Executive Summary

Civilisation is assembled matter. Every artefact, institution, and network of coordinated agents represents thermodynamic work accumulated against the Second Law; work that, once performed, must be continuously maintained or it decays. The total stock of this accumulated order is denoted Σ, measured in assembly-steps (Sharma et al., 2023). Maintaining Σ requires a continuous power input whose irreducible floor is set by the aggregate decay rate δ, the specific exergy cost of maintenance ξ, and the Second Law efficiency of the conversion chain η_II. Every watt consumed is ultimately dissipated as waste heat, which must radiate from the planet's finite surface according to the Stefan–Boltzmann relation. This creates a technology-independent ceiling on the assembly stock any planetary civilisation can sustain:

The ceiling exists, is calculable from known physical constants and measurable planetary parameters, and, at historical growth rates, is not distant. Its existence and finiteness cannot be eliminated by any technology, any fuel source, or any institutional reform; but its value depends on operable parameters (ε, A, T_hab, Γ). The ceiling can be widened through engineering; it cannot be removed.

The trajectory toward the ceiling is not accidental. Lotka's maximum power principle, derived in the competitive viability analysis as a consequence of geometric persistence (T2); the principle that entities with larger viability kernels survive more perturbations, as a statistical consequence of the monotonicity of survival probability under kernel inclusion, and that larger kernels retain more measure after perturbation; drives competitive dissipative structures sharing a finite energy gradient to saturate that gradient. The composite coupling Γ(t) = μδξ/η_II links total civilisational power to the assembly stock: P = Γ(t) · Σ. The monetary shadow of this relation has been confirmed empirically: Garrett et al. (2022) documented a stable linear proportionality between global energy consumption and cumulative economic output, holding at approximately 5.9 milliwatts per 2019 US dollar across five decades of data (1970–2019). The empirical stability of this ratio is a predicted consequence of the competitive viability equilibrium derived in the multi-agent analysis. Efficiency improvements, rather than reducing aggregate consumption, are recycled into expanded demand through the Jevons mechanism, confirmed empirically across lighting, computation, and thirty industrial sectors. Financial architecture accelerates the approach: compound interest on endogenous money requires real growth that translates, through the coupling, into minimum energy flows exceeding the maintenance floor. The result is a system that moves toward the ceiling with statistical regularity.

The decomposition of Γ identifies the complete control surface between civilisation and the waste heat constraint. Each parameter falls into one of three compliance classes. Fixed (Tier 1): the Stefan–Boltzmann constant, the positivity of δ and ξ, the bound η_II < 1, set by fundamental physics and immovable. Stiff (Tier 2): the aggregate attractor values of the maximum power principle, Jevons recycling, and competitive exclusion; these resist change with enormous force and no civilisation-scale precedent ∃ for overriding them, though the physics does not forbid it. Compliant (Tier 3): the real interest rate, governance timescales, debt architecture, and the financial compulsions that inflate the metabolic multiplier μ; reformed repeatedly in recorded history. The ceiling is physics. The timeline to the ceiling is scenario. The distinction between the two is this essay's core analytical contribution.

The viability kernel, the region in state space from which at least one admissible control trajectory avoids all constraint violations indefinitely (Aubin, 1991), is contracting. The constraints form a nested temporal hierarchy ordered by decision windows: the greenhouse constraint (decades), the financial constraint (decades to a century), the evolutionary ratchet (continuous), and the waste heat ceiling (approximately 300 years at 2.3% annual growth, conditional on sustained exponential growth and current Γ). Each constraint, if left unaddressed, tightens those below it. The greenhouse problem reduces effective emissivity, lowering the ceiling before waste heat itself becomes significant. The debt architecture steepens the growth trajectory. The ratchet ensures that under noncooperative dynamics, the trajectory proceeds at maximum power regardless of Tier 3 reforms. Success at each tier buys time for the next; failure cascades downward. The kernel shrinks from the outside in: compliant constraints bind first and yield to compliant means; the fixed constraint binds last and requires structural transformation that may not be achievable from within the current mode of civilisation.

The full ceiling formula contains eight parameters with real operability, partitioned between the numerator (ε, A, T_hab, α) and the denominator (the four components of Γ). The combinatoric of planetary stabilisation (nostos; the reduction of Γ) and space-industrial expansion (kleos; the expansion of A) with temporal ordering and awareness yields seven strategic orientations that exhaust the space of possible responses. Of these seven, three are viable or possibly viable; those that pursue Γ-reduction with sufficient magnitude. Two probably fail. Two fail on every trajectory the framework can resolve. The default trajectory is non-viable; every viable trajectory requires Γ-reduction of a magnitude that has no historical precedent; and the parameter that most sharply discriminates viable from non-viable strategies is μ, the metabolic multiplier whose reduction requires civilisation-scale coordination without historical precedent.

Each strategy produces a distinct observational signature in the infrared, generating a richer resolution of Fermi's paradox than any binary filter permits. The galaxy is silent not because survival is impossible, but because the viable region of the allocation space is small and the default trajectory misses it. The strategy of ignorance; in which a civilisation never develops the analytical synthesis connecting maintenance thermodynamics to waste heat ceilings to viability geometry; is probably the modal outcome galaxy-wide, and its observational signature (a brief brightening, then silence) is indistinguishable from that of pure kleos.

The formal apparatus that identifies the danger also delimits the arena within which the danger operates. That arena is large. Current global primary energy consumption stands at approximately 20 TW; the waste heat ceiling at the reference habitability threshold is approximately 6,800 TW. The current civilisation occupies less than one per cent of the thermodynamic corridor that physics permits on this planet, at this distance from this star, with this atmosphere. More than 99 per cent of the assembly stock that Earth's radiative budget can sustain; at current Γ, without any expansion of the radiating area; lies in the future.

The correct objective is not maximum instantaneous stock, which rewards overshoot and collapse, but the expected integral of maintained assembly over time, subject to viability: the stock-years functional V(x). Under this objective, collapse trajectories are strictly dominated; temporary contraction is admissible if it raises the reachable long-run frontier; growth is desirable wherever it increases the integral. The objective is not "less." It is "more, durably." Each stock component's value is its shadow contribution to V; complexity matters only instrumentally, insofar as it preserves low-coupling support functions, deepens optionality, or expands the boundary conditions within which the coupled system can persist.

The planetary ceiling is the surface-bound special case of a general source–reservoir–stock–sink viability condition requiring that exergy demand not exceed available supply and that entropy export not exceed sink capacity. The general model admits intermediate exergy reservoirs; temporal batteries; whose inventory spans proven fossil reserves (approximately 50,000 EJ), through technology-contingent nuclear reserves with breeder reactors (approximately 450,000 EJ), to deep reserves including oceanic uranium with breeders exceeding 10⁸ EJ. These batteries cannot fuel the default trajectory to the ceiling; they run out long before it is reached. But they are enormous relative to the energy required for trajectory correction at the current position, which is the strategically relevant comparison. Their allocation between throughput acceleration and trajectory correction is a Tier 3 decision; an institutional choice, not a physical constraint.

The aggregate variable Σ conceals functionally distinct stock classes whose shadow values under V differ sharply: support stock (the low-coupling biospheric substrate), control stock (sensing, computation, institutions, coordination mechanisms), boundary stock (collectors, radiators, launch infrastructure; the only class that raises the ceiling without bound), reservoir stock (temporal batteries providing transition manoeuvre energy), and burden stock (high-maintenance complexity whose main effect is to raise Γ and narrow the corridor). The technosphere spans all five classes. It is not an unconditional good to be maximised; it is a catalytic control layer whose legitimacy is instrumental.

The essay's contribution is the construction of a navigational instrument: the source–reservoir–stock–sink model, fitted to the planetary special case, with the viability kernel mapped, the control surface decomposed, the competitive equilibrium derived, the stock roles classified, and the objective stated. The terrain is severe and the default trajectory is lethal. None of that is retracted. But the viable region is not empty; the corridor is measured in orders of magnitude, not decades; the batteries are loaded; and, for the first time in the trajectory, the instruments exist to see the walls before hitting them. The corridor is open. The allocation has not yet been made. Quantitative calibration of all Earth-specific parameters is consolidated in Appendix A.

Nomenclature

State Variables

Physical Parameters (Tier 1: Immutable)

T_hab = Habitability threshold (K): maximum equilibrium temperature compatible with organised civilisation. Biological wet-bulb floor is fixed; effective civilisational threshold has moderate operability through engineering

Thermodynamic Coupling (Derived, Tier 1 + Tier 2)

Dynamical Variables

Climate System

Braking boundary

Σ^*(T) = Viability boundary in (Σ,T) space: maximum stock from which braking to C = 0 avoids constraint violation (assembly-steps)

Empirical Corroboration (Garrett, Tier 3 Monetary Shadow)

Solvency

Multi-Agent Viability (introduced in §4.3)

The geometric persistence principle

Biosphere–Technosphere Coupling (introduced in §2.1)

Value Functional and General Viability (introduced in §5.4)

V(x₀) — Stock-years value functional: expected integral of maintained assembly over time, subject to viability

Ḃ_req, Ḃ_avail — Required and available exergy flow rates (W)

R_i — Reservoir stock for reservoir i (J)

I_i — Reservoir recharge rate for reservoir i (W)

Key Relations

Part 1 — The Arena

1.1 The Cosmological Reference Frame

The universe began in a state of extraordinarily low entropy. This is not a minor footnote in cosmology. It is, in the precise assessment of Roger Penrose, the most important unexplained fact about the initial conditions of the cosmos. Roughly 13.8 billion years ago, the matter and energy that would become everything now observed was compressed into a configuration of almost perfect uniformity, gravitationally smooth, thermally homogeneous, and staggeringly far from the disordered equilibrium toward which the Second Law of Thermodynamics demands all isolated systems evolve. The entire history of the universe since that moment; the formation of galaxies, stars, planets, oceans, organisms, and civilisations, has been the progressive dissipation of that initial gradient. Structure is not built against entropy. Structure is what entropy production looks like when gradients are large and dissipation is constrained by geometry and kinetics.

This essay is about one particular dissipative structure: human civilisation. But to understand the constraints that govern civilisation's future, one must first understand the gradient it rides.

The relevant gradient for life on Earth is not the total entropy budget of the universe. It is a local, astrophysical gradient defined by two temperatures: the surface temperature of the Sun, approximately 5,800 K, and the temperature of deep space, approximately 2.7 K (set by the cosmic microwave background, the cooled remnant of the Big Bang itself). The Sun delivers energy to Earth at a rate of roughly 1.2 × 10^17 watts. But the critical point is not the quantity of energy. The First Law of Thermodynamics guarantees that energy is conserved, Earth radiates back to space almost exactly what it receives. The critical point is the quality. Solar photons arrive at high temperature, few in number, each carrying substantial energy. Earth re-radiates the same total energy as infrared photons at much lower temperature, vastly more numerous, each carrying far less energy. The entropy of outgoing radiation is enormously greater than the entropy of incoming radiation. This entropy difference is the gradient. It is the thermodynamic window through which every living system on this planet operates.

A simple way to quantify this: the entropy flux entering the Earth system from sunlight is approximately

while the entropy flux leaving is approximately

where T_⊕ ≈ 255 K is Earth's effective radiative temperature as seen from space. The difference, roughly 4.5 × 10^14 watts per kelvin, is the entropy production rate of the entire Earth system. It is the budget within which photosynthesis, weather, ocean circulation, the water cycle, and every biological and industrial process on the planet must operate. Nothing on Earth creates order. Everything on Earth borrows order from the Sun and pays for it by exporting entropy to space.

Civilisation is a particular strategy for exploiting this gradient. It is a dissipative structure in the precise sense defined by Prigogine: an open, far-from-equilibrium system that maintains its internal organisation by continuously processing free energy (exergy) and exporting entropy to its surroundings. A city does not violate the Second Law any more than a hurricane does. Both are sustained by throughput. Cut the gradient, remove the energy source or block the entropy sink, and both collapse toward equilibrium. The difference is that civilisation has learned to access gradients beyond the immediate solar flux: fossil hydrocarbons (stored Cretaceous sunlight), fissile elements (remnants of supernova nucleosynthesis), and potentially fusion fuels (primordial hydrogen). Each of these represents a reservoir of low-entropy energy that civilisation can process to maintain and expand its structure.

The question this essay addresses is not whether civilisation can find new sources of exergy. It is whether the sink can absorb the consequences.

Every watt of power that civilisation consumes, from any source whatsoever, is ultimately dissipated as waste heat. This is not a statement about fossil fuels. It is not about carbon dioxide or the greenhouse effect. It is a direct consequence of the Second Law. A nuclear reactor produces heat. A wind turbine converts kinetic energy to electricity that is then degraded to heat through resistance, friction, computation, and every other end use. A fusion reactor, should one operate commercially, produces heat. This waste heat must be radiated to space from the top of the atmosphere, and the rate at which it can be radiated is governed by the Stefan-Boltzmann law:

where ε is the effective emissivity, σ is the Stefan-Boltzmann constant (5.67 × 10^- 8 W m^- 2 K^- 4), A is the planetary radiating area, and T_eq is the equilibrium temperature. If civilisation adds power P to the planetary energy balance, the equilibrium temperature must rise until outgoing radiation matches incoming solar plus anthropogenic flux. This is pure physics. It is source-independent, technology-independent, and policy-independent.

The cosmological reference frame, then, is this: civilisation ∃ within a thermodynamic window defined by a hot source above and a cold sink below. The source determines the maximum rate at which exergy can be accessed. The sink determines the maximum rate at which entropy can be exported. The first constraint is a question of fuel and technology. The second is a question of geometry and radiative physics, and it is far harder to negotiate with. As civilisation's power consumption grows, it presses against the sink constraint, not because of which fuels it burns, but because of how much total power it dissipates. The warming that results is not a consequence of atmospheric chemistry. It is a consequence of the Second Law operating at planetary scale.

This essay develops that argument in full. The assembly framework and maintenance physics establish civilisation as a heat engine and derive the empirical relationship between energy consumption and accumulated structure. The trajectory analysis examines the mechanisms that lock civilisation onto an exponential energy path; the Jevons Paradox, historical inertia, debt-driven growth, and the compound ratchet they form together. The viability analysis confronts the hard limits: the waste heat ceiling, the thermodynamic costs of information, and the formal mathematics of survival under constraint. The final part asks what follows, whether the constraint set admits any feasible trajectory, or whether thermodynamic mortality is the default outcome for technological civilisations in this universe.

The laws of thermodynamics do not negotiate. They do not yield to innovation, capital allocation, or political will. They define the boundaries of the possible. Everything that follows in this essay operates within those boundaries.

1.2 System Boundaries

The previous section established the cosmological setting: a universe born in a state of extraordinarily low entropy, dissipating toward equilibrium through the formation and dissolution of structure. Within that setting, civilisation ∃ as a local dissipative structure, sustained by the gradient between a hot source (the Sun, at 5,800 K) and a cold sink (deep space, at 2.7 K), and constrained ultimately by the rate at which waste heat can be radiated from the top of the atmosphere. That constraint is source-independent, it applies to any energy technology, any fuel, any economic system.

But to make that intuition rigorous, a set of thermodynamic primitives must be established. This section defines the formal tools that the rest of the essay deploys: the entropy balance for open systems, the distinction between entropy export and entropy generation, the concept of exergy and its destruction, and most critically, a principle for selecting the correct system boundary at civilisational scale.

The entropy balance

Every thermodynamic argument begins with a choice of system boundary. Everything inside is "the system." Everything outside is "the environment." The boundary determines what counts as internal process and what counts as exchange. Get the boundary wrong and the accounting is wrong. For an open system, one that exchanges both energy and matter with its surroundings, the entropy balance takes the form:

where Ṡᵢₙ is the entropy carried into the system across the boundary (by heat, radiation, or mass flows), Ṡₒᵤₜ is the entropy carried out, and Ṡ_gen is the entropy generated internally by irreversible processes. The Second Law imposes one absolute constraint on this balance:

Internal entropy generation is never negative. This is the operational content of the Second Law for open systems, formalised by Prigogine in his 1945 doctoral thesis and presented definitively in his 1977 Nobel Lecture. The decomposition is often written in differential form as dS = d_eS + d_iS, where d_eS is the entropy exchanged with the environment (which can be positive or negative) and d_iS ≥ 0 is the entropy produced internally by irreversible processes.

The critical subtlety, one that causes persistent confusion in energy policy, is that heat rejection and irreversibility are not the same thing. A perfectly reversible Carnot engine rejects heat to its cold reservoir. It must: this is required by the Second Law even in the ideal case. But a reversible engine generates zero entropy (Ṡ_gen = 0). The heat rejection is thermodynamically necessary; the irreversibility is not. In a real engine, Ṡ_gen > 0 because of friction, finite-temperature-difference heat transfer, turbulence, mixing, and every other departure from the reversible ideal. These departures are the true thermodynamic cost. Heat rejection is the price of operating between two reservoirs. Irreversibility is the tax on operating imperfectly.

This distinction matters at civilisational scale because it separates two different problems. The first problem, waste heat, is inescapable. Any power consumed within the Earth system is ultimately degraded to low-grade heat that must be radiated to space. The second problem, irreversibility, is reducible in principle. Better engines, more efficient processes, and cleverer engineering can reduce Ṡ_gen per unit of useful output. The question is whether that reduction can be achieved fast enough and far enough to matter. The analysis of evolutionary dynamics and inertia will show that it cannot.

Exergy: what civilisation actually consumes

A common confusion runs through nearly all public discourse about energy. Energy is conserved, the First Law guarantees this. The Earth system receives energy from the Sun and radiates the same total quantity back to space. No energy is lost. None is created. But civilisation is not powered by energy. It is powered by exergy: the fraction of energy that is available to do useful work, given the temperature and composition of the surrounding environment.

The concept traces to J. Willard Gibbs (1873) and Hermann von Helmholtz (1882). The term itself was coined by the Slovenian engineer Zoran Rant in 1956, from the Greek ex (out of) and ergon (work). The formal definition: exergy is the maximum useful work obtainable as a system is brought reversibly into equilibrium with its reference environment. For thermal energy at temperature T_H in an environment at temperature T_C, the maximum extractable work is bounded by the Carnot efficiency:

The higher the source temperature relative to the sink, the greater the fraction of heat that can be converted to work. The lower the ratio, the less work is available. At T_H = T_C, the exergy is zero, there is thermal energy, but none of it can do anything useful. This is equilibrium: the state of maximum entropy for a given total energy.

Unlike energy, exergy is destroyed by every irreversible process. Friction destroys exergy. Mixing destroys exergy. Finite-temperature heat transfer destroys exergy. Chemical reactions destroy exergy (unless they are run reversibly, which no real process achieves). The Gouy-Stodola theorem, independently derived by Louis Gouy in 1889 and Aurel Stodola in 1898, quantifies this destruction:

where X_dest is the exergy destroyed, T_0 is the reference environment temperature, and Ṡ_gen is the entropy generated by irreversible processes. This is a remarkably clean result. It says that penalising entropy generation and penalising exergy destruction are mathematically equivalent operations, linearly related by the ambient temperature. Entropy is the physicist's currency. Exergy is the engineer's. They are exchangeable at a fixed rate.

This matters because it gives a single, objective measure of thermodynamic cost. When civilisation burns fuel, smelts ore, runs a data centre, or heats a building, the exergy consumed is real and measurable. The entropy generated is real and measurable. The financial cost is a human convention that may or may not reflect the physical reality. The thermodynamic cost is not a convention. It is set by the laws of physics and cannot be negotiated, deferred, or externalised. It can only be paid.

The boundary renormalisation principle

With the entropy balance and exergy framework in hand, the most consequential question in civilisational thermodynamics can now be addressed: where does the boundary belong?

In textbook thermodynamics, the boundary is a pedagogical choice. It is drawn around the piston, or the turbine, or the refrigeration cycle, and the flows across it are analysed. The environment is assumed to be an infinite, fixed-temperature reservoir, a heat bath that accepts whatever entropy the system exports without changing its own state. This is a reasonable assumption for a machine in a room. The room absorbs the waste heat. Its temperature rises negligibly. The analysis holds.

But there is a regime in which this assumption fails, and it fails silently. When the activity of the system materially perturbs the sink it assumed was fixed, the entire analysis becomes unreliable. The room is no longer an infinite reservoir if the machine heats it by ten degrees. The regional environment is no longer a fixed sink if the city changes its local climate. And the atmosphere is no longer a passive boundary condition if the civilisation's total power dissipation begins to alter the planetary radiative balance.

This leads to a principle that is simple to state and ruthless in its implications:

If the activity materially perturbs the sink assumed fixed, the boundary is wrong. Expand it until a sink is reached that cannot be materially perturbed.

Call this the boundary renormalisation principle. It is not a law of thermodynamics — it is a methodological rule that follows from the requirement that the entropy balance be self-consistent. It scales naturally with the size and power of the system under analysis. For a machine, the room is the correct sink. The machine cannot materially perturb the room's temperature. For a factory, the local environment is the sink. For a city, the regional atmosphere and hydrosphere. For a planetary civilisation — one whose total power dissipation is measurable against the solar input — the only sink that cannot be materially perturbed is deep space itself. The boundary that closes the energy balance is the top-of-atmosphere radiative interface: the surface from which the Earth system radiates infrared photons into the 2.7 K void.

This is not an arbitrary choice. It is the unique boundary at which the accounting becomes honest. Below it, every "sink" is actually an intermediate reservoir that the system can and eventually will perturb. Above it, there is no practically recoverable gradient. Radiation emitted to space is, in any realistic engineering regime, gone. The top-of-atmosphere interface is where civilisation's entropy account closes.

For a spacecraft — a civilisation that has left the planetary surface — the equivalent boundary is the radiator surface. The sink is still deep space. The physics is identical. Only the geometry changes.

The hard constraint: radiative geometry

Once the boundary is fixed at the top of the atmosphere, the physics of heat rejection becomes a geometry problem. The rate at which the Earth system can radiate energy to space is governed by the Stefan-Boltzmann law:

where ε is the effective emissivity of the Earth system (accounting for atmospheric absorption and re-emission), σ is the Stefan-Boltzmann constant (5.67 × 10^- 8 W m^- 2 K^- 4), A is the radiating area (the surface area of the top of the atmosphere, approximately 5.1 × 10^14 m^2), and T is the effective radiative temperature.

At steady state, energy in must equal energy out:

where P_☉ is the absorbed solar flux (approximately 1.2 × 10^17 W) and P is the total power dissipated by civilisation from non-solar sources. If P is small compared to P_☉, the equilibrium temperature shifts negligibly. If P grows — as it has, exponentially, for two centuries — then T_eq must rise to restore balance. The relationship between added power and temperature rise is nonlinear (T scales as the fourth root of total flux), but it is monotonic and it is absolute. There is no technology that changes it. There is no fuel that avoids it. There is no efficiency gain that eliminates it, because the waste heat is a consequence of power consumption itself, not of any particular conversion pathway.

The implications are developed in detail in the temporal hierarchy of constraints (Section 4.1). But the point to register here is structural: the sink constraint is fundamentally a constraint on area, temperature, and emissivity. The Earth has a fixed radiating area. Its emissivity is set by atmospheric physics. The only variable that can adjust is temperature. And temperature is the variable that determines whether the planet remains habitable.

This is the thermodynamic cage within which the rest of the essay operates. Civilisation sits between a source it can choose (solar, fission, fusion, fossil — though not all are equivalent in their secondary effects) and a sink it cannot choose. The source determines how much exergy is accessible. The sink determines how much total power can be dissipated before the equilibrium temperature exceeds the boundaries of habitability. The gap between these two constraints — and the rate at which civilisation is closing it — is the subject of this essay.

Part 2 — The Machine

2.1 Civilisation as Assembled Matter

Part 1 established the arena: a universe born in a state of extraordinarily low entropy, dissipating toward equilibrium through the formation and dissolution of structure. Within that arena, civilisation occupies a thermodynamic window defined by a hot source (the Sun, at 5,800 K) and a cold sink (deep space, at 2.7 K), bounded ultimately by the rate at which waste heat can be radiated from the top of the atmosphere. That boundary is source-independent, it applies to any energy technology, any fuel, any economic system.

This section asks a more specific question: what is civilisation, physically? Not what it does. Not what it produces. Not what it means. What it is, stated in the language of thermodynamics, measurable in principle and undeniable in fact.

The answer turns out to be surprisingly precise, and the precision matters. Everything that follows in this essay, the maintenance derivation (Section 2.2), the waste heat ceiling, the Landauer floor, the trajectory analysis of Part 3, the viability mathematics (Section 4.2) of Part 4, depends on getting this definition right. A vague definition ("civilisation is complex") yields vague conclusions. A physics-native definition yields physics-native constraints. The constraints are the point.

Dissipative structures and the price of order

The classical Second Law, formulated by Rudolf Clausius in 1865, states that the entropy of an isolated system never decreases:

For isolated systems, those exchanging neither energy nor matter with their surroundings, the arrow points inexorably toward equilibrium, the state of maximum disorder. But civilisation is not an isolated system. It is an open system, exchanging both energy and matter with the biosphere and, ultimately, radiating waste heat to the 2.7 K cold sink of deep space.

The decisive extension of the Second Law to such systems was formalised by Ilya Prigogine (1917–2003) in his 1945 doctoral thesis and presented definitively in his 1977 Nobel Lecture. Prigogine decomposed the total entropy change of an open system into two terms:

where d_iS ≥ 0 is the entropy produced internally by irreversible processes; friction, chemical reactions, heat conduction, mixing, and d_eS is the entropy exchanged with the environment, which can be negative. The total entropy of the system can therefore decrease (dS < 0) provided the system exports enough entropy to its surroundings: | dₑS | > dᵢS. As Prigogine stated in his Nobel Lecture: the entropy production inside the system is always positive or zero, but entropy transfer across the boundary has no sign restriction.

