Essay III

Thermodynamically Sustained Autopoiesis in an (∞)-Topos

Life-like autonomy is a least fixed point of feasible production together with a greatest fixed point of viable dynamics. Organizational closure is sustained by thermodynamic openness, not by isolation.

Contents60 sections

Abstract

Let $E$ be an $(\infty)$ -topos and $P^{\otimes}$ an internal symmetric monoidal $(\infty)$ -category of open physical processes, with operad $O^{\otimes}$ . For a structured system $X$ , let $E_{X}$ be its primitive endogenous operad, $R_{X}$ its requirement operad, and $Φ_{X}$ its production operator, with least fixed point $G_{X} = μ Φ_{X}$ .

Thermodynamics enters through an internal ordered commutative monoid, or quantale, $Q$ of costs, and a lax symmetric monoidal valuation $Θ : P^{\otimes} \to B Q^{\otimes}$ . In physical models $Θ$ may record work, exergy consumption, free-energy dissipation, entropy production, or a tuple of these. The state object $S_{X}$ carries a viability subobject $V_{X} ↪ S_{X}$ . Admissible dynamics determine a monotone operator $Ψ_{X}$ on $Sub (S_{X})$ , whose greatest fixed point $V_{X} = ν Ψ_{X}$ is the viability kernel: the states from which $X$ can remain viable indefinitely.

Thermodynamically sustained autopoiesis is the conjunction

R_{X} \subseteq μ Φ_{X}^{th} and V_{X} \neq = 0,

where $Φ_{X}^{th}$ retains only thermodynamically feasible realizations. Life-like autonomy is the coexistence of a least fixed point of recursive production and a greatest fixed point of viable dynamics. Cells, viruses, and ordinary computers are then separated by how endogenous generation, environmental coupling, viable maintenance, and thermodynamic expenditure sit relative to one another — not by information-processing capacity alone.

1The ambient physical (∞)-topos

Let $E$ be an $(\infty)$ -topos containing internal objects of real quantities, among them $R$ , $R_{\geq 0}$ , and $R_{\geq 0} = [0, \infty]$ . Objects and morphisms may depend on spacetime, environment, experimental context, or scale. Thermodynamic quantities are variable objects of the topos, not globally fixed scalars.

2Open physical processes

Let $P^{\otimes} \in CAlg (Cat_{\infty} (E))$ be an internal symmetric monoidal $(\infty)$ -category. Objects are physical interfaces. A morphism $p : A \to B$ is an open process. The tensor product $p \otimes q$ is parallel composition, and $q \circ p$ is sequential composition. Open reaction networks sit in this pattern: they are morphisms, and composition joins outputs of one network to inputs of another.

3The cost object

Definition 3.1

Internal thermodynamic quantale

Let $Q$ be an internal commutative quantale in $E$ . The basic model is $Q = ([0, \infty], \geq, +, 0)$ , with the reversed order of Lawvere’s convention: $a \leq_{Q} b$ if and only if $a \geq_{R} b$ . Addition is the monoidal operation. The point of the reversal is that enrichment records cost, and a cheaper process is a “better” morphism.

Lawvere’s observation that metric spaces are categories enriched in $([0, \infty], \geq, +, 0)$ is the cost semantics used below.

4Thermodynamic valuation

Two quantities must be kept apart. For a concrete physical realization $p$ , entropy production or dissipated work is an actual physical number. For an abstract operation $f$ , one may instead ask for the greatest lower bound of that cost over all realizers. The first is an additive valuation. The second is a Lawvere cost.

Definition 4.1

Realized dissipation

A thermodynamic valuation is a lax symmetric monoidal functor $Θ : P^{\otimes} \to B Q^{\otimes}$ . For a process $p$ , write $Θ (p) \in Q$ . Depending on the model, $Θ (p)$ may be dissipated work $W_{diss} (p)$ , consumed exergy $B (p)$ , environmental temperature times entropy production $T_{0} Σ (p)$ , or a tuple of such quantities.

