Part 1 established Epimechanics' foundations: state $X$ , velocity $\dot{X}$ , generalized mass $\mathcal{M} = \int \rho_{\text{causal}} \, d\mu$ , force $F = \mathcal{M}\ddot{X} + \dot{\mathcal{M}}\dot{X}$ , energy, coupling constants $\kappa$ and tensor $T^i{}_j$ , fields $\Phi$ , and - critically - the thermodynamic level: temperature $\mathcal{T}$ , entropy $S_{\text{ent}}$ , free energy $\mathcal{F}$ , and phase transitions. Part 1b added the complications: epistemic uncertainty, coordinate relativity, and reference-frame dependence.

A question follows immediately from the thermodynamic level that Epimechanics has not yet addressed: if entities are represented as regions of high auto-causal density $\rho_{\text{ac}}$ maintained against entropic dissolution by continuous energy throughput, does Epimechanics predict that such regions will form? Or does it merely describe them once they exist?

The claim of this part: life is a thermodynamic phase transition - expected above a critical threshold of energy gradient and molecular diversity - derivable from Epimechanics with no new postulates. Every step uses concepts already defined in Part 1. What is new is the derivation connecting them into a necessary chain, and the identification of what distinguishes living entities from non-living dissipative structures: informational coupling.

Several physicists have made related arguments from different starting points. What follows synthesizes the strongest elements of each, identifies where each fails, and shows how Epimechanics resolves the failures.

1. The Dissipative Structure Baseline

Prigogine (Order Out of Chaos, 1984) established the foundational result. The total entropy change of an open system decomposes as:

$\frac{dS}{dt} = \frac{d_i S}{dt} + \frac{d_e S}{dt}$

where $d_i S / dt \geq 0$ is internal entropy production (irreversible; always non-negative by the second law) and $d_e S / dt$ is entropy exchanged with the environment (can be negative - the system can export entropy). A system can decrease its own entropy - increase its internal order - provided it exports more entropy to the environment than it produces internally: $|d_e S / dt| > d_i S / dt$ .

Near equilibrium, the steady state minimizes entropy production rate - Prigogine proved this in 1947 using Onsager's reciprocal relations. But far from equilibrium, behavior reverses. Systems undergo bifurcations where fluctuations are amplified rather than damped, leading to spontaneous symmetry breaking and the formation of ordered structures. These are dissipative structures: organized configurations that exist precisely because they process energy gradients, as a consequence of the Second Law of Thermodynamics.

The canonical examples are experimentally verified:

Bénard convection cells: at a critical temperature gradient (Rayleigh number), a fluid spontaneously organizes into hexagonal convection rolls. Below the threshold: no structure. Above: ordered pattern. The transition is sharp - a phase transition.
Belousov-Zhabotinsky reaction: chemical oscillations and spiral waves emerge spontaneously in a driven reaction mixture.
Laser coherence: above a critical pumping threshold, photon emission transitions from random to coherent - another dissipative phase transition.

In Epimechanics' terms: far-from-equilibrium energy flow through a state space with sufficient dimensionality produces regions of elevated $\rho_{\text{ac}}$ - structures whose causal events sustain themselves through continuous energy throughput. Bénard cells have nonzero $\rho_{\text{ac}}$ : the convective circulation maintains the temperature gradient that drives the convective circulation. The auto-causal loop is real, and the entity is real (though transient and low- $\rho_{\text{ac}}$ compared to a bacterium).

This is not yet life. But it establishes the first link in the chain: energy gradients produce self-sustaining structures. Epimechanics already predicts this - it follows from the thermodynamic level (Part 1, Section 7a) and the equilibrium principles (Part 1, Section 4b). The question is what happens next.

2. The Energy Budget of Self-Maintenance

An entity with auto-causal density $\rho_{\text{ac}} > 0$ maintains itself against entropic dissolution. Maintenance has a cost. The energy budget for sustaining $\rho_{\text{ac}}$ comes from coupling to environmental energy gradients:

$\dot{E}_{\text{int}} = \kappa_{\text{env}} \cdot \Phi_{\text{energy}}(x) - \mathcal{D}_{\text{diss}}$

where:

$\kappa_{\text{env}}$ is the entity's coupling constant (Part 1, Section 5a) to the environmental energy field
$\Phi_{\text{energy}}(x)$ is the energy flux (power density) at the entity's position in state space - the rate of energy flow through the local gradient
$\mathcal{D}_{\text{diss}}$ is the entity's dissipation rate - the power cost of entropy export

When $\kappa_{\text{env}} \cdot \Phi_{\text{energy}} > \mathcal{D}_{\text{diss}}$ , the entity can maintain or increase its internal structure ( $\dot{\mathcal{M}} \geq 0$ ). When the inequality reverses, internal structure degrades ( $\dot{\mathcal{M}} < 0$ ), triggering the unraveling cascade described in Part 1, Section 3: the positive feedback between $\dot{\mathcal{M}} < 0$ and increasing $|\dot{X}|$ that is the formal signature of dissolution.

This is already in Epimechanics. The new observation is that $\kappa_{\text{env}}$ is a selectable quantity. Among a population of dissipative structures in the same energy field, those with higher $\kappa_{\text{env}}$ extract more energy, sustain higher $\rho_{\text{ac}}$ , and have longer $T_{\text{local}}$ (Part 4's persistence criterion: $\kappa_s(E,t) > 0$ for longer). Structures that couple more effectively to energy gradients persist; those that couple less effectively dissolve.