This is the physics that permits local islands of order; cells, organisms, cities, civilisations, to exist in a universe trending toward heat death. They do not violate the Second Law. They satisfy it, by exporting more entropy than they generate internally, at the cost of continuous throughput of free energy. The moment the throughput stops; the gradient is cut, the fuel runs out, the Sun dims, the system relaxes toward equilibrium. The order was never free. It was rented.

Prigogine called such systems dissipative structures: open, far-from-equilibrium configurations that maintain their internal organisation by continuously dissipating free energy. The canonical examples are Bénard convection cells (ordered hexagonal patterns that emerge spontaneously in a fluid heated from below), the Belousov–Zhabotinsky reaction (chemical oscillations maintained by reagent throughput), and biological organisms (which maintain their structure through metabolic processing of nutrients). Each exists only so long as the thermodynamic gradient that sustains it persists, and each collapses toward equilibrium when the gradient is removed.

The intellectual lineage runs through Ludwig von Bertalanffy (open systems theory, 1932), Erwin Schrödinger (What is Life?, 1944, in which he proposed that organisms "feed on negative entropy"), Lars Onsager (reciprocal relations in irreversible thermodynamics, 1931, Nobel Prize 1968), and culminates in Prigogine's demonstration that new ordered states can emerge spontaneously in far-from-equilibrium systems, that the Second Law, far from forbidding structure, provides the driving force for its creation through dissipative self-organisation (Prigogine and Nicolis, 1977).

Civilisation is the largest and most complex dissipative structure on Earth. Its physical substrate; roads, buildings, power grids, communication networks, factories, vehicles, cultivated land, managed forests, domesticated organisms, and the organised bodies and minds of eight billion humans, represents an accumulation of thermodynamic work held in a state of low entropy against the relentless push of the Second Law. The question is how to measure that accumulation in physics-native units, rather than the monetary proxies that economics provides.

The thermodynamic maintenance stock: an axiomatic derivation

The Prigogine framework establishes that civilisation is a dissipative structure maintained far from equilibrium by continuous entropy export. The question that follows immediately is quantitative: how should the accumulated physical substrate of that structure be measured? Economics provides monetary aggregate; GDP, capital stock, cumulative wealth, but these are denominated in units of human convention. A dollar of GDP tells how much monetary value was transacted; it tells nothing about the physical complexity of the structure that was built, maintained, or destroyed in the process. A billion dollars of financial derivatives and a billion dollars of semiconductor fabrication represent the same economic magnitude but vastly different physical commitments. What the argument requires is a physics-native aggregate measure of civilisational complexity, one whose properties are dictated by thermodynamics, not by accounting convention. The question is: what properties must such a measure have?

Four axioms are sufficient. Each follows from a named principle of thermodynamics applied to the class of thermodynamically unstable structures maintained by civilisational power throughput.

Axiom A1 (Extensiveness). The total maintenance burden of a civilisation is the sum of the maintenance burdens of its components. This follows from the First Law: energy is a conserved, extensive quantity, and the energy costs of maintaining independent subsystems are additive. Maintaining a bridge and a data centre costs the maintenance cost of the bridge plus the maintenance cost of the data centre. No coupling term arises unless the subsystems share maintenance pathways, and even then, the total energy expenditure remains additive over the combined system. Therefore the aggregate measure must be extensive: additive across the ensemble.

Axiom A2 (Strict positivity of unit maintenance cost). Every component of the technosphere is thermodynamically unstable relative to its environment. Steel is unstable relative to iron oxide in the presence of oxygen and water. Semiconductors are unstable relative to amorphous silicon in the presence of thermal fluctuations. Concrete is unstable relative to calcium carbonate in the presence of atmospheric CO₂. Restoring any degraded component to its functional state requires strictly positive exergy input, a consequence of the Gouy-Stodola theorem, which states that the minimum work required to reverse a spontaneous process equals the exergy destroyed by that process. The maintenance cost per unit of the aggregate measure is therefore strictly positive: denoting this specific exergy cost ξ, the axiom gives ξ > 0. No component of the technosphere can be maintained for free.

Axiom A3 (Monotonic decay without maintenance). When maintenance power is withdrawn from the aggregate, the measure decreases monotonically. This is the Second Law operating on thermodynamically unstable structures in contact with a higher-entropy environment: left unattended, steel corrodes, concrete spalls, circuits degrade, roads crack, code rots. The direction is universal; only the rate varies. Characterising the aggregate rate of decay by a parameter δ > 0, and denoting new construction by C(t), the axiom gives the stock balance directly:

This is a consequence of the Second Law, not a modelling choice. Any measure of accumulated thermodynamic work in unstable structures must satisfy an equation of this form: construction adds to the stock, entropic decay subtracts from it, and the rate of subtraction is proportional to the stock.

Axiom A4 (Operational partitionability). The measure admits a clean partition between objects maintained by civilisational power throughput and objects maintained by other energy flows, biospheric, geological. The partition criterion is a counterfactual test: if civilisation's power throughput were reduced to zero, would the object disassemble on a timescale shorter than its geological relaxation time? If yes, it belongs to the civilisational aggregate. If no, it belongs to the biosphere or geosphere and is maintained by non-civilisational energy flows. This axiom is not derived from a single thermodynamic identity; it is a consequence of the system boundary choice established in the boundary analysis, combined with the physical fact that different energy flows maintain different classes of structure. It ensures the measure counts only the structures whose maintenance cost enters the civilisational thermodynamic balance.

The formal object. Any aggregate measure satisfying A1–A4 is a thermodynamic maintenance stock, denoted Σ, with units of maintenance-equivalent operations. The axioms determine the measure's mathematical structure: extensive (A1), strictly positive unit cost (A2), governed by the stock balance (A3), and cleanly partitioned from non-civilisational assembly (A4). Any specific counting procedure that satisfies A1–A4 yields the same maintenance floor, the same power–assembly coupling structure, the same waste heat ceiling, and the same viability geometry. The framework depends on the axioms, not on any particular instantiation of the counting procedure. This is the central methodological point: what follows requires Σ to satisfy A1–A4; it does not require any specific method of computing Σ for a given object.

Assembly Theory as empirical instantiation

Sharma et al. (2023), publishing in Nature under the title "Assembly Theory Explains and Quantifies Selection and Evolution," provide a specific instantiation of the thermodynamic maintenance stock that satisfies A1–A4. Their framework rests on two primitives. The assembly index, a, counts the minimum number of joining operations required to construct an object from its basic building blocks, a physical property of the object, independent of economic valuation. The copy number, n, counts instances of a given object type within the system. From these, the assembly of an ensemble can be computed. The framework is deliberately scale-free: the "basic building blocks" and "joining operations" can be instantiated at any level — atoms joined by chemical bonds, components joined by welds, modules joined by electrical connections.

Assembly Theory contributes four things beyond what A1–A4 alone provide. First, a concrete, experimentally grounded counting procedure that has been validated at the molecular scale. Second, the empirical threshold (a ≈ 15) distinguishing biotic from abiotic products, molecules with assembly index above this value are never observed in abiotic samples, providing a crisp, experimentally confirmed demarcation between the products of selection and the products of chance. Third, scale-free extensibility from molecules to civilisational structures, which ensures the counting procedure does not break at the macroscopic level. Fourth, connection to the broader theoretical literature on selection, evolution, and the accumulation of complexity.

The logical relationship is explicit: A1–A4 are necessary and sufficient for every result that follows. Assembly Theory is one sufficient instantiation that provides additional empirical content. If Assembly Theory did not exist, the argument would stand on A1–A4 alone. If Assembly Theory is rejected at civilisational scale, a position some critics hold, the rejection does not touch the axioms, and the axioms carry the entire load.

Garrett's monetary shadow

A second, independent empirical confirmation comes from Garrett's programme of work (Garrett, 2011, 2012, 2015; Garrett et al., 2022). The aggregate data show a remarkably stable proportionality between global power consumption P and cumulative economic wealth W: P ≈ λW, with λ ≈ 5.9 mW per 2019 USD, stable over the observational record (1970–2019) despite large variations in energy mix, technology, and economic structure. This is the monetary shadow of the physical relation P = Γ(t)·Σ derived in this essay, where λ = Γ(t)/κ and κ is the monetary conversion factor. The empirical stability is a predicted consequence of the competitive viability equilibrium; the derivation does not depend on it.

The physically derived relation P = Γ(t)Σ subsumes this empirical observation. Cumulative wealth W is a monetary proxy for the thermodynamic maintenance stock: W ≈ κΣ, where κ is a monetary conversion factor, a Tier 3 artefact reflecting units of account, not physics. Garrett's coupling constant decomposes as λ = Γ(t)/κ, an empirical composite reflecting both the physical coupling and the monetary measurement convention. The stability of λ is evidence that Γ(t) has been approximately constant over the observational period, a significant empirical fact, but not a foundational one. Garrett provides calibration and confirmation. He is not foundational.

Three definitions

With the source/sink boundary from the system boundary analysis and the axiomatic framework A1–A4, three definitions can now be stated with precision.

Definition 1. The system boundary is the top-of-atmosphere radiative interface, as established by the boundary renormalisation principle. Everything within this boundary is part of the Earth system. Everything outside is the sink.

Definition 2. The thermodynamic maintenance stock, Σ, is the total maintenance-equivalent operations embodied in all objects within the system boundary that are maintained by civilisational power throughput:

where the sum runs over all object types within the boundary whose maintenance is powered by civilisational energy flows, σ_i is the maintenance-equivalent operation count of type i, and n_i is its copy number. Any counting procedure satisfying A1–A4 yields an equivalent Σ. When Assembly Theory is used as the instantiating framework, σ_i corresponds to the assembly index a_i; the simplified linear form is used because the argument requires only the total accumulated maintenance-equivalent operations, the stock of thermodynamic work held against decay, not the selection-theoretic content captured by more elaborate formulations. Σ has units of assembly-steps.

Definition 3. An object is "maintained by civilisational power throughput" if and only if its persistence depends on exergy flows routed through the technosphere. Equivalently: if civilisation's power throughput were reduced to zero, would the object disassemble on a timescale shorter than its geological relaxation time? If yes, it is part of Σ. If no, it is part of the geosphere or biosphere and is maintained by non-civilisational energy flows (solar-driven weathering cycles, biological metabolism independent of the technosphere, geothermal processes).

This third definition requires elaboration, because the system boundary contains more than one kind of assembled matter.

Partitioning the boundary

Within the top-of-atmosphere radiative interface, three categories of matter coexist.

The geosphere; rock, ocean, atmosphere in their abiotic configurations, has structure but not assembly in the sense of the thermodynamic maintenance stock. Its order is a product of gravitational, thermal, and chemical processes operating over geological time. It does not require civilisational power to persist. Axiom A4 excludes it: the counterfactual test is trivially satisfied, the geosphere would not disassemble if civilisational power throughput ceased. It is not part of Σ. (Sharma et al. (2023) provide independent empirical confirmation of this partition: abiotic chemical processes do not produce molecules with assembly index exceeding approximately 15, establishing an experimentally grounded threshold between geospheric structure and the assembly products of selection.)

The biosphere; every living organism and the ecological structures they collectively maintain, has enormous assembly. A single bacterium has an assembly index in the thousands. A forest ecosystem embodies billions of high-assembly objects. But the biosphere is maintained by solar-driven photosynthesis and trophic energy flows that do not pass through the technosphere. It would persist (and in most configurations, flourish) if civilisation ceased to exist. It is not part of Σ.

The technosphere; every road, building, vehicle, semiconductor, cable, turbine, ship, hospital, server, and piece of tooling that civilisation has constructed, plus the agricultural systems, managed forests, domesticated organisms, and engineered landscapes that depend on civilisational energy for their persistence, is the domain maintained by civilisational power throughput. Without that throughput, every component of the technosphere begins to decay: steel corrodes, concrete spalls, circuits degrade, roads crack, cultivated land reverts. The rate of decay varies, but the direction is universal. This is Σ.

The partition is not perfectly clean. Agriculture reshapes biological systems using civilisational energy (tractors, fertiliser, irrigation). Domesticated animals are biological organisms whose current forms depend on selective breeding sustained by technospheric infrastructure. Cities create microclimates that alter local biosphere dynamics. The boundary between technosphere and biosphere is fuzzy at the margins. But it is sharp at the core: a semiconductor fabrication plant is unambiguously part of Σ; an unmanaged coral reef is unambiguously not. For the purposes of this essay, marginal cases do not affect the argument. The maintenance cost scales with Σ, and the dominant components of Σ; the built environment, industrial plant, transportation networks, energy infrastructure, digital systems, are unambiguously technospheric.

This partition criterion does the essential work. It asks a single, operationally testable question: what maintains this object? If the answer is civilisational power throughput, it is part of Σ and its maintenance cost enters the thermodynamic balance. If the answer is anything else geological processes, biological metabolism, solar-driven cycles, it is outside Σ and does not contribute to the maintenance burden that ultimately drives the waste heat constraint.

Relative magnitudes

The partition criterion establishes what belongs to Σ and what does not. It does not establish how much of each category exists. An order-of-magnitude comparison reveals a quantitative relationship between the biosphere and the technosphere that has consequences for the analysis that follows.

The biosphere's assembly stock can be estimated from first principles. Bar-On, Phillips and Milo (2018) measured total living biomass at approximately 550 Gt C, dominated by plants (~450 Gt C), corresponding to roughly 10^18 kg of wet biomass. Biological matter is high-assembly: a functioning cell is a product of ~ 10⁴–10^5 coordinated assembly operations; membrane construction, protein folding, metabolic network assembly, genome replication machinery. A multicellular organism compounds these across differentiated tissues. A forest ecosystem compounds them across trophic levels and symbiotic networks. At even a conservative average of 10^3 assembly-steps per functional molecular unit, applied across the roughly 10^22 functional units per kilogram of biomass, the total biospheric assembly stock falls in the range Σ_bio ~ 10^38 – 10^40 assembly-steps. The precision of these estimates is unimportant. What matters is the order of magnitude, and the order of magnitude is robust: any plausible refinement of the individual parameters shifts the total by factors, not by the orders of magnitude required to challenge the inequality that follows.

The technosphere's assembly stock can be estimated by two independent routes. The first is mass-based. Elhacham et al. (2020) reported total human-made mass at approximately 30 Tt (3 × 10^13 kg) as of 2020, roughly equal to total living biomass by weight, but radically different in assembly content. The vast majority by mass is concrete, aggregates, brick, and asphalt, materials with assembly indices in the range aᵢ ~ 10–10^2. High-assembly components (semiconductors at aᵢ ~ 10⁵–10^6; pharmaceuticals at ~ 10³–10^4) constitute a negligible fraction of total technospheric mass. The mass-weighted average assembly index is perhaps aᵢ ~ 10²–10^3. The resulting estimate: Σ_tech ~ 10³⁴–10^37 assembly-steps. The second route provides a consistency check through Garrett's empirical coupling. Global power throughput of approximately 20 TW sustains an assembly stock at composite coupling Γ_tech ≈ P/Σ_tech. With Σ_tech ~ 10³⁵ assembly-steps, Γ_tech ~ 10⁻ ²³ W per assembly-step, a plausible order of magnitude for the power cost of maintaining one joining operation against entropic decay. The two routes converge on the same range: Σ_tech ~ 10³⁴–10^37 assembly-steps. Wide, but the upper bound remains well below the biosphere's lower bound.

Σ_bio exceeds Σ_tech by conservatively three to six orders of magnitude. The biosphere is the planet's dominant assembly stock. The technosphere is a recent, thin layer of structure built atop, and partially at the expense of the biosphere's far deeper stock.

The power throughput comparison is equally striking. The biosphere maintains its vastly larger assembly stock on approximately 130 TW of photosynthetic capture (Kleidon, 2012). The technosphere maintains its far smaller stock on approximately 20 TW. The implied composite couplings differ by orders of magnitude: Γ_bio ll Γ_tech. The biosphere maintains vastly more assembly per watt than the technosphere does. This is not surprising: 3.5 billion years of selection have optimised Γ_bio far beyond anything the technosphere has achieved in its few millennia. But the quantitative gap is striking. These estimates are order-of-magnitude, and deliberately so. The argument that follows depends on the inequality Σ_bio ≫ Σ_tech and on the ratio Γ_tech/Γ_bio ≫ 1. Both inequalities are robust across any plausible refinement of the individual estimates. The underlying facts, biology is more massive and more assembly-efficient than technology, are uncontroversial. This disparity in coupling, and the competitive interaction between the two stocks that share a common planetary surface, has consequences for the trajectory and viability analyses that follow.

The maintenance requirement

The Second Law guarantees that every high-assembly-index structure is thermodynamically unstable relative to its environment. Left without maintenance, steel oxidises, concrete carbonates, polymers degrade, circuits fail, roads crack, roofs leak, code rots. The rate varies, a Roman aqueduct decays more slowly than a smartphone, but the direction is universal. Every component of the technosphere is losing assembly at all times, at a rate determined by its material properties, its environment, and its exposure to the elements.

This means Σ has a non-negotiable maintenance cost. To sustain a given assembly stock against entropic decay, civilisation must continuously invest exergy in repair, replacement, and upkeep. The aggregate rate of decay across the entire technospheric ensemble is characterised by a single parameter, δ the aggregate decay rate, measured in yr⁻¹: the fraction of the assembly stock that disassembles per unit time. For the US economy, where depreciation data are most complete, δ is approximately 3–5% per year, meaning the technosphere loses 3–5% of its assembly annually and must replace it just to hold Σ constant.

The minimum power required to counteract entropic decay across the full assembly stock is therefore:

where δ is the aggregate decay rate (yr⁻¹), ξ is the specific exergy cost per assembly-step, the minimum exergy required to execute one joining operation against the local entropy gradient (Joules / assembly-step), η_II is the aggregate Second Law efficiency of civilisation's conversion processes, and Σ is the total assembly stock.

The parameter ξ varies by material and process but is always strictly positive. This positivity is not a contingent fact. The Gouy-Stodola theorem guarantees that every real thermodynamic process dissipates exergy in proportion to the entropy it generates; achieving ξ = 0 would require reversible maintenance at infinite slowness, which is no maintenance at all. The positivity of ξ is doing real work in the argument: it ensures that the maintenance floor is genuinely positive, not merely non-negative.

The inclusion of η_II in the denominator captures the gap between the theoretical minimum and what civilisation actually achieves. Current aggregate Second Law efficiency is perhaps 10–15% (Cullen and Allwood, 2010). Real maintenance therefore costs substantially more than the thermodynamic minimum. The floor exists regardless.

The inequality, not equality, is important. It states that there exists an irreducible minimum maintenance cost, a floor below which the assembly stock cannot be sustained. Real maintenance exceeds this floor because real processes are thermodynamically irreversible (they waste exergy), because maintenance is often performed imperfectly, and because the coordination costs of maintenance across a complex system add their own overhead. But the floor exists, and it is set by the Second Law. It cannot be eliminated by technology, policy, or institutional design. It can only be approached.

The existence of the floor is all that the waste heat ceiling analysis (Section 2.3) requires. If maintenance power is bounded below by a quantity that scales with Σ, and if that power is ultimately dissipated as waste heat from a finite radiating surface, then there exists a maximum sustainable Σ_max. The rest is arithmetic.

No legislature can repeal δ. No central bank can defer it. No restructuring agreement can reduce ξ to zero. Concrete carbonates at the rate concrete carbonates. Steel oxidises at the rate steel oxidises. The Second Law does not read the Financial Times.

This is what economic "depreciation" actually is. Not a financial convention. Not an accounting rule. It is the Second Law of Thermodynamics operating on the physical substrate of civilisation, continuously, universally, and without appeal.

The empirical weight of maintenance

The abstract claim that maintenance is non-negotiable can be given concrete, measurable content. The data are extensive, consistent, and sobering.

National accounts. The US Bureau of Economic Analysis reports that Consumption of Fixed Capital, the national accounts measure of depreciation, was $4.14 trillion in 2023, approximately 15.4% of GDP. This is a monetary shadow of δΣ: the fraction of the economy's physical stock that must be replaced each year simply to prevent net disassembly. The figure has been rising in relative terms for decades, from roughly 10% of GDP in 1960 to 15% in 2023, consistent with a technosphere shifting toward higher-turnover components (electronics, software, short-lived consumer goods).

Infrastructure condition. The American Society of Civil Engineers graded US infrastructure C- in its 2021 Report Card, estimating an investment gap of $2.59 trillion over ten years to bring systems to adequate condition. This gap is the physical manifestation of deferred maintenance: assembly that has been lost to decay and not yet restored. Deferred maintenance does not defer decay. It accumulates a debt payable in assembly-steps, compounding at rate δ.

The maintenance share is rising. In 2023, US governments spent $625.8 billion on transportation and water infrastructure, of which 56.7% ($355 billion) went to operations and maintenance of existing systems. Only 43.3% went to new capacity. For every dollar spent building something new, $1.31 was spent preventing something old from falling apart. This ratio increases as Σ grows, because the maintenance burden scales with the stock while the capacity for new construction is constrained by available surplus above maintenance.

Embodied energy. The maintenance burden has a physical dimension that financial accounts only approximate. Data from the University of Bath's Inventory of Carbon and Energy (Hammond and Jones, 2011) and the IEA's tracking of global energy flows suggest that roughly 30–50% of global primary energy consumption is devoted to maintaining existing infrastructure and capital stock, replacing worn components, heating and cooling buildings, resurfacing roads, running maintenance operations across the industrial base. The remainder goes to new construction, transport, services, and the metabolic needs of the population. The maintenance fraction is expected to rise as the global technosphere matures and as emerging economies build stock that then requires upkeep.

These are not edge cases or anomalies. They are the central tendency of any system that accumulates complex structure over time. The maintenance fraction rises because δΣ grows with Σ, and the surplus available for new construction (C(t) - δΣ) shrinks as a share of total effort. This is not a failure of management or policy. It is a thermodynamic identity.

Tainter and the diminishing returns to complexity

Joseph Tainter, in The Collapse of Complex Societies (1988), generalised the maintenance problem into a theory of civilisational dynamics that maps directly onto the assembly framework developed here. Tainter's core insight, arrived at through historical analysis of the Roman Empire, the Maya, and the Chaco culture, among others is that civilisations face a universal trajectory of diminishing marginal returns to increasing complexity.

Tainter's argument proceeds in three steps. First, societies are "problem-solving organisations" that respond to challenges by adding institutional and physical complexity. A drought is met with an irrigation system. A military threat is met with a standing army. An information bottleneck is met with a bureaucracy. Each response adds structure, adds assembly, in the terms of this essay.

Second, because the easiest and most rewarding solutions are adopted first, each subsequent increment of complexity yields proportionally less benefit. The first road connects two cities and enables trade that previously was impossible. The hundredth road connects two suburbs and saves commuters five minutes. The marginal return on complexity investment declines monotonically, even as the marginal maintenance cost remains constant or increases.

Third, and this is where Tainter's historical analysis converges with the thermodynamic framework, there comes a point at which the maintenance cost of accumulated complexity exceeds the surplus available to sustain it. At this point, further investment in complexity yields negative net returns. The civilisation has overshot its carrying capacity in complexity-space. What follows is simplification, voluntary if the society can manage a controlled descent, involuntary and catastrophic if it cannot. The Roman Empire did not fall to barbarians. It fell to the maintenance costs of its own infrastructure, military, bureaucracy, and territorial extent, and the barbarians simply occupied the space that opened up when the maintenance could no longer be sustained.

The convergence with the framework of this essay is exact. Tainter's "diminishing returns to complexity" is the observation that ∂(service delivery)/∂Σ decreases while ∂(maintenance cost)/∂Σ = δξ/η_II remains constant. At the crossover point, the civilisation is investing more in maintaining its existing stock than it is gaining from the services that stock provides. Tainter documented this crossover in at least a dozen historical cases and argued it is a universal feature of complex societies.

The convergence is not metaphorical. Tainter is describing, in the language of historical sociology, the same phenomenon that the Second Law describes in the language of physics: the inescapable overhead of maintaining ordered structure in a universe that favours disorder. The vocabulary differs. The mathematics is identical.

The dynamics of ΣThe assembly stock evolves according to a differential equation whose simplicity belies the severity of its consequences:

where C(t) is the construction rate, the rate at which civilisation performs new joining operations, adding assembly to the technosphere (assembly-steps / yr), and δΣ is the rate of entropic decay, the continuous disassembly imposed by the Second Law.

This is the thermodynamic skeleton of economic growth, stated without any economic assumptions. Growth is the net accumulation of assembly. Recession is the net loss. Stagnation is the exact balance between construction and decay. Every macroeconomic debate about growth, investment, and depreciation has this equation at its physical foundation, whether the participants acknowledge it or not.

Four consequences follow directly from Equation (5).

First: growth creates its own overhead. Every unit of new assembly added to Σ increases the future maintenance burden by δ per unit time. The larger the stock, the more power must be devoted to maintaining it, leaving less surplus for further construction. Growth is not self-sustaining. It is self-burdening.

Second: there is no pause button. A civilisation cannot suspend its assembly stock in some low-power standby mode while it considers its options. The decay term δΣ runs continuously. If C(t) drops to zero, if all construction ceases, the assembly stock decays exponentially:

The system cannot be frozen. It is either being maintained or it is falling apart. There is no intermediate state.