5Sequential and parallel composition

For independently composed realized processes one expects additivity, $Σ (p \otimes q) = Σ (p) + Σ (q)$ and $Σ (q \circ p) = Σ (p) + Σ (q)$ , up to the bookkeeping of the thermodynamic model. For minimum costs, exact additivity is too strong. What survives is the triangle inequality $c (q \circ p) \leq c (p) + c (q)$ .

Proposition 5.1

Lawvere cost inequality

Define $c (f) := in f_{p ⊨ f} Θ (p)$ , where $p ⊨ f$ means that the physical process $p$ realizes the abstract operation $f$ . Then $c (g \circ f) \leq c (f) + c (g)$ .

Proof. If $p ⊨ f$ and $q ⊨ g$ , then $q \circ p ⊨ g \circ f$ , so $c (g \circ f) \leq Θ (q \circ p)$ . Under additive accounting, $Θ (q \circ p) = Θ (p) + Θ (q)$ . Taking the infimum over realizers yields the inequality. Lax additivity gives the same bound.

End of proof.

Minimum thermodynamic cost is an enriched geometry on the space of realizable operations.

6Nonequilibrium free energy

For biochemistry the relevant thermodynamics is nonequilibrium thermodynamics. Open chemical reaction networks admit energy and entropy balances in which nonequilibrium Gibbs free energies act as potentials. Rao and Esposito show that the entropy production of an open network decomposes into a change of potential and the work supplied through external chemostats and nonconservative forces.

At constant environmental temperature $T_{0}$ , a schematic splitting of supplied free energy is

W_{in} = Δ F + T_{0} Σ + W_{out}, Σ \geq 0.

In rates, for an open material system,

W ˙_{in} = F ˙ + J_{out}^{F} + T_{0} Σ ˙ + W ˙_{out} .

The potential depends on the ensemble and on the conserved quantities. The categorical theory needs only an additive or lax-additive valuation $Θ$ . The displays above are the intended physical reading of that valuation, not an extra axiom.

7Entropy production

Definition 7.1

Entropy production

For each process $p : A \to B$ , let $Σ (p) \in R_{\geq 0}$ be its entropy production. The second law is the internal proposition $Σ (p) \geq 0$ . When the interface accounting includes every exchanged entropy, sequential composition adds: $Σ (q \circ p) = Σ (p) + Σ (q)$ . For independent subsystems, parallel composition adds: $Σ (p \otimes q) = Σ (p) + Σ (q)$ .

8Exergy

Energy is too coarse a currency, because energy is conserved. What an autonomous system consumes is available free energy relative to its environment: exergy.

Definition 8.1

Exergy consumption

Let $B (p)$ denote the exergy consumed by a process $p$ . At environmental temperature $T_{0}$ , irreversibility destroys exergy at the rate $T_{0} Σ$ . Thus $Σ > 0$ is irreversible degradation of useful free-energy resources.

9Thermodynamically feasible operations

Definition 9.1

Budget feasibility

Fix a budget $b \in Q$ . An operation $f$ is $b$ -feasible when $c (f) \leq b$ . Write $O_{\leq b} ↪ O$ for the collection of $b$ -feasible operations. Budgets compose by the monoidal structure $b_{1} \oplus b_{2}$ . For scalar exergy, $b_{1} \oplus b_{2} = b_{1} + b_{2}$ .

10Budget-indexed operadic composition

If $f : (y_{1}, \dots, y_{n}) \to z$ costs at most $b_{0}$ and each $g_{i}$ costs at most $b_{i}$ , then $f \circ (g_{1}, \dots, g_{n})$ has a realization costing at most $b_{0} + \sum_{i} b_{i}$ .

Theorem 10.1

Operadic cost inequality

c (f \circ (g_{1}, \dots, g_{n})) \leq c (f) + i \sum c (g_{i}) .

Proof. Choose realizers arbitrarily close to the respective infima and substitute them operadically. Costs of the resulting process are additive or lax-additive. Taking infima yields the inequality.

End of proof.

The process $(\infty)$ -operad therefore carries a cost enrichment.

11The thermodynamic production doctrine

Definition 11.1

Feasible production

Let $β_{X}$ be the admissible environmental exergy supply of $X$ . Define $Φ_{X}^{β_{X}} (A)$ , also written $Φ_{X}^{th} (A)$ , to be the suboperad of operations which can be produced using operations in $A$ , which satisfy the physical constraints of $X$ , and which possess realizers compatible with the budget $β_{X}$ . The recursively generated operad is $G_{X}^{th} := μ Φ_{X}^{th}$ .

