NOTE-424: Time, Structure and Fluctuations

Contribution

The lecture is a survey of the Brussels school’s results, not a new result. It puts three things in one place:

• the thermodynamic stability theory of Glansdorff and Prigogine (LIT-537), with the second-order excess entropy δ²S as a Lyapounov function and the excess entropy production as the far-from-equilibrium stability test;
• the chemical case, through the Brusselator and its Hopf and “Turing” bifurcations (detailed in Nicolis and Prigogine, LIT-524);
• the stochastic case: near an instability the law of large numbers fails and fluctuations become long-range correlated.

It closes with a programme (§§6–7) for deriving irreversibility from dynamics through a non-unitary, “star-unitary” transformation.

Key insight

Equilibrium thermodynamics has potentials whose extrema are attractors, and near equilibrium the linear theory keeps one: entropy production is minimized at the steady state, and every fluctuation decays. Far from equilibrium that guarantee lapses, because the sign of the excess entropy production depends on the kinetics. Where it goes negative, a fluctuation can grow into a new macroscopic state that is maintained only by the flows through the system. That state is the dissipative structure. Its form depends on global features such as size, shape and boundary conditions, and which branch it takes at a bifurcation depends on which fluctuation happened. So the lecture locates “history” in physics at bifurcations.

Assumptions

• Local equilibrium (§2, p. 265): entropy depends on the same variables outside equilibrium as at equilibrium. The lecture says the explicit entropy production P = d_iS/dt = Σ_ρ J_ρ X_ρ ≥ 0 (2.3) “can only be established in some neighborhood of equilibrium” (p. 266).
• Linear regime: J_ρ = Σ L_ρρ′ X_ρ′ (2.5), with Onsager symmetry L_ρρ′ = L_ρ′ρ (2.6).
• Strictly linear regime for minimum entropy production: “the phenomenological coefficients L_ρρ′ may be treated as constants” (p. 266).
• Macroscopic description for the stability theory: δ²S < 0, from the Gibbs conditions C_v > 0, χ > 0 and Σ μ_γγ′ δN_γ δN_γ′ > 0 (3.2, p. 268), is assumed to hold “in the range of macroscopic description” also around non-equilibrium states (p. 269).
• Chemistry: mass-action kinetics, with the initial and final products A, B, D, E of the Brusselator “maintained constant” (p. 271). That is, the system is open to matter.
• Stochastic model (§5): birth-and-death Markov chains with a master equation (5.6) whose transition rates are non-linear in occupation numbers.
• Microscopic programme (§7): the existence of a positive operator M with Ω = tr ρ†Mρ ≥ 0 and dΩ/dt ≤ 0 (7.1–7.2) is assumed. The lecture states this “is certainly not always possible” (p. 279) and ties it to the spectrum of the Liouville operator (Misra).

Key results

• Minimum entropy production (§2, pp. 266–267). “For steady states sufficiently close to equilibrium entropy production reaches its minimum. Time-dependent states (corresponding to the same boundary conditions) have a higher entropy production.” It “requires even more restrictive conditions than the linear relations (2.5)”. It “is strictly valid only in the neighborhood of equilibrium”, and far from equilibrium behaviour can be “even opposite to that indicated by the theorem”. Entropy production is a Lyapounov function “in the strictly linear region around equilibrium” only (p. 268). No proof is given; ref. 6 is Prigogine 1945.
• Near-equilibrium stability (§3). S = S₀ + δS + ½δ²S (3.1). δ²S < 0 (3.2), and ½ ∂δ²S/∂t = Σ J_ρ X_ρ = P > 0 (3.4). So δ²S is a Lyapounov function near equilibrium “independently of the boundary conditions”, and it “ensures the damping of all fluctuations”.
• Far-from-equilibrium criterion (§3, p. 269). Around a non-equilibrium steady state ½ ∂δ²S/∂t = Σ δJ_ρ δX_ρ (3.5), the “excess entropy production”, which “has generally not a well-defined sign”. If Σ δJ_ρ δX_ρ ≥ 0 for all t > t₀ (3.6), δ²S is a Lyapounov function and the state is stable. This is a sufficient condition. Its violation is not shown to imply instability.
• Convection (§2, §3). Beyond a critical adverse temperature gradient the Bénard layer convects, and “the entropy production is then increased as the convection provides a new mechanism of heat transport” (p. 267). Internal convection “cannot be generated from a state at rest which is at thermodynamic equilibrium” (p. 270).
• Chemistry (§4). Autocatalytic steps are “necessary (but not sufficient)” for instability of the thermodynamic branch (p. 271).
  • The Brusselator: A → X, 2X + Y → 3X, B + X → Y + D, X → E (4.1). With unit rate constants, dX/dt = A + X²Y − BX − X and dY/dt = BX − X²Y (4.2). The steady state X₀ = A, Y₀ = B/A (4.3) is unstable when B > B_c = 1 + A² (4.4), beyond which there is a limit cycle (a Hopf bifurcation) whose frequency, unlike Lotka–Volterra’s, is fixed by the macroscopic parameters: “a chemical clock”.
  • With diffusion (4.5) there are non-uniform steady states, which the lecture calls the “Turing bifurcation” after Turing 1952 (LIT-535), and chemical waves.
  • Dissipative structures generally require the system’s size to exceed a critical value (p. 272).
• Fluctuations (§5). For a non-linear chain (5.8) the stationary distribution is not Poisson (Nicolis and Prigogine 1971). Near the critical point of the Schlögl model the variance ⟨(δX)²⟩ grows as a higher power of the volume than the first (p. 275). In a box model of the Brusselator, spatial correlations become long-range near the instability (Figs. 5.1–5.3), and oscillation onset is described as time-symmetry breaking (p. 277).
• Microscopic irreversibility (§§6–7).
  • Ω = tr ρ†ρ is conserved by Liouville dynamics (6.6–6.7), so it is not a Lyapounov function.
  • Writing M = (Λ⁻¹)†Λ⁻¹ (7.3) and ρ̃ = Λ⁻¹ρ (7.5) gives a new equation of motion iρ̃_t = Φρ̃ with Φ = Λ⁻¹LΛ (7.6–7.7). Its even part under L-inversion is the “entropy production” (7.17–7.18).
  • “Microscopic entropy production” is the commutator i(ML − LM) (7.9).
  • The link back to macroscopic δ²S is established “at least in the linear region” (p. 283, Theodosopulu, Grecos and Prigogine 1978).