Third: the maintenance power is bounded below by the stock. Since every unit of Σ decays at rate δ, and restoring each assembly-step costs at least ξ joules at efficiency η_II, the minimum power to hold Σ constant isP_maint ≥ (δξ/η_II) · Σ. The relationship holds regardless of the composition of Σ, whether it is embodied in Roman aqueducts or semiconductor fabrication plants, because δ and ξ are aggregate properties of the technospheric ensemble, not features of any individual component.

Fourth: Σ is, in principle, measurable. Unlike "wealth" or "GDP," Σ is defined in physical units, assembly-steps, or equivalently, the total depth of joining operations embodied in the technosphere. Its measurement is a formidable practical challenge (it would require a comprehensive inventory of the global technosphere at component level), but it is a well-posed physical quantity. The empirical bridge is provided by Garrett's finding that total power consumption is proportional to accumulated economic wealth (P ≈ λ W), which is itself a monetary proxy for assembly (W ≈ κΣ). The physical relation P = Γ(t) · Σ, derived in the maintenance analysis (Section 2.2), subsumes the empirical observation, with λ = Γ(t)/κ absorbing the monetary conversion. This bridge between the physics developed here and the empirical data available from national accounts is examined in detail in the maintenance derivation that follows.

The aggregate δ of roughly 1–4% per year conceals a spectrum. Different components of Σ decay at vastly different rates, spanning nearly five orders of magnitude.

Roman aqueducts still carry water after two millennia, implying δ ~ 10⁻⁴yr⁻¹ for well-constructed masonry in a benign environment. Heavy civil infrastructure (bridges, dams, tunnels) has δ ~ 0.005 - 0.01yr⁻¹. Residential and commercial buildings:δ ~ 0.01 - 0.02yr⁻¹. Industrial machinery:δ ~ 0.05 - 0.10yr⁻¹. Vehicles:δ ~ 0.10 - 0.15yr⁻¹. Consumer electronics:δ ~ 0.20 - 0.33yr⁻¹ (a smartphone is functionally obsolete within 3–5 years). Software: effectively continuous, requiring constant updating against security vulnerabilities, dependency changes, and platform evolution.

This spectrum has a consequence that strengthens the maintenance argument rather than weakening it. As civilisation shifts toward higher-turnover components, a shift that has been accelerating for a century, the aggregate δ rises. The technosphere's average half-life is shortening. Each generation of technology requires more frequent replacement. The maintenance burden per unit of Σ is increasing, not decreasing, even as the total Σ grows. The aggregate δ is not a fixed parameter; it is a composition-weighted average that shifts upward as the technospheric mix tilts toward silicon, software, and short-lived consumer goods.

The implication is counterintuitive. A civilisation that shifts from concrete and steel to silicon and software does not become "lighter" in any thermodynamic sense. It becomes faster, its metabolic rate increases, its maintenance cycle accelerates, and its power requirement per unit of maintained assembly rises. The "dematerialisation" that some economists celebrate as evidence of decoupling from physical constraints is, in thermodynamic terms, an increase in δ, exactly the opposite of what would be needed to relax the waste heat ceiling. The possibility of deliberately engineering δ downward, choosing durable infrastructure over disposable, is examined in the maintenance derivation and the viability analysis (Section 4.2).

What this section has established

The argument is now complete for Part 2's first pillar. Four results have been stated, and none of them depend on economics, finance, or any assumption about human behaviour.

First, civilisation is a dissipative structure (Prigogine, 1977) whose physical substrate is assembled matter; infrastructure, networks, machines, institutions, organisms, maintained against the Second Law by continuous exergy throughput.

Second, the thermodynamic maintenance stock Σ is derived from four axioms (A1–A4) that follow from the First Law, the Gouy-Stodola theorem, and the Second Law. It is partitioned from the biosphere and geosphere by the operational criterion of A4: would it disassemble without civilisational power input? Sharma et al. (2023) provide one specific empirical instantiation; Garrett et al. (2011) provides a second, monetary shadow. The framework depends on the axioms, not on either.

Third, Σ decays at rate δ imposed by the Second Law, and each assembly-step costs a strictly positive exergy ξ to restore at efficiency η_II. The maintenance power is therefore bounded below:P_maint ≥ (δξ/η_II) · Σ.

Fourth, the dynamics of Σ are governed by d Σ/dt = C(t) - δΣ. Growth creates overhead. There is no pause button. Collapse occurs when the maintenance budget falls below δΣ.

This is the floor. The thermodynamic tax on civilisation's existence. It rises in absolute terms as Σ grows, consuming an ever-larger fraction of total power, and it is this power, ultimately dissipated as waste heat, that brings civilisation to the ceiling.

The question that follows immediately is quantitative. How much power does civilisation actually require per unit of Σ? The maintenance floor(δξ/η_II) · Σ is a lower bound. Real power consumption exceeds it by a factor μ, the metabolic multiplier that captures everything civilisation does beyond bare maintenance: construction, transport, computation, governance, defence, culture. The maintenance derivation (Section 2.2) establishes the full coupling P = Γ(t) · Σ where Γ(t) = μδξ/η_II, and examines how the individual components of Γ sit at different tiers of the ontological hierarchy, with consequences for what can and what cannot be changed.

The question that follows after that is terminal. If Σ requires continuous power, and that power is ultimately dissipated as waste heat on a finite planetary surface, then what is the maximum sustainable Σ_max? The waste heat ceiling analysis answers this question. The answer is finite, calculable, and closer than it appears.

2.2 The Maintenance Requirement

The previous section established the thermodynamic skeleton. Civilisation is assembled matter, a dissipative structure whose physical substrate is the total assembly stock Σ, partitioned from the biosphere and geosphere by a single operational criterion: if the object would disassemble without civilisational power input, it belongs to Σ. The stock decays at an aggregate rate δ, imposed by the Second Law and resisted only by continuous expenditure of exergy. The dynamics are governed by the assembly balance:

The quantity δΣ is a disassembly rate, assembly-steps lost per unit time. It is not yet a power. Assembly-steps per year and joules per second are dimensionally distinct. To convert the maintenance obligation into a power requirement demands two further quantities: the exergy price of each reassembly operation, and the thermodynamic efficiency with which that exergy is delivered. This section derives that conversion, producing a lower bound on civilisational power consumption that scales with Σ and depends on no economic quantity whatsoever.

The exergy cost of reassembly

Every maintenance operation; re-curing concrete, replacing a corroded beam, re-etching a degraded circuit, resurfacing a road, involves performing thermodynamic work against the direction favoured by the Second Law. The minimum exergy required for each such operation is set by the free energy difference between the degraded configuration and the functional one: the bond energies that must be re-established, the activation barriers that must be overcome, the transport work required to deliver materials to the site of repair. This quantity is ξ the specific exergy cost per assembly-step of maintenance, measured in joules per assembly-step restored.

The value of ξ varies across the ensemble of technospheric structures. Replacing a degraded weld on a steel bridge demands a different exergy expenditure per assembly-step than re-depositing a thin-film layer on a photovoltaic cell. But two properties of ξ are universal.

First, ξ is strictly positive. There is no zero-cost maintenance in a universe governed by the Second Law: restoring order from disorder requires work, and work requires exergy. This positivity is not an empirical observation that might be overturned by future data. It is a consequence of the Gouy-Stodola theorem (Gouy, 1889; Stodola, 1898), which establishes that every irreversible process destroys exergy in proportion to the entropy it produces. Since decay is irreversible, its reversal cannot be free.

Second, ξ has a positive lower bound, the minimum reversible work of assembly, analogous to the Landauer limit for computation (Landauer, 1961). No process, however cleverly engineered, can perform maintenance below this floor. In practice, real maintenance operations exceed it by orders of magnitude, but the floor's existence is guaranteed by thermodynamics.

With the disassembly rate δΣ (assembly-steps per year) and the exergy cost per step ξ (joules per assembly-step), the minimum exergy expenditure rate for maintenance is δξΣ (joules per year). But real processes do not deliver exergy at perfect thermodynamic efficiency. The aggregate Second Law efficiency η_II, the ratio of the minimum thermodynamic work required for a given task to the actual energy consumed in performing it, is always strictly less than unity. For the aggregate civilisational conversion chain, from primary energy extraction through every intermediate step to final end-use, Ayres and Warr (2009) reconstructed century-long exergy accounts for the United States, finding that aggregate η_II rose from roughly 2.5% in 1900 to approximately 11% by the 1970s, where it has since stagnated. Brockway et al. (2014) confirmed the plateau at around 11% for the US and documented a rise from 9% to 15% in the UK over 1960–2010. The global aggregate is estimated at roughly 10–12%.

The critical structural point: η_II < 1 always. Every real conversion chain wastes exergy. The gap between the theoretical minimum and the actual consumption is not a market failure or an engineering shortcoming that might be closed by incremental improvement. It is a consequence of operating finite-rate processes in a world where the Carnot limit is never reached.

Dividing the exergy expenditure rate by the efficiency yields the maintenance power floor:

This is a lower bound, not an equality. It represents the power cost of performing exactly the maintenance work required to hold Σ constant, assuming no other civilisational activity whatsoever. No real civilisation operates at this limit. The inequality states only that the maintenance power cannot fall below it.

Beyond maintenance: the full metabolic cost

Real civilisation does more than maintain. It constructs new assembly, transports materials, processes information, delivers services, governs, defends, and grows. The total power consumed exceeds the bare maintenance floor by a factor that reflects the scope of non-maintenance activity.

The metabolic multiplier captures this ratio: total civilisational power to the subset devoted to maintenance. A civilisation performing nothing but upkeep; no new construction, no services, no growth, has μ = 1. In practice, μ exceeds unity because the system also builds, transports, computes, governs, and wages war. The physical definition is μ = P / P~maint~, where P~maint~ includes all power expenditure whose removal would cause net disassembly of the existing stock: not only the replacement of worn components (captured in national accounts as depreciation) but also repair, infrastructure operations, the energy overhead of sustaining supply chains, and the service activity required to keep assembled systems functional. Under this physical accounting, maintenance-equivalent activity represents roughly 40–60% of total economic throughput, placing μ in the range of approximately 1.7–2.5. A common but misleading proxy is the national-accounts depreciation share: US Consumption of Fixed Capital was approximately 15% of GDP in 2023 (Bureau of Economic Analysis, 2023), which would suggest μ ≈ 6–7 if depreciation were the whole of maintenance. It is not. Depreciation captures only the formal replacement of capital goods; it excludes repair expenditure, operational upkeep, energy costs of sustaining existing systems, and the large service sector whose function is to keep assembled infrastructure operational. The depreciation-share estimate is better understood as a lower bound on the maintenance fraction, and therefore an upper bound on μ, not as the primary calibration. The value of μ is not fixed. It varies with institutional structure, competitive pressure, and the evolutionary dynamics examined in the trajectory analysis of Part 3. But μ is always at least 1, because a civilisation that cannot even maintain its stock is in terminal decline.

With these two quantities, the total power consumption of civilisation is:

and the lower bound, the irreducible minimum, achievable only by a civilisation performing maintenance alone, is equation (2) above.

The composite coupling

The four factors in the numerator and denominator of equation (3) are independently motivated, dimensionally consistent, and physically transparent. Each has its own ontological status and its own susceptibility to change, a point that will matter when the analysis turns to intervention in the implications. Where notational economy requires a shorthand in later sections, the composite is written as:

so that the power-assembly coupling takes the compact form P = Γ(t) · Σ. The explicit time-dependence is a permanent reminder that Γ is a variable product of four time-varying quantities, not a fixed parameter.

Each constituent sits at a different tier of the ontological hierarchy, and the compliance of each, the degree to which civilisational action can shift its value, varies accordingly.

δ and ξ have fixed Tier 1 floors: the Second Law guarantees δ > 0 (assembled matter decays) and ξ > 0 (reversing decay costs exergy). These floors cannot be moved by any technology consistent with thermodynamics. Above the floor, both parameters are stiff engineering quantities, materials science can reduce δ through durability design and lower ξ through more efficient repair processes, but competitive dynamics and Jevons recycling resist sustained reductions at civilisational scale.

η_II has a fixed Tier 1 ceiling: η_II < 1 always, by the Second Law. The aggregate value is stiff, governed by Tier 2 evolutionary dynamics — competitive adoption selects for the conversion efficiency that maximises power throughput, and the Jevons mechanism recycles efficiency gains into expanded demand. Historical improvements in η_II are real but have been absorbed by growth; the aggregate has plateaued near 11% for decades.

μ is partially compliant. Its value is shaped by Tier 3 institutional incentives; interest rates, governance horizons, military expenditure, discretionary consumption, all of which are reformable in principle. But it is also shaped by Tier 2 competitive dynamics: agents that invest surplus in growth outcompete those that do not, selecting for μ values well above the maintenance-only floor of unity. The Tier 3 component is compliant; the Tier 2 component is stiff.

The four components of Γ, together with four parameters in the ceiling numerator: the effective emissivity ε, radiating area A, the habitability threshold Thab, and the planetary albedo α, constitute the eight-parameter control surface governing the distance between the maintenance floor and the waste heat ceiling. Their compliance ranges from fully fixed (the Stefan–Boltzmann constant, the positivity of δ) through stiff (the evolutionary attractor governing the aggregate values) to partially compliant (the institutional parameters shaping μ). This compliance spectrum is the control-surface map of the civilisational predicament; the implications analysis returns to it in full.

The aggregate decay rate δ is not a single number but a stock-weighted mean across an ensemble of components with vastly different lifetimes. Reinforced concrete in a dam (δ ≈ 0.01yr^- 1, half-life approximately 70 years) and a silicon transistor in a consumer device (δ ≈ 0.2yr^- 1, half-life approximately 3.5 years) contribute to the same δΣ with weights proportional to their share of total assembly. The aggregate δ is therefore a function of the technosphere's material composition. A civilisation whose stock consists predominantly of long-lived infrastructure has a lower aggregate δ, and therefore a lower maintenance power floor, than one whose stock is dominated by short-lived electronics, software-dependent systems, and rapidly obsolescing consumer goods.

Decay rate engineering

This compositional dependence has a direct and underappreciated consequence: deliberately engineering for durability, choosing low-δ infrastructure over high-δ alternatives, reduces the composite coupling Γ(t), since δ enters the numerator of equation (4). A lower Γ means less power required per unit of assembly stock, which widens the viable corridor between the maintenance floor and the waste heat ceiling established in the following section. Durability engineering is a Tier 2 and Tier 3 intervention, a matter of materials choice, design standards, and institutional incentives, with direct Tier 1 consequences, because it changes the rate at which the Second Law claims its tax. The historical trend, however, runs in the opposite direction. As the technosphere shifts from concrete, steel, and masonry toward silicon, polymers, and software-defined systems, the stock-weighted δ is rising. Aggregate half-lives are shortening. Each generation of technology demands more frequent replacement, increasing the maintenance burden per unit of Σ even as total Σ grows. The civilisational metabolism is accelerating, precisely the wrong direction for a system approaching a thermal ceiling.

What this section has established

The result claims that civilisational power consumption scales with the assembly stock through the composite Γ(t). It does not claim that the composite is constant. The specific value of Γ(t) depends on the technological base (which determines ξ and η_II), the material composition of the stock (which determines δ), and the institutional and evolutionary dynamics that set μ, all of which vary with time, geography, and the stage of civilisational development. Whether the composite is empirically stable, and why, is a trajectory question properly addressed in Part 3, where the data and the evolutionary mechanisms that govern these parameters are examined.

What the derivation does guarantee, by the Second Law alone, is that Γ(t) is positive and bounded below. Since δ > 0 (assembled matter decays), ξ > 0 (reversing decay costs exergy), η_II < 1 (no real process is reversible), and μ ≥ 1 (civilisation does at least maintenance), there exists an irreducible minimum power cost per unit of assembly stock. This minimum cannot be engineered to zero. It cannot be legislated away. It cannot be avoided by switching energy sources, changing economic systems, or adopting any technology consistent with the laws of thermodynamics.

This is all the ceiling argument requires. The waste heat analysis that follows establishes that the power sustaining Σ is ultimately dissipated as waste heat from a finite planetary surface, creating a maximum sustainable assembly stockΣ_\max. That calculation needs only the existence and positivity of the power-assembly coupling, not its constancy, not its precise numerical value, and not any economic quantity. The floor established here, combined with the ceiling established next, defines the thermodynamic corridor within which any planetary civilisation must operate. The corridor's width depends on the numerical value of Γ(t); its existence depends only on the Second Law.

2.3 The Waste Heat Ceiling

The preceding sections established the two foundational results of Part 2. The assembly framework defines civilisation as a stock of ordered structure Σ maintained against entropic decay by continuous exergy throughput, with the balance equation

governing its evolution: construction C(t) adds assembly; decay δΣ subtracts it. The maintenance derivation established that the minimum power required merely to prevent net disassembly, to hold Σ constant, satisfies the inequality

where δ is the aggregate decay rate, ξ is the specific exergy cost per unit of assembly maintenance, and η_II is the aggregate Second Law efficiency of the conversion chain. The composite δξ/η_II is strictly positive, there is no zero-cost maintenance in a universe governed by the Second Law, and it bounds the minimum power demand from below regardless of the energy source, the conversion technology, or the institutional architecture that organises the maintenance activity. Actual power consumption exceeds this floor because civilisation does more than maintain: it constructs, transports, computes, defends, and grows. The maintenance derivation captured this through the metabolic multiplier μ ≥ 1, giving the total power relation

where Γ(t) = μδξ/η_II is a time-varying composite whose value depends on technology, material choices, and the evolutionary dynamics examined in Part 3. The ceiling argument that follows requires only one property of this composite: that it is bounded below by a strictly positive quantity, δξ/η_II > 0. That bound is guaranteed by the Second Law.

This section completes the thermodynamic picture by asking a question that follows directly from the maintenance cost: where does the power go?

The answer is dictated by the Second Law. Every watt of power that civilisation consumes, from any source whatsoever, is ultimately dissipated as waste heat. This is not a statement about inefficiency. It is not a commentary on fossil fuels or atmospheric chemistry. It is a consequence of the First and Second Laws operating together: energy is conserved, but exergy is destroyed in every real process, and the destroyed exergy manifests as heat at ambient temperature. A nuclear reactor produces heat. A wind turbine converts kinetic energy to electricity that is subsequently degraded to heat through resistance, friction, computation, and every other end use. A fusion reactor, should one operate commercially, produces heat. This waste heat must be radiated to the 2.7 K void of deep space as infrared photons, at a rate governed by geometry and temperature. A perfectly efficient civilisation, one that extracted the theoretical maximum of useful work from every joule of primary energy, would still reject the Carnot fraction as waste heat, and the useful work itself would eventually degrade to heat through the final services it performs.

The waste heat must go somewhere. On a planet, "somewhere" means the top-of-atmosphere radiative interface established in the system boundaries discussion as the system boundary. This is where the maintenance cost of civilisation meets the radiative capacity of a finite surface, and the meeting defines an absolute ceiling on the assembly stock that any planetary civilisation can sustain.

The planetary energy balance

At steady state, the energy entering the Earth system must equal the energy leaving it. The outgoing flux is governed by the Stefan-Boltzmann law:

where ε is the effective emissivity of the Earth system (accounting for atmospheric absorption and re-emission; ε ≈ 0.62 for the present atmosphere), σ is the Stefan-Boltzmann constant (5.67 × 10⁻⁸ W/m² K^- 4), A is the planetary radiating area (≈ 5.1 × 10¹4 m²), and Tₑq is the effective equilibrium temperature.

The incoming flux consists of absorbed solar radiation P_☉ ≈ 1.2 × 10¹⁷W (approximately 120,000 TW), plus any non-solar power dissipated by civilisation, P. The equilibrium condition is:

If P = 0, the system radiates only the solar input and the equilibrium temperature settles at the value consistent with the current atmospheric greenhouse. If P > 0, the equilibrium temperature must rise until the outgoing radiation matches the augmented total. The relationship is monotonic and absolute. There is no technology that changes it. There is no fuel that avoids it. There is no efficiency gain that eliminates it, because the waste heat is a consequence of power dissipation itself, not of any particular conversion pathway.

Solar energy and the thermal budget: three cases

The relationship between solar power and the waste heat ceiling requires more careful treatment than a blanket exemption. Three cases arise, distinguished by the thermodynamic pathway through which the energy enters the planetary budget.

Case 1: Surface solar, matched albedo. A photovoltaic panel whose reflectivity matches the surface it replaces produces no net thermal addition. The intercepted photons were already destined for the planetary heat budget; the panel merely extracts useful work en route to their eventual dissipation as heat. This is the dominant case at current and near-future deployment scales, and it is the basis on which solar energy is correctly exempted from the waste heat ceiling. Nuclear fission, nuclear fusion, geothermal extraction beyond natural background rates, and any hypothetical energy source that liberates previously sequestered energy all represent net additions to the planetary thermal load. The constraint applies to every non-solar watt.

Case 2: Surface solar, albedo mismatch. A dark photovoltaic panel replacing a lighter surface; sand, snow, light soil, a high-albedo roof, absorbs more solar flux than the surface it replaced. The difference,

where α denotes albedo and f_covered the fractional area covered, is a net thermal addition. At current global PV deployment, the effect is small relative to total anthropogenic forcing. At deployment scales required to power a multi-hundred-TW civilisation, it could become locally significant, particularly in desert installations where the albedo difference between sand (α ≈ 0.35–0.40) and silicon panels (α ≈ 0.10–0.15) is large. The albedo effect does not invalidate solar energy's exemption from the waste heat ceiling; it narrows the exemption and introduces a deployment-geometry constraint that pure conversion analysis misses.

Case 3: Beamed solar from off-planet collection. Energy collected by orbital solar arrays beyond the Earth-intercepted flux and transmitted to the surface as microwave or laser constitutes a net addition to the planetary thermal budget, identical in its thermodynamic effect to any non-solar source. This case is analysed in the Fermi discussion but must be stated here for completeness: beamed solar energy from space is subject to the waste heat ceiling in exactly the same way as fission, fusion, or geothermal.

The three-case partition strengthens the framework by eliminating an apparent loophole. Case 1 remains dominant at terrestrial scales. Case 2 is a deployment-design consideration. Case 3 becomes relevant only to civilisations pursuing the space-industrial bootstrap discussed in Part 5.

This is the source-independence that distinguishes the waste heat ceiling from the greenhouse problem. The greenhouse effect is a function of atmospheric composition, it can, in principle, be eliminated by changing which fuels are burned or by removing offending gases from the atmosphere. The waste heat ceiling is a function of total power dissipation. It cannot be eliminated by switching fuels. It cannot be reduced by cleaning emissions. It can only be respected by limiting the total non-solar power that civilisation dissipates on the planetary surface.

The ceiling can now be stated as a bound on the assembly stock.

From the maintenance derivation, the total power demand of civilisation satisfies P = Γ(t) · Σ, with the irreducible lower bound P ≥ (δξ/η_II) · Σ holding when μ = 1 the limiting case where civilisation does nothing beyond bare maintenance. This is the most generous possible assumption. Any activity beyond maintenance (new construction, transport, computation, biological sustenance) increases μ above unity, consuming additional power and tightening the constraint.

Theorem T1 (Finite thermodynamic corridor).

The habitability constraint requires that the equilibrium temperature remain below some critical threshold Tₕab, the temperature beyond which the biosphere collapses and sustained human existence becomes impossible. This threshold is not a single number; it depends on regional climate, agricultural capacity, and the wet-bulb temperature limit for mammalian thermoregulation. But it exists, and it is finite. For the purpose of bounding the problem, take Tₕab as the global mean temperature at which large-scale habitability failure begins, a value informed by the literature discussed below.

The maximum tolerable civilisational power is the P that pushes Tₑq to Tₕab:

This is the central result of Part 2. Note the inequality implicit in the expression: it gives an upper bound on Σ_max, evaluated at the minimum possible Γ (when μ = 1). At any realistic μ > 1, the effective ceiling is lower. The expression states that for any planetary civilisation, regardless of its energy source, its economic system, its institutional design, or its cultural values, the assembly stock cannot exceed the ratio of the planet's surplus radiative capacity to the minimum specific maintenance cost of the civilisational structure. The numerator is pure radiative physics: how much additional power the planet can radiate before overheating. The denominator is pure thermodynamics: the irreducible exergy cost per unit of assembly, set by decay, material properties, and conversion efficiency.

The bound is conservative in two distinct senses. First, it assumes μ = 1 that civilisation does nothing beyond maintenance. Any activity beyond maintenance consumes additional power, which means the effective ceiling on Σ at any given moment is lower than the theoretical maximum. A growing civilisation, one with C(t) > δΣ, which is to say, one that is building new structure, consumes power well above the maintenance floor and therefore hits the thermal constraint at a smaller assembly stock. Second, the derivation assumes that power scales linearly with assembly through the composite Γ(t). If the actual relationship is superlinear, if each additional unit of assembly costs more to maintain due to coordination overhead, network complexity, or supply-chain fragility, then the true Σ_max is lower than the bound. If there are economies of scale, the true Σ_max is higher, but still finite. The ceiling exists regardless of the functional form. Only its numerical value depends on it.

Several further features of this result deserve emphasis.

First, the ceiling formula is technology-independent in the sense that it applies to any energy source (provided it is non-solar); it does not privilege one conversion pathway over another. A civilisation of semiconductor fabs and a civilisation of steam engines face the same form of constraint. But the value of Σ_max depends on the composite coupling Γ = μδξ/η_II, which is itself partially a function of engineering choices. A civilisation that builds more durably (lower δ), maintains more efficiently (higher η_II), and operates with less overhead (lower μ) has a higher ceiling. The technology-independence is in the form of the constraint, not in its magnitude.

