression, i.e

    Function regression ThenumericalexperimentsperformedontheXORand MNIST tasks demonstrated classification with quantum networks. Wenowturntoregression, i.e. fittingacontin- uous mapping rather partitioning the input space using both presented platforms. This is a natural benchmark for thermodynamic networks, as the steady-state output potential is by defaul...

    work page doi:10.13039/501100011033 2023

  12. [12]

    Kaspar, B

    C. Kaspar, B. J. Ravoo, W. G. van der Wiel, S. V. Weg- ner, and W. H. Pernice, Nature594, 345 (2021)

    work page 2021

  13. [13]

    L. G. Wright, T. Onodera, M. M. Stein, T. Wang, D. T. Schachter, Z. Hu, and P. L. McMahon, Nature601, 549 (2022)

    work page 2022

  14. [14]

    Marković, A

    D. Marković, A. Mizrahi, D. Querlioz, and J. Grollier, Nature Reviews Physics2, 499 (2020)

    work page 2020

  15. [15]

    Tanaka, T

    G. Tanaka, T. Yamane, J. B. Héroux, R. Nakane, N. Kanazawa, S. Takeda, H. Numata, D. Nakano, and A. Hirose, Neural Networks115, 100–123 (2019)

    work page 2019

  16. [16]

    Mohseni, P

    N. Mohseni, P. L. McMahon, and T. Byrnes, Nature Re- views Physics4, 363 (2022)

    work page 2022

  17. [17]

    Conte, E

    T. Conte, E. DeBenedictis, N. Ganesh, T. Hylton, J. P. Strachan, R. S. Williams, A. Alemi, L. Al- tenberg, G. Crooks, J. Crutchfield,et al., arXiv preprint arXiv:1911.01968 (2019)

    work page arXiv 1911

  18. [18]

    Seifert, Reports on progress in physics75, 126001 (2012)

    U. Seifert, Reports on progress in physics75, 126001 (2012)

    work page 2012

  19. [19]

    Aifer, K

    M. Aifer, K. Donatella, M. H. Gordon, S. Duffield, T. Ahle, D. Simpson, G. Crooks, and P. J. Coles, npj Unconventional Computing1, 13 (2024)

    work page 2024

  20. [20]

    Moroder, F

    M. Moroder, F. C. Binder, and J. Goold, Digitally opti- mized initializations for fast thermodynamic computing (2026)

    work page 2026

  21. [21]

    Lipka-Bartosik, K

    P. Lipka-Bartosik, K. Donatella, M. Aifer, D. Melanson, M. Perarnau-Llobet, N. Brunner, and P. J. Coles, in2024 IEEE International Conference on Rebooting Computing (ICRC)(IEEE, 2024) pp. 1–13

    work page 2024

  22. [22]

    P. J. Coles, C. Szczepanski, D. Melanson, K. Donatella, A. J. Martinez, and F. Sbahi, in2023 IEEE Inter- national Conference on Rebooting Computing (ICRC) (IEEE, 2023) pp. 1–10

    work page 2023

  23. [23]

    Aifer, S

    M. Aifer, S. Duffield, K. Donatella, D. Melanson, P. Klett, Z. Belateche, G. Crooks, A. J. Martinez, and P. J. Coles, Thermodynamic bayesian inference (2024), arXiv:2410.01793 [cond-mat.stat-mech]

    work page arXiv 2024

  24. [24]

    Whitelam, Physical Review Letters136, 037101 (2026)

    S. Whitelam, Physical Review Letters136, 037101 (2026)

    work page 2026

  25. [25]

    Whitelam and C

    S. Whitelam and C. Casert, Nature Communications (2026)

    work page 2026

  26. [26]

    Rolandi, P

    A. Rolandi, P. Abiuso, P. Lipka-Bartosik, M. Aifer, P. J. Coles, and M. Perarnau-Llobet, Energy-time- accuracy tradeoffs in thermodynamic computing (2026), arXiv:2601.04358 [cond-mat.stat-mech]

    work page arXiv 2026

  27. [27]

