Contribution
Bounds of this kind were known for linear networks (Maes and Netočný) and for catalytic ones, where the network is still a graph (the authors’ own earlier work). The paper proves them for arbitrary stoichiometry. Without any kinetic information, a network’s topology and its chemostatted driving fix a region of concentration space, the thermodynamic space, that contains every stationary state any choice of rate constants can produce (eqs. 20, 24, 29). It then uses the region to bound bistability, chiral imbalance, self-assembly yield and pattern contrast, to falsify candidate topologies, and to infer a minimum driving from measured concentrations.
Key insight
A steady current through a reaction must close into a cycle (an elementary flux mode) that runs from fuel to waste, and every reaction along that cycle pushes forward. So no step can take more of the cycle’s free energy than the whole cycle has (eqs. 15–17). Turning affinities into concentrations, the stationary value of any product of concentrations must lie between what it would be at equilibrium along the least favourable pathway and along the most favourable one (eq. 28). Driving widens the space; it does not choose a point in it.
Assumptions
- Open, dilute, mass-action networks, all reactions reversible, with chemostatted species at fixed concentrations (§III A). The proofs use only the local detailed balance J⁺/J⁻ = exp(A/RT) (eq. 8), “not … the specific form of the kinetics” (p. 5).
- A stationary state with nonzero concentrations exists. “Our framework does not capture the case of persistent oscillations, but it holds for CRNs exhibiting multi-stability” (p. 2).
- Fixed chemostats. The driving is a constant budget. How long it lasts is not considered.
- Conformal decomposition of the steady flux into elementary flux modes (EFMs), a standard result (§III B).
Key results
- Affinity bounds (eq. 20). min(0, min_{E⁻ρ} A^Y_e/e_ρ) ≤ A^ss_ρ ≤ max(0, max{E⁺_ρ} A^Y_e/e_ρ), over EFMs containing ρ, normalised by the number of times ρ is used. Equality holds at equilibrium and for reactions in no cycle.
- Chemical probe (§IV A, eq. 24). An auxiliary reaction ρ̂ compatible with the conservation laws, infinitely slow so it carries no flux but has a finite affinity, “analogous to a voltmeter in an electric circuit” (p. 9). Its affinity obeys a two-sided bound from the EFMs it closes.
- Thermodynamic space (eq. 29). K^min_{π} ≤ Π_i (x^ss_i)^{S_iρ} ≤ K^max_{π}, with K_π = exp(−Δ_π G/RT) for each pathway π converting the species on one side of ρ (or ρ̂) into the other. Each bound “can be saturated … when the dynamics is kinetically dominated by the corresponding reaction pathway” (p. 3). At equilibrium the two coincide, and “a non-zero width of the thermodynamic space is a signature of non-equilibrium conditions and a prerequisite for the emergence of complex behaviors” (p. 3).
- Topology test (§IV C, Fig. 4). A measured dimer amplification above Δμ/RT cannot come from the network without 2B ⇌ C, whose EFM uses the driven step twice and raises the ceiling to 2Δμ/RT. Exceeding a ceiling falsifies a topology; a missing reaction that does not re-use the driving is invisible.
- Schlögl (§V A, eq. 36). exp(−(μ°_X − μ_B)/RT) ≤ x^ss ≤ exp(−(μ°_X − μ_A)/RT). Both stable and the unstable fixed point lie inside; at Δ_e G = 0 the bounds “coincide, collapsing the thermodynamic space and precluding bistable solutions” (p. 13). Checked: with k₁± = 8, k₂± = 1, b = 0.02, a = 0.8 the fixed points are 0.0234, 0.1787 and 0.5979, inside [0.02, 0.8]; at a = 0.5 there is one, 0.0218; at a = b the space and the fixed point are both 0.02.
- Chiral symmetry breaking (§V B, eq. 39). |ln(r/s)| ≤ Δμ/RT, saturated as the dimerisation becomes fast; “the emergence of perfect homochirality … necessitates an infinite driving force” (p. 15).
- Self-assembly and patterns (§V C–D). Trimer amplification is bounded by 3Δμ (eq. 46); reaction–diffusion contrast by tanh(Δμ/2mRT) (eq. 54), approached as D_u/D_v → 0.