Second,Σ_max is finite and calculable. The quantities in the expression are all either known physical constants (σ), measurable planetary parameters (ε, A,P_☉), or empirically estimable system properties (δ, ξ, η_II, Tₕab). The ceiling is not a vague future concern. It is a number, an upper bound on the total ordered complexity that this planet can thermodynamically support.

Third, the decomposed denominator reveals what a civilisation can and cannot control. The decay rate δ is a material property: build with durable materials and δ falls; build to discard and δ rises. The specific exergy cost ξ depends on the physical processes required for maintenance. The Second Law efficiency η_II captures the quality of the conversion chain. A civilisation that builds to last, maintains with precision, and converts energy efficiently has a lower denominator and therefore a higher ceiling. But no combination of these improvements drives the denominator to zero. The positivity of δξ/η_II is guaranteed by the Second Law, and it is this positivity, not the specific numerical value, that makes the ceiling finite. The compliance spectrum introduced in the maintenance derivation classifies these parameters: δ and ξ are constrained to be strictly positive (fixed, Tier 1) but their magnitudes respond to engineering choices (stiff, Tier 2); η_II is bounded above by unity (fixed) but its current aggregate value sits far below that bound (stiff); μ is partially compliant (Tier 3) insofar as institutional and strategic choices determine the ratio of total to maintenance power. The ceiling is physics. The distance to it is partially a matter of engineering and governance.

Fourth, the numerator contains one partially controllable parameter: the effective emissivity ε. Because ε captures the net effect of the atmosphere on outgoing longwave radiation, it is sensitive to atmospheric composition. A thicker greenhouse blanket reduces ε, lowering the radiative capacity and tightening the ceiling. Conversely, an atmosphere stripped of greenhouse gases raises ε toward unity, increasing Σ_max. This interaction means that the greenhouse problem and the waste heat problem are not independent: unchecked greenhouse gas accumulation reduces ε and thereby lowers the waste heat ceiling before the waste heat itself becomes significant. The two constraints compound rather than substitute.

Fifth, the planetary albedo α enters the ceiling through the absorbed solar flux P_☉ = S(1 - α)π R^2, where S is the solar constant and R the planetary radius. Reducing albedo, darker surfaces from large-scale photovoltaic deployment, deforestation, ice loss, increases P_☉ and tightens the ceiling. Increasing albedo, stratospheric aerosol injection, marine cloud brightening, high-albedo surface materials, reduces P_☉ and loosens it. The distinguishing feature of α as a control parameter is speed. Stratospheric aerosol injection is deployable on timescales of years, faster than any other lever on the control surface. This makes albedo modification the strategic reserve: the intervention available when the other controls are too slow and time must be bought. Its limitations; continuous maintenance requirement, governance complexity, and the fact that it addresses only the thermal budget without reducing the maintenance obligation δΣ or the power coupling Γ, are real, but do not diminish its unique temporal advantage.

Sixth, the habitability threshold Tₕab enters the numerator with fourth-power sensitivity, giving it very high leverage over Σ_max. The biological wet-bulb floor, approximately 308 K for sustained human survival, is fixed physics and sets an absolute upper bound on Tₕab. But the effective civilisational threshold, the temperature at which critical infrastructure fails, agriculture becomes unviable, or coordinated industrial existence degrades, sits well below this biological limit and has real engineering operability. Raising the effective Tₕab through heat-adapted agriculture, thermal management of the built environment, and infrastructure resilience widens the thermal headroom available for civilisational activity. The gain, however, feeds back through Σ: climate-controlled buildings, hardened infrastructure, and engineered crop systems are themselves assembled structures that add to the stock and its maintenance burden. The net headroom gain after the induced Σ increase is the decision-relevant quantity, and it is smaller than the gross Tₕab shift alone would suggest.

The eight-parameter control surface

The ceiling formula Σ_max = ( εσ A T_hab^4 - P_☉ )/Γ thus contains eight parameters with real operability, partitioned cleanly between the numerator and the denominator. The ceiling-numerator group, ε, A, Tₕab, and α (the last entering through P_☉) determines the size of the thermal headroom available to the civilisation. The coupling-denominator group μ, δ, ξ, and η_II (composing Γ) determines the rate at which civilisational activity consumes that headroom. Both classes matter, and both contain parameters with meaningful engineering or institutional compliance. The numerator group includes one parameter (A) that can in principle move the ceiling upward without bound, and one (α) that can buy time faster than any other lever. The denominator group contains the coupling whose reduction is, as the viability analysis will establish, a necessary condition for any survivable strategy. The manifold analysis develops the complete control-surface characterisation; sensitivity, operability, and directionality for all eight parameters, and maps each to the seven strategic orientations available to any civilisation facing this ceiling.

Sensitivity of the ceiling to operable parameters

The following table illustrates the response of the ceiling power P_ceiling = εσ A_s( T_hab^4 - T_0^4 )and the relative maximum assembly stock Σ_max = P_ceiling / Γ to variation in the two parameters with the highest leverage: Tₕab (which enters with fourth-power sensitivity) and Γ (which enters linearly in the denominator).

Computed with:

T₀ = 288 K.

Two features are immediate. The fourth-power sensitivity to Tₕab means that doubling Δ T from 4 to 8 K roughly doubles P_ceiling. The linear sensitivity to Γ means that halving Γ doubles Σ_max. The table is a no-feedback Planck calculation; feedbacks would reduce the effective Tₕab achievable for any given scenario, tightening all entries.

Albedo engineering and the thermal budget

The partial controllability of the numerator extends beyond atmospheric composition. Deliberate albedo modification; stratospheric aerosol injection, marine cloud brightening, or space-based reflectors, could increase the effective outgoing radiation without reducing power throughput. In the framework of the energy balance equation, these interventions modify the thermal budget on the left-hand side, effectively raising Σ_max without expanding A. Stratospheric aerosol injection, the most studied of these approaches, operates by increasing the planetary albedo, reflecting a greater fraction of incoming solar radiation and thereby freeing radiative headroom for civilisational waste heat. Marine cloud brightening achieves a similar effect through cloud microphysics over ocean surfaces. Space-based reflectors at the L1 Lagrange point could in principle modulate the effective solar constant directly.

In the three-tier ontology, albedo engineering classifies as a partial Tier 1 lever: it modifies a parameter adjacent to the hard physics of radiative balance without violating any thermodynamic law. The limitations, however, are significant. Albedo engineering addresses only the thermal budget, it does nothing to reduce the maintenance obligation δΣ or the power coupling Γ. The interventions themselves require continuous maintenance, adding to δΣ and to the civilisational power demand. And they do not eliminate the waste heat constraint; they delay it. A civilisation that increases its albedo to buy radiative headroom and then fills that headroom with additional growth arrives at the same ceiling, merely later. Albedo engineering buys time. It does not solve the problem.

The current position

At present, global primary energy consumption stands at approximately 20 TW (2.0 × 10¹3 W), based on the Energy Institute's 2024 Statistical Review reporting approximately 620 EJ for 2023. Against a solar absorption budget of approximately 120,000 TW, this represents a fraction of roughly 1.7 × 10⁻⁴ about one part in six thousand. Globally averaged, anthropogenic heat flux amounts to approximately 0.04 W/m², which is roughly one per cent of the current greenhouse gas radiative forcing of approximately 2.9 W/m², and approximately 0.01% of mean absorbed solar irradiance at the surface.

By any global metric, waste heat is currently negligible. The assembly stock Σ is far below Σ_max. The ceiling is distant.

Two observations should discipline any resulting complacency.

The first is that waste heat is already significant locally. Peak grid-cell values in Flanner's (2009) global analysis of anthropogenic heat flux exceed 200 W/m² in central Tokyo, comparable to peak solar irradiance on a cloudy day. Direct urban-scale measurements confirm fluxes of 200–400 W/m² in the densest metropolitan cores (Ichinose et al., 1999; Sailor and Lu, 2004). Western Europe averages 0.68 W/m². These are not projections; they are measurements. Cities are already thermodynamic objects in their own right, dissipating at intensities that materially perturb their local radiative balance. The urban heat island effect is not merely an artefact of concrete and asphalt geometry; it is, in significant part, a consequence of concentrated power dissipation. The ceiling is global, but the approach to it is local and uneven. Some regions will encounter thermal constraints long before the planetary average becomes problematic.

Expressed as a global mean flux, the waste heat ceiling corresponds to the additional anthropogenic dissipation that raises Tₑq to Tₕab. For a habitability threshold of Tₕab ≈ 292 K, roughly 4 K above the current approximately 288 K baseline, the no-feedback (Planck) ceiling is approximately 12–13 W/m² of globally averaged anthropogenic heat flux, computed from the linearised Planck sensitivity 4εσ T₀³ ≈ 3.2 W/m² K⁻¹. This is the direct-heating value, consistent with the linearised formula used in the three-horizons table below. Climate feedbacks, water vapour amplification, ice-albedo response, would amplify the warming from any given flux, meaning the effective ceiling accounting for feedbacks binds at a lower flux, perhaps 5–7 W/m² depending on the equilibrium climate sensitivity. The no-feedback calculation is therefore the conservative (upper) bound on the ceiling flux; the true effective ceiling is tighter. Even the no-feedback value is roughly four to five times the current greenhouse gas radiative forcing of approximately 2.9 W/m². The exact value depends on Tₕab; the figure serves as an order-of-magnitude anchor for comparison with the climate forcing literature, not as a precise prediction.

Waste-heat flux derivation. Define the planetary surface area Aₛ = 4π R² ≈ 5.1 × 10^14 m² and the global-mean waste-heat flux F_w = P/Aₛ. The linearised no-feedback temperature response to added power P is Δ T ≈ ( T_0/4 ) · ( P/P_☉ ), where P_☉ ≈ 1.2 × 10¹⁷W is absorbed solar power and T₀ = 288 K. Equivalently, F_w = Δ T · 4εσ T₀³ ≈ 3.2 · Δ T W/m² (for ε ≈ 0.6, T₀ = 288 K). The conversion rule is: 1 W/m² (global mean over Aₛ) ≈ 510 TW. Feedbacks amplify; the effective climate sensitivity exceeds the Planck value, so the effective ceiling flux is lower than 3.2 · Δ T. The no-feedback calculation gives the conservative upper bound on the ceiling flux.

The second is that the distance to the ceiling is measured not in absolute watts but in doubling times. Global primary energy consumption has grown at approximately 2.3% per year over the last two centuries. At this rate, the doubling time is roughly 30 years. The number of doublings between the current 20 TW and the threshold at which waste heat becomes a first-order forcing is small. Exponential growth closes large gaps with deceptive speed.

The approach: three horizons

The following projections illustrate the thermal consequences of sustained exponential growth in total power dissipation. They assume a continuation of the historical 2.3% annual growth rate, not as a prediction, but as a baseline against which any deviation must be measured.

Horizon Power Fraction of P_☉Thermal consequence

~100 yr ~200 TW ~0.17% Avg. heat flux ~0.4 W/m². Flanner (2009): continental-scale warming of 0.4–0.9 K in GCM simulations. Secondary forcing, but non-negligible alongside greenhouse effect.

~250 yr ~6,200 TW ~5.2% Δ T ≈ 3.7 K from direct heating alone, independent of atmospheric composition. Combined with any residual greenhouse warming, equatorial regions approach sustained habitability limits.

~400 yr ~200,000 TW >100% of P_☉Civilisational power dissipation exceeds total absorbed solar flux. Tₑq ≈ 360 K (87°C). Water boils at low elevations within decades. Complex life is extinct. The planet is sterilised.

The linearised perturbation formula for the equilibrium temperature response to a fractional increase f in total planetary heat loading is:

For f ≈ 0.052 and Tₑq ≈ 288 K, this yields Δ T ≈ 3.7 K at the 250-year mark, from direct heating alone, independent of any greenhouse gases whatsoever. This is the temperature increase that would occur if civilisation had solved the carbon problem entirely, eliminated every molecule of excess CO₂, perfected atmospheric management, but continued to grow its energy consumption at the historical rate using perfectly clean sources. The warming is intrinsic to the power throughput itself.

It bears emphasis: these projections assume perfectly clean energy. No carbon emissions. No pollution. No environmental damage of any conventional kind. Just energy, consumed and dissipated, as the Second Law requires.

A necessary caveat on timescales. The projections above are not predictions. They are consequences of sustained exponential growth at the historical rate of approximately 2.3% per year, assuming current Γ. Both parameters are partially controllable, they are not physical constants. The 2.3% growth rate is a historical observation, a Tier 2/3 parameter shaped by competitive dynamics and institutional architecture, not a law of nature. Global population growth is already decelerating toward stabilisation. Per-capita energy demand could plateau. At 1% annual growth, the timescales in the table approximately double. At zero net growth, steady-state maintenance of the current Σ, waste heat remains at approximately 0.017% of solar flux indefinitely, and the ceiling never bites. The coupling Γ(t) = μδξ/η_II is likewise adjustable through engineering: more durable materials reduce δ; better conversion processes raise η_II; reduced overhead lowers μ. A factor-of-two reduction in Γ doubles Σ_max. The physics guarantees that the ceiling ∃ and is finite. The growth rate and the trajectory of Γ determine when, and whether, it is reached. The evolutionary dynamics examined in Part 3 explain why both parameters have historically resisted deliberate reduction, but resistance is not impossibility. The projections map what happens if nothing changes. They are a baseline, not a destiny. Abbot and Malani (2025) demonstrated this sensitivity formally in a coupled economic–physical model, showing that within plausible parameter ranges the binding time of the waste heat constraint varies from fewer than 100 years to beyond a millennium, depending on the coupling between GDP growth, total factor productivity, and power consumption. The essay's three-tier framework explains both why the historical parameters have been sticky, Tier 2 competitive dynamics select for growth-maximising trajectories, as the evolutionary analysis of Part 3 establishes, and why they are not immutable: Tier 3 institutional architecture shapes the selection landscape, as Hanley's pre-1970 divergence confirms. The physics guarantees the ceiling exists; the trajectory toward it is scenario-dependent.

Independent convergence

These are not speculative calculations by marginal theorists. The waste heat constraint has been derived independently by researchers across multiple disciplines, and the results are remarkably consistent.

Murphy (2022), professor of physics at UC San Diego, presented the core argument in Nature Physics. His conclusion: at 2.3% annual energy growth, waste heat renders Earth uninhabitable in roughly 400 years. At the same rate, total human energy consumption would equal the Sun's entire luminosity (3.8 × 10²6 W) in approximately 1,350 years, and the Milky Way's luminosity in approximately 2,500 years. These latter figures are not predictions. They are demonstrations of the absurdity of sustained exponential growth against physical law.

Balbi and Lingam (2025) provided the most rigorous treatment to date in their analysis published in Astrobiology, incorporating atmospheric feedbacks, biological temperature tolerances, and regional habitability thresholds that a simple Stefan-Boltzmann projection omits. Their finding: at just 1% annual energy growth, less than half the historical rate, loss of habitable conditions occurs within approximately 1,000 years from the onset of the exponential phase. A temperature increase of approximately 6 K drives global biodiversity collapse. At 12 K above baseline, large regions become uninhabitable for placental mammals, including humans. The Balbi-Lingam analysis is more conservative than Murphy's and for that reason more sobering: even at growth rates that most economists would consider stagnant, the ceiling arrives within a millennium.

Chaisson (2008) reached the same conclusion from a different direction entirely. Working within his energy rate density framework, which tracks the quantity Phi_m(watts per kilogram) across cosmic history, Chaisson warned that civilisation could encounter a thermodynamic growth limit of roughly 3 K of direct heating within 1,000 years, dictated solely by the Second Law. His framework provides additional context: the energy rate density of modern technological society (~50 W kg⁻¹) already exceeds the Sun's mass-specific luminosity (~2 × 10⁻⁴ W kg⁻¹) by a factor of 250,000. The trajectory of increasing energy intensity appears to be a general evolutionary tendency of complex dissipative structures, from galaxies (~10^- 5 W kg⁻¹) through stars, planets, and biota to technology, but it must eventually confront the radiative capacity of whatever surface the structure inhabits.

The convergence across these independent analyses; an atmospheric physicist, a climate physicist, an astrobiologist, and an astrophysicist, working with different methods, different assumptions, and different disciplinary frameworks, is itself evidence that the result is robust. The waste heat ceiling is not an artefact of any particular model. It is a consequence of the Stefan-Boltzmann law applied to a finite radiating body, and no competent physicist who examines the problem arrives at a different conclusion.

The power-assembly coupling Γ(t) that appears in the denominator of Σ_max is not an assumed constant. Garrett et al. (2022) confirmed its monetary shadow empirically, finding that global power consumption tracks cumulative economic output with a stable coefficient λ ≈ 5.9 mW per 2019 USD over the period 1970–2019. This stability is a predicted consequence of the competitive viability equilibrium derived in Part 4, not a premise of the ceiling argument. Hanley (2025) extended the analysis to show that this coupling was not stable before 1970, demonstrating that the attractor value of Γ can shift. The decomposition of Γ into its constituent physical components, and the implications for the trajectory toward the ceiling, are developed in the analysis of evolutionary dynamics in Part 3.

Calibration of the ceiling

The ceiling formula contains eight parameters with real operability. Their current best estimates, plausible ranges, and relative contributions to uncertainty can be consolidated into a single reference block. The exercise is not intended to produce a precise number for Σₘax, the uncertainties are too wide for that, but to demonstrate that the ceiling is a measured quantity with bounded uncertainty, not an unmeasured abstraction.

The numerator parameters govern the thermal headroom, the gap between current radiative output and the maximum tolerable output.

The effective emissivity ε captures the net transparency of the atmosphere to outgoing longwave radiation. Its current value is approximately 0.61, accounting for the present greenhouse gas inventory. Under aggressive decarbonisation with carbon dioxide removal, ε could recover toward 0.65. Under continued fossil fuel combustion on a business-as-usual trajectory, it falls toward 0.55 or below. The parameter is partially controllable and enters the ceiling linearly.

The radiating area A is 5.1 × 10^14^ m² the surface area of the top of the atmosphere. It is fixed for any planetary civilisation and enters linearly. Only off-planet radiative infrastructure can change it.

The habitability threshold T~hab~ is the parameter with the highest leverage, entering the numerator with fourth-power sensitivity. Its value is not a single number but a spectrum of consequences. At a 4 K anomaly above the pre-industrial baseline (T~hab~ ≈ 292 K), widespread agricultural failure and regional wet-bulb exceedance begin. At 8 K (≈ 296 K), large-scale wet-bulb exceedance renders substantial equatorial and tropical and areas uninhabitable for unassisted outdoor labour. At 10–12 K (≈ 298–300 K), the mammalian thermoregulation limit is approached globally. The essay's worked example uses 298 K. A civilisation that defines habitability conservatively, as the preservation of current agricultural systems, faces a significantly tighter ceiling than one that defines it as bare mammalian survival.

The planetary albedo α enters through the absorbed solar flux P~⊙~ = S~0~(1 − α)πR^²^, currently approximately 121,500 TW at α ≈ 0.30. Albedo modification through stratospheric aerosol injection or marine cloud brightening could raise α to 0.33–0.35, reducing P~⊙~ and thereby increasing the thermal headroom by a few thousand terawatts. The parameter has moderate leverage and moderate operability.

The ceiling power, the maximum additional civilisational dissipation before T~eq~ reaches T~hab~ is P~max~ = εσAT~hab~^⁴^ − P~⊙~. At the conservative threshold (ΔT = 4 K, ε = 0.61): approximately 7,000 TW. At the central estimate (ΔT = 8 K): approximately 14,000 TW. At the upper bound (ΔT = 10 K): approximately 18,000 TW. Current civilisational power consumption is approximately 20 TW. The ratio P~max~/P~current~ the remaining headroom measured in multiples of current consumption, therefore ranges from roughly 350× to 900×, corresponding to 8–10 doublings at the historical growth rate and a timeline of 250–300 years.

The denominator parameters govern the coupling between assembly stock and power demand.

The aggregate decay rate δ is a stock-weighted mean across the technospheric ensemble. Component values span nearly five orders of magnitude, from approximately 0.05% per year for heavy masonry to 30% or more for consumer electronics. The current aggregate, estimated from national accounts depreciation data and consistent with Garrett's framework, is approximately 1.5–2.5% per year. The aggregate is rising as the technosphere shifts toward higher-turnover components. δ enters the ceiling linearly through the composite coupling.

The specific exergy cost ξ the minimum exergy to execute one joining operation, varies by material and process but is bounded below by the free energy of the relevant chemical or physical transformation. Aggregate estimation is the least constrained parameter in the framework. Its value is implicitly absorbed into the Garrett coupling when the monetary proxy is used, but an independent physical estimate would strengthen the framework. This is the parameter most in need of dedicated empirical work.

The aggregate Second Law efficiency η~II~ is currently estimated at 10–15% globally (Cullen and Allwood, 2010). It enters the denominator inversely: higher efficiency reduces the power required per unit of assembly maintained. The theoretical ceiling is unity. The practical ceiling, set by the Curzon-Ahlborn efficiency at maximum power output, is substantially lower, perhaps 30–50% for the aggregate. The Jevons mechanism established in Part 3 predicts that efficiency gains are recycled into throughput expansion, holding the aggregate coupling approximately stable even as component efficiencies improve.

The metabolic multiplier was calibrated in the maintenance derivation (§2.2) at approximately 1.7–2.5, based on the physical definition and the estimate that maintenance-equivalent activity; depreciation, repair, infrastructure operations, supply-chain sustenance, and the service overhead of keeping assembled systems functional, represents roughly 40–60% of total economic throughput. This is the most operationally significant parameter in the denominator; the one whose reduction the essay identifies as the discriminant between viable and non-viable strategies; and it is also the parameter most strongly governed by Tier 2 competitive dynamics and Tier 3 institutional architecture.

The composite coupling Γ(t) = μδξ/η~II~ can be estimated empirically without decomposing it. Garrett's empirical finding P ≈ λW is the monetary shadow of the physical relation P = Γ(t)·Σ, where λ = Γ(t)/κ. The Garrett coupling λ ≈ 5.9 mW per 2019 US dollar of accumulated wealth, stable within measurement uncertainty over the period 1970–2019, provides the best-constrained empirical anchor for the denominator as a whole. Its stability over fifty years of data, during which every component of the composite changed individually, is the empirical signature of the competitive viability equilibrium derived in Part 4. Its instability in the pre-1970 period, documented by Hanley (2025), demonstrates that the equilibrium value is landscape-dependent, not fixed.

The sensitivity structure is dominated by two parameters. In the numerator, T~hab~ enters with fourth-power sensitivity: the difference between a 4 K and a 10 K habitability threshold changes the ceiling power by a factor of roughly 2.5. In the denominator, μ enters as a direct multiplier on the entire coupling: halving μ doubles the maximum sustainable assembly stock at any given ceiling power. The remaining parameters contribute linearly or inversely and, while important for quantitative precision, do not dominate the uncertainty. The question of how much civilisation can sustain reduces, at first order, to two quantities: how much warming is tolerable and how much of its power budget civilisation devotes to activities beyond maintenance.

The nature of the constraint

It is worth pausing to appreciate what kind of limit this is, because it is qualitatively unlike every other constraint in the environmental policy literature.

Carbon emissions can, in principle, be eliminated. Air pollution can be filtered. Biodiversity loss can, theoretically, be reversed. Resource depletion can be addressed through substitution or recycling. Ocean acidification can be mitigated by removing CO₂ from the atmosphere. Every environmental problem that dominates contemporary discourse is, at root, a problem of kind, of which specific externalities accompany energy use, and of whether those externalities can be managed. They are problems of the exhaust stream's composition.

The waste heat ceiling is a problem of quantity. It does not depend on the energy source. It does not depend on the cleanliness of the conversion process. It does not depend on whether every step in the energy chain has been optimised to the thermodynamic ideal. It depends only on the total non-solar watts dissipated on the planetary surface.

This means the ceiling cannot be addressed by any of the standard tools of environmental policy. Carbon taxes are irrelevant to it. Renewable energy mandates do not touch it, and solar is the one source that does not contribute to the waste heat budget in its dominant conversion pathway, though it faces its own scaling limits related to and area, intermittency, material throughput, and the albedo effects discussed above. Efficiency improvements do not reduce total power dissipation; as the trajectory analysis of Part 3 will demonstrate, they historically increase it. Geoengineering proposals that target atmospheric composition address the greenhouse effect but not the waste heat ceiling, except insofar as albedo modification buys limited thermal headroom as discussed above. No technology changes the Stefan-Boltzmann law. No policy negotiates with it. No market prices it.

The only interventions that address the waste heat ceiling directly are: limiting the total non-solar power that civilisation dissipates on the planetary surface, reducing Γ to raise the assembly stock sustainable at any given power level, or expanding the radiating surface area A beyond the planet, which requires space-based infrastructure at a scale that does not yet exist. All three responses will be examined in later sections. Here, the point is structural. The ceiling exists. It is calculable. It is absolute. And at historical growth rates, it is not distant.

The viable region

From below: the maintenance floor. The Second Law guarantees that assembled matter decays at rate δ. To sustain any given Σ, civilisation must continuously supply power at or above( δξ/η_II ) · Σ. If the power supply falls below this threshold, Σ declines, structures simplify, networks fragment, institutions lose coherence. Below some minimum viable stock Σ_min, the interdependencies within the technosphere produce cascading failure: the loss of one subsystem accelerates the decay of those that depend on it. This is the collapse boundary.