    Lipka-Bartosik, M

    P. Lipka-Bartosik, M. Perarnau-Llobet, and N. Brunner, Science advances10, eadm8792 (2024)

    work page 2024

  28. [28]

    S. Bai, J. Z. Kolter, and V. Koltun, Deep equilibrium models (2019), arXiv:1909.01377 [cs.LG]

    work page arXiv 2019

  29. [29]

    H. L. Smith,Monotone dynamical systems: an introduc- tion to the theory of competitive and cooperative systems: an introduction to the theory of competitive and coopera- tive systems, 41 (American Mathematical Soc., 1995)

    work page 1995

  30. [30]

    X. Liu, X. Han, N. Zhang, and Q. Liu, Advances in Neu- ral Information Processing Systems33, 15427 (2020)

    work page 2020

  31. [31]

    Runje and S

    D. Runje and S. M. Shankaranarayana, inInternational Conference on Machine Learning(PMLR, 2023) pp. 29338–29353

    work page 2023

  32. [32]

    Chen and W

    D. Chen and W. Ye, inProceedings of the third ACM international conference on AI in finance(2022) pp. 70– 78

    work page 2022

  33. [33]

    Esaki, Physical review109, 603 (1958)

    L. Esaki, Physical review109, 603 (1958)

    work page 1958

  34. [34]

    Esposito, U

    M. Esposito, U. Harbola, and S. Mukamel, Reviews of modern physics81, 1665 (2009)

    work page 2009

  35. [35]

    Baiesi, C

    M. Baiesi, C. Maes, and B. Wynants, Physical review letters103, 010602 (2009)

    work page 2009

  36. [36]

    W. G. Van der Wiel, S. De Franceschi, J. M. Elzerman, T. Fujisawa, S. Tarucha, and L. P. Kouwenhoven, Re- views of modern physics75, 1 (2002)

    work page 2002

  37. [37]

    M. C. Reed, A. Lieb, and H. F. Nijhout, Bioessays32, 422 (2010)

    work page 2010

  38. [38]

    Cybenko, Mathematics of Control, Signals, and Sys- tems2, 303 (1989), aDS Bibcode: 1989MCSS....2..303C

    G. Cybenko, Mathematics of Control, Signals, and Sys- tems2, 303 (1989), aDS Bibcode: 1989MCSS....2..303C

    work page 1989

  39. [39]

    Hornik, Neural networks4, 251 (1991)

    K. Hornik, Neural networks4, 251 (1991)

    work page 1991

  40. [40]

    D. E. Rumelhart, G. E. Hinton, and R. J. Williams, na- ture323, 533 (1986)

    work page 1986

  41. [41]

    Scellier and Y

    B. Scellier and Y. Bengio, Frontiers in Computational Neuroscience11, 24 (2017)

    work page 2017

  42. [42]

    Lipka-Bartosik, G

    P. Lipka-Bartosik, G. Blasi, J. Lalueza Puértolas, G. Haack, M. Perarnau-Llobet, and N. Brunner, Ther- moNet: A code library for physics-based computation with non-equilibrium steady states,https://github. com/patryk-lb/ThermoNet(2026)

    work page 2026

  43. [43]

    Büttiker, Physical review letters57, 1761 (1986)

    M. Büttiker, Physical review letters57, 1761 (1986)

    work page 1986

  44. [44]

    Blasi, S

    G. Blasi, S. Khandelwal, and G. Haack, Phys. Rev. Res. 6, 043091 (2024)

    work page 2024

  45. [45]

    Imry and R

    Y. Imry and R. Landauer, Reviews of Modern Physics 71, S306 (1999)

    work page 1999

  46. [46]

    Y. V. Nazarov and Y. M. Blanter,Quantum transport: introduction to nanoscience(Cambridge university press, 2009)

    work page 2009

  47. [47]

    Datta,Electronic Transport in Mesoscopic Systems (Cambridge university press, 1997)