Claims

id	claim	strength	support
C1	Near equilibrium δ²S is a Lyapounov function, so fluctuations are damped	strong (standard), derivation cited	(3.2), (3.4); Glansdorff & Prigogine ch. V
C2	Minimum entropy production holds only for steady states in the strictly linear regime with constant L_ρρ′	strong as a statement of scope; theorem not proved here	p. 266, ref. 6
C3	Far from equilibrium the excess entropy production has no fixed sign; ΣδJδX ≥ 0 suffices for stability	moderate: the sufficiency follows from (3.2) and (3.5); generality of (3.5) is cited	(3.5)–(3.6), ch. VII of ref. 7
C4	Autocatalytic steps are necessary but not sufficient for chemical instability of the thermodynamic branch	asserted, with the Brusselator as a worked case	p. 271
C5	The Brusselator steady state is unstable for B > 1 + A²	strong (linear stability, checkable)	(4.4)
C6	Dissipative structures are “giant fluctuations” amplified past an instability and stabilized by exchange with the outside	interpretive	p. 267
C7	Near a non-equilibrium instability the law of large numbers fails and correlations become long-range	moderate: model calculations and simulations cited	§5, refs 17, 19, 20
C8	Bifurcation introduces “history” into physics	interpretive	p. 273
C9	Irreversibility can be derived from dynamics by a non-unitary transformation, without coarse-graining	weak here: a programme, with existence of M assumed and the macroscopic link shown only near equilibrium	§§6–7, p. 283

Concepts

• dissipative structure: “a new type of dynamic states of matter” that irreversible processes may produce (p. 263). In Bénard convection it is “a new supermolecular order … which corresponds basically to a giant fluctuation stabilized by exchanges of energy with the outside world” (p. 267). The lecture always links three aspects: “the function as expressed by the chemical equations, the space-time structure, which results from the instabilities, and the fluctuations, which trigger the instabilities” (p. 272).
• thermodynamic branch: the family of steady states continuous with equilibrium, from which dissipative structures bifurcate (p. 272).
• excess entropy production: Σ δJ_ρ δX_ρ, the perturbation of the entropy production about a steady state (3.5).
• order through fluctuations: the amplification of a fluctuation at an instability into a macroscopic state (p. 272).
• local equilibrium: entropy outside equilibrium depends on the same variables as at equilibrium (p. 265).
• star-unitary transformation: the non-unitary Λ connecting the dynamical and “thermodynamic” representations (7.13–7.16).