From above: the waste heat ceiling. The power required to maintain Σ is dissipated as waste heat from a finite planetary surface. When the total power dissipation( μδξ/η_II ) · Σ, encompassing both maintenance and non-maintenance activity, raises the equilibrium temperature past the habitability threshold, the biosphere collapses and civilisation with it. This is the thermal boundary, and it defines Σ_max.

The space between Σ_minand Σ_max is the viable region, the range of assembly stocks within which civilisation can, in principle, persist indefinitely on a single planet. The floor is set by the minimum complexity required for coordinated existence. The ceiling is set by the radiative geometry of the planet and the thermodynamic properties of the civilisational stock. Both are physics. Neither is negotiable.

The viable region is not small. At current values P ≈ 20 TW against a surplus radiative capacity measured in thousands of terawatts, there is room. But the region is finite, and its upper boundary is fixed. The viability analysis formalises this gap as the braking boundary Σ^*(T), the surface inside the naive ceiling beyond which even maximum braking cannot prevent constraint violation, and demonstrates that the true viable region is substantially narrower than the arithmetic distance between the maintenance floor and the waste heat ceiling suggests.

Civilisation can grow within it. It cannot grow past it. The question that remains, and that the rest of this essay addresses, is whether the trajectory of growth is compatible with remaining inside the viable region, and if not, what mechanisms drive the trajectory toward the ceiling, what the timescales are, and whether any feasible intervention can redirect it.

That question requires understanding not just where the wall is, but why civilisation appears unable to stop driving toward it. The mechanisms are the subject of Part 3. But before turning to dynamics, one further piece of the thermodynamic machine must be placed: the physics of information processing, which closes the most common proposed escape route from the waste heat accounting. That is the subject of the next section.

2.4 Information Entropy and the Landauer Floor

The waste heat ceiling sets an absolute upper bound on the assembly stock Σ that any planetary civilisation can sustain, determined by radiative geometry and independent of energy source. The question of why civilisation might approach that ceiling rather than stabilising safely below it is the subject of Part 3. Before turning to dynamics, however, one further piece of the thermodynamic machine must be established: the physics of information, which addresses the most common proposed escape route from the waste heat accounting.

The most common objection to this framework, indeed, the single objection that recurs most reliably in policy discussions, corporate sustainability reports, and technology journalism, runs as follows: civilisation is transitioning from a material economy to an information economy. As GDP shifts from steel and concrete to software and services, the coupling between economic output and energy consumption will weaken. The waste heat wall will recede. Civilisation will dematerialise.

This objection deserves the most serious treatment this essay can provide, because it contains a kernel of truth wrapped in a fundamental misunderstanding of physics. The kernel of truth is that the composition of economic output has changed dramatically. Global internet traffic has increased by roughly four orders of magnitude since 2000. The proportion of GDP attributable to information services in advanced economies has risen from single digits to over a third. The information flux component of the service vector has grown explosively relative to all other components.

The fundamental misunderstanding is the assumption that information is immaterial. It is not. Information is physical. This was established theoretically by Szilard (1929), formalised by Brillouin (1951), given its definitive thermodynamic statement by Landauer (1961), and has been experimentally confirmed with increasing precision ever since. The implications for the dematerialisation thesis are devastating.

Shannon, Brillouin, and the bridge between entropies

In 1948, Claude Shannon published "A Mathematical Theory of Communication," defining the entropy of an information source as:

where pᵢ is the probability of the i-th message. Shannon chose the word "entropy" on the advice of John von Neumann, who reportedly told him: "You should call it entropy, for two reasons. In the first place your uncertainty function has been used in statistical mechanics under that name, so it already has a name. In the second place, and more important, nobody knows what entropy really means, so in a debate you will always have the advantage."

The anecdote is famous and often treated as evidence that the connection between information entropy and thermodynamic entropy is merely formal, a mathematical analogy, not a physical identity. This interpretation is wrong. Brillouin demonstrated in his 1951 paper and in his 1956 monograph Science and Information Theory that the two entropies are not merely analogous but connected by a precise conversion factor. Acquiring one bit of information about a physical system requires dissipating at least:

of free energy, where k_B is the Boltzmann constant (1.38 × 10⁻²³ J/K) and T is the temperature of the environment. At room temperature (T = 300 K), this gives:

This is the Brillouin limit on measurement. It establishes that information entropy and thermodynamic entropy are connected by the factor k_BTln2 per bit, not metaphorically, but through the physics of measurement, recording, and erasure. The connection was anticipated by Szilard's 1929 analysis of Maxwell's Demon, which showed that the demon's measurements must generate at least as much entropy as the sorting operation saves. Brillouin completed the argument: information is not free. It is purchased with thermodynamic work, at a price set by the Second Law.

Landauer's Principle: the irreducible floor

Landauer sharpened Brillouin's insight into a precise operational statement. In his 1961 paper "Irreversibility and Heat Generation in the Computing Process," published in the IBM Journal of Research and Development, he proved that any logically irreversible computational operation, any operation that destroys information, such as erasing a bit or merging two computational paths into one, must dissipate at least k_BTln2 of energy as heat. This is not an engineering limitation. It is a consequence of the Second Law applied to information processing.

The argument is elegant. Consider a single bit in a known state (0 or 1). Erasing it, resetting it to a standard state, reduces the information entropy of the bit by exactly one bit, or k_Bln2 in thermodynamic units. By the Second Law, this reduction in the system's entropy must be compensated by an equal or greater increase in the entropy of the environment. The minimum environmental entropy increase corresponds to heat dissipation of:

per bit erased. This is the Landauer limit. It is the thermodynamic floor of computation. No amount of engineering ingenuity, no future material science breakthrough, no quantum mechanism can reduce the energy dissipated per irreversible logical operation below this value. It is as fundamental as the Carnot limit for heat engines, and as unbreakable.

Experimental confirmation

For half a century after Landauer's paper, the principle remained a theoretical result, accepted by physicists but untested in the laboratory, because the energies involved are so small that thermal noise overwhelms the signal. That changed in the 2010s with a series of increasingly precise experiments.

Bérut, Arakelyan, Petrosyan, Ciliberto, Dillenschneider, and Lutz (2012) provided the first direct experimental verification, published in Nature. They used a colloidal silica bead (2 μm diameter) trapped in a modulated double-well potential created by focused laser beams, immersed in water at room temperature. The bead in one well or the other constituted a physical bit. By slowly lowering and raising the barrier between wells, they performed controlled erasure cycles and measured the heat dissipated into the surrounding water via the bead's Brownian trajectory. Their result: the mean dissipated heat converged to k_BTln2 from above as the erasure protocol was made slower (more quasi-static), confirming Landauer's bound to within experimental precision. Faster erasure dissipated more; no erasure dissipated less. The Second Law held.

Jun, Gavrilov, and Bechhoefer (2014) refined the measurement using a feedback trap, an electrokinetic system that tracks a colloidal particle in real time and applies computed forces to confine it. Their apparatus achieved erasure closer to the quasi-static limit and confirmed the Landauer bound with tighter error bars. Gavrilov and Bechhoefer (2016) subsequently measured the full probability distribution of dissipated heat during erasure, verifying that the distribution satisfies the Jarzynski equality and that the mean dissipation approaches the Landauer limit as expected.

Hong, Lambson, Dhuey, and Bokor (2016) demonstrated Landauer-limit erasure in a nanomagnetic system, a single-domain nanomagnet whose magnetisation direction encodes one bit. They showed that field-driven erasure of the bit dissipated energy approaching k_BTln2, confirming the principle in a solid-state context directly relevant to digital logic.

The experimental situation is now settled. Landauer's principle is not a conjecture or an approximation. It is a verified physical law, confirmed across multiple experimental platforms (colloidal, electrokinetic, nanomagnetic), at energies spanning several orders of magnitude. The floor is real.

The reversible computing caveat

A necessary clarification: Charles Bennett (1973) proved that any computation can in principle be performed using only logically reversible operations, operations that preserve all intermediate information and therefore need not erase anything. Since no information is destroyed, the Landauer limit does not apply, and a fully reversible computer can in principle approach zero dissipation per logical step. This result is correct and important. It does not, however, invalidate the floor for four reasons. First, most useful computation is not logically reversible; the overwhelming majority of real workloads; database writes, hash functions, lossy compression, neural network training, involve irreversible operations as a matter of logical structure, not engineering choice. Second, input preparation and output measurement are inherently irreversible: data must be written into the machine and results read out, and both processes erase prior states at the Landauer cost. Third, the practical gap between current hardware and the Landauer limit spans nine orders of magnitude, and the trajectory analysis demonstrates that competitive dynamics will fill that gap with demand rather than bank it as savings. Fourth, even a fully reversible computation embedded in an irreversible world, receiving inputs and delivering outputs, must pay the Landauer price at the boundary. Reversible computing shifts the floor for a subset of internal operations. It does not shift it for the system.

The gap and its Jevonian fate

Here is where the dematerialisation thesis encounters its first quantitative problem. Current state-of-the-art processors dissipate approximately 10^- 12 joules per logical operation (of order one picojoule). The Landauer limit at room temperature is approximately 3 × 10^- 21 joules per operation. The ratio is roughly 3 × 10⁸ current hardware operates approximately a billion times above the theoretical minimum.

To the techno-optimist, this looks like extraordinary headroom. If even a fraction of that nine-order-of-magnitude gap could be closed, vastly more computation per watt becomes possible, and the energy cost of the information economy would shrink toward insignificance. The dematerialisation thesis would be vindicated.

But the trajectory analysis has already explained what happens when efficiency improves by orders of magnitude. The evolutionary ratchet, grounded in Lotka's maximum power principle and the empirically observed stability of the power–assembly coupling Γ(t), predicts that efficiency gains in any component of the service vector are captured by the system and recycled into aggregate growth. The history of computation already follows exactly this pattern, at a scale and speed that dwarfs every prior Jevonian cycle.

In 1971, the first commercially available microprocessor (the Intel 4004) performed roughly 90,000 operations per second and consumed 0.5 watts. A modern data centre GPU (such as the NVIDIA H100) performs approximately 10¹5 operations per second and consumes 700 watts. Energy per operation has fallen by a factor of roughly 10^10, ten billion. This is among the most dramatic efficiency improvements in the history of technology.

Total computational energy consumption has not fallen. It has exploded. Global data centre electricity consumption was approximately 200 TWh in 2010, roughly 460 TWh in 2022, and the International Energy Agency projects it could exceed 945 TWh by 2030 (IEA, 2024). Goldman Sachs (2024) estimates that data centre power demand may reach 1,000 TWh globally by 2030. To place this in context, 1,000 TWh per year corresponds to an average continuous power draw of approximately 114 GW, roughly equivalent to the total electricity consumption of Japan.

Between 1990 and 2020, computational efficiency improved by roughly six orders of magnitude. Total computation increased by roughly eight orders of magnitude. Total computational energy consumption increased by roughly two orders of magnitude. The rebound was not 100 per cent. It was approximately 130 per cent, super-Jevons backfire, where each unit of efficiency gained produced more than one unit of additional consumption. Koomey's Law (Koomey, Berard, Sanchez, and Wong, 2011), which documents the historical doubling of computations per kilowatt-hour every 1.6 years, is sometimes cited as evidence of progress toward sustainability. It is the opposite. It is the efficiency trend that enables the demand explosion. Koomey's Law is the computational instantiation of the Jevons Paradox.

The pattern is identical to Saunders and Tsao's (2012) finding on artificial lighting: a roughly 3,000-fold improvement in luminous efficacy from tallow candles to LEDs accompanied by a roughly 40,000-fold increase in total light consumption, with the world consistently spending approximately 0.72 per cent of GDP on illumination. The technology changes. The efficiency improves. The total dissipation increases. The maximum power principle selects for throughput, not parsimony.

Artificial intelligence as thermodynamic accelerant

The acceleration in computational energy demand is not uniform across the information sector. It is concentrated overwhelmingly in one application: artificial intelligence. AI represents the most energy-intensive frontier of the most rapidly growing component of civilisational power consumption, and its scaling dynamics are incompatible with any plausible dematerialisation trajectory.

Training a single frontier large language model now requires an estimated 50 GWh of electricity, roughly the annual consumption of 5,000 American households, compressed into a training run of several months (Patterson et al., 2022; de Vries, 2023). Each subsequent generation of frontier model has demanded approximately 3–10 times more compute than its predecessor, a scaling trend documented by Sevilla et al. (2022) across six decades of AI research. The trend shows no sign of saturating. If anything, it is steepening: the compute required for frontier models has been growing at roughly 4× per year since 2010, a rate that exceeds Moore's Law by a substantial margin.

Training, however, is a one-time cost per model. The ongoing energy expense is inference, the act of running a trained model to generate outputs. Every query to a large language model, every AI-generated image, every automated coding suggestion involves a forward pass through billions of parameters, each requiring thousands of floating-point operations, each dissipating energy as heat. The IEA estimates that a single query to a large language model consumes roughly ten times the electricity of a conventional search engine query (IEA, 2024). As AI is integrated into search engines, productivity tools, coding environments, scientific instruments, and consumer applications, the aggregate inference load scales with the number of users, the number of queries per user, and the size of the models being served.

This creates a compounding dynamic that is precisely the structure described by the evolutionary ratchet. AI increases the productivity of labour and capital, it raises the rate of return Y/W in the Garrett framework. Higher Y/W means faster growth of W (via dW/dt = (Y/W)W - δ W), which through the physical relation P = Γ(t) · Σ drives higher power consumption. Simultaneously, the computational infrastructure required to deliver the AI services itself demands power that grows faster than the broader economy. The sector that was supposed to demonstrate dematerialisation is instead the sector with the tightest energy coupling and the fastest growth rate.

The projections are stark. At 20 per cent annual growth from a 2025 base of approximately 50 GW average global data centre power draw: at 25 years (circa 2050), approximately 4,800 GW, or 4.8 TW, roughly one quarter of current total global primary power consumption; at 50 years (circa 2075), approximately 460 TW, exceeding current global primary power by a factor of 25; at 75 years (circa 2100), approximately 44,000 TW, more than 35 per cent of the total solar radiation absorbed by the Earth.

The last figure is plainly impossible on the surface of a habitable planet, and it is reached in 75 years, not the roughly 400 years projected for the broader economy in the waste heat analysis (Section 2.3). Twenty per cent annual growth will not be sustained for 75 years, precisely because the waste heat ceiling, the power generation capacity, and the resource constraints described elsewhere in this essay will intervene. The point is not that this trajectory will be realised. The point is that the information sector is not an escape from the thermodynamic framework. It is, on current trends, an acceleration toward its limits. The sector that was supposed to decouple growth from energy is instead driving civilisation toward the ceiling faster than any other activity.

Reversible computing and Bennett's result

The reversible computing caveat established above warrants fuller treatment, because the theoretical result is genuinely important even as its practical implications are routinely overstated.

Bennett (1973) proved that any computation can, in principle, be performed using only logically reversible operations. A reversible computer never erases intermediate results; it preserves them, allowing the computation to be "uncomputed", run backward to recover the input from the output. Since no information is destroyed, the Landauer limit does not apply. In principle, a reversible computer can operate with arbitrarily low energy dissipation per logical step, approaching zero in the quasi-static limit.

There are three reasons why this escape route does not change the practical conclusion.

First, the thermodynamic trade-off. Reversible computation trades energy for time and memory. A reversible Turing machine must store all intermediate results, which means its memory requirements grow with the length of the computation. Bennett (1989) showed that the space-time trade-off is fundamental: reducing energy dissipation below the Landauer limit requires either more time (slower computation) or more memory (more physical substrate), or both. Fredkin and Toffoli (1982) established the theoretical framework for reversible logic gates, and subsequent work by Frank (2005) and others has quantified the practical overhead. The memory cost of full reversibility can be exponential in the depth of the computation for general algorithms, though Bennett's "pebbling" strategies reduce this to polynomial in many cases.

Second, the maximum power principle selects against reversibility. This is the critical point, and it connects directly to the evolutionary ratchet analysis. A reversible computer that approaches the zero-dissipation limit must operate quasi-statically, infinitely slowly. This is precisely analogous to a Carnot engine at maximum efficiency: thermodynamically ideal but producing zero power. Under Lotka's maximum power selection, the system does not optimise for minimum dissipation. It optimises for maximum throughput. A civilisation (or a firm, or a research programme) that computes at the reversible limit computes infinitely slowly and is outcompeted by one that computes irreversibly at maximum speed. The competitive dynamics drive computation toward the maximum-power configuration, not the minimum-dissipation configuration. In practice, this means hardware will be operated far above the Landauer limit, dissipating excess heat in exchange for faster results.

The analogy to heat engines is exact. Real heat engines operate at roughly 30–50 per cent of Carnot efficiency, not because engineers are incompetent but because the Curzon-Ahlborn efficiency (Curzon and Ahlborn, 1975), the efficiency at maximum power output, is substantially below the Carnot limit. The system is not optimised for thermodynamic perfection. It is optimised for power. The same selection pressure, operating through market competition, research funding allocation, and institutional incentives, keeps real computers operating orders of magnitude above the Landauer limit for as long as competitive dynamics persist.

Third, even the Landauer limit does not save the dematerialisation thesis. Suppose, counterfactually, that a civilisation could build processors operating exactly at the Landauer limit with no speed penalty. At room temperature and 10^30 operations per second, a plausible estimate for the computational demands of a mature AI-integrated civilisation, the power requirement would be:

This is a single large nuclear power plant, almost manageable. But 10^30 operations per second is roughly what current global data centres perform today. A civilisation that has closed the nine-order-of-magnitude gap to the Landauer limit will not hold computation constant at current levels. It will compute nine orders of magnitude more. That is the lesson of Jevons, of Garrett, and of three centuries of empirical data on the relationship between efficiency and consumption. The Landauer limit is not a ceiling on demand. It is the floor beneath the next Jevonian expansion.

Quantum computing: a different limit, not an escape

Quantum computing is sometimes invoked as a further escape route, on the grounds that quantum parallelism allows certain problems to be solved with exponentially fewer operations than classical computation requires. This is correct for specific problem classes, Shor's algorithm for factoring, Grover's algorithm for unstructured search, and certain quantum simulation tasks. It does not change the thermodynamic picture for three reasons.

First, quantum computers still dissipate energy. The operations are different (unitary rotations of quantum states rather than classical logic gates), but error correction, measurement, and the maintenance of cryogenic environments all consume power. Current quantum processors require dilution refrigerators operating at millikelvin temperatures, consuming tens of kilowatts to maintain the cryogenic environment for a processor performing far fewer useful operations per second than a consumer laptop. The energy per useful quantum operation is, at present, vastly higher than the energy per useful classical operation.

Second, the Landauer limit applies to quantum measurement and error correction. Every time a quantum computation produces a classical output, which it must, if the result is to be used, the measurement collapses quantum superposition into a definite classical state, an irreversible process that dissipates at least k_BTln2 per bit of classical output. Quantum error correction, essential for any computation longer than the decoherence time of the physical qubits, generates classical syndrome information that must be processed and erased. The thermodynamic cost of this erasure is bounded by Landauer's principle (Reeb and Wolf, 2014).

Third, and most importantly, the maximum power argument applies with equal force. Even if quantum computing dramatically reduces the energy per solution for certain problem classes, the competitive dynamics will drive the total volume of problems attempted upward until the energy budget is consumed. A quantum advantage that makes protein folding simulations a thousand times cheaper does not reduce the energy spent on protein folding. It increases the number of proteins folded by a factor of a thousand, or more, because problems previously too expensive to attempt become feasible. The Jevonian expansion applies to quantum speedups just as it applies to classical efficiency gains.

Thermodynamic inflation in the information sector

The information sector is subject to its own form of thermodynamic inflation. The computational demands of maintaining a given level of civilisational information service grow over time, independent of any expansion in the service itself.

Software complexity increases monotonically. Each generation of operating system, database engine, security protocol, and application framework is larger and more computationally expensive than its predecessor. This is not waste or poor engineering. It reflects the genuine expansion of the attack surface that must be defended, the interoperability requirements that must be met, the regulatory compliance that must be verified, and the accumulated legacy interfaces that must be maintained. Wirth's Law, the empirical observation that software bloat absorbs hardware gains (Wirth, 1995), is a specific instance of thermodynamic inflation operating in information space.

AI model size is growing faster than the efficiency of inference. The compute required to run a frontier model at interactive speed has been increasing at a rate that outpaces improvements in hardware efficiency (Sevilla et al., 2022). If model capability is the service and inference compute is the exergy cost, then the cost per unit of service is rising, not falling. This is the information-sector analogue of declining ore grades: the "ore", the computational substrate from which AI capability is extracted, is getting richer in output but more expensive in energy per unit of output as models scale.

Data volumes grow exponentially. The global datasphere is projected to reach approximately 175 zettabytes by 2025 and continues to expand at roughly 25 per cent per year (Reinsel, Gantz, and Rydning, 2018). Storing, indexing, securing, backing up, and transmitting this data requires energy that scales with volume. Even if the energy per byte of storage falls (which it has, dramatically), the total energy consumed by storage grows because the volume growth rate exceeds the efficiency improvement rate. This is Jevons operating on data, exactly as it operates on light, on transport, and on every other service component.

Information is physical, therefore information is thermodynamic

The deeper point, and the one that closes the logical circle of this section, is that the dematerialisation thesis rests on a category error. It treats information as if it exists in a realm separate from matter and energy, as if software runs on nothing, as if data is weightless, as if a shift from manufacturing steel to manufacturing algorithms represents an escape from thermodynamic constraint.

Szilard, Brillouin, Landauer, and Bennett proved otherwise. Every bit stored requires a physical substrate. Every bit erased dissipates heat. Every computation, however efficient, generates entropy. The information flux is not a weightless alternative to material or transport flux. It is another physical channel through which civilisation processes exergy and exports entropy to the 2.7 K sink of deep space. And it is the channel whose throughput is growing faster than any other.

The power–assembly coupling Γ(t) does not care whether the assembly stock it couples to is embodied in steel mills or server farms. It couples to the total assembly stock Σ, regardless of composition. The waste heat ceiling is not softened by the information economy. If anything, it is hardened, because the information sector is growing faster than the material sectors it was supposed to replace, and because the thermodynamic floor on its energy consumption, Landauer's limit, merely sets the starting line for the next Jevonian expansion.

Part 3 — The Trajectory

3.1 Maximum Power and the Evolutionary Ratchet

Part 2 established the machine. Civilisation is assembled matter maintained against the Second Law by continuous exergy throughput. The power required scales with the accumulated assembly stock, coupled by the composite Γ(t) = μδξ/η_II: the relation P = Γ(t) · Σ. That power is dissipated as waste heat from a finite planetary surface, which sets a maximum sustainable assembly stock Σ_maxbeyond which the equilibrium temperature exceeds the habitability threshold. Between the maintenance floor δΣ, below which the stock decays, and the waste heat ceiling Σ_max, above which the planet overheats, lies the viable region.

The viable region is large. At current power consumption of approximately 20 TW against a solar absorption budget of 120,000 TW, civilisation occupies a small fraction of the available space. The waste heat ceiling is centuries away at historical growth rates. There is, in principle, ample room to stabilise.

The question that Part 3 addresses is why civilisation does not stabilise. Why does the trajectory tend toward the ceiling rather than settling at a sustainable level within the viable region? The answer involves three mechanisms, ordered in this and the following two sections by their ontological hardness, by how deeply they are embedded in the physics and how resistant they are to deliberate override.

This section presents the hardest mechanism: the evolutionary ratchet. It is not a human institution. It is not a policy failure. It is the outcome of competitive selection among dissipative structures drawing on a common energy gradient, and it has been operating for as long as life has existed. The pattern is observed across biological and economic systems with remarkable consistency; its formal derivation from competitive viability geometry follows in the competitive viability analysis (§4.3). What this section presents is the empirical case that motivates that derivation.

Lotka's principle

In 1922, the mathematician and biophysicist Alfred Lotka published two papers in the Proceedings of the National Academy of Sciences that stated a principle so fundamental it has never been refuted, only ignored. Lotka observed that natural selection, operating on any population of self-organising systems competing for available energy, favours configurations that maximise the rate of energy throughput, not the efficiency of energy conversion, but the speed at which energy is captured, transformed, and dissipated (Lotka, 1922a; 1922b).

The distinction between maximum efficiency and maximum power is critical. Consider a heat engine operating between a hot reservoir at temperature T_H and a cold reservoir at T_C. The Carnot efficiency, the theoretical maximum fraction of heat that can be converted to work, is:

But a Carnot engine achieves this efficiency only in the reversible limit: infinitely slow operation, zero power output. A real engine trades thermodynamic efficiency for speed. It accepts higher irreversibility, largerṠ_gen, in exchange for greater power output per unit time. The power output Ẇof an endoreversible engine operating at finite rate between the same reservoirs is maximised at the Curzon–Ahlborn efficiency (Curzon and Ahlborn, 1975):

which is always less than η_Carnot. The engine at maximum power wastes more exergy per cycle than the engine at maximum efficiency, but it produces more useful work per unit time, and it is time, not thermodynamic perfection, that selection acts upon.

Lotka's insight was that this trade-off is not confined to engineered heat engines. It applies to any self-replicating system competing for a finite energy gradient. An organism that captures energy faster, even at lower thermodynamic efficiency, can reproduce faster, defend territory more effectively, and outcompete slower but more efficient rivals. A firm that converts capital into revenue faster outgrows one that converts it more carefully. A nation that industrialises faster achieves military and economic dominance over one that industrialises more sustainably. The selection pressure is universal. It operates on organisms, on ecosystems, on firms, on nations, and on civilisations. It favours maximum power, not maximum efficiency.