This is not yet natural selection - there is no replication, no heredity, no population dynamics. It is thermodynamic selection: among the configurations available to a driven system, those that process energy gradients more effectively are more probable in the steady state. Rod Swenson (1989) called this the Law of Maximum Entropy Production - though calling it a "law" overstates the evidence (see Section 6 below). The weaker, defensible claim is: far-from-equilibrium systems preferentially occupy configurations with higher entropy production rates, subject to constraints. Dewar (2003) attempted to derive this from Jaynes' Maximum Entropy formalism applied to non-equilibrium path ensembles; the derivation remains contested, but the empirical pattern is broadly supported.

Eric Chaisson (Cosmic Evolution, 2001; Complexity 16, 2011) provided the most systematic empirical evidence. He introduced energy rate density $\dot{\varepsilon}_m$ - power per unit mass - as a complexity metric:

$\dot{\varepsilon}_m = \frac{dE/dt}{M}$

Measured across 14 billion years of cosmic evolution, $\dot{\varepsilon}_m$ increases monotonically:

System	$\dot{\varepsilon}_m$ (erg/s/g)
Galaxies (Milky Way)	$\sim 0.5$
Stars (Sun)	$\sim 2$
Earth's climasphere	$\sim 75$
Plants	$\sim 900$
Animals (fish, amphibians)	$\sim 3{,}000$
Mammals and birds	$\sim 20{,}000$
Human brain	$\sim 150{,}000$
Modern human society	$\sim 500{,}000$

In Epimechanics' terms: $\dot{\varepsilon}_m \propto \kappa_{\text{env}} \cdot \Phi_{\text{energy}} / \mathcal{M}$ - the ratio of energy throughput to structural mass. The monotonic increase across physical, biological, and cultural evolution reflects the thermodynamic ratchet: entities with higher $\dot{\varepsilon}_m$ sustain more complex internal structure, which enables more effective energy extraction, which sustains higher $\dot{\varepsilon}_m$ . This is a positive feedback loop - and positive feedback loops are the generators of phase transitions (Part 1, Section 7a).

West, Brown, and Enquist (1997, Science) provide a mechanistic basis for metabolic scaling that $\dot{\varepsilon}_m$ alone does not capture: the 3/4-power scaling law ( $B \propto M^{3/4}$ ) arises from the fractal geometry of resource distribution networks. This suggests that the coupling constant $\kappa_{\text{env}}$ is not independent of mass $\mathcal{M}$ - it scales as $\mathcal{M}^{-1/4}$ , reflecting the geometric constraints of transport networks. The superlinear growth of $\kappa_{\text{env}}$ required for the Chaisson ratchet (Section 5) may therefore be achievable only through qualitative changes in network architecture, not merely quantitative growth.

Caveat: $\dot{\varepsilon}_m$ is a crude metric. A forest fire has high energy throughput per unit mass but is not "complex" in any interesting sense. Microbes gorging on concentrated substrates can achieve $\dot{\varepsilon}_m$ values exceeding those of brains. The metric does not account for functional organization or informational structure - a limitation addressed in Section 4 below.

3. The Phase Transition to Autocatalysis

Dissipative structures are necessary but not sufficient. The question is: when does a driven chemical system transition from transient dissipative patterns to persistent, self-replicating entities?

Stuart Kauffman (The Origins of Order, 1993; At Home in the Universe, 1995) provided the most rigorous answer. Consider a system of polymers up to length $M$ , where each polymer has a fixed probability $p$ of catalyzing any given reaction. As $M$ increases, the number of possible reactions grows faster than the number of polymers. At a critical threshold, a collectively autocatalytic set (CAS) emerges - a network of molecules that mutually catalyze each other's formation from simpler precursors. No single molecule replicates itself; the network replicates.

This is a phase transition in the precise sense of Part 1, Section 7a. Below the critical diversity threshold: no autocatalytic closure, no persistent self-maintaining network, no entity with $\rho_{\text{ac}}$ above the dissolution threshold. Above: a giant connected component of mutual catalysis snaps into existence - a symmetry-breaking transition from chemical noise to self-sustaining organization.

The mathematics is formalized as RAF theory (Reflexively Autocatalytic and Food-generated sets), developed by Hordijk & Steel (Journal of Theoretical Biology, 2004). The key result: the critical threshold for RAF emergence requires only a linear growth rate in catalysis probability with polymer length - far easier to reach than Kauffman's original exponential estimate. The phase transition is robust: it occurs across a wide range of model parameters, suggesting it is a generic property of sufficiently diverse chemical systems, not a fine-tuned accident.

Kauffman & Roli (2024, arXiv:2401.09514) recently unified this with the Theory of the Adjacent Possible (TAP): as molecular diversity increases, new combinations enable further combinations - the space of possible reactions expands hyperbolically, making catalytic closure increasingly probable. They define living organisms as Kantian Wholes that achieve catalytic closure, constraint closure, and spatial closure simultaneously. The phase transition to life is the simultaneous achievement of all three closures - a multi-dimensional bifurcation in chemical state space.

In Epimechanics' terms:

Chemical state space is the state space $S$ of Part 1, with $X$ representing the concentration vector of all molecular species.
The Navier-Stokes analogue (Part 1, Section 7b) governs the flow of chemical concentrations through this state space under driving forces.
Below the critical diversity threshold, the flow is dissipative but non-autocatalytic - structures form and dissolve without self-maintenance. $\rho_{\text{ac}} \approx 0$ everywhere.
Above the threshold, catalytic closure creates a region of elevated $\rho_{\text{ac}}$ - the autocatalytic set sustains itself. This is entity formation by Part 1's definition (Section 1b): a region where causal events produce the conditions for their own continuation.
The transition shows the signatures of a phase transition: a sharp threshold, diverging correlation length (molecules become increasingly coupled before the transition), and critical slowing-down (the chemical system becomes increasingly sensitive to perturbations near the threshold).