12Thermodynamic organizational closure

Definition 12.1

Feasible closure

The system $X$ is thermodynamically organizationally closed when $R_{X} \subseteq G_{X}^{th}$ .

This is strictly stronger than $R_{X} \subseteq G_{X}$ . Formal generability does not imply energetic feasibility. A system may satisfy the first inclusion and fail the second. It then has an abstract self-production network that it cannot energetically sustain.

13State objects

Let $S_{X} \in E$ be the state object of $X$ . A point of $S_{X}$ packages the state variables that matter: concentrations, temperature, chemical potentials, the boundary $\partial X$ , and whatever else the model treats as state.

14The viability subobject

Definition 14.1

Viability

A viability condition is a subobject $V_{X} ↪ S_{X}$ . Internally, $V_{X} (s)$ means that the state $s$ is compatible with continued existence of $X$ as a system of the specified organizational type. Viability may constrain membrane integrity, temperature, chemical potential, ATP/ADP ratio, pH, structural damage, error rates, voltage, or other variables.

Homeostasis is a viability problem. Critical variables are held in restricted ranges despite disturbances, and holding them there costs continuing free energy.

15Admissible controls

For $s \in S_{X}$ , let $Adm_{X} (s)$ be the space of thermodynamically admissible endogenous process-families available at $s$ . Each $u \in Adm_{X} (s)$ produces a successor $T_{X} (s, u) \in S_{X}$ . The deterministic notation may be replaced by a stochastic kernel. The lattice argument is unchanged.

16The viability predecessor

For $A ↪ S_{X}$ , declare $s \in Pre_{X} (A)$ if and only if some admissible endogenous process $u$ satisfies $T_{X} (s, u) \in A$ . Define

Ψ_{X} (A) := V_{X} \cap Pre_{X} (A) .

A state lies in $Ψ_{X} (A)$ when it is viable and can be driven, admissibly, into $A$ .

17Monotonicity

Lemma 17.1

The viability operator is monotone

If $A \leq B$ , then $Ψ_{X} (A) \leq Ψ_{X} (B)$ .

Proof. A process taking $s$ into $A$ also takes $s$ into $B$ whenever $A \leq B$ . Thus $Pre_{X} (A) \leq Pre_{X} (B)$ . Intersection with $V_{X}$ preserves the inequality.

End of proof.

18The viability kernel

The subobject lattice $Sub (S_{X})$ is complete. A monotone endomap has a greatest fixed point.

Definition 18.1

Viability kernel

Set $V_{X} := ν Ψ_{X}$ . This is the viability kernel of $X$ . A state lies in $V_{X}$ precisely when some indefinitely extendable sequence of thermodynamically admissible processes keeps the system inside $V_{X}$ .

Theorem 18.2

Fixed-point equation

V_{X} = V_{X} \cap Pre_{X} (V_{X}) .

Proof. Unfold $Ψ_{X} (V_{X}) = V_{X}$ .

End of proof.

19Two fixed points

The theory now contains two recursive constructions of opposite variance. The least fixed point $μ Φ_{X}^{th}$ asks which organizational operations the system can recursively generate. The greatest fixed point $ν Ψ_{X}$ asks from which states the system can continue indefinitely without leaving its viability region.

20Thermodynamically sustained autopoiesis

Definition 20.1

Thermodynamic autopoiesis

The system $X$ is thermodynamically autopoietic, written $TA (X)$ , when

R_{X} \subseteq μ Φ_{X}^{th} and ν Ψ_{X} \neq = 0.

The inequality $0$ denotes the initial subobject. The kernel is required to be inhabited.

21Why both fixed points are required

A self-copying chemical process might meet the operadic condition and immediately exhaust its free energy. Then $R_{X} \subseteq μ Φ_{X}$ while $ν Ψ_{X} = 0$ . A passive crystal might persist for a very long time inside a stable viability region and possess no recursively self-producing operad. Then $ν Ψ_{X} \neq = 0$ does not yield $R_{X} \subseteq μ Φ_{X}$ .