Howard Odum spent four decades developing Lotka's principle into a general theory of ecological and economic energetics. In a foundational 1955 paper with Richard Pinkerton in American Scientist, Odum and Pinkerton demonstrated that biological systems operate at an optimal intermediate efficiency, typically around 50% of Carnot, to achieve maximum power output. This is not a deficiency. It is the configuration that selection has found to be optimal under competition. Systems that operate closer to the Carnot limit produce less power and are outcompeted. Systems that operate further below it waste too much exergy on internal irreversibility and are also outcompeted. The maximum power configuration sits at a specific point on the efficiency–power curve, and it is this point, not the Carnot limit, that characterises real dissipative structures in competitive environments (Odum and Pinkerton, 1955).

Odum later reformulated this as the Maximum Empower Principle: systems self-organise to maximise power intake, energy transformation, and those uses that reinforce production and resource intake (Odum, 1996).

From observation to derivation

The maximum power pattern is observed with sufficient consistency, across ecosystems, economies, and technological domains, to demand a formal explanation rather than merely an empirical catalogue. The ecological and thermodynamic literature has debated which extremal principle governs the self-organisation of dissipative structures: maximum power (Lotka, 1922a; Odum, 1996), maximum entropy production (Dewar, 2003; Kleidon, 2010), or maximum efficiency under varying constraint regimes. The most significant recent institutional recognition came in the 2023 Royal Society theme issue "Thermodynamics 2.0" in Philosophical Transactions A, where Hall and McWhirter (2023) presented new evidence that free market mechanisms function according to maximum power selection, and Rees (2023) cited Lotka's principle in asking whether Homo sapiens is "unsustainable by nature."

This essay does not import the maximum power principle as an axiom from ecology. It treats the observed pattern, competitive dissipative structures converging on maximum throughput, as the empirical phenomenon requiring explanation. The formal derivation of this result, showing that gradient saturation follows from the geometry of viability kernels under competitive allocation, without invoking any selection mechanism beyond the principle that persistence is inclusion in the viable region, is developed in the viability analysis. What follows here is the empirical case that motivates the derivation.

The key result, established formally in the competitive viability analysis (§4.3), is that gradient saturation (maximum total dissipation) is a proposition of competitive viability dynamics: agents competing for shares of a finite energy gradient generically saturate that gradient because unused capacity is unstable, any agent that captures surplus weakly expands its viability kernel and weakly contracts competitors'. The maximum power principle, on this account, is not an ecological axiom but a corollary of competitive geometry on a bounded gradient. Its stiffness, its resistance to unilateral departure, is what places it at Tier 2 of the ontological hierarchy: not immutable physics, but a game-theoretic equilibrium requiring coordinated multi-agent departure to override.

The competitive viability derivation (§4.3) predicts that competitive dissipative structures sharing a finite gradient converge on a Nash equilibrium in which the coupling between power throughput and accumulated structure is stable. This prediction has an empirical signature: the observed stability of the power–assembly coupling. Garrett's empirical finding P ≈ λ W, where λ ≈ 5.9 mW per 2019 US dollar, is the monetary shadow of the physical relation P = Γ(t) · Σ, where λ = Γ(t)/κ absorbs both the physical composite Γ(t) and the monetary conversion factor κ that maps assembly stock to dollar-denominated wealth.

Garrett's fifty-year dataset (1970–2019) shows λ holding approximately constant. The physical interpretation: Γ(t) sits at a competitive equilibrium. Under current competitive conditions, the ratio of power throughput to accumulated structure has converged on the value that maximises the system's rate of expansion. Any civilisation that reduced Γ(t) below this value, that attempted to sustain more assembly per watt, would grow more slowly and be outcompeted by one that maintained the higher metabolic rate. This is precisely the behaviour predicted by the competitive viability result: the equilibrium coupling is the one at which no agent can unilaterally improve its kernel by deviating.

Consider the dynamics. Civilisation's assembly stock Σ requires power P = Γ(t) · Σ. The stock grows when construction exceeds decay:

In Garrett's monetary framework, the rate of return on accumulated wealth is α = Y/W, where Y is annual economic output and W is cumulative wealth. An improvement in the thermodynamic efficiency of any component of the civilisational system, a more efficient engine, a better insulated building, a lower-loss transmission line, reduces the exergy cost per unit of service delivery. In standard economic terms, this raises productivity. In the Garrett framework, it increases α: more output per unit of accumulated wealth.

Among competing agents drawing on a common energy gradient, firms competing for market share, nations competing for geopolitical influence, technologies competing for adoption, those that achieve higher α grow faster. Their output Y is larger relative to their maintenance burden δ W, leaving a greater surplus for net accumulation. They add to Σ faster. They capture a larger share of available exergy. They dominate.

But the surplus does not reduce P. It increases W. And since P = Γ(t) · Σ or equivalently, in monetary terms, P ≈ λ W, the increase in W drives a proportional increase in P. The efficiency gain has been captured by the system and converted into growth. Γ remains stable because it represents the competitive equilibrium, the ratio of power throughput to accumulated structure at which no agent improves its viability by deviating.

This is the thermodynamic explanation for why Garrett's empirical λ has not changed in five decades of data despite massive improvements in the energy efficiency of virtually every technology in the civilisational portfolio. The improvements are real. They increase α. And the increase in α is immediately captured by the competitive dynamic and recycled into faster growth of W, which restores P/W to its equilibrium value. The system absorbs efficiency gains the way an ecosystem absorbs a nutrient pulse: not by shrinking, but by growing until the new resource base is fully exploited.

Garrett, Grasselli, and Keen (2022) formalised this connection in "Lotka's wheel and the long arm of history," demonstrating that the constancy of λ over fifty years of global data is consistent with Lotka's evolutionary principle operating at civilisational scale. The global economy behaves as a maximum power system, not a maximum efficiency one.

The attractor can shift

The stability of Γ under current conditions must not be mistaken for permanence. Hanley (2025) extended the empirical analysis before 1970 and found that the power–wealth ratio was not stable in the pre-1970 period, the attractor value shifted at least once in recorded history, coinciding with the structural transformation of the global financial architecture following the collapse of the Bretton Woods system and the subsequent expansion of fiat money creation. This observation is significant for two reasons.

First, if Γ were fixed by physics alone, a Tier 1 constant, it could not have changed. That it did change demonstrates that the equilibrium is set by the competitive landscape: the institutional architecture, financial structure, and incentive environment within which dissipative agents operate. The viability analysis formalises this as landscape dependence of the attractor: the equilibrium value of Γ depends on the competitive parameters {Γᵢ, δᵢ, αᵢ} across agents, which are themselves shaped by institutional architecture (Tier 3). The stiffness of the attractor, its resistance to departure, is Tier 2. The location of the attractor, which value of Γ the system converges to, is Tier 3.

Second, Hanley's data provides direct evidence that the competitive equilibrium has shifted at least once when the landscape changed, precisely the behaviour predicted by the landscape-dependence result. Under current competitive dynamics, globalised markets, fiat monetary expansion, short-horizon institutional incentives, Γ is empirically stable and the system is locked onto a maximum-power trajectory. But the current selection landscape is not the only possible one. If the competitive environment were altered through coordination, institutional reform, or a fundamental restructuring of the incentive architecture, the attractor could move. The engineering headroom within Γ = μδξ/η_II each component independently adjustable, provides the physical space for such a shift. Accessing that space requires overriding the competitive dynamics that currently select for the present value: a cooperative game solution in which all N agents simultaneously accept kernel contraction on binding commitment. Reshaping the game so that competition converges to a lower Γ is a different and more tractable problem. The degrees of freedom exist. The question is whether the selection landscape can be reshaped to reach them.

The Jevons Paradox: aggregate result and specific channel

The implication can now be stated precisely. The Jevons Paradox, the observation that improvements in the efficiency of resource use tend to increase rather than decrease total consumption of that resource, is not an anomaly, not a market failure, and not a correctable policy deficiency. It is a consequence of competitive dynamics operating through the power–assembly coupling.

Two levels of the Jevons result must be distinguished.

Aggregate result (Tier 2). Efficiency gains are recycled into throughput expansion at the civilisational level. This holds regardless of economic system and follows as a corollary of the competitive viability result derived in the competitive viability analysis (§4.3): an efficiency improvement increases the construction efficiency αᵢ, expanding the velocity set available to each agent; competition then recycles the gain into throughput until the gradient constraint binds again. The aggregate result, that efficiency gains do not reduce total dissipation, is a property of the competitive equilibrium itself. It holds in market economies, command economies, and any hybrid: wherever multiple agents compete for shares of a finite energy gradient, efficiency gains expand capacity rather than reduce load.

Specific channel (Tier 3). The mechanism through which the aggregate result manifests varies with institutional architecture. In market economies, it operates through price-mediated demand expansion: lower marginal cost of energy services stimulates consumption, price signals attract investment, and the macroeconomic multiplier converts micro-level savings into system-wide growth. In command economies, it operates through surplus reallocation: the central planner directs efficiency-freed resources to priority sectors. For competing technologies, it operates through adoption migration: the more efficient technology displaces incumbents and expands the accessible market. The channel is reformable. The aggregate result is not.

The implication for policy is precise. Tier 3 interventions, energy taxes, throughput caps, efficiency mandates, can redirect the Jevons channel: they alter which sectors grow, which technologies are adopted, which agents capture the surplus. They cannot suppress the aggregate effect without solving the Tier 2 coordination problem identified in the competitive viability analysis (§4.3). This distinction strengthens both the diagnosis and the policy analysis. It is not that efficiency policy is useless. It is that efficiency policy operates on the channel, not the aggregate, and the aggregate is what reaches the ceiling.

William Stanley Jevons first documented the phenomenon in 1865. In The Coal Question, he observed that James Watt's improvements to the steam engine, which dramatically reduced the coal required per unit of useful work, had not reduced Britain's coal consumption. They had increased it, enormously.

For 160 years, mainstream economics has treated Jevons's observation as a curiosity that applies in some sectors under some conditions but which can be overcome by sufficiently ambitious policy. The standard view, codified in reports by the International Energy Agency and in the integrated assessment models used by the IPCC, assumes that energy efficiency improvements translate more or less directly into reduced energy demand, with a modest "rebound" that erodes a fraction of the savings. A typical estimate from micro-level studies places the combined direct and indirect rebound at 10–30% of expected savings (Gillingham, Rapson, and Wagner, 2016). Policy can work with this. If efficiency saves 100 units of energy in engineering terms, actual savings are 70–90 units. Manageable.

The problem is that this analysis is confined to partial equilibrium. It examines one technology, one market, one set of consumers in isolation. It does not and cannot capture the macroeconomic rebound: the structural transformation of the entire economic system in response to a generalised improvement in the productivity of energy. And it is at this macro scale, the scale at which the power–assembly coupling operates, the scale at which competitive selection acts, that the Jevons Paradox ceases to be a paradox and becomes a structural consequence of the competitive equilibrium.

The derivation is straightforward. From the physical relation P = Γ(t) · Σ and its monetary shadow P ≈ λ W, and from the definition of the return rate α = Y/W, the growth rate of power consumption follows:

where the last step uses the monetary equivalence and the fact that net wealth accumulation is dW/dt = Y - δ W = (α - δ)W. An efficiency improvement that raises α directly increases dP/dt. The growth rate of energy consumption is an increasing function of the rate of return on wealth. Make the economy more efficient, raise α, and energy consumption grows faster, not slower. This is not a rebound of 10–30%. It is structural backfire embedded in the dynamics of the system.

The empirical record

The theoretical prediction is confirmed by the data at every scale that has been examined.

The most comprehensive assessment of economy-wide rebound is the meta-analysis by Brockway, Sorrell, Semieniuk, Heun, and Court, published in Renewable and Sustainable Energy Reviews in 2021. They analysed 33 studies that attempted to measure rebound at the national or global scale, using econometric estimation, computable general equilibrium models, and structural decomposition analysis. Their central finding: economy-wide rebound effects typically exceed 50%, with many estimates clustering near or above 100%. They found no historical precedent for absolute decoupling of energy use from GDP at the global level, and warned that global energy scenarios, including those underpinning IPCC mitigation pathways, may systematically underestimate future energy demand (Brockway et al., 2021).

The case of artificial lighting provides the longest and cleanest empirical test. Saunders and Tsao examined the history of lighting across six continents and five major technology transitions; tallow candles, whale oil lamps, gas lighting, incandescent bulbs, fluorescent tubes, and LEDs, spanning over 300 years. They found 100% rebound: despite a roughly 3,000-fold improvement in luminous efficacy (lumens per watt), the world consistently spends approximately 0.72% of GDP on illumination. Every efficiency gain was absorbed entirely by expanded demand for light. Total energy consumed for lighting did not fall at any point in the historical record. It grew, monotonically (Saunders and Tsao, 2012).

This is not a peculiarity of lighting. Saunders's empirical analysis of 30 US industrial sectors over 1960–2005 found an average short-term rebound of 126%, backfire, meaning that efficiency improvements were associated with an increase in total energy consumption exceeding the engineering savings. Individual sectors ranged from −16% (a rare case of net savings) to 378% (extreme backfire). The distribution was heavily right-skewed: most sectors showed rebound well above 50%, and a substantial fraction exhibited full backfire (Saunders, 2013).

Brockway, Heun, Santos, and Barrett extended the analysis using a nonlinear dynamic model calibrated to global data from 1900 to 2018, over a century of industrial history. Under a scenario maximising GDP growth, energy efficiency investment roughly doubled but produced no decrease in primary energy use growth, because economy-wide rebound effects dominated. Even under a scenario explicitly designed to minimise energy use, a 3.5-fold increase in efficiency investment only minimised, but did not eliminate, continued growth in energy consumption (Brockway et al., 2024).

Stern's review of the general equilibrium modelling literature reached the same conclusion: "some recent general equilibrium studies find large rebound, around 100%" (Stern, 2020).

The case of computation is the most extreme and the most recent. Between 1971 and 2024, the energy cost per floating-point operation fell by roughly ten orders of magnitude, from approximately 10^- 2 joules per operation (Intel 4004) to approximately 10^- 12 joules per operation (modern GPU). Over the same period, total computational demand grew by approximately twelve orders of magnitude, and total energy consumed by computation grew by approximately two orders of magnitude. The rebound was not 100%. It was approximately 120%, super-Jevons backfire, where each order of magnitude of efficiency improvement produced more than one order of magnitude of additional demand. The sector that was supposed to dematerialise the economy is instead the sector with the tightest coupling between efficiency gain and demand growth, growing faster in energy terms than any other component of civilisational throughput.

The decoupling illusion

The Jevons mechanism, grounded in the competitive viability result and confirmed by the empirical record, demolishes the central hope of mainstream climate-energy policy: that efficiency improvements can enable continued economic growth while reducing total energy consumption. This is the doctrine of "absolute decoupling."

The key distinction is between relative and absolute decoupling, and between different quantities being "decoupled." Energy intensity of GDP, the ratio P/Y, has indeed declined in most developed economies by 1–2% per year. This is relative decoupling: more GDP per unit of energy. But absolute global energy consumption has continued to rise, because GDP growth has outpaced intensity improvements. The physical framework explains why: reducing P/Y increases Y (by making production cheaper), which adds to W (via the integral), which increases P (via P = Γ(t) · Σ). Efficiency drives growth, not savings.

Haberl and colleagues examined 835 peer-reviewed articles on decoupling in a systematic review published in Environmental Research Letters in 2020. Their finding: relative decoupling is common for materials and greenhouse gas emissions, but not for useful exergy a quality-based measure of energy use. The actual thermodynamic throughput of the economy, the useful work performed, has never decoupled from GDP at the global level (Haberl et al., 2020). This is precisely what the competitive viability framework predicts.

National-level apparent decoupling is substantially explained by trade effects. When a developed country offshores its manufacturing, its domestic energy consumption falls while its consumption of energy-intensive imports rises. The energy has not disappeared; it has been relocated. Moreau, Vuille, and Toussaint (2018) found that embodied energy in EU imports reached 81% of direct final energy consumption, largely offsetting domestic efficiency gains.

The Breakthrough Institute identified 32 countries that achieved absolute decoupling of both territorial and consumption-based CO₂ emissions from GDP since 2005. This is genuine, but it concerns CO₂, not energy. The distinction is critical. CO₂ can decouple through fuel switching, gas replaces coal, renewables replace gas, without any reduction in total energy throughput. The competitive viability framework does not predict that carbon emissions are permanently locked to wealth. It predicts that energy consumption is. Decarbonisation of the energy supply, reducing the carbon intensity, is physically possible. Reducing the power–assembly coupling Γ(t) under current competitive conditions is not, absent either wealth destruction or a fundamental shift in the selection landscape.

The inter-sphere gradient

The competitive dynamics established in this section, agents competing for shares of a common energy gradient, efficiency gains recycled into expansion, gradient saturation as a geometric consequence of competitive allocation, are not confined to the technosphere's internal economy. The technosphere and the biosphere are both dissipative structures drawing on the same planetary energy budget within the same radiative boundary. They share surface area, material inputs, and the atmospheric and oceanic systems that regulate the conditions under which both operate. The maximum power principle and competitive exclusion apply at this inter-sphere level with the same force they exert within the technosphere.

Land-use conversion is competitive exclusion operating at the ecosystem level: forest cleared for agriculture or industry is gradient share transferred from Σ_bio to Σ_tech. Ocean acidification and eutrophication are chemical interference with the biosphere's metabolic pathways, the equivalent, at the inter-sphere level, of one firm disrupting a competitor's supply chain. Habitat fragmentation reduces the biosphere's capacity to maintain its own assembly stock, just as market consolidation reduces the viability of smaller competitors. The mechanism is the same. The scale is different. This is the essay's own Tier 2 dynamics operating at the biosphere–technosphere interface, not an ecological concern imported for moral effect.

The Jevons mechanism already derived in this section operates with particular force at this interface. The Green Revolution is the canonical example: improvements in agricultural efficiency, higher yields per hectare, did not reduce the total biospheric footprint of agriculture. They enabled population growth, dietary upgrading, and the conversion of land nominally "freed" by higher yields to other technospheric uses. Precision agriculture, genetically modified crops, and industrial aquaculture repeat the pattern. Each efficiency gain expands the technosphere's reach into biospheric territory rather than reducing it. This is the same rebound dynamic established above for energy efficiency, operating at the ecological boundary. The Jevons mechanism predicts that technospheric efficiency improvements in biological resource extraction will not voluntarily reduce biosphere displacement. They will accelerate it, because competitive selection rewards the agents that convert the efficiency gain into expansion.

The critical consequence of inter-sphere competition is the burden-transfer mechanism. When the technosphere displaces biospheric assembly, it does not simply remove Σ_bio from the planetary ledger. It inherits a maintenance obligation. The biosphere provides services; atmospheric regulation, water purification, soil formation, nutrient cycling, pollination, flood attenuation, climate stabilisation, that the technosphere depends upon for its own functioning. These services are powered by P_bio at the biosphere's low composite coupling Γ_bio.

When those services degrade, the technosphere must replace them with engineered infrastructure: desalination plants for water purification, synthetic fertiliser production for nutrient cycling, mechanical pollination systems, engineered flood defences, direct air capture for carbon removal. Each replacement adds to Σ_tech. Each addition draws maintenance power at Γ_tech, which is orders of magnitude higher than Γ_bio. The quantitative basis for this inequality, Σ_bio ≫ Σ_tech and Γ_bio ≪ Γ_tech, is established in the assembly stock analysis.

The net thermodynamic effect is doubly adverse. Total planetary assembly stock decreases, because the technosphere cannot replace biospheric assembly one-for-one at equivalent depth; the biosphere's assembly is the product of 3.5 billion years of evolutionary optimisation, and engineered replacements are crude functional substitutes, not assembly-equivalent reconstructions. Simultaneously, total power demand increases, because the replacement assembly operates at the technosphere's far higher Γ_tech. What the biosphere maintained at its own metabolic expense, powered by solar-driven photosynthesis outside the technospheric energy budget, the technosphere must now maintain at its own. The planet becomes simultaneously less complex and hotter. This is the thermodynamic signature of a self-defeating competitive strategy.

Stated precisely: biosphere displacement transfers maintenance obligations from a low-Γ dissipative structure to a high-Γ one. The transfer accelerates the approach to the waste heat ceiling, not by increasing the assembly stock the technosphere wants to maintain, but by increasing the assembly stock it must maintain just to preserve its existing functional capacity.

The consequence for the viability analysis is direct. The waste heat ceiling computed in the preceding part is an outer bound, the constraint surface that binds last, derived from the Stefan–Boltzmann relation alone. The actual viability kernel sits strictly inside this outer bound. Biosphere degradation contracts the kernel from within: it raises the maintenance floor by transferring obligations to the high-Γ technosphere without raising the waste heat ceiling, which is set by physics. The distance between the current state and the kernel boundary is therefore shorter than the purely thermodynamic model suggests. The "zone of false security" identified in the viability analysis is wider than the waste-heat-only calculation implies. The viability formalism makes this precise; here the point is qualitative. The crude hard ceiling is not the binding constraint. The actual constraint set is tighter, and biosphere displacement through the burden-transfer mechanism is one of the processes that tightens it.

The ontological status of the ratchet

This section has presented the maximum power pattern as the first and hardest mechanism driving civilisation toward the waste heat ceiling. The empirical evidence; Lotka's original observation, Odum's ecological confirmation, Garrett's civilisational-scale data, Hanley's historical extension, the Jevons record across lighting, industry, and computation, is extensive and convergent. The formal derivation from competitive viability geometry, developed in the viability analysis, establishes the pattern as a corollary of competition on a bounded gradient rather than an ecological axiom imported into a thermodynamic argument.

The maximum power principle, so derived, belongs to the second tier of the three-tier ontology: harder than human conventions (Tier 3, which can be reformed by institutional action), but softer than fundamental thermodynamics (Tier 1, which cannot be altered by any action). Its stiffness comes not from physics but from game theory: it is a competitive equilibrium, and departure requires coordinated multi-agent action, not unilateral reform.

The distinction matters for what follows. A civilisation could, in principle, override the maximum power attractor. It could deliberately choose to operate below maximum power; to forgo growth, to ∩ its energy throughput, to accept competitive disadvantage in exchange for long-term viability. This would require overriding competitive selection: ensuring that no sub-unit of the civilisation, no firm, no nation, no technology, defects from the collective restraint to capture a competitive advantage. It would require, in other words, a cooperative game solution of a kind that has no precedent in the history of any species, any ecosystem, or any civilisation on this planet.

The maximum power principle is not a cage from which escape is forbidden by physics. It is a current so strong that no swimmer has yet reached the far bank. The Second Law says the river exists. The competitive viability result says which way it flows. Whether civilisation can swim against it is an open question, but one whose answer depends on mechanisms (institutional, financial, political) examined in the amplifiers analysis, and whose formal structure is analysed in the viability framework.

What is not open is the consequence of failing to override it. A civilisation that remains on the maximum power trajectory will reach the waste heat ceiling within the timeframe conditioned on the current growth rate and coupling, a timeframe that is scenario, not physics. The evolutionary ratchet does not create the ceiling. The Second Law and the Stefan–Boltzmann relation do that. What the ratchet does is ensure that the trajectory points at the ceiling rather than away from it. It is the answer to the question posed at the opening of Part 3: why does civilisation not simply choose to stabilise within the viable region?

It does not choose to stabilise because stabilisation is not selected for. Growth is. Every efficiency improvement that could, in principle, reduce the power demand of the existing stock is instead captured by competitive dynamics and recycled into further expansion of the stock. The result is a system that moves toward the ceiling with the same statistical inevitability with which a gas expands into a vacuum. Individual molecules could, in principle, all move in the same direction and compress spontaneously. They do not, because the statistics of large numbers overwhelm any coordinated fluctuation. Individual agents within civilisation could, in principle, all agree to limit their energy throughput. They do not, because the statistics of competition overwhelm any uncoordinated restraint.

The analogy to gas expansion is more exact than it first appears. A gas fills its container not because molecules "choose" to disperse, but because dispersed configurations occupy overwhelmingly more of the accessible phase space than concentrated ones. The system does not select for dispersal. The geometry of the accessible region ensures it. Competitive dissipative structures saturate their available energy gradient for the same geometric reason: configurations that leave gradient uncaptured occupy less of the viable state space than configurations that capture it. The viability analysis makes this precise: the viable region under competitive allocation is geometrically larger for gradient-saturating configurations, and perturbations redistribute state vectors into the regions of largest measure. No selection mechanism is required beyond the geometry itself.

The next section examines the second mechanism: the inertia of the accumulated stock, which ensures that even if the ratchet could be overridden, the system's response time is measured in decades to centuries, far longer than the decision timescales of any existing institution. The integral structure of the stock, Σ = ∫₀ᵗ [C(τ) − δΣ(τ)] dτ is what makes the viability kernel path-dependent and the state vector slow to redirect.

3.2 Inertia and the Integral

The previous section documented the empirical pattern: civilisation's trajectory toward the waste heat ceiling is consistent with evolutionary dynamics selecting for configurations that maximise energy throughput, with the Jevons Paradox as its observable signature. The ratchet turns in one direction. Efficiency gains are captured by competitive selection and recycled into further growth. Why this pattern is so robust, why it resists coordinated override, is a question the viability analysis will address formally. The observation itself is what matters here: the ratchet turns, and overriding it has no precedent at civilisational scale.