- Data (§VI). GroEL–MDH concentrations: native dimer to monomer stays at equilibrium even with ATP, while native to misfolded is displaced and lies within the space (Fig. 9b–c). A photoacid: a minimum light-driven rate inferred from concentrations recovers 75–85% of the true one (Fig. 9f).
Claims
| id | claim | strength | support |
|---|---|---|---|
| C1 | Stationary reaction affinities are bounded by extremal EFM cycle affinities per use | strong | eqs. 15–20, derivation from conformal decomposition |
| C2 | Stationary concentration products lie between the extremal pathway equilibrium constants | strong | eqs. 26–29; my Schlögl check |
| C3 | The space collapses to a point at zero driving, so bistability and symmetry breaking need driving | strong | eq. 29 at A^Y_e = 0; Schlögl check |
| C4 | Each bound is saturated when one pathway dominates kinetically | moderate | argued, shown for chirality (eq. 40) and patterns; not proved in general |
| C5 | A concentration outside the space falsifies a candidate topology | strong as a one-sided test | §IV C, one worked example |
| C6 | Chaperone and photoacid data lie within the predicted spaces | moderate | two datasets from the literature, model taken as given |
Concepts
- Elementary flux mode (EFM) — a support-minimal vector in ker S_X: a minimal set of reactions that leaves internal species unchanged (§III B).
- Cycle affinity A^Y_e — the free energy the chemostats lose per turn of an EFM (eq. 14), independent of internal concentrations.
- Chemical probe — an infinitely slow virtual reaction used to read the affinity between species the network does not connect directly.
- Thermodynamic space — the region of stationary concentrations compatible with the network’s topology and driving, for all kinetics.
Connections
It uses the open-network framework of Rao and Esposito (LIT-1889, NOTE-1464), whose emergent cycles are its driven EFMs, and the hypergraph cycle geometry of Dal Cengio et al. (LIT-1897, NOTE-1476). It generalises Maes and Netočný’s bound for linear networks and Liang and Pigolotti’s flux-ratio bound (eq. 33). Prigogine’s question, which structure appears, is THEORY-692’s; this paper bounds where any can appear and does not choose.
Bearing on the record
- A13’s use of it: half right, and the other half belongs to another paper. The registration says A13 used it for “the intersection of the dynamically viable region with what thermodynamic driving permits”. That is what the paper gives (THEORY-1261). A13 also used it for “an organization … can still have a finite lifetime because its driving is exhausted”. The paper says nothing of that: its chemostats are fixed and its states stationary. Exhaustion of driving, and the time at which it happens, is Remlein, Esposito and Avanzini (LIT-1887, NOTE-1468). The LIT’s “Why this is here” should say so.
- CASE-189, rewritten. The probe ∅ → c₁ has two pathways, from fuel through the catalysed reaction and from waste through reversed decay. So (d′/d)[W] ≤ c₁ ≤ (k/k′)[S]. In my rewriting ([S] = 3, unit rates) every fixed point, stable or not, lies in [[W], 3] for [W] = 0.01, 0.08 and 1. The lower edge is the extinct state’s floor, set by the waste; the upper is the maintained state’s ceiling, set by the fuel. The space is empty of width when [W] = 3, which is why the rewriting has a single equilibrium there. The construction’s bistability needs a width it does not have below 3.52 RT (NOTE-1464).
- CLAIM-273 and THEORY-692. The paper makes concrete THEORY-692’s point that thermodynamics bounds and does not select. Driving fixes a region; which state in it is occupied is “kinetically dominated” (p. 3). It is no support for maximum entropy production (THEORY-671) or for driven adaptation (THEORY-412) as selection principles, and no evidence against them.
- CLAIM-478. The bounds are attained in time-scale-separation limits: the Schlögl upper bound when reaction 1 is much faster than reaction 2 (p. 13), the chiral bound as dimerisation becomes fast (eq. 40). A fast pathway reaches pseudo-equilibrium and fixes the concentrations the slow one must work against. That is the claim’s first half in a chemical form, with a bound on how far the separation can carry it.
- No instruction for machine-learning practice.
Limitations
- Stationary states only. Oscillations, chaos and transients are outside it, as the authors say (p. 2, p. 20).
- Necessary, not sufficient. Inside the space says nothing about whether a state is reached, stable or unique.
- Saturation conditions are argued from examples, not proved.
- EFM enumeration grows combinatorially (App. B); the bounds for large networks depend on targeted searches the paper does not supply.
- Appendices skimmed.