22Living systems are not thermodynamically closed

A living cell generally satisfies $E ˙_{in} > 0$ , $J_{matter} \neq = 0$ , and $Σ ˙ > 0$ . The relevant condition is not thermodynamic closure.

Nonequilibrium thermodynamics treats cellular processes as dissipative systems held away from equilibrium by sustained chemical or energetic driving. There is no contradiction between the two fixed points and that fact. The driving is what lets the least fixed point stay feasible and the greatest fixed point stay inhabited.

23Maintenance cost

Define the minimal average maintenance power of $X$ by

m_{X} := π in f T \to \infty lim sup \frac{1}{T} \int_{0}^{T} B ˙_{π} (t) d t,

where $π$ ranges over viable policies whose trajectories remain in $V_{X}$ .

Definition 23.1

Strictly positive maintenance

The system $X$ has strictly positive maintenance cost when $m_{X} > 0$ .

For nonequilibrium target states this is the expected situation. Maintaining a prescribed distribution away from equilibrium requires continuing work and continuing dissipation. Explicit lower bounds of this form are known for Markov systems.

24An energetic obstruction

Theorem 24.1

Insufficient exergy empties the kernel

Suppose $m_{X} > 0$ and the asymptotic available exergy power satisfies $P_{X} < m_{X}$ . Then $V_{X} = 0$ .

Proof. If $s_{0} \in V_{X}$ , there is an indefinitely viable policy $π$ . By definition of $m_{X}$ ,

T \to \infty lim sup \frac{1}{T} \int_{0}^{T} B ˙_{π} (t) d t \geq m_{X} .

The environment supplies only asymptotic power $P_{X} < m_{X}$ , so the required cumulative exergy eventually exceeds the available cumulative exergy. The trajectory cannot continue indefinitely.

End of proof.

25The cumulative form

If every viable trajectory requires at least $ε > 0$ of free energy on each interval of duration $Δ t$ , then after $n$ intervals the required exergy is at least $n ε$ . An indefinitely viable trajectory therefore requires infinite cumulative throughput. Perpetual biological organization does not require a finite store large enough to last forever. It requires a persistent supply.

26Isolation cannot pay the maintenance

Corollary 26.1

Nonequilibrium maintenance

If $V_{X}$ is separated from thermodynamic equilibrium and maintaining states in $V_{X}$ has a strictly positive dissipation lower bound, then a finite-energy isolated realization of $X$ cannot remain viable indefinitely. Symbolically, $V_{X} \cap E_{eq} = \emptyset$ and $m_{X} > 0$ imply that a finite isolated resource forces $V_{X} = 0$ .

Autonomous organization is compatible with thermodynamics because the autonomy is organizational, not an energetic isolation.

27The operadic maintenance burden

For each required operation $r \in R_{X}$ , set $c_{X} (r) := in f {Θ (p) ∣ p ⊨ r}$ . A generating family $R = {r_{i}}$ has a schematic burden $\sum_{i} λ_{i} c_{X} (r_{i})$ , where $λ_{i}$ is the required execution frequency. For a continuous reaction network the same idea becomes an integral, or a flux–force pairing.

28Flux and force

For a chemical reaction network with currents $J_{ρ}$ and affinities $A_{ρ}$ , the entropy-production rate has the standard form

Σ ˙ = \frac{1}{T} ρ \sum J_{ρ} A_{ρ} \geq 0.

The operad says which compositions exist. The valuation says what they cost. Rao and Esposito connect the conservation laws and the topology of an open network with the decomposition of thermodynamic forces and of entropy production.

29Host coupling

Let $V$ be a virus and $H$ a host, interacting across an interface $I$ . The coupled process operad is generated by $E_{V} ⨿_{I} E_{H}$ . The valuation extends to mixed virus–host trees.

30Cost of a mixed tree

Lemma

Cost of a mixed tree

Let $T$ be a finite operadic tree with vertices $v$ , each assigned a physical process of cost $c_{v}$ . Then $c (T) \leq \sum_{v \in T} c_{v}$ .