But suppose it could be overridden. Suppose a civilisation achieved what no population of competing dissipative structures has ever achieved: a collective decision to redirect its trajectory away from maximum power. How fast could the redirection take effect?

The answer is: slowly. Generationally slowly. The assembly stock Σ is not a flow variable that responds immediately to changes in policy or intent. It is a stock variable, an integral, and integrals have inertia. The energy demand of the present is not set by the decisions of the present. It is set by the accumulated decisions of the entire past.

This is the second mechanism locking civilisation onto its trajectory, and it is ontologically distinct from the first. The evolutionary ratchet explains why the system drives toward the ceiling. Inertia explains why it cannot turn quickly, even in the hypothetical case where the will to turn exists.

The integral structure

Recall from the assembly derivation (Section 2.1) the dynamics of the assembly stock:

where C(t) is the construction rate, new joining operations performed per unit time, and δ is the aggregate decay rate imposed by the Second Law. The assembly stock at any time t is therefore:

where Σ₀ is the initial stock. This is a convolution integral with an exponential kernel of time constant τ_δ = 1/δ. Every increment of past construction contributes to the present stock, weighted by how much of it has survived decay. Recent construction contributes almost fully. Ancient construction contributes only insofar as it has been continuously maintained, replaced component by component, like the ship of Theseus, with the network node persisting even as the physical substrate turns over.

This integral structure has a geometric consequence that the viability analysis will make precise. Because Σ(t) is a cumulative quantity, the weighted sum of all past construction, it cannot be instantaneously redirected. The current value of Σ constrains the reachable set at t + Δ t: the achievable velocity dΣ/dt is bounded by the construction capacity C(t) and the decay rate δ. In the language of the viability framework developed in the formal analysis, this means the viability kernel is path-dependent. A civilisation deep inside the kernel, with low Σ relative to Σ_max and ample construction margin, has a wide velocity set and many admissible trajectories. A civilisation near the kernel boundary has a narrow velocity set and few. The integral acts as a geometric constraint on the rate of state-space traversal, and the formal apparatus of the viability analysis provides the precise characterisation of how this constraint shapes the family of admissible trajectories.

In Garrett's economic framework, the corresponding quantity is cumulative wealth:

where Y is annual real GDP, the monetary proxy for the construction rate C(t). The rate of change of W is governed by:

The physical coupling P = Γ(t) · Σ maps to Garrett's empirical finding P ≈ λ W through the conversion factor κ, where λ = Γ(t)/κ. Today's power demand is set not by today's production but by the integral of all past production, discounted by decay. Garrett, Grasselli, and Keen (2022) called this "the long arm of history." The phrase is precise. Every road poured, every hospital opened, every shipping lane charted, every city electrified reaches forward through time to grasp the energy budget of the present.

The time constant of this integral is:

where α = Y/W is the rate of return on accumulated wealth, empirically averaging approximately 2.3% per year. This gives τ ≈ 43 years. But τ measures the ratio of stock to flow, the time current production would take to reproduce the entire stock from scratch. The effective inertial timescale is longer, because W is dominated by the cumulative contribution of the last several time constants: roughly 50–100 years of accumulated structure. This is the timescale on which civilisation's thermodynamic character can change. It is measured in generations, not electoral cycles.

Limiting cases

The inertia becomes concrete when examined at the boundaries.

Case 1: Zero net construction. Suppose civilisation agreed to cease all net addition to the assembly stock, to produce only enough to offset decay, so that C(t) = δΣ and dΣ/dt = 0. In Garrett's economic terms, this is the steady-state condition Y = δ W. No new roads. No new factories. No new networks. Only replacement of what entropy has degraded.

Even in this scenario, the power demand P = Γ(t) · Σ remains at its current value indefinitely. The maintenance burden of the existing stock is not discretionary. It is the thermodynamic cost of continued existence at the current level of complexity. A civilisation that stops growing does not stop consuming energy. It consumes exactly as much as before — every watt directed to fighting the Second Law on the structures already built.

This is the steady-state economy as envisioned by Daly (1977), and its thermodynamic content is stark. Steady state is not low energy. It is current energy, sustained indefinitely, with every joule devoted to maintenance and none to expansion. If the current power throughput already places the system on a trajectory toward the waste heat ceiling, and the ceiling derivation (Section 2.3) established that at approximately 20 TW it does, on a timescale of centuries, then steady state merely converts an exponential approach into an asymptotic one. The ceiling is still there. The maintenance cost still generates waste heat. The system stabilises at a fixed distance from the ceiling rather than accelerating toward it, but only if 20 TW of continuous dissipation is compatible with long-term habitability. At current levels, it is. At the levels implied by several more decades of growth before stabilisation, the margin narrows considerably.

Case 2: Zero production. Now consider the opposite extreme, a theoretical limit, not a policy proposal. If all production ceased entirely (C(t) = 0, equivalently Y = 0), the assembly stock would not reset. It would decline exponentially at rate δ:

Through the coupling P = Γ(t) · Σ, power demand would track this decline. But δ is of order 1–2% per year for the global aggregate (Garrett, 2014). At 2% annual decay, the stock halves in approximately 35 years. Power demand halves with it. But "halves" is not "disappears." After a full decade of zero production, more than 80% of the original assembly stock, and 80% of the original power demand, persists.

This mathematical limit is somewhat misleading in isolation, because a world with Y = 0 is a world with no food production, no water treatment, no electricity generation. In practice, the withdrawal of production does not produce a graceful exponential descent. It produces cascading failure. The interdependencies within Σ, the fact that hospitals require electricity, which requires fuel supply chains, which require functioning ports, which require maintained channels, mean that decay in one subsystem accelerates decay in others. The aggregate δ is itself a function of the system's coherence. In an intact civilisation, coordinated maintenance holds δ to its baseline value. In a fragmenting one, δ spikes. The historical examples are instructive: the collapse of the Western Roman Empire, the Black Death, the dissolution of the Soviet Union. In each case, energy consumption fell not because anyone chose to reduce it, but because the civilisational structure that demanded it was physically dismantled or depopulated, and the effective decay rate rose sharply.

Case 3: Sustained contraction. Between these extremes lies the most policy-relevant scenario: a deliberate, managed reduction in the construction rate C(t) such that Σ declines at a controlled rate. Suppose global GDP fell by 50% and was sustained at that level for a full decade. The integral W would be reduced by the difference between actual cumulative production and what would have been produced under the counterfactual, a substantial absolute figure, but a modest fraction of W itself, because W has been accumulating for centuries. A rough estimate: current W is of order 3 × 10^15 2019 USD. A decade of halved GDP (approximately 5 × 10^13 per year foregone) would reduce W by roughly 5 × 10^14 about 17% of the total. Through the coupling, power demand would fall by 17%. The system would still demand more than 80% of its current energy throughput after a decade of economic depression worse than any in recorded history.

The point is structural. The energy demand of civilisation has a time constant measured in decades to centuries. Policy operates on the timescale of the flow Y. The constraint operates on the timescale of the stock Σ. The mismatch is not political. It is mathematical.

Temporal batteries

The integral argument reveals a strategic dimension to the planet's remaining exergy reserves. Fossil fuels and fissile materials; coal, oil, gas, uranium, thorium, are not merely energy sources. They are temporal batteries: intermediate exergy reservoirs storing gradients accumulated over geological and stellar timescales. Coal seams represent hundreds of millions of years of photosynthetic capture, compressed and preserved. Uranium ores embody nucleosynthetic processes in long-dead stars. These reserves constitute a one-time endowment of low-entropy potential, irreplaceable on any timescale relevant to civilisation.

Their strategic significance lies not in their total magnitude but in the timing of their exploitation. The integral structure of Σ means that the trajectory through state space, the path civilisation traces toward or away from the viability kernel, is shaped by the cumulative sequence of construction decisions, not by the instantaneous rate. A civilisation that exhausts its temporal batteries during an uncontrolled exponential growth phase has committed the resulting stock integral: the assembled matter exists, demands maintenance at P = Γ(t) · Σ, and cannot be wished away. The degrees of freedom for subsequent course correction narrow with every increment of Σ built using non-renewable gradients without a viable replacement supply chain in place. Conversely, a civilisation that deploys these reserves strategically, financing the space-industrial bootstrap required for radiating area expansion, or funding the coordination institutions necessary for Tier 2 parameter adjustment, retains greater manoeuvrability within the viability kernel. The temporal batteries are steering reserves. Burning them for undirected growth is the thermodynamic equivalent of spending seed corn. The full inventory; nuclear, chemical, thermal, and gravitational, and the allocation calculus against the viability corridor are developed in the closing analysis (§5.4).

The decay spectrum

Different components of Σ decay at vastly different rates. Heavy civil infrastructure; aqueducts, bridges, dams, rail beds, depreciates slowly, on timescales of 50–100 years. Roman aqueducts still carry water after two millennia. Industrial plant operates for 20–40 years. Digital infrastructure, servers, fibre optics, software systems, turns over on cycles of 3–7 years. A smartphone becomes e-waste in three. Human capital, in the form of tacit knowledge and institutional memory, can be lost in a single generation if not actively transmitted.

This spectrum has a consequence that strengthens the inertia argument rather than weakening it. As civilisation shifts toward higher-turnover components, as the digital and service economy grows relative to heavy industry, the weighted-average δ rises. A higher δ means a larger fraction of each year's construction C(t) must be devoted to replacing what has degraded, leaving less available for net accumulation. The system must run faster merely to stay in place. This is Tainter's (1988) diminishing returns to complexity, restated in the language of assembly dynamics: the more sophisticated and rapidly evolving the structure of Σ becomes, the greater the maintenance power it demands per unit of stock.

The implication is counterintuitive. A civilisation that shifts from concrete and steel to silicon and software does not become "lighter" in any thermodynamic sense. It may become lighter per unit of GDP, the energy intensity of new production may fall, but the maintenance metabolism of the total stock accelerates. The power-assembly coupling Γ(t) captures the aggregate relationship; the internal composition of Σ determines how hard the system must work to prevent its own dissolution. A high-δ civilisation is not more flexible than a low-δ one. It is more fragile, in the precise sense that interruptions to its energy supply produce faster structural degradation.

Carbon lock-in is assembly lock-in

The climate policy literature has developed the concept of "carbon lock-in", the observation that fossil fuel infrastructure, once built, tends to operate for its full economic lifetime. A coal plant commissioned in 2024 will likely burn coal until 2054 or beyond. A gas pipeline laid today constrains energy choices for 40–60 years. Unruh (2000) formalised this as "techno-institutional lock-in," and the concept has been central to the argument for early action on emissions.

The observation is correct but narrow. It identifies a symptom while missing the underlying condition. In the framework of this essay, carbon lock-in is a special case of assembly lock-in, the general principle that every structure added to Σ creates a maintenance obligation that persists for the lifetime of the structure.

The deeper lock-in operates through the integral. Every dollar of real production generated during the fossil-fuel era added to W. That accumulated W now demands maintenance power P = Γ(t) · Σ regardless of what energy source delivers it. Even a complete transition to zero-carbon energy, nuclear, solar, wind, satisfies only the carbon constraint. It does not reduce P, because P is coupled to Σ, not to the carbon intensity of the energy supply. The civilisational structure built by fossil fuels demands the same wattage whether those watts come from coal or from fission.

This distinction is routinely collapsed in mainstream energy analysis. The International Energy Agency publishes scenarios in which total final energy consumption peaks and declines through efficiency gains alone, while GDP continues to grow. Within the Garrett framework, this is physically impossible unless the power-assembly coupling Γ(t) itself changes, an event for which there is no empirical evidence across five decades of data. What efficiency gains actually do, as the maximum power analysis documented, is accelerate the growth of W by enabling more production per unit of energy, which then feeds back through the integral to demand still more energy.

Decarbonisation changes the source of the energy. It does not and cannot change the quantity, unless the financial architecture is simultaneously reformed, a point that the amplifiers analysis (Section 3.3) will examine. The climate policy discourse treats the problem as one of substitution: replace dirty joules with clean joules and the problem is solved. The thermodynamic framework reveals substitution as necessary but insufficient. The clean joules must still be produced, in quantity P = Γ(t) · Σ, and W grows with every year of positive production. Solving the carbon problem without solving the growth problem merely converts the greenhouse constraint into the waste heat constraint, trading a near-term crisis driven by atmospheric chemistry for a longer-term crisis driven by the Second Law.

Time constants and transition speed

The inertia quantified above sets a hard floor on how fast civilisation can change its thermodynamic trajectory without involuntary collapse. The relevant comparison is between the system's time constant and the time constants of the institutions attempting to steer it.

The assembly stock turns over on a timescale of τ_δ = 1/δ ≈ 50–100 years. The wealth integral responds to changes in its integrand Y with a lag proportional to τ = W/Y ≈ 43 years. Atmospheric CO₂ has a residence time measured in centuries. Ocean thermal mass equilibrates on timescales of decades to centuries.

Against these, the control timescales: a parliamentary term is 4–5 years. A corporate planning horizon is 5–10 years. An infrastructure investment cycle is 20–30 years. The Paris Agreement targets 2050, approximately 25 years from its signing. All of these are significantly shorter than τ. They operate on the timescale of the flow Y, not the stock Σ. Policy can modulate the rate of new construction. It cannot rapidly alter the integral of all past construction.

The rate of energy source substitution is further constrained by the growth rate of the total energy supply. Civilisation cannot simultaneously grow W (which demands more P) and reduce the fossil fraction of P unless the clean fraction grows faster than total P. Given that total P has historically grown at approximately 2.3% per year, and given the capital intensity and deployment timescales of nuclear and renewable infrastructure, the required growth rate of clean energy to achieve meaningful substitution while maintaining economic growth is extraordinarily high. Smil's (2010; 2017) extensive historical work on energy transitions documents that past transitions; wood to coal, coal to oil, oil to gas, each required 50–70 years to reach majority share, precisely because of the infrastructure stock embedded in Σ.

This is not a counsel of despair. The rate of renewable deployment has exceeded most projections. Solar photovoltaic costs have fallen by a factor of roughly 100 since 2000. But the thermodynamic framing insists on a distinction that the policy discourse elides: the question is not whether clean energy can be deployed fast enough to reduce the carbon intensity of the energy supply (it may), but whether it can be deployed fast enough to replace the total energy throughput demanded by an exponentially growing W (it almost certainly cannot, absent changes to the growth dynamics themselves). The first is a race against the greenhouse constraint. The second is a race against the waste heat ceiling. They operate on different timescales, and the second is the one with no engineering workaround.

The setup for what follows

The inertia of the assembly stock interacts with the evolutionary ratchet to produce a system that is doubly locked. The ratchet ensures that the trajectory aims at the ceiling, competitive selection drives maximum power throughput. The integral ensures that the trajectory cannot turn quickly, the stock responds to changes in the flow with lags measured in decades. Together, they describe a physical system whose behaviour is largely determined by its own accumulated past, responsive to present decisions only at the margins and only with substantial delay.

But the physical mechanisms described in this section and the previous one do not, by themselves, explain the full rate of approach to the ceiling. They explain why civilisation tends toward maximum power and why it cannot rapidly change course. They do not explain why the rate of approach accelerates, why the construction rate C(t) does not simply match the pace that the evolutionary ratchet and the available energy gradient would naturally produce, but is amplified beyond it by institutional mechanisms that pull future exergy into the present.

Those mechanisms; endogenous money creation, compound-interest debt, and the governance structures that operate on timescales mismatched to the physics, are the subject of the amplifiers analysis. They are ontologically distinct from everything discussed so far. The Second Law is physics (Tier 1). Lotka's principle is an empirical pattern whose formal basis the competitive viability analysis (§4.3) will establish as a competitive equilibrium result (Tier 2). The integral structure of Σ is a mathematical consequence of accumulation, inheriting its character from the Tier 1 maintenance obligation. The financial system is a human convention (Tier 3). It is real, consequential, and deeply entrenched, but it is, in principle, subject to reform in ways that the Second Law and Lotka's principle are not.

3.3 The Amplifiers: Money, Debt, and Governance

Everything in this section is about human institutions. Not one claim that follows depends on the Second Law, on Lotka's principle, or on any result from physics. The mechanisms described here are powerful, consequential, and, in the lived experience of every person on Earth, utterly real. But they are conventional. They were invented. They can, in principle, be reformed, restructured, or abolished. Debt has been forgiven, repeatedly, throughout recorded history. Monetary systems have been constructed, operated for decades, and then abandoned. The gold standard was adopted and discarded. The Bretton Woods architecture lasted three decades before being dismantled by executive decision. Negative real interest rates have persisted for extended periods. Periodic debt jubilees were a structural feature of Mesopotamian civilisation for two millennia, not a crisis response but a routine institutional practice (Hudson, 2018).

This matters because the previous two sections identified mechanisms that are not conventional. The maximum power ratchet operates through the statistics of competition among dissipative structures, it is physical in origin, even though it acts through the medium of complex adaptive systems. The inertia of the integral argument is a mathematical property of the assembly balance, it follows from the stock-flow structure of Σ and cannot be legislated away. These mechanisms would drive civilisation toward the waste heat ceiling even if every institution described in this section were dismantled overnight.

What the financial and governance architecture does is accelerate the approach. It converts a physical tendency into a compulsion. It adds a second engine to a vehicle that was already moving in one direction. The trajectory toward the ceiling exists without compound interest. With compound interest, the trajectory steepens. The distinction is not academic. It determines which interventions are possible and which are futile, a classification that the viability analysis (Section 4.2) will formalise as the boundary between hard and soft constraints in the viability kernel.

Four amplifiers deserve analysis: endogenous money creation, which acts as a temporal pump pulling exergy from the future into the present; compound interest on the resulting debt stock, which imposes a growth floor on the physical economy; the Minsky cycle, which reconciles virtual and real wealth through periodic crisis; and the governance frequency mismatch, which prevents the control vector from operating at the timescale required to modulate the trajectory. Following their analysis, the empirical evidence for their aggregate effect, drawn from the work of Garrett (2011, 2012, 2015) and Hanley (2025), is examined.

The Temporal Pump: Endogenous Money Creation

The standard textbook account of banking, that banks intermediate between savers and borrowers, lending out deposits received, is wrong. This is not a heterodox claim. The Bank of England stated it without equivocation in its 2014 Quarterly Bulletin: "Whenever a bank makes a loan, it simultaneously creates a matching deposit in the borrower's bank account, thereby creating new money" (McLeay, Radia, and Thomas, 2014). The process is endogenous. Money is not a fixed stock parcelled out by central authority. It is created by the act of lending and destroyed by the act of repayment.

The thermodynamic implication is immediate. When a bank issues a loan of magnitude L, it creates purchasing power that did not previously exist. The borrower can now command labour and materials, can now direct exergy flows, to construct new physical structure: a factory, a power station, a housing estate, a data centre. Each addition increases Σ, the assembly stock. Through the maintenance coupling established in the maintenance derivation (Section 2.2), each addition permanently increases the power demand of civilisation by

where ΔΣ is the increment to the assembly stock financed by the loan, and Γ(t) = μδξ/η_II is the composite coupling of metabolic multiplier, decay rate, specific exergy cost, and Second Law efficiency. The energy commitment is immediate and, barring decay, perpetual.

But the loan does not merely create a one-time addition to Σ. It simultaneously creates an obligation: the borrower must repay L plus interest over the loan's maturity. If the annual interest rate is r and the maturity is n years, the total repayment under compound interest is

For the borrower to meet this obligation without default, sufficient economic output Y must be generated to service the debt. Since output requires energy; Keen, Ayres, and Standish (2019) demonstrated that energy carries a production exponent of approximately 0.3, an order of magnitude above the 0.03–0.07 assigned by standard Cobb–Douglas estimation, the repayment obligation R is itself a claim on future exergy flows.

This is the temporal pump. Credit creation pulls exergy from the future into the present. It creates a vacuum, immediate purchasing power, immediate command over energy and materials, by drawing against production that has not yet occurred. The mechanism is extraordinarily useful. It solves a genuine coordination problem: it enables investment in long-gestation projects, power grids, transport networks, industrial plant, that no individual could fund from accumulated savings. The entire industrial revolution, and every major economic expansion since, has been financed by this temporal displacement. Every civilisation that has discovered credit has grown faster than those that did not, which is itself a Lotka-type selection result operating at the institutional level: credit-using civilisations outcompete non-credit civilisations because they can mobilise exergy faster.

But the pump has a cost that appears on no balance sheet. Each loan commits the future to being larger, in energetic terms, than the present. Each act of credit creation is a bet that the economy will grow sufficiently to generate the surplus exergy needed for repayment. When that bet fails at scale, the result is not merely financial. It is thermodynamic. The claims on future energy cannot be honoured because the future energy does not exist.

Frederick Soddy, the Nobel laureate in chemistry who turned to economics in the 1920s, identified this structural tension with a clarity that has not been surpassed:

Debts are subject to the laws of mathematics rather than physics. Unlike wealth, which is subject to the laws of thermodynamics, debts do not rot with old age and are not consumed in the process of living. On the contrary, they grow at so much per cent per annum. (Soddy, 1926)

Real wealth, the assembly stock Σ, decays at rate δ. It is subject to the Second Law. Virtual wealth, the financial claims denominated in money and debt, obeys no such law. A dollar deposited at compound interest doubles, redoubles, and grows without bound according to the mathematics of exponential functions. There is no friction term, no decay rate, no entropy production in this process. The numbers simply grow.

The two systems are yoked together by the assumption that financial claims can always be redeemed against physical assets. That assumption holds so long as the physical economy grows at least as fast as the financial claims upon it. When it does not, the claims become orphaned, directed at goods, services, and energy that do not exist and cannot be produced. The result is inflation (the monetary system devaluing claims to match physical reality), default (claims destroyed outright), or crisis (some combination, typically accompanied by significant human suffering). Herman Daly recovered Soddy's economic thought in a landmark article in History of Political Economy (Daly, 1980), and Steve Keen has spent three decades formalising the dynamics (Keen, 2011; Keen, 2017).

The crucial observation for this essay is that the temporal pump is a convention, not a law of nature. A civilisation could, in principle, operate without endogenous money creation. Many pre-modern civilisations did. Islamic finance prohibits interest-bearing debt entirely and substitutes equity participation and profit-sharing (Usmani, 2002). Full-reserve banking proposals, from Soddy himself through Irving Fisher's "100% Money" (Fisher, 1935) to the Chicago Plan Revisited (Benes and Kumhof, 2012), would eliminate the money-creation function of commercial banks. The fact that credit creation is dispensable in principle does not mean it is easily dispensed with. It means it belongs in the category of soft constraints, constraints whose position in the viability space depends on institutional parameters that humans can change. The distinction between "dispensable in principle" and "easily removed" is critical and will be formalised in the viability analysis.

The Growth Floor: Compound Interest and the Solvency Condition

The temporal pump creates debt. Compound interest makes debt grow. Together, they impose a minimum growth rate on the physical economy, a floor below which the financial system destabilises and the coordination layer that enables complex economic activity disintegrates.

Let D(t) represent the aggregate stock of debt across the global economy, and let r represent the average real interest rate. If serviced without net repayment, the debt stock grows as

This is exponential growth with doubling time ln2/r. At a real interest rate of 3%, the doubling time is approximately 23 years. At 5%, roughly 14 years.

Now compare this to the growth dynamics of the assembly stock, governed by the assembly balance:

where C(t) is the construction rate and δ is the decay rate. For the financial system to remain solvent, for virtual wealth to remain redeemable against real wealth, the growth rate of real wealth must at minimum match the growth rate of financial claims. Since the monetary proxy W tracks Σ through the relation W ≈ κΣ (where κ is the monetary valuation per unit of assembly), the solvency condition can be stated:

This inequality demands that annual economic output exceed the sum of the real interest rate and the physical decay rate, multiplied by the entire accumulated wealth stock.

The equation contains two terms of radically different ontological status. This difference is the sharpest illustration of the three-tier framework in the entire essay.

The decay rate δ is Tier 1: hard physics. It is the Second Law operating on assembled matter. No legislature can repeal it. No central bank can defer it. No restructuring agreement can reduce it. Concrete carbonates at the rate concrete carbonates, regardless of what the Federal Reserve does with interest rates. The term δ W represents the irreducible thermodynamic tax on civilisation's existence, the maintenance power that must be supplied simply to prevent Σ from degrading. Even a civilisation that abolished every financial institution, cancelled every debt, and operated without money of any kind would still face the requirement Y ≥ δ W to maintain its physical stock.

The interest rate r is Tier 3: human convention. It is a parameter of a game that humans invented and can change. Debt can be forgiven. It has been, on scales ranging from individual bankruptcy to the London Debt Agreement of 1953, which wrote off roughly half of West Germany's external obligations and restructured the remainder on concessionary terms. Interest rates can be set to zero, as Japan demonstrated for two decades. They can be set below zero, as the European Central Bank and several Scandinavian central banks demonstrated in the 2010s. A civilisation that restructured its monetary system could in principle eliminate r from the solvency condition entirely, leaving only the physical floor Y ≥ δ W.

The solvency condition thus nests a reformable convention inside an immutable law. The compound term (r + δ) adds a Tier 3 parameter to a Tier 1 parameter, and the sum governs the system's growth requirement. Removing r;,through jubilee, restructuring, or architectural reform, reduces the growth floor to the thermodynamic minimum. Removing δ is not an option available to any civilisation operating within the Second Law. Building the central beam of a civilisational analysis on Y ≥ (r + δ)W, treating r and δ as though they have equivalent ontological status, would be a category error. They do not have equivalent status. One is a law of nature. The other is a rule of a game. This essay does not make that error. The solvency condition appears here, in a section explicitly demarcated as conventional, precisely because r is conventional. The load-bearing structure of the argument, the arena, the machine, and the first two sections of the trajectory analysis, depends on nothing but thermodynamics and evolutionary dynamics.