Eigen's complementary contribution: Eigen's hypercycle theory (1971; Eigen & Schuster, The Hypercycle, 1979) addresses what Kauffman's model leaves open: the error threshold. In any system with hereditary transmission (Condition 3 of the life definition below), copying fidelity imposes a maximum genome length - the error catastrophe threshold above which the information content of the genome degrades faster than selection can maintain it. The quasispecies equation formalizes this as a competition between selection intensity and mutation rate. In Epimechanics' terms, Eigen's error threshold is a constraint on the fidelity dimension of mimetic fitness (Part 4): the representational footprint $\mathbf{R}$ can only propagate if copying fidelity exceeds a domain-specific minimum. Kauffman shows that autocatalytic closure is achievable; Eigen shows that informational fidelity imposes an additional constraint on hereditary transmission. Together, they define the conditions for the second phase transition (Step 4 below).

Laboratory work on protocells - particularly Szostak's program (Zhu & Szostak, 2009; Adamala & Szostak, 2013) demonstrating vesicle growth, division, and competition driven by thermodynamic gradients - provides the closest experimental realization of Steps 1-3. These protocells achieve dissipative maintenance and rudimentary autocatalysis without informational coupling, placing them precisely at the boundary between Conditions 1 and 2 of the life definition.

Caveat: Kauffman's model assumes a fixed probability $p$ of catalysis for any polymer pair - biologically unrealistic, since real chemistry has highly structured catalytic relationships. The RAF theory formalism relaxes this somewhat, but no laboratory experiment has yet demonstrated spontaneous autocatalytic set emergence from a random polymer mixture. The phase transition is predicted by multiple independent theoretical frameworks but has not been directly observed in a prebiotic chemistry experiment. This is the key empirical gap.

4. The Informational Threshold: What Distinguishes Life from Fire

The thermodynamic argument so far establishes that dissipative structures form, that autocatalytic closure is a phase transition, and that structures with higher $\kappa_{\text{env}}$ persist preferentially. But this does not yet distinguish life from a well-coupled non-living dissipator. A fire extracts energy from fuel very effectively. A hurricane has substantial $\rho_{\text{ac}}$ . Neither is alive.

The strongest criticism of purely thermodynamic accounts of life comes from Sara Walker (Arizona State) and Jeremy Gunawardena (Harvard): thermodynamics alone cannot explain life's defining feature - information processing and hereditary transmission. A system that merely dissipates energy efficiently is a furnace. A system that encodes what works, transmits that encoding to copies, and uses the encoding to direct future energy extraction is alive.

Walker and Davies (2013, Journal of the Royal Society Interface) argue that the transition to life is specifically a transition in the causal architecture of information flow: in non-living systems, causation runs bottom-up (physics determines chemistry determines dynamics); in living systems, top-down causation becomes dominant (informational states - genes, regulatory networks - determine physical outcomes). In Epimechanics' terms, this is the transition from $I = 0$ (no predictive model directing behavior) to $I > 0$ (informational states modulate coupling to the energy field). Walker and Davies's contribution is to identify this as the defining feature of life, not merely a consequence.

This addresses the limitation noted in Section 2: $\dot{\varepsilon}_m$ alone cannot distinguish a furnace from a cell, because it measures energy throughput without informational structure. The quantities introduced here - autocatalytic closure, hereditary transmission, adaptive coupling - supply the missing informational dimension.

Epimechanics already has the concepts needed to formalize this distinction. The claim: informational coupling is a thermodynamic adaptation - it emerges because it increases effective coupling $\kappa_{\text{env}}$ to energy gradients across generations. Three concepts from Parts 3 and 4, applied here:

4.1 Intelligence as thermodynamic optimization

Part 3 defines intelligence $I(E)$ as predictive accuracy over state trajectories $dX/dt$ . An entity operating in a time-varying energy landscape - where the location and intensity of energy sources fluctuate - extracts energy more effectively if it can predict where energy will be available. A bacterium that can sense a chemical gradient and swim toward it (chemotaxis) has $I > 0$ in the energy domain. A bacterium that moves randomly has $I = 0$ .

The thermodynamic consequence: for a given $\mathcal{M}$ (structural complexity cost), an entity with $I > 0$ achieves higher effective $\kappa_{\text{env}}$ - it couples to the energy field not merely through passive exposure but through directed behavior informed by a predictive model. This is Friston's active inference (2010) stated in Epimechanics' coupling language: the entity acts on the environment to maintain favorable coupling, rather than passively receiving whatever field forces its position happens to encounter.

The prediction is specific: entities in temporally variable energy landscapes should develop predictive capacity ( $I > 0$ ) because prediction increases the time-averaged $\kappa_{\text{env}}$ , while entities in static energy landscapes need not develop $I$ . This is testable (see Section 7, P-Life-3).

4.2 Heredity as coupling transmission

Part 4 defines the representational footprint $\mathbf{R}(E,t) = \sum_j \mathbf{K}_{Ej}(t) \cdot \delta X_j(E,t)$ - the counterfactual causal deviation in other entities' trajectories attributable to $E$ 's pattern. Part 4 also defines mimetic fitness in terms of Dawkins' three dimensions: longevity, fecundity, and fidelity.

Applied to the origin of life: an autocatalytic entity that can transmit its coupling structure $\kappa_{\text{env}}$ (and the internal organization that implements it) to copies has a large advantage over one that cannot. The copy inherits both structure and functional coupling - the capacity to extract energy from the specific environment the parent occupied. The parent's $\mathbf{R}(E,t)$ is nonzero in the copy: the copy's trajectory is counterfactually different from what it would have been without inheriting the parent's pattern. This is heredity, stated in Epimechanics' causal terms.