Proof. Induct on the tree. The single-vertex case is immediate. At an internal vertex, apply Theorem 10.1 to the root and to the child subtrees.

End of proof.

31Two ways a host can help

It may happen that $r \in / G_{V}^{th}$ while $r \in G_{V ⋆ H}^{th}$ . There are two different reasons. The host may supply an operation that $V$ does not have. Or the host may supply the energetic conditions under which an otherwise inaccessible operation becomes feasible. Coupling enlarges both the feasible process algebra and the available thermodynamic budget.

32Viral thermodynamic dependence

A virus can therefore satisfy $R_{V} ⊈ μ Φ_{V}^{th}$ while $R_{V} \subseteq μ Φ_{V ⋆ H}^{th}$ . This is stronger than bare causal dependence on a host. The host closes an operadic deficit and a thermodynamic deficit at once.

33Coupling is not a free lunch

Theorem 33.1

No thermodynamic free lunch

Suppose host supplementation makes a previously unavailable operation $r$ feasible, so $r \in / G_{V}^{th}$ and $r \in G_{V ⋆ H}^{th}$ . Every physical realization of $r$ in $V ⋆ H$ still satisfies $Σ (r) \geq 0$ .

Proof. The pushout enlarges the space of allowed compositions. It does not change the second-law clause of the valuation. Every realized composite is a process of $P$ , and $Σ (p) \geq 0$ for every such process.

End of proof.

Supplementation redistributes an energetic burden. It does not delete it.

34Subsystem cost is not total cost

From the virus’s point of view, $H$ is an environment supplying machinery and free energy. From the point of view of $V ⋆ H$ , those contributions are internal and must be paid for in the same thermodynamic ledger.

Apparent autonomy must not be obtained by pushing maintenance across an unstated boundary.

35Industrial coupling of computers

Let $C$ be a conventional computer and $I$ the industrial system required for its continued physical reproduction. The pushout $E_{C} ⨿_{I} E_{I}$ contains silicon purification, photolithography, chip fabrication, power generation, replacement, and assembly. The isolated computer does not realize these processes. Thus $R_{C}^{phys} ⊈ μ Φ_{C}^{th}$ , while one may have $R_{C}^{phys} \subseteq μ Φ_{C ⋆ I}^{th}$ .

36The dependency analogy, and its limit

The structural analogy is not “virus $\sim$ software.” A tighter analogy pairs the virus with its host against the computer with its industrial ecology.

Both exhibit a subsystem whose physical continuation depends on a much richer external process algebra. This remains an analogy under a coarse dependency functor. The full objects differ.

37The cell, thermodynamically

Let $B$ be a bacterium. The modeling hypothesis is $R_{B} \subseteq μ Φ_{B}^{th}$ and $ν Ψ_{B} \neq = 0$ , provided the environment supplies nutrients, electron donors or acceptors, light, or some other free-energy source in an admissible way. Then $TA (B)$ . The cell is not energetically self-sufficient.

That is the stronger distinction. Essay I separated closure from substrate. Essay II separated closure from a snapshot of available operations. The present condition separates a self-maintaining converter of free energy from a device whose fabrication remains outside its own operad.

38The thermodynamic invariant

Write

T (X) = (R_{X}, μ Φ_{X}^{th}, ν Ψ_{X}, m_{X}, Θ_{X}) .

With material typing included,

T_{K} (X) = (Supp (X), E_{X}, R_{X}, μ Φ_{X}^{th}, ν Ψ_{X}, m_{X}, Θ_{X}) .

39Thermodynamic equivalence

Definition 39.1

Equivalence

An equivalence $X ≃ Y$ of thermodynamically structured systems preserves $E$ , $R$ , $Φ^{th}$ , the viability subobject $V$ , the operator $Ψ$ , and the valuation $Θ$ .

40Invariance

Theorem 40.1

Equivalence preserves both fixed points

If $X ≃ Y$ thermodynamically, then $F_{*} (μ Φ_{X}^{th}) ≃ μ Φ_{Y}^{th}$ and $F_{*} (ν Ψ_{X}) ≃ ν Ψ_{Y}$ . Consequently $TA (X)$ if and only if $TA (Y)$ .