Connections

• Turing 1952 (LIT-535), read. The lecture names the stationary diffusive instability of the Brusselator the “Turing bifurcation” and cites Turing as “the first to notice the possibility of such bifurcations in chemical kinetics” (p. 272). Turing had no thermodynamic stability criterion, but he did say that his patterns need “a continual supply of free energy” (Turing p. 66).
• The Belousov–Zhabotinsky reaction (LIT-531), read. It is not named in the lecture. The “chemical clock” of §4 is the Brusselator, a model, and the experimental chemical example the lecture gestures at is “oscillatory enzymes”. Zhabotinsky’s own review credits Nicolis and Prigogine 1977 with the point that sustained oscillations need distance from equilibrium.
• Cross and Hohenberg 1993 (LIT-527), skimmed. They file “dissipative structures” (citing Nicolis and Prigogine) as one name for macroscopic steady-state structure under constant non-equilibrium conditions. In their conclusion they quote Glansdorff and Prigogine’s own admission that “the search for a universal kinetic potential has proved to be unsuccessful” (1971, p. 108), and find no evidence for a global minimization principle beyond a perturbative one near threshold. That matches the lecture’s own restriction of minimum entropy production to the strictly linear regime.
• Observational entropy (LIT-082), read. The two stand on opposite sides of one question. LIT-082 makes entropy relative to a measurement, that is, a coarse-graining. The lecture rejects coarse-graining as the source of irreversibility (“Can dissipative structures be the result of mistakes?”, p. 279) and looks for a Lyapounov functional in the dynamics itself.
• Thermodynamics of prediction (LIT-327) and THEORY-026. No overlap. The lecture’s entropy production is the macroscopic local-equilibrium sum ΣJ_ρX_ρ. LIT-327 computes dissipation for a stochastic system driven through a Markov kernel, a different regime and a different quantity, so this reading neither supports nor tests THEORY-026.
• Emergence surveys (LIT-021, LIT-141). NOTE-007 records that LIT-021’s history of emergence names Prigogine. The lecture supplies what that name stands for. LIT-141’s “self-organisation” characteristic, which it counts as subjective, is here given an objective criterion, but only for the onset (instability of a steady state), not for the organisation that follows.
• Primitive metabolic cycles (LIT-036), read. It is a modern instance of the lecture’s chemistry: a non-equilibrium, catalytic cycle whose homogeneous state goes unstable, giving aggregation or oscillation. It uses linear stability of a homogeneous state, as the lecture does, and no entropy-production principle.
• Teleology and the definition of life (LIT-192, LIT-211). See Bearing on the record.

Bearing on the record

• LIT-192’s demarcation dispute. The lecture’s definition is the dissipative-structure side of the line Nahas and Sachs discuss. Its criterion is physical: an open system, held away from equilibrium, whose steady state has lost stability, maintained by flows across its boundary. Bénard cells and the Brusselator meet it. So does a candle flame on the same terms (my inference: the lecture does not discuss flames), and so, by the lecture’s own §1, do living systems. Nothing in the lecture separates teleological systems from other dissipative structures. That supports the reading in NOTE-094 that the demarcation has to be drawn by something added to thermodynamics (closure of constraints, Deacon’s teleodynamics), not read off it.
• LIT-211’s “Dissipative Self-Organizing Systems” cluster. The 21 definitions in it lean on exactly this picture. The lecture shows what the picture does and does not supply. It supplies a necessary physical condition: far from equilibrium, flux-maintained, instability-born order. It does not supply a sufficient one for life, since the canonical dissipative structures (convection rolls, chemical clocks) are not alive.
• No THEORY is indicated by this reading alone. A candidate, “being a dissipative structure in Prigogine’s sense is necessary but not sufficient for being alive or goal-directed”, would need the parallel filings on the life-as-dissipation literature to source it. Those filings now source THEORY-072, which holds that organisms differ from dissipative structures by closure among differentiated constraints, and cites this lecture for the structures themselves.
• No ML instruction. Nothing here belongs in the anthology.

Limitations

• A lecture, not a derivation. Minimum entropy production, the δ²S stability results and the general form (3.5) are stated and cited to ref. 6 and Glansdorff and Prigogine, not proved.
• The far-from-equilibrium criterion is one-sided. (3.6) is sufficient for stability. The lecture does not claim its failure gives instability, and gives no positive principle selecting which dissipative structure appears; it hands that to the kinetics and to fluctuations.
• “Functional order” in towns and living systems is asserted in §1, not analysed. No biological system is modelled.
• The microscopic programme is incomplete by its own account. The existence of M is assumed, and the identification with δ²S is shown only for small deviations from equilibrium and only for conserved quantities (p. 283).

Open questions

• Is there a variational principle, of any form, for steady states far from equilibrium? The lecture offers none. This is the question the parallel filings on entropy-production principles take up.
• Does the failure of (3.6) imply instability in the systems the lecture names, or only permit it?
• Does a positive entropy operator M exist for physically realistic (non-integrable, large) systems, and does the star-unitary picture reach beyond the linear regime?

Corrections

• none to a seeded skim (there was no seed)
• Citation. The Science reprint is titled “Time, Structure, and Fluctuations” (with a serial comma), Science 201(4358):777–785, published September 1978 per Crossref. The lecture PDF prints it without the comma. The brief’s citation is otherwise right.
• Bénard convection. The lecture describes the Bénard instability as buoyancy-driven, a fluid layer heated from below “in a constant gravitational field” (p. 267). Cross and Hohenberg (LIT-527, §II.B.3) note that Bénard’s own 1900 open-dish hexagons were later shown (Pearson 1958) to be driven by temperature-dependent surface tension. The lecture’s idealization is Rayleigh’s problem, not Bénard’s experiment.
• Reference slips in the printed lecture. Ref. 7 spells the first author “Glandsdorff”; ref. 13 prints “Thorn” for Thom; ref. 3 gives the monograph’s title as “Thermodynamics of Structure, Stability and Fluctuations”, where the book’s title is “Thermodynamic Theory of Structure, Stability and Fluctuations”.