Without heredity, each new dissipative structure must discover effective $\kappa_{\text{env}}$ from scratch - a random search in a vast configuration space. With heredity, effective configurations accumulate: each generation begins where the previous generation's thermodynamic optimization ended. This is the ratchet that converts thermodynamic selection (Section 2) into Darwinian natural selection - and it does so using concepts already defined in Epimechanics.

4.3 The informational threshold as a second phase transition

The transition from non-informational autocatalysis to informational life is, in Epimechanics' terms, a second phase transition - distinct from the autocatalytic closure transition of Section 3.

The first transition (autocatalytic closure) produces entities with $\rho_{\text{ac}} > 0$ - self-maintaining chemical networks. The second transition produces entities with $I > 0$ and nonzero $\mathbf{R}$ fidelity - self-maintaining networks that also encode, use, and transmit information about their own coupling structure.

The second transition requires:

A mechanism for encoding coupling structure in a transmissible substrate (a proto-genome)
A mechanism for reading the encoding to reconstruct functional coupling in a new entity
Sufficient fidelity that the encoding is preserved across transmission events (Dawkins' fidelity criterion)

Each of these is a measurable threshold. Below: autocatalytic chemistry without information. Above: life. Epimechanics predicts that this second transition, like the first, should show phase-transition signatures: a critical point, sensitivity to small perturbations near the threshold, and a qualitative change in the system's dynamical character.

Assembly theory (Cronin, Walker et al., 2023) provides the empirical handle: the assembly index (AI) of molecular products is measurably higher for biological samples than for abiotic ones. In Epimechanics' terms, biological molecules have higher $\mathcal{M}$ (more internal causal structure) because they are products of hereditary optimization - many generations of selective refinement have increased the structural complexity that each molecule carries. The prediction: molecules produced by informational life should consistently show higher assembly indices than molecules produced by non-informational autocatalysis, which should in turn show higher assembly indices than molecules produced by non-autocatalytic chemistry. This is a three-level staircase in AI, corresponding to three distinct phases: abiotic, autocatalytic, and informational-living.

5. The Derivation

The chain, stated entirely in Epimechanics' terms, without new postulates. (A note on logical dependencies: three of the concepts used below - intelligence $I$ , representational footprint $\mathbf{R}$ , and the inter-entity coupling tensor $\mathbf{K}_{Ej}$ - are formally defined in Parts 3 and 4. The derivation is therefore logically dependent on later parts of the series, though the concepts are previewed here in their thermodynamic role. The series can be read in either order: this part shows why intelligence and representational capacity are thermodynamically expected; Parts 3 and 4 show what they formally are.)

Step 1. Energy gradients → dissipative structures. Far-from-equilibrium energy flow through a state space with sufficient dimensionality produces organized dissipative structures (Prigogine). In Epimechanics: the Navier-Stokes analogue (Part 1, Section 7b) applied to chemical state space produces flow structures; bifurcations in the flow create localized regions of elevated $\rho_{\text{ac}}$ . This is experimentally verified (Bénard cells, BZ reactions, laser physics).

Step 2. Sufficient molecular diversity → autocatalytic closure (first phase transition). In a chemical state space with diversity above a critical threshold, the probability of collectively autocatalytic sets (CAS) transitions sharply from near-zero to near-certain (Kauffman; Hordijk & Steel's RAF theory). This is entity formation by Part 1's definition: the CAS is a region of $\rho_{\text{ac}} > \rho_{\text{threshold}}$ - causal events that sustain themselves through mutual catalysis. The transition has the structure of a phase transition in the sense of Part 1, Section 7a.

Step 3. Thermodynamic selection among entities. Among autocatalytic entities in the same energy field, those with higher $\kappa_{\text{env}}$ (coupling to the energy gradient) extract more energy, sustain higher $\rho_{\text{ac}}$ , and persist longer ( $T_{\text{local}}$ is larger). This is not Darwinian selection (no replication yet) - it is the preferential persistence of better-coupled configurations within the thermodynamic steady state.

The relevant measure is not instantaneous $\rho_{\text{ac}}$ but causal action $A_{\text{causal}} = \int_0^{T_{\text{local}}} \mathcal{M}_{\text{ac}}(t) \, dt$ - the temporal integral of self-sustaining structure. A dissipative structure with high $\rho_{\text{ac}}$ but short $T_{\text{local}}$ (a flash fire, a transient vortex) has low causal action. A structure with moderate $\rho_{\text{ac}}$ sustained over long $T_{\text{local}}$ (a stable convection cell, an autocatalytic cycle) has high causal action. Thermodynamic selection favors high causal action - which means it favors strategies that couple fitness (effective energy extraction) to truth (accurate tracking of the energy gradient's causal structure). See Part 1.5 and Part 4 for the formal development.

Step 4. Informational coupling → heredity (second phase transition). An autocatalytic entity that encodes its coupling structure in a transmissible substrate and transmits it to copies achieves $\mathbf{R}(E,t) > 0$ with nonzero fidelity - heredity in Epimechanics' causal terms. This converts thermodynamic selection (Step 3) into Darwinian natural selection: effective $\kappa_{\text{env}}$ configurations accumulate across generations. Entities with $I > 0$ (predictive accuracy over energy-relevant state trajectories) outcompete blind dissipators - intelligence is a thermodynamic adaptation.