Proof. The least-fixed-point claim is the transfinite comparison used for the unweighted production operator: equivalence conjugates $Φ^{th}$ and preserves the colimits that build $μ Φ$ . For the viability operator, equivalence induces an isomorphism $F_{*} : Sub (S_{X}) \sim Sub (S_{Y})$ intertwining $Ψ_{X}$ and $Ψ_{Y}$ . Greatest fixed points are transported to greatest fixed points. Preservation of $R$ gives the equivalence of $TA$ .

End of proof.

41Thermodynamic separation

Theorem 41.1

Cell and isolated computer

Suppose $TA (B)$ for a bacterium $B$ and $\neg TA (C)$ for an isolated computer $C$ . Then $B \neq ≃ C$ as thermodynamically structured systems.

Proof. Theorem 40.1.

End of proof.

42The virus is a different object from the infected cell

Suppose $TA (B)$ , while an isolated virus satisfies $R_{V} ⊈ μ Φ_{V}^{th}$ . Then $B \neq ≃ V$ , again by Theorem 40.1, since thermodynamic autopoiesis fails for $V$ . The inclusion $R_{V} \subseteq μ Φ_{V ⋆ H}^{th}$ does not disturb the conclusion. The systems $V$ and $V ⋆ H$ are different.

43Changing the boundary changes the object

Let $X ↪ Y$ be an enlargement of the system boundary. In general $E_{X} \subseteq E_{Y}$ , and the available exergy may satisfy $β_{X} \leq β_{Y}$ . Under compatible production doctrines, $μ Φ_{X}^{th} \subseteq μ Φ_{Y}^{th}$ . A system can fail autonomy at one boundary and satisfy it at another. This is not a paradox. The object has changed.

44No self-sufficiency claim without a boundary

For a computer the tower of boundaries runs from the machine, to the machine plus a power grid, to the machine plus a factory, to the machine plus an industrial economy. For a virus it runs from the virion, to the virion plus a host, to the virion plus host plus a transmission ecology. For a bacterium it runs from the cell to the cell plus its nutrient environment.

Every open physical system depends on an environment. The question that distinguishes them is which organizational operations are internal to the chosen boundary.

45Feedstock is not machinery

Suppose an organism imports glucose. The molecule is a resource object. The organism already contains the processes that carry it through transport, glycolysis, respiration, ATP production, and biosynthesis. When a computer receives a newly fabricated processor, much of the process algebra that produced the processor lies outside the computer. External dependence has a type.

46Two operads, not one pile of matter

Let $F_{X}$ be the operad of permitted feedstock transformations and $M_{X}$ the operad of machinery-producing transformations. Autopoietic closure asks, approximately, that

R_{X}^{machinery} \subseteq μ Φ_{X} (E_{X}),

and does not ask that every material particle required by $X$ already lie in $X$ . Organisms do not fail autonomy by eating.

47What a bacterium imports

A bacterium may import carbon, nitrogen, phosphate, and free energy. It uses recursively produced machinery of its own to turn those imports into proteins, lipids, nucleic acids, ribosomes, membranes, and daughter cells. The intended condition is not the absence of external inputs.

48The ordinary computer has the opposite architecture

For a conventional computer the corresponding physical modes look like external machinery plus external feedstock yielding replacement components. The computer itself performs little or none of ore extraction, silicon purification, wafer fabrication, lithography, packaging, or mechanical assembly. The machinery-producing suboperad is external. This is a stronger claim than the observation that the computer needs electricity.

49The theory does not ban artificial life

Nothing here defines life by carbon chemistry. Suppose an artificial system $A$ had endogenous processes for resource acquisition, refining, component fabrication, repair, assembly, and reproduction of its controller, under a sustainable energy throughput. If $R_{A} \subseteq μ Φ_{A}^{th}$ and $ν Ψ_{A} \neq = 0$ , then $TA (A)$ . The predicate is structural. A future machine could satisfy it. Satisfying it is a different claim from computation, learning, self-modifying software, or behavioral copying.

50Autopoiesis does not classify up to equivalence

Even if $TA (A)$ and $TA (B)$ both hold, $A ≃ B$ does not follow. The supports may differ, $Supp (A) \neq ≃ Supp (B)$ . The operads may differ while sharing the same closure truth value. Thermodynamic autopoiesis is an invariant, not a complete classifier. Essay I already separated the substrate obstruction from the organizational one. Both survive the thermodynamic enrichment.