    S. Datta,Electronic Transport in Mesoscopic Systems (Cambridge university press, 1997)

    work page 1997

  48. [48]

    Y. M. Blanter and M. Büttiker, Physics Reports336, 1 (2000)

    work page 2000

  49. [49]

    L. P. Kouwenhoven, C. M. Marcus, P. L. McEuen, S. Tarucha, R. M. Westervelt, and N. S. Wingreen, in Mesoscopic electron transport(Springer, 1997) pp. 105– 214

    work page 1997

  50. [50]

    Evers, R

    F. Evers, R. Korytár, S. Tewari, and J. M. Van Ruiten- beek, Reviews of Modern Physics92, 035001 (2020)

    work page 2020

  51. [51]

    J. P. Sun, G. I. Haddad, P. Mazumder, and J. N. Schul- man, Proceedings of the IEEE86, 641 (1998)

    work page 1998

  52. [52]

    Biele and R

    R. Biele and R. D’Agosta, Entropy21, 752 (2019)

    work page 2019

  53. [53]

    LeCun, L

    Y. LeCun, L. Bottou, Y. Bengio, and P. Haffner, Pro- ceedings of the IEEE86, 2278 (2002)

    work page 2002

  54. [54]

    Ghosh, M

    S. Ghosh, M. G. Baltussen, N. M. Ivanov, R. Haije, M. Jakstaite, T. Zhou, and W. T. Huck, Chemical re- views124, 2553 (2024)

    work page 2024

  55. [55]

    Kushwaha, and J.-L

    A.Pandi, M.Koch, P.L.Voyvodic, P.Soudier, J.Bonnet, M. Kushwaha, and J.-L. Faulon, Nature communications 10, 3880 (2019)

    work page 2019

  56. [56]

    Deteix, T

    S.Okumura, G.Gines, N.Lobato-Dauzier, A.Baccouche, R. Deteix, T. Fujii, Y. Rondelez, and A. J. Genot, Nature 610, 496 (2022). 13

    work page 2022

  57. [57]

    U. T. Bornscheuer, G. Huisman, R. Kazlauskas, S. Lutz, J. Moore, and K. Robins, Nature485, 185 (2012)

    work page 2012

  58. [58]

    Cornish-Bowden,Fundamentals of enzyme kinetics (John Wiley & Sons, 2013)

    A. Cornish-Bowden,Fundamentals of enzyme kinetics (John Wiley & Sons, 2013)

    work page 2013

  59. [59]

    J. D. Murray and J. D. Murray,Mathematical biology: II: spatial models and biomedical applications, Vol. 18 (Springer, 2003)

    work page 2003

  60. [60]

    R. A. Copeland,Enzymes: a practical introduction to structure, mechanism, and data analysis(John Wiley & Sons, 2023)

    work page 2023

  61. [61]

    Opala, S

    A. Opala, S. Ghosh, T. C. Liew, and M. Matuszewski, Physical Review Applied11, 064029 (2019)

    work page 2019

  62. [62]

    A. I. Luppi, Y. Sanz Perl, J. Vohryzek, H. Ali, P. A. Mediano, F. E. Rosas, F. Milisav, L. E. Suárez, S. Gini, D. Gutierrez-Barragan,et al., Nature Neuroscience , 1 (2026)

    work page 2026

  63. [63]

    T. Dou, S. Kumara, J. Burns, E. Sigler, P. Girdhar, D. Petty, G. Milburn, J. Plested, and M. Woolley, arXiv preprint arXiv:2604.10861 (2026)

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  64. [64]

    R. Li, Y. Gong, H. Huang, Y. Zhou, S. Mao, Z. Wei, and Z. Zhang, Advanced Materials37, 2312825 (2025)

    work page 2025

  65. [65]

    Marković and J

    D. Marković and J. Grollier, Applied physics letters117 (2020)

    work page 2020

  66. [66]

    Suderman and G

    S. Suderman and G. Rubino, Interpretable rule-based learning in an autonomous thermodynamic network (2026), to appear on arXiv

    work page 2026

  67. [67]