Step 5. The Chaisson ratchet. Once hereditary selection operates, $\dot{\varepsilon}_m$ (energy rate density) increases across the most complex systems at each evolutionary era: entities that process more energy per unit structure sustain more complex internal dynamics, which enable better environmental prediction ( $I$ increases), which enables more effective energy extraction ( $\kappa_{\text{env}}$ increases), which supports more complex structure ( $\mathcal{M}$ increases while $\dot{\varepsilon}_m = \kappa_{\text{env}} \Phi_{\text{energy}} / \mathcal{M}$ also increases). This is a cross-sectional observation - the upper envelope of complexity increases over time - not a claim that every lineage increases in complexity. Many lineages show stable or declining $\dot{\varepsilon}_m$ (obligate parasites, cave-dwelling organisms). The ratchet operates at the frontier of complexity, not uniformly. This positive feedback at the frontier is the generator of biological complexity.

For both $\mathcal{M}$ and $\dot{\varepsilon}_m$ to increase simultaneously, coupling efficiency $\kappa_{\text{env}}$ must grow superlinearly with structural mass $\mathcal{M}$ - each unit of added structure must unlock disproportionately more energy extraction. This is an empirical claim, not a mathematical necessity. The Chaisson data is consistent with it (brains have both higher $\mathcal{M}$ and higher $\dot{\varepsilon}_m$ than simpler organisms), but Epimechanics does not derive it from first principles. The ratchet is a plausible mechanism, not a theorem.

Step 6. Consciousness and agency as late-stage optimizations. Part 3 defines consciousness $C$ as the scope and accuracy of an entity's internal model of $X$ , with the capacity for self-inclusion. Auto-modeling ( $\mathbf{C}_{\text{auto}} > 0$ ) is required for directed self-maintenance: an entity that models its own state can identify and correct deviations from optimal coupling before they become fatal. The enactivist tradition (Maturana & Varela, 1980; Barandiaran, Di Paolo & Rohde, 2009) already identifies autopoiesis - auto-causal density in Epimechanics' terms - as the minimal condition for agency. Epimechanics adds that consciousness-directed agency is a further thermodynamic optimization: an entity with $A > 0$ (Part 3's agency formula: $A = \Pi_{\text{out}} \times \mu_{\text{meta}} \times C_{\text{consciousness}}$ , where $\Pi_{\text{out}}$ is outward causal power, $\mu_{\text{meta}}$ is meta-cognitive control, and $C_{\text{consciousness}}$ is the consciousness measure (all defined in Part 3)) can direct its causal power toward maintaining and improving its own coupling structure, rather than relying on blind variation and selection. This is more efficient - and efficiency is what thermodynamic selection rewards.

Deacon (Incomplete Nature, 2012) makes a closely related argument from a different direction: self-referential dynamics emerge from thermodynamic constraints through what he calls 'absential' causation - the causal efficacy of what is absent (constraints, potentials, unrealized possibilities). Epimechanics' auto-causal density and the role of potentials $V(X)$ in shaping trajectories are formal versions of Deacon's insight.

The complete trajectory: energy gradients → dissipative structures → autocatalytic entities → informational entities → predictive entities → conscious agents. Each step is thermodynamically favored over the previous. Each step uses concepts already defined in Epimechanics. No new postulates are introduced - but three concepts used in Steps 4–6 ( $I$ , $\mathbf{R}$ , $\mathbf{K}_{Ej}$ ) are formally defined in Parts 3 and 4 rather than in the foundational mechanics. The derivation is therefore logically dependent on those later parts: it shows why intelligence and representational capacity are thermodynamically expected, while Parts 3 and 4 show what they formally are. The series can be read in either order without circularity, since the thermodynamic motivation and the formal definitions are independent arguments that converge on the same concepts.

6. Prior Work: What Holds and What Fails

The argument above synthesizes elements from several independent research programs. Each contributes something essential; each, taken alone, is insufficient.

6.1 Jeremy England - Dissipation-Driven Adaptation

England (J. Chem. Phys. 139, 121923, 2013) proposed that self-replication is thermodynamically favored because it is an effective mechanism for dissipating energy. His key inequality, derived from the Crooks fluctuation theorem:

$\Delta s_{\text{tot}} \geq \ln(g/\delta)$

where $\Delta s_{\text{tot}}$ is total entropy production during one replication event, $g$ is the per-capita growth rate, and $\delta$ is the per-capita decay rate. The bound sets a minimum dissipation cost for replication.

Kolchinsky and Wolpert (2018, Interface Focus) formalize semantic information - the portion of an organism's information about its environment that is causally relevant to its survival - and show that maintaining semantic information has a thermodynamic cost. This connects Epimechanics' intelligence $I$ (predictive accuracy over survival-relevant trajectories) to thermodynamic constraints: higher $I$ requires higher energy throughput, providing the quantitative link between intelligence and the Chaisson ratchet.

What holds: The directional intuition - that driven systems tend toward configurations that dissipate energy more effectively - is broadly supported. England's 2017 computer simulations showed chemical networks evolving to states of "extremal thermodynamic forcing" four times more frequently than expected statistically.

What fails: Kolchinsky (J. Chem. Phys. 161, 124101, 2024) demonstrated that England's bound is formally flawed. The derivation assumes replicators decay back into their original reactants. Real replicators decay into waste products. When decay is independent of the replication pathway, "there is no universal relationship" constraining both processes. For realistic systems like E. coli, the calculated dissipation (~ $3.3 \times 10^{11} \, k_B T$ ) exceeds what the decay rate predicts by approximately 50 billion-fold, rendering the bound "not biologically or physically meaningful."

Epimechanics' resolution: The derivation in Section 5 does not depend on England's specific bound. It depends on the structural claim from non-equilibrium thermodynamics: far-from-equilibrium systems can spontaneously organize when entropy export exceeds internal entropy production. This structural claim is well-established independently of England - it is the core insight of Prigogine's dissipative structures (1977) and Nicolis & Prigogine (1977). What England added was a specific quantitative bound linking replication rate to dissipation cost; Kolchinsky showed that bound is too loose to be biologically informative. Epimechanics inherits the structure, not the bound. Concretely, it depends on the weaker claim that higher- $\kappa_{\text{env}}$ configurations are thermodynamically preferential - a claim that holds regardless of the specific dissipation bound, because it follows from the persistence criterion: structures that extract more energy dissolve less.