51Viability as a modality

Let $j : Ω \to Ω$ be a Lawvere–Tierney topology encoding a chosen notion of viable observability, with left-exact localization $L_{j} : E \to E_{j}$ . Two systems may become indistinguishable after viability localization, $L_{j} X ≃ L_{j} Y$ , without being equivalent in the ambient topos.

A 2026 treatment of autopoiesis proposes viability-localized self-production, with a Lawvere–Tierney topology, an explicit production and repair structure, boundary production, and viability conditions. The fixed-point construction here is complementary. Rather than only localizing at viable observations, it computes the invariant set of states that support indefinitely extendable viable dynamics.

52Localization and the kernel together

One may ask both for $L_{j} (R_{X}) \subseteq L_{j} (μ Φ_{X}^{th})$ and for $ν Ψ_{X} \neq = 0$ . The first is closure up to viability-relevant equivalence. The second is the existence of actual viable trajectories. Write $VTA (X)$ for the conjunction. It is stronger than either half.

53A viability localization is not a physical equivalence

Suppose a computational model $C$ reproduces every viability-relevant observable of a cell $B$ , so $L_{j} C ≃ L_{j} B$ . It still does not follow that $C ≃ B$ . The missing inference is reflection of equivalences. Unless $L_{j}$ is conservative on the relevant objects, $L_{j} C ≃ L_{j} B$ is only a localized equivalence.

54A scale of equivalence

The notions of sameness in play, from coarsest to finest, are behavioral equivalence, computational equivalence, dependency equivalence, operadic equivalence, thermodynamic-operadic equivalence, and materially typed physical equivalence. None of the reverse implications is automatic. Each step remembers structure the previous step was allowed to forget, and forgetting is exactly where conservativity fails.

55The full invariant

Define

I (X) = (Supp (X), E_{X}, R_{X}, μ Φ_{X}^{th}, V_{X}, ν Ψ_{X}, Θ_{X}, m_{X}, Her (X), Lin (X)) .

The tuple records substrate, process algebra, recursive generability, thermodynamic feasibility, viability, maintenance cost, heredity, and lineage.

56Full invariance

Theorem 56.1

Equivalence preserves the full invariant

If $X ≃ Y$ as fully thermodynamically structured physical systems, then $I (X) ≃ I (Y)$ .

Proof. Each coordinate is defined from structure the equivalence preserves. Support is invariant under typed equivalence. The operads $E$ and $R$ are preserved by definition. Least production fixed points are preserved by conjugacy of $Φ$ . Greatest viability fixed points are preserved by conjugacy of $Ψ$ . The valuation and the maintenance cost are preserved by thermodynamic equivalence. Heredity and lineage are part of the structured equivalence. The tuple is therefore preserved.

End of proof.

57Main separation theorem

Theorem 57.1

Thermodynamic-operadic separation

Let $B$ , $V$ , and $C$ be a bacterium, a virus, and a conventional computer. Assume

R_{B} \subseteq μ Φ_{B}^{th}, ν Ψ_{B} \neq = 0,

while $R_{V} ⊈ μ Φ_{V}^{th}$ but $R_{V} \subseteq μ Φ_{V ⋆ H}^{th}$ for some host $H$ , and $R_{C}^{phys} ⊈ μ Φ_{C}^{th}$ but $R_{C}^{phys} \subseteq μ Φ_{C ⋆ I}^{th}$ for some industrial environment $I$ . Then $B \neq ≃ V$ and $B \neq ≃ C$ as thermodynamically structured systems. If furthermore $I (V) \neq = I (C)$ , then $V \neq ≃ C$ .

Proof. The first two statements are invariance of thermodynamic autopoiesis, Theorem 40.1. For the third, an equivalence would identify the full invariant by Theorem 56.1.

End of proof.

58The diagram

The theory has three layers. Exergy $β_{X}$ enters from the environment. Endogenous operations generate a feasible closure, which is asked to contain the requirements. States evolve by $Ψ_{X}$ , and autonomy additionally requires the viability kernel to be inhabited.