    Momeni, B

    A. Momeni, B. Rahmani, B. Scellier, L. G. Wright, P. L. McMahon, C. C. Wanjura, Y. Li, A. Skalli, N. G. Berloff, T. Onodera,et al., Nature645, 53 (2025)

    work page 2025

  68. [68]

    Lopez-Pastor and F

    V. Lopez-Pastor and F. Marquardt, Physical Review X 13, 031020 (2023)

    work page 2023

  69. [69]

    P. L. McMahon, Nature Reviews Physics5, 717 (2023)

    work page 2023

  70. [70]

    Mujal, R

    P. Mujal, R. Martínez-Peña, J. Nokkala, J. García-Beni, G. L. Giorgi, M. C. Soriano, and R. Zambrini, Advanced Quantum Technologies4, 2100027 (2021)

    work page 2021

  71. [71]

    Opala, K

    A. Opala, K. Tyszka, M. Kędziora, M. Furman, A. Rah- mani, S. Świerczewski, M. Ekielski, A. Szerling, M. Ma- tuszewski, and B. Piętka, Room temperature exciton- polariton neural network with perovskite crystal (2024), arXiv:2412.10865 [physics.optics]

    work page arXiv 2024

  72. [72]

    Martínez-Peña, G

    R. Martínez-Peña, G. L. Giorgi, J. Nokkala, M. C. So- riano, and R. Zambrini, Physical Review Letters127, 100502 (2021)

    work page 2021

  73. [73]

    D. H. Wolpert, Journal of Physics A: Mathematical and Theoretical52, 193001 (2019)

    work page 2019

  74. [74]

    D. H. Wolpert, J. Korbel, C. W. Lynn, F. Tasnim, J. A. Grochow, G. Kardes, J. B. Aimone, V. Balasub- ramanian, E. De Giuli, D. Doty, N. Freitas, M. Marsili, T. E. Ouldridge, A. W. Richa, P. Riechers, E. Roldan, B. Rubenstein, Z. Toroczkai, and J. Paradiso, Pro- ceedings of the National Academy of Sciences121, 10.1073/pnas.2321112121 (2024)

    work page doi:10.1073/pnas.2321112121 2024

  75. [75]

    Chattopadhyay, A

    P. Chattopadhyay, A. Misra, T. Pandit, and G. Paul, Reports on Progress in Physics88, 086001 (2025)

    work page 2025

  76. [76]

    de Oliveira Junior, J

    A. de Oliveira Junior, J. B. Brask, and R. Chaves, PRX Quantum6, 10.1103/phkv-wrsd (2025)

    work page doi:10.1103/phkv-wrsd 2025

  77. [77]

    Campbell, I

    S. Campbell, I. D’Amico, M. A. Ciampini, J. Anders, N. Ares, S. Artini, A. Auffèves, L. Bassman Oftelie, L. P. Bettmann, M. V. S. Bonança, T. Busch, M. Campisi, M. F. Cavalcante, L. A. Correa, E. Cuestas, C. B. Dag, S. Dago, S. Deffner, A. Del Campo, A. Deutschmann- Olek, S. Donadi, E. Doucet, C. Elouard, K. Ensslin, P. Erker, N. Fabbri, F. Fedele, G. Fiu...

    work page 2026

  78. [78]

    Goold, M

    J. Goold, M. Huber, A. Riera, L. d. Rio, and P. Skrzypczyk, Journal of Physics A: Mathematical and Theoretical49, 143001 (2016)

    work page 2016

  79. [79]

    Vinjanampathy and J

    S. Vinjanampathy and J. Anders, Contemporary Physics 57, 545 (2016)

    work page 2016

  80. [80]

    Manzano, G

    G. Manzano, G. Kardes, E. Roldan, and D. H. Wolpert, Physical Review X14, 10.1103/physrevx.14.021026 (2024)

    work page doi:10.1103/physrevx.14.021026 2024

Showing first 80 references.