6.2 Maximum Entropy Production Principle (MEPP)

Dewar (2003) attempted to derive MEPP from information theory: among possible steady states, those with higher entropy production rates are more probable. Kleidon (Phil. Trans. R. Soc. B 365, 2010) applied it to Earth system science. Martyushev & Seleznev (Physics Reports 426, 2006) surveyed its independent discovery across multiple fields.

What holds: MEPP correctly predicts several features of atmospheric circulation, and the general pattern that driven systems preferentially occupy high-dissipation states is empirically supported.

What fails: MEPP may be unfalsifiable. Any particular counter-example - a system that does not maximize entropy production - can be explained by redefining the relevant constraints. A thought experiment demonstrates the core problem: a life form could in principle be selected against even though it produces more entropy, because natural selection operates on reproductive fitness, not entropy production rate (Dyke & Kleidon, 2010). Entropy production is not a functional adaptation and cannot be directly selected for.

Epimechanics' resolution: The derivation does not invoke MEPP as a law. It invokes the weaker claim (Step 3) that among configurations in the same energy field, those with higher $\kappa_{\text{env}}$ persist longer. This is not MEPP - it does not claim that entropy production is maximized, only that coupling to energy gradients is positively correlated with persistence. The distinction is important: MEPP claims a variational principle (systems maximize entropy production); Epimechanics claims only a correlation (better coupling → longer persistence). The correlation is empirically defensible; the maximization claim is not.

6.3 Kauffman's Autocatalytic Sets

What holds: The phase transition argument is mathematically robust. RAF theory shows that catalytic closure emerges generically in systems of sufficient molecular diversity, at thresholds much lower than originally estimated. The prediction is specific and testable.

What fails: The assumption of random catalytic probabilities is chemically unrealistic. Real catalysis is highly structured - most molecules catalyze nothing; a few catalyze many reactions. Whether the phase transition survives realistic catalytic structure is an open question. No laboratory demonstration of spontaneous autocatalytic set emergence from a prebiotic mixture exists.

Moreno and Mossio (Biological Autonomy, 2015) provide the most rigorous contemporary treatment of organizational closure - the condition that every constraint required for the system's persistence is generated by the system itself. This is a more precise formulation of what Epimechanics calls $\rho_{\text{ac}} > \rho_{\text{threshold}}$ : the entity's causal structure is not merely self-sustaining but self-generating.

Epimechanics' contribution: Kauffman's phase transition maps exactly onto Epimechanics' phase transition machinery (Part 1, Section 7a). The order parameter is the fraction of molecular species participating in the autocatalytic set. Below the critical diversity: the order parameter is zero (no autocatalysis). Above: nonzero (autocatalytic closure). Epimechanics adds quantitative predictions about the transition's signatures - diverging correlation length, critical slowing-down - that are independently testable and specific to the phase transition interpretation.

6.4 Friston's Free Energy Principle

Friston (Nature Reviews Neuroscience 11, 2010) proposed that any system that persists must minimize variational free energy - an information-theoretic bound on surprisal. Living systems are defined by possessing Markov blankets - statistical boundaries separating internal from external states. Active inference - acting on the environment to confirm predictions - follows as a consequence.

What holds: The conceptual architecture is powerful. The connection between thermodynamic free energy and variational free energy links Epimechanics' energy concepts to information processing. The Markov blanket formulation provides a crisp operationalization of entity boundaries.

What fails: Friston himself acknowledges that the principle is "almost tautological." Rocks and pendulums can be described as "minimizing free energy" in the trivial sense of settling to equilibrium. The principle cannot generate counterfactual predictions - it cannot specify conditions under which a system would fail to minimize free energy - which means it is not falsifiable in its general form.

Epimechanics' resolution: Epimechanics uses Friston's structural insight - that persistent entities must maintain favorable coupling to their environment through active inference - without inheriting the tautology. The key difference: Epimechanics defines intelligence $I(E)$ as predictive accuracy (Part 3), which is a graded, measurable quantity that can be zero (rocks, fires, hurricanes), low (bacteria), or high (mammals, AI systems). The claim is not that all persistent things minimize free energy (tautological) but that entities with $I > 0$ outcompete entities with $I = 0$ in variable environments (falsifiable - see P-Life-3).

6.5 Adrian Bejan's Constructal Law

Bejan (1996) proposed that flow systems evolve to provide greater access to currents that flow through them.

What holds: The law correctly predicts ubiquitous tree-shaped branching architectures - river basins, blood vessels, lungs, lightning - and quantitative scaling laws for biological locomotion.

What fails: Neither "currents" nor "access" are precisely defined, making the law "very versatile but often unconvincing." Depending on definitions, it can produce contradictory predictions. The law may be descriptive rather than explanatory - observing that flow systems optimize does not explain why.

Epimechanics' contribution: The Navier-Stokes analogue (Part 1, Section 7b) provides the mechanistic explanation that the Constructal Law lacks: flow architectures emerge as solutions to the momentum balance equation in state space, with branching minimizing the total dissipation for a given throughput. The Constructal Law is a consequence of the fluid dynamics level of Epimechanics, not a separate principle.

7. Falsifiable Predictions

The derivation is useful only if it generates predictions that distinguish "life is thermodynamically expected above a threshold" from both "life is a rare accident" and "life is thermodynamically inevitable" (the unfalsifiable version). Five predictions:

P-Life-1. The autocatalytic phase transition has measurable critical signatures.