E_{X} ⟶ β_{X} ↓ μ Φ_{X}^{th} ↓ S_{X} ↓ Ψ_{X} ν Ψ_{X} \subseteq V_{X} ⟵ R_{X}

The top asks what can be produced. The horizontal arrows ask what is required. The bottom asks whether production can sustain a viable trajectory.

59The defining equations

Thermodynamically sustained autopoiesis is $R_{X} \subseteq G_{X}^{th}$ and $V_{X} \neq = 0$ . A host-dependent replicator fails the first inclusion for $G_{X}^{th}$ and satisfies it for $G_{X ⋆ H}^{th}$ . An energetically impossible system has $V_{X} = 0$ , whatever its formal operad says.

Autopoietic

$R_{X} \subseteq G_{X}^{th}$ and $V_{X} \neq = 0$

Host-dependent

$R_{X} ⊈ G_{X}^{th}$ , but $R_{X} \subseteq G_{X ⋆ H}^{th}$

Energetically impossible

$V_{X} = 0$

60Conclusion

The resulting distinction is stronger than substrate alone and stronger than computation alone. The structures that matter are material support, the endogenous process operad, recursive productive closure, thermodynamic feasibility, viability, and the typing of environmental coupling.

The biological pattern is not “information processor.” It is closer to the following chain. Environmental matter and free energy enter a recursively self-generated process algebra. That algebra regenerates itself. The trajectory remains in a viable nonequilibrium region. In symbols, $R_{X} \subseteq μ Φ_{X}^{th}$ and $ν Ψ_{X} \neq = 0$ . The thermodynamic cost of maintaining the condition cannot in general vanish: $m_{X} > 0$ .

The organization is causally closed in the relevant organizational sense, and thermodynamically open. Thermodynamic openness is what lets organizational closure persist.

For the cell, a large part of the machinery required to construct, repair, regulate, and reproduce the system lies in $μ Φ_{B}^{th}$ . For an ordinary computer, the crucial fabrication and maintenance operations lie in $μ Φ_{C ⋆ I}^{th}$ only after the industrial environment has been adjoined.

A future artificial system could cross the boundary drawn here. Demonstrating that would require more than computation, learning, self-modifying software, or behavioral self-replication. One would have to establish $R_{A} \subseteq μ Φ_{A}^{th}$ and $ν Ψ_{A} \neq = 0$ , under an explicit physical boundary and a complete thermodynamic accounting. Only then would the separation proved in this edition fail to distinguish that system from an organism.

References

1Jacob Lurie, Higher Algebra. Symmetric monoidal $(\infty)$ -categories and $(\infty)$ -operads.
2John C. Baez and Blake S. Pollard, “A compositional framework for reaction networks.” Reviews in Mathematical Physics 29 (2017), 1750028.
3John C. Baez, Brendan Fong, and Blake S. Pollard, “A compositional framework for Markov processes.” Journal of Mathematical Physics 57 (2016), 033301.
4Riccardo Rao and Massimiliano Esposito, “Nonequilibrium thermodynamics of chemical reaction networks: wisdom from stochastic thermodynamics.” Physical Review X 6 (2016), 041064. Nonequilibrium free energies, chemical driving, and entropy production for open networks.
5Riccardo Rao and Massimiliano Esposito, “Conservation laws shape dissipation.” New Journal of Physics 20 (2018), 023007.
6Jordan M. Horowitz, Kevin Zhou, and Jeremy L. England, “Minimum energetic cost to maintain a target nonequilibrium state.” Physical Review E 95 (2017), 042102. The maintenance-cost invariant $m_{X}$ .
7F. W. Lawvere, “Metric spaces, generalized logic, and closed categories.” Rendiconti del Seminario Matematico e Fisico di Milano 43 (1973), 135–166. Enrichment over $([0, \infty], \geq, +)$ , and the inequality $c (g \circ f) \leq c (f) + c (g)$ .
8The 2026 topos-theoretic treatments of autopoiesis cited in the essay — viability localization, production and repair, and boundary maintenance — are complementary to the two-fixed-point criterion used here, not substitutes for it.