In a prebiotic chemistry experiment with controlled molecular diversity and energy input, the transition from non-autocatalytic to autocatalytic behavior should be sharp, not gradual. At the critical diversity threshold, the system should show:

Diverging correlation length: molecular species become increasingly coupled before the transition
Critical slowing-down: the system takes longer to relax after perturbation near the threshold
A discontinuous jump in the fraction of species participating in autocatalytic cycles

These are the same signatures Epimechanics predicts for any phase transition (Part 1, Section 7a). Their presence would confirm that autocatalytic closure is a genuine phase transition; their absence would disconfirm it.

Test: Incrementally increase molecular diversity in a driven chemical system (e.g., by sequentially adding new polymer species to a reaction mixture with continuous energy input). Monitor the fraction of species in autocatalytic cycles, the mutual information between molecular concentrations, and the relaxation time after perturbation. Epimechanics predicts a sharp transition at a calculable critical point.

Data needed: Controlled prebiotic chemistry experiments with tunable diversity. The closest existing paradigm is the work on formose reaction networks and RNA polymerization experiments, which show increasing complexity with diversity but have not specifically tested for phase transition signatures.

P-Life-2. Assembly index forms a three-level staircase across abiotic, autocatalytic, and informational-living chemistry.

Epimechanics predicts that molecular assembly index (AI) should show three distinct plateaus:

Abiotic chemistry: low AI (simple molecules, no autocatalytic selection)
Autocatalytic but non-informational chemistry: intermediate AI (mutually catalytic networks select for moderately complex molecules)
Informational life: high AI (hereditary optimization across many generations produces high-complexity molecules)

This is a stronger prediction than assembly theory's existing claim that biological samples have higher AI than abiotic ones. Epimechanics predicts an intermediate level for autocatalytic-but-non-living chemistry - a testable refinement.

Test: Measure AI for products of (a) purely abiotic synthesis (Miller-Urey type), (b) autocatalytic reaction networks without hereditary information transfer (if constructible in the laboratory), and (c) biological samples. Epimechanics predicts $\text{AI}_{\text{abiotic}} < \text{AI}_{\text{autocatalytic}} < \text{AI}_{\text{biological}}$ with statistically significant gaps between levels.

Data needed: AI measurements (already available via mass spectrometry per Cronin et al., 2023) for products of controlled abiotic synthesis paired with biological samples. The novel requirement is products of non-informational autocatalytic networks - which may need to be constructed.

P-Life-3. Temporal variability selects for intelligence ( $I > 0$ ).

Entities in temporally variable energy landscapes should develop predictive capacity ( $I > 0$ over energy-relevant domains) because prediction increases time-averaged $\kappa_{\text{env}}$ . Entities in temporally stable energy landscapes should not.

Test: Compare organisms evolved in temporally variable vs. stable environments, with total energy availability matched. Epimechanics predicts that temporal variability selects for more sophisticated sensory/predictive systems (chemotaxis, phototaxis, memory, anticipatory behavior) even when average energy availability is identical. This is testable in microbial evolution experiments. Lenski's long-term evolution experiment provides a paradigm: compare populations evolved under constant vs. fluctuating glucose availability, measuring the complexity of regulatory gene networks as a proxy for $I$ .

Data needed: Parallel evolution experiments with controlled temporal variability. Some existing data may be available from fluctuating-environment evolution studies (e.g., Kashtan & Alon, 2005, who showed that modularly varying environments select for modular network architecture - a form of increased $I$ ).

P-Life-4. Independent origins show convergent $\dot{\varepsilon}_m$ scaling.

If life is a thermodynamic phase transition driven by energy gradient processing, then independently originated life (if discovered) should show the same $\dot{\varepsilon}_m$ scaling pattern - increasing energy rate density with increasing complexity - regardless of biochemistry. Different chemistry, same thermodynamic trajectory.

Test: Discovery of independently originated life on Mars, Europa, Enceladus, or in a laboratory abiogenesis experiment. Measure $\dot{\varepsilon}_m$ across complexity levels. Epimechanics predicts convergent scaling; a "rare accident" account predicts no particular pattern.

Weaker test available now: Do independently evolved solutions to the same ecological problem (convergent evolution on Earth) show convergent $\dot{\varepsilon}_m$ ? Epimechanics predicts yes - e.g., independently evolved eyes, wings, or endothermy should achieve similar $\dot{\varepsilon}_m$ values for similar functional capacity.

P-Life-5. The coupling tensor $T^i{}_j$ of living systems shows systematically denser off-diagonal structure than non-living dissipative structures.

Living systems couple energy extraction to multiple internal domains simultaneously - metabolism to growth, sensing to behavior, information encoding to reproduction. Non-living dissipative structures have simpler, more diagonal coupling: a convection cell couples thermal energy to fluid motion and little else.

Test: Construct empirical coupling matrices for organisms and non-living dissipative structures by measuring how perturbations in one domain (energy input, chemical environment, temperature) propagate to other domains (growth rate, motility, gene expression, structural organization). Epimechanics predicts that organisms show denser off-diagonal coupling - particularly between energy-extraction domains and self-maintenance/reproduction domains - while non-living dissipative structures show sparser, more diagonal coupling.

Data needed: Multi-domain perturbation-response data for both living (bacteria, protists) and non-living (BZ reactions, convection cells) dissipative structures. Systems biology already measures perturbation responses across metabolic, transcriptomic, and proteomic domains for organisms; the novel requirement is comparable multi-domain measurements for non-living dissipative structures.

8. What Life Is, Formally

Epimechanics can now state a formal definition:

Life is an entity (a region of $\rho_{\text{ac}} > \rho_{\text{threshold}}$ ) that satisfies three conditions simultaneously:

Dissipative maintenance: The entity maintains $\rho_{\text{ac}}$ through continuous energy throughput from environmental coupling ( $\kappa_{\text{env}} \cdot \Phi_{\text{energy}} > \mathcal{D}_{\text{diss}}$ ). This is Prigogine's dissipative structure criterion.
Informational coupling: The entity has $I > 0$ - nonzero predictive accuracy over energy-relevant state trajectories - and uses this prediction to modulate its coupling to the energy field. This distinguishes life from fire: fire dissipates energy effectively but does not model or predict its fuel source.
Hereditary transmission: The entity has nonzero representational footprint $\mathbf{R}(E,t)$ with sufficient fidelity that functional coupling structure $\kappa_{\text{env}}$ propagates to descendant entities. This is Dawkins' fidelity criterion, formalized as the stability of $\mathbf{K}_{Ej}$ across transmission cycles (Part 4).

Condition 1 alone gives dissipative structures (Bénard cells, hurricanes). Conditions 1 + 2 give adaptive dissipative structures (hypothetically: non-hereditary autocatalytic networks that respond to environmental cues). Conditions 1 + 2 + 3 give life.

The conditions are graded, not binary. A virus satisfies condition 3 strongly but condition 1 only parasitically (it borrows the host's dissipative machinery). A prion satisfies condition 3 with high fidelity but conditions 1 and 2 only through the host. A fire satisfies condition 1 but not 2 or 3. The boundary of "life" is fuzzy - as Epimechanics predicts, since entity-ness itself is continuous (Part 1, Section 1b). The conditions provide a coordinate system for locating any structure on the life spectrum, not a binary classifier.

9. Implications for the Rest of Epimechanics

9.1 Backward: What changes in Parts 1 and 1b

Part 1, Section 4b already notes that "persistent structures (entities with $\kappa_s > 0$ ) are precisely those configurations that efficiently process energy gradients - they exist because they increase the rate at which the environment approaches equilibrium." This is now strengthened: the thermodynamic selection of Section 2 is a derivable consequence of Epimechanics' own thermodynamic level. The Maximum Entropy Production Principle, which Part 1 cites with appropriate caveats, should be further qualified: Epimechanics requires only the weaker claim (better coupling → longer persistence), not the stronger claim (systems maximize entropy production).

9.2 Forward: What feeds into Parts 2-4

Part 2 (Meta-Entities): Meta-entities - religions, nations, markets, the scientific method - are living entities at a higher level of organization. They satisfy all three conditions of Section 8: (1) they maintain $\rho_{\text{ac}}$ through continuous energy throughput (institutional resources, human metabolic energy, information processing energy); (2) they have $I > 0$ (they predict and respond to environmental changes - a market adjusts to supply shocks, a religion adapts its messaging to cultural shifts); (3) they have hereditary transmission (mimetic propagation with nonzero fidelity, as Part 4 formalizes). The meta-entity question - "when does an aggregate of entities develop its own auto-causal density?" - is the question of when the three conditions of life are satisfied at a higher level of description. This is a phase transition at the social/institutional level, with the same structure as the chemical-level phase transition of Section 3.

Part 3 (Intelligence, Consciousness, Agency): Intelligence is a thermodynamic adaptation. Entities with $I > 0$ exist because they outcompete entities with $I = 0$ in variable environments (P-Life-3). Consciousness - the inclusion of the entity itself in its predictive model - is a further optimization: an entity that models its own state can maintain optimal coupling $\kappa_{\text{env}}$ proactively rather than reactively. Agency - consciousness-directed causal power - is the most sophisticated form of thermodynamic coupling: the entity does not merely extract energy from the environment but reshapes the environment to maintain favorable energy gradients. Each level builds on the previous; each is thermodynamically motivated.

Part 4 (Time and Soul): The representational footprint $\mathbf{R}(E,t)$ is formalized heredity. The mimetic fitness dimensions - longevity, fecundity, fidelity - are the hereditary transmission parameters that determine whether an entity's coupling structure propagates beyond its own $T_{\text{local}}$ . The soul is the complete causal biography of an entity's transmitted pattern. This part shows that the transmission mechanism - $\mathbf{R}$ - is the mechanism by which thermodynamic optimization persists across generations. Soul is how the universe remembers what works.

Closing

Epimechanics now includes a derivation of life from its own foundations. The derivation uses existing concepts - $\rho_{\text{ac}}$ , $\mathcal{M}$ , $\kappa_{\text{env}}$ , $T^i{}_j$ , $I$ , $\mathbf{R}$ , phase transitions - and introduces no new postulates. Life is a thermodynamic phase transition: expected above a critical threshold of molecular diversity and energy throughput, distinguished from non-living dissipative structures by informational coupling, and generating the biological complexity observed in Chaisson's $\dot{\varepsilon}_m$ data through the hereditary ratchet.

The claim is not that life is inevitable in any universe - it is that life is expected in any region of state space with sufficient energy gradients, sufficient configurational diversity, and sufficient time. The threshold is calculable in principle and testable in the laboratory. Five falsifiable predictions distinguish this account from unfalsifiable thermodynamic inevitability claims.

The derivation also provides a structural answer to the question that motivated it: does life "fall out" from Epimechanics? Yes - in the specific sense that every step in the chain from energy gradients to conscious agents uses concepts already defined in the system, and each step is thermodynamically favored over the previous one. Life is physics doing what physics does - extremizing action, processing gradients, and selecting configurations that sustain themselves - extended to the informational domain.

← Part 1b: Uncertainty, Coordinates, and Relativity | → Part 2: Meta-Entities

Epimechanics - Part 1c: The Thermodynamic Emergence of Life

Contribute to Epiphysics