That said, the solvency condition is not arbitrary. It is deeply entrenched for reasons that are themselves partially thermodynamic. Credit creation solves a real coordination problem. Long-gestation infrastructure projects, the kind that build Σ most durably, require capital mobilisation on timescales and at magnitudes that savings alone cannot provide. Every civilisation that has discovered credit has outgrown those that have not. This is Lotka-type selection operating at the institutional level: the financial architecture that maximises the rate of exergy capture and deployment is the architecture that spreads. Compound interest emerged independently in Sumer, in Rome, in medieval Europe, and in modern capitalism, not because of a shared cultural inheritance but because it is the mathematical structure that most efficiently pulls future energy into present construction. The convention is not accidental. It is the institutional expression of the maximum power principle.

The practical consequences of the solvency condition are three. First, it imposes a growth floor on energy consumption that is independent of technology, policy preference, or environmental awareness. Since Y requires energy, through the production function established by Keen, Ayres, and Standish (2019) and the physical coupling P = Γ(t) · Σ, the condition Y ≥ (r + δ)W translates directly into a minimum energy flow that includes the additional exergy required to service debt obligations above the thermodynamic maintenance floor. The precise magnitude depends on the structure of the financial system, but its sign is always positive and it grows with the debt-to-wealth ratio D/W. Second, the growth imperative accelerates with scale. As Σ grows, the maintenance burden δΣ grows in absolute terms. As D accumulates, the interest burden rD grows in absolute terms. Both are proportional to their respective stocks, which means both increase even if their rates remain constant. The system is not merely growing; it is growing into an ever-larger commitment to future energy consumption. Third, the timescale mismatch identified at the close of the inertia analysis is structurally dangerous. The physical system has a time constant τ ≈ 1/δ of 50–100 years. Debt maturities are measured in years to decades. Compound interest operates on an annual or sub-annual cycle. The financial system can generate claims on future energy far faster than the physical system can deliver the energy to honour them.

The Minsky Correction as Thermodynamic Reconciliation

The dynamics of debt accumulation in a physically bounded economy are not smooth. They are punctuated by discontinuities, moments at which the gap between virtual wealth and real wealth closes violently. The economist who understood this best was Hyman Minsky, whose Financial Instability Hypothesis describes an endogenous cycle in credit markets (Minsky, 1986).

Minsky identified three phases. In the first (hedge finance), borrowers can service both interest and principal from operating income. The condition Y ≥ (r + δ)W is comfortably satisfied, with surplus available for net accumulation. In the second (speculative finance), borrowers can service interest but must refinance principal, relying on continued access to credit markets. The system is solvent but fragile: Y is approximately equal to (r + δ)W, with no margin. In the third (Ponzi finance), borrowers can service neither interest nor principal from income and rely entirely on asset price appreciation to remain solvent. The condition Y < (r + δ)W is violated; the gap is papered over by the creation of additional debt, which is to say, by pulling still more energy forward from an increasingly encumbered future.

The Minsky moment, the point at which confidence collapses and the Ponzi structure implodes, is in thermodynamic terms, the moment when the physical system can no longer supply the energy flow required by the financial commitments layered upon it. Virtual wealth snaps back into alignment with real wealth, destroying financial claims until the debt-to-wealth ratio returns to a level the energy supply can actually support. Keen formalised this in a series of models demonstrating that Minsky dynamics are endogenous to any system that combines interest-bearing debt with physically bounded production (Keen, 2011; Keen, Grasselli, and Garrett, 2022).

The 2008 financial crisis was a localised demonstration. Global debt-to-GDP ratios reached approximately 350% by 2024 (Institute of International Finance), a level sustainable only with continued real growth. But continued real growth requires continued energy growth through the maintenance coupling, which drives continued waste heat rejection, which tightens the temperature constraint established in the ceiling derivation (Section 2.3). The two constraints are anti-correlated on the timescale that matters: relaxing the debt constraint requires growth in Σ that accelerates the approach to the ceiling, while constraining total power to respect the ceiling requires a steady-state or contracting economy that is structurally incompatible with debt service under current financial architecture.

This is the signature of a system whose conventional constraints amplify the physical ones. The evolutionary ratchet drives the trajectory toward the ceiling. The debt architecture steepens the trajectory by requiring faster growth than the ratchet alone would produce. And when the steepened trajectory becomes unsustainable, the correction is not gradual adjustment but discontinuous collapse of the coordination layer, a Minsky moment that destroys effective Σ as the institutional architecture maintaining wealth coherence disintegrates.

Critically, none of this is inevitable in the way the Second Law is inevitable. The Minsky cycle is a property of a specific financial architecture, one built on endogenous money creation, fractional reserves, and compound interest. A different architecture would produce different dynamics. Whether those different dynamics would be sufficient to avoid the waste heat ceiling is a separate question, addressed in the implications analysis (Part 5). The point here is narrower: the financial system is a powerful amplifier of a trajectory that would exist without it, because Lotka's principle drives maximum power throughput regardless of whether compound interest exists. Debt makes the problem worse and faster. It does not create the problem.

The Governance Frequency Mismatch

The third amplifier is not financial but temporal. It concerns the relationship between the clock speed of political decision-making and the clock speed of the physical system those decisions are supposed to govern.

The civilisational time constant, the characteristic timescale over which the assembly stock Σ responds to changes in its drivers, is τ ≈ 1/δ, which for δ in the range of 1–2% per year gives τ on the order of 50–100 years. Atmospheric CO₂ accumulates with an effective residence time measured in centuries. Ocean thermal equilibration operates on comparable timescales. The waste heat constraint tightens monotonically with accumulated Σ, which itself responds to changes in C(t) with generational lag. These are the timescales of the physical system.

The political system operates on fundamentally shorter timescales. Electoral cycles in democracies run 2–5 years. The median tenure of a finance minister is under three years. Policy reversals, on carbon pricing, energy subsidies, infrastructure programmes, occur with each change of government. Corporate planning horizons are 5–10 years. Sovereign debt maturities range from months to 30 years, with the majority under 10.

The result is a catastrophic frequency mismatch. The control vector u(t) oscillates on a 2–5 year political cycle. The financial layer amplifies these oscillations through pro-cyclical credit dynamics on a 7–10 year cycle. But the state variables that determine civilisational viability, Σ, cumulative CO₂, atmospheric temperature, the waste heat loading, respond on 50–100 year timescales. The control signal oscillates orders of magnitude faster than the system it is attempting to steer.

In control theory, this pathology is well understood. A controller that oscillates faster than the plant's response time produces either no net effect (the oscillations average out) or destabilising resonance. In the civilisational case, the averaging-out interpretation is the charitable reading: successive governments cancel each other's carbon pricing, infrastructure programmes are announced and defunded, energy strategies are adopted and reversed, and the net perturbation to the multi-decadal trajectory is negligible. The physical system, Lotka's wheel, turning with the inertia of all accumulated history, barely registers the intervention.

The Australian carbon pricing mechanism provides a compact illustration. Introduced in July 2012 at A$23 per tonne of CO₂-equivalent, it covered approximately 60% of national emissions and was linked to the EU Emissions Trading System. It operated for two years. In July 2014, the incoming government repealed it. The entire policy cycle; design, legislation, implementation, operation, repeal, fell within a single electoral period. The atmospheric CO₂ concentration, which stood at approximately 394 ppm when the mechanism was introduced and 399 ppm when it was repealed, did not register the intervention at any detectable scale.

This is not a uniquely Australian phenomenon. The United States joined and withdrew from the Paris Agreement within a single presidential term. European carbon prices collapsed from approximately 30 to 3 euros between 2008 and 2013 due to over-allocation of permits, then took nearly a decade to recover. In each case, the policy instrument oscillated on timescales that the physical system cannot resolve.

Monetary policy faces the same structural limitation. Central banks target price stability and maximum employment on horizons of 1–3 years. Each adjustment implicitly modulates the rate of energy dissipation, lower rates stimulate borrowing, borrowing funds construction, construction increases Σ, and Σ demands maintenance power via P = Γ(t) · Σ. Higher rates do the reverse, but only within bounds set by the Minsky dynamics: raise rates too aggressively and the debt structure destabilises. The central bank is operating a thermostat for the heat engine without a thermometer that reads in the correct units. It adjusts the rate of energy dissipation using proxy signals, the Consumer Price Index, unemployment, purchasing managers' surveys, that are at best loosely coupled to the physical variables that determine system viability.

Fiscal policy, taxation, public expenditure, regulation, can in principle redirect the composition of exergy flows. A sufficiently high carbon tax shifts relative prices toward lower-carbon energy sources. Public investment in nuclear or renewable infrastructure accelerates the energy transition. Capital allocation rules can favour long-duration, low-carbon assets. These are real levers with real effects. But they operate on the composition ofP, not its magnitude. Recall that Γ(t)has not exhibited significant secular decline over fifty years of data, despite dramatic changes in the composition of the global energy system and the structure of economic output. Fiscal policy can change what kind of energy civilisation uses. The evidence examined in the following subsection suggests it cannot easily change how much, because the total is set by the accumulated stock Σ through a composite coupling that does not respond to price signals, tax rates, or regulatory frameworks on policy-relevant timescales.

The governance amplifier, then, is this: the decision-making architecture operates on timescales too short to modulate the trajectory, with instruments that act on the composition rather than the magnitude of energy throughput, within a financial system that penalises any attempt to reduce the magnitude. Each of these limitations is conventional. Electoral cycles are design choices. Central bank mandates are statutory constructions. Tax policy is legislation. None is a law of physics. All can be reformed.

But "can be reformed" is a weaker claim than it appears. The frequency mismatch is not merely a feature of democratic governance. It is a feature of any governance system operating within a competitive international order. A nation that unilaterally extends its planning horizon, that commits to multi-decadal energy policy at the cost of short-term economic competitiveness, faces adverse selection. Capital migrates to jurisdictions with shorter horizons and higher returns. Export markets are lost to competitors whose energy costs are lower because their environmental constraints are looser. This is Lotka's principle operating at the level of nation-states: the polity that maximises its rate of energy capture outcompetes the one that exercises restraint. International coordination could, in principle, resolve this, the Paris Agreement was an attempt, but coordination among 195 sovereign nations on timescales of decades, against the incentive structure just described, has no historical precedent of success.

The Empirical Signature: Garrett's Ratio and the Hanley Transition

The preceding subsections described the amplifiers in mechanistic terms. The question that follows naturally is whether the mechanisms leave an empirical trace, whether the lock-in they predict is visible in aggregate data. It is. The trace is precise, quantitative, and spans half a century.

The competitive viability equilibrium derived in Part 4 predicts that the aggregate coupling between power throughput and accumulated structure should be approximately stable wherever the system sits at or near gradient saturation. This prediction has an independent empirical confirmation. Timothy Garrett, working from a thermodynamic model of civilisation developed from 2009 onward, documented that total global primary energy consumption P is related to accumulated global economic wealth W by a ratio that is remarkably stable over the period 1970–2019 (Garrett, 2011, 2012, 2015; Garrett et al., 2022). Garrett's empirical finding P ≈ λW, where λ ≈ 5.9 mW per 2019 US dollar, is the monetary shadow of the physical relation P = Γ(t)·Σ derived in this essay, where λ = Γ(t)/κ. The ratio held approximately constant across four decades of radical structural transformation; the information revolution, China's industrialisation, the collapse of the Soviet Union, the 2008 financial crisis, the shale gas revolution, and the beginning of large-scale renewable deployment. Through all of it, the ratio of power consumption to accumulated wealth did not systematically shift. The variation was contained within an envelope of approximately ± 4%. The essay's derivation does not depend on this finding. The physical relation P = Γ(t)·Σ follows from the maintenance axioms (A1–A4) and the definition of Σ; the competitive viability analysis (Part 4) explains why Γ should be stable at equilibrium; Garrett's data provide the empirical corroboration.

In the framework of this essay, the ratioP/Wis not a fundamental physical constant. It is the observable aggregate of the composite Γ(t) = μδξ/η_II, modulated by the monetary conversion factorκthat translates between physical assembly (Σ) and its economic shadow (W). The stability of the ratio does not mean that each component of the composite is individually constant. It means that the product of the components is approximately stationary, that changes in one factor are compensated by changes in others within the envelope of measurement uncertainty.

Why should a composite of four independently varying physical and institutional quantities remain stable? The answer lies precisely in the amplifiers described in the preceding subsections. The debt-growth coupling, combined with endogenous money creation, creates a regime where the economy must grow at a rate sufficient to service existing obligations. That growth requirement translates directly into an energy requirement through the physical maintenance cost of the assembly stock. The financial system does not merely accompany civilisational energy use, it enforces a specific ratio between accumulated structure and power throughput by penalising any deviation with financial instability. When energy use falls below the ratio implied by existing debt commitments, the solvency condition Y ≥ (r + δ)W is violated, triggering the Minsky reconciliation described above. When energy use rises above it, the surplus is captured by further credit creation, which raises Σ and re-establishes the ratio at a higher absolute level. The system oscillates around the attractor, never straying far from it, because the financial architecture provides both a floor (debt service) and a ceiling (credit exhaustion) that bracket the ratio tightly.

The stability of the Garrett ratio is, in this reading, the signature of the financial amplifier operating at full lock-in. It is the empirical fingerprint of a system in which the institutional architecture has become sufficiently integrated, sufficiently global, and sufficiently self-reinforcing to enforce a uniform growth compulsion across the entire world economy.

The most detailed published challenge to the stability of this ratio appeared in a 2025 preprint by Brian Hanley, submitted to Earth System Dynamics. Hanley's contribution is, for the purposes of this essay, more valuable than Garrett's original finding, because it identifies precisely when the lock-in engaged.

Hanley assembled replicate datasets for energy consumption and GDP extending back to 1 CE and found that the ratio P/W diverges substantially from Garrett's values before approximately 1970. His energy estimates for the pre-industrial period are considerably lower than those in Garrett's supplementary dataset, which Hanley argued were back-projected from post-1970 equations rather than independently estimated. If Hanley's replicate data are correct, and his methodology is transparent and reproducible, then the ratio was not stable before the modern financial system fully integrated. It underwent a transition, converging to its current value sometime during the mid-twentieth century.

This is not a challenge to the argument of this essay. It is a confirmation of it.

The instability of the ratio before industrialisation is exactly what the three-tier ontology predicts. Before the modern financial system engaged, before global credit markets, before fiat currency, before the institutional apparatus that enforces the solvency condition across sovereign borders, the composite Γ(t)fluctuated because the institutional amplifiers were weaker, less globally integrated, and periodically disrupted by jubilees, defaults, and regime changes. The pre-industrial global economy was overwhelmingly biomass-powered, with energy flows tightly coupled to local ecology rather than to a globally integrated market. Every component of the composite would have had different values and different dynamics in this regime. The decay rate δwas dominated by biological and agricultural structures with shorter lifespans. The specific exergy costξwas set by pre-industrial joining processes. The Second Law efficiency η_II was far lower wind, water, and muscle rather than turbines and generators. And the metabolic multiplierμwas constrained by the limited capacity of pre-modern institutions to coordinate activity beyond the local.

The convergence to a stable ratio in the post-1970 period marks the point at which the global financial system became sufficiently integrated to enforce a uniform growth compulsion worldwide. The collapse of the Bretton Woods system in 1971, paradoxically, completed the integration by removing the gold constraint on credit creation and allowing the endogenous money mechanism to operate at full scale across the entire global economy. From that point forward, the financial amplifier was fully engaged. The composite Γ(t) stabilised, not because the physics changed, but because the institutional architecture locked the components into a regime where their product was held approximately constant by the self-reinforcing dynamics of debt-financed growth.

Garrett's later work (Garrett, 2015; Keen, Grasselli, and Garrett, 2022) explicitly connects the stability of the ratio to the structure of credit creation and debt service, which is precisely the mechanism described in the first half of this section. The data and the mechanism are consistent. Together, they establish that the financial architecture is not merely a surface phenomenon but a powerful amplifier that shapes the trajectory ofPand Σ on decadal timescales. The Garrett ratio is the thermometer reading. The amplifiers are the thermostat.

Hanley's conclusion, carefully stated, frames the matter correctly: "As presented, the long arm of history hypothesis is falsified. However, the thermodynamic argument in prior work is compelling, and in 1970, a radical change in slope of W/E occurs. From 1970 forward, the long-arm of history hypothesis as presented appears probable." The qualification matters. The "long arm of history", the claim that a single ratio has held since antiquity, does not survive Hanley's data. But the thermodynamic argument does not require it to. What the argument requires is that the current coupling is real, is driven by identifiable mechanisms, and is entrenched by institutional architecture that cannot be reformed without structural change to the global financial system. All three claims survive. Hanley's data strengthen them by identifying the institutional transition that brought the coupling into existence.

One further empirical refinement deserves note. Garrett's ratio treats all primary energy equally, but declining Energy Return on Investment means an increasing fraction of gross consumption is consumed by the energy sector itself. Brockway and colleagues (2019) calculated that global fossil fuel EROI at the finished fuel stage fell roughly 23% between 1995 and 2011, from approximately 8:1 to 6:1. Aramendia and colleagues (2024) extended this to the useful energy stage, finding fossil fuels' useful-stage EROI at approximately 3.5:1. The implication is that the stable ratio, measured against gross energy, may be absorbing a declining EROI within its envelope, the system must run faster on the gross energy treadmill to maintain the same net output. This does not refute the stability but refines its interpretation and makes the outlook more severe, not less: the effective coupling between net energy and accumulated structure may be tightening even as the gross ratio holds steady.

What the Amplifiers Do and Do Not Explain

The amplifiers, endogenous money creation, compound interest, the governance frequency mismatch, and the institutional lock-in whose empirical signature Garrett measured and Hanley dated, together convert a physical tendency into an institutional compulsion. The maximum power ratchet establishes that civilisation tends toward the configuration that maximises energy throughput. The amplifiers ensure that it must accelerate toward that configuration or face financial collapse, institutional crisis, and the disintegration of the coordination layer that maintains Σ.

The distinction between tendency and compulsion maps directly onto the three-tier ontology that this essay proposes and that the viability analysis formalises, specifically through the solvency-constrained admissible control set U_adm(Σ,D) derived in §4.2, which makes exact the difference between the physical maintenance obligation δ W(Tier 1) and the debt-service requirement rD(Tier 3).

The first tier is hard physics: the maintenance obligation δΣ, the waste heat ceiling Σ_max, the Landauer floor, the irreversibility of exergy destruction. These define the arena and cannot be altered by any intervention, physical, institutional, or cognitive. They are the geometry of the manifold within which civilisation exists.

The second tier is evolutionary dynamics: Lotka's maximum power principle, the Jevons recycling of efficiency into demand, the competitive selection among dissipative structures for energy throughput. These are physical in origin, they follow from the statistics of competition in a finite environment, but they operate through the medium of complex adaptive systems rather than through fundamental law. They explain why the system's trajectory tends toward the ceiling rather than away from it. They are extremely difficult to overcome but are not, in the strictest sense, laws of physics. A sufficiently coordinated collective action could in principle override them, in the same way that an organism can override metabolic imperatives through conscious choice, with difficulty, temporarily, and at a cost. No civilisation has achieved this at the required scale.

The third tier is human conventions: the monetary architecture, governance structure, property rights, and institutional design described in this section. These are rules of games that humans invented. They powerfully shape the rate and direction of the trajectory, but they can be changed by deliberate action. Debt is in this category. So is the structure of the corporation, the design of the tax system, and the length of the electoral cycle.

The stability of the Garrett ratio is an empirical fact about the current financial regime. It is not a law of thermodynamics. A civilisation with different monetary architecture would exhibit a different ratio. But changing the ratio requires changing the financial architecture, and the financial architecture is entrenched for reasons that are themselves partially thermodynamic, credit creation solves a real coordination problem and has been selected for by Lotka-type competition at the institutional level. The ratio is conventional in principle and tenacious in practice. The Hanley transition confirms its institutional character. The Garrett stability confirms its lock-in.

The critical conclusion is this: removing the amplifiers would slow the trajectory but would not eliminate the ceiling, the maintenance cost, or the evolutionary ratchet. A civilisation that restructured its financial system, that replaced compound-interest debt with equity participation, adopted full-reserve banking, extended governance timescales to match physical ones, and coordinated internationally to prevent competitive undercutting, would have removed the third-tier compulsion. It would still face the second-tier tendency. And it would still face the first-tier wall.

The question "but what if we reformed the financial system?" can now be answered with precision. Reform would remove the most powerful amplifier, slow the trajectory, and widen the viability kernel, the set of states from which survival remains possible. It would buy time. It would not buy escape. The predicament is softened but not resolved, because the predicament is not created by compound interest. It is created by the Second Law, the finite radiative surface, and the competitive dynamics of dissipative structures. The amplifiers make it worse and faster. The physics makes it real.

This distinction, between the merely difficult (institutional reform, which has ample historical precedent) and the physically impossible (repealing the Second Law, which has none), is the most important conceptual contribution of this essay. It is the basis on which the viability analysis assesses the viability of civilisation's current trajectory, and the implications analysis identifies which parameters most widen the kernel of survivable futures.

Part 4 — The Viability Problem

4.1 The Temporal Hierarchy of Constraints

The preceding parts established a set of constraints on civilisational existence. The maintenance floor P_maint = ( δξ/η_II ) · Σ sets the minimum power throughput below which the assembly stock decays. The waste heat ceiling Σ_max sets the maximum assembly stock the planet can thermally sustain. The evolutionary ratchet drives the trajectory toward that ceiling at maximum power. Inertia prevents the trajectory from being redirected on anything shorter than generational timescales. The financial amplifiers accelerate the approach. And the Landauer floor closes the dematerialisation escape route by demonstrating that information processing is physical, thermodynamically bounded, and, in its current trajectory, the fastest-growing component of civilisational power demand.

Each of these results was derived independently. What has not yet been established is how they interact temporally, which constraints bind first, which bind last, and what the consequences are of addressing them in the wrong order or failing to address them at all.

This matters because the constraints do not all operate on the same timescale. A civilisation that treats them as a single undifferentiated crisis will misallocate its finite control capacity. A civilisation that understands their temporal ordering can, in principle, buy time on the nearer constraints while building the structural capacity to address the further ones. The distinction between these two strategies is the difference between a managed trajectory through the viability kernel and an uncontrolled exit from it.

Decision windows and constraint bites

Two distinct temporal concepts govern each constraint. The decision window is the period within which action must be initiated to avoid future constraint violation, it concerns response time, not impact time. The constraint bite is the point at which the constraint actually becomes binding on the state vector, when the physical parameter crosses its threshold. These can differ by decades or centuries. Conflating them overstates the urgency of near-term constraints and understates the inescapability of long-term ones.

The constraints identified in this essay separate into four categories, each with its own decision window, constraint bite, and position in the three-tier ontology. They are ordered here by the urgency of their decision windows, because this ordering determines the sequence in which they must be addressed.

The greenhouse constraint

The most immediate decision window belongs not to waste heat but to atmospheric composition. The current atmospheric CO₂ concentration of approximately 425 ppm (NOAA, 2024) produces a radiative forcing of roughly 2.9 W/m² relative to pre-industrial levels. The associated temperature anomaly, approximately 1.3 K above the 1850–1900 baseline as of 2024 (WMO, 2024), is already altering precipitation patterns, accelerating ice sheet dynamics, and increasing the frequency of extreme heat events. At current emission trajectories, the 1.5 K threshold will be crossed within the decade and 2.0 K before mid-century.

The greenhouse constraint is categorically different from the waste heat ceiling established in the earlier analysis. It is a constraint on the composition of the entropy exported by the civilisational heat engine, not on the quantity of power throughput. Waste heat is source-independent: every joule dissipated on the planetary surface contributes to it regardless of origin. The greenhouse effect is source-dependent: it is driven specifically by the atmospheric accumulation of CO₂, CH₄, N₂O, and other radiatively active gases, overwhelmingly from fossil fuel combustion and land use change. A civilisation that replaced its entire primary energy supply with zero-carbon sources would eliminate the greenhouse forcing while leaving its waste heat trajectory untouched.

In the language of the three-tier ontology, the greenhouse constraint is partially compliant. The laws of radiative transfer that govern the greenhouse effect are fixed, Tier 1 physics. But the source of the forcing, the combustion of fossil hydrocarbons, is an engineering and institutional choice that belongs to Tier 3 and is therefore compliant: it can be changed. The thermodynamic maintenance floor δΣ demands power; it does not demand carbon. A civilisation that delivers its maintenance power from fission, fusion, or concentrated solar thermal satisfies P = Γ(t) · Σ without adding a single molecule of CO₂ to the atmosphere.

The decision window is decades. The remaining carbon budget consistent with limiting warming to 1.5–2.0 K is being consumed now, and atmospheric CO₂ has a complex removal profile: roughly half of a pulse is absorbed within 30 years by ocean and terrestrial sinks, but a long tail of approximately 20% persists for millennia (Archer et al., 2009). The constraint bite, the point at which reduced effective emissivity ε materially tightens Σ_max, arrives roughly a century from the present, governed by the ocean's thermal equilibration timescale of 30–100 years for upper layers and centuries to millennia for the deep ocean (Hansen et al., 2011). The distinction matters: decisions about energy source substitution must be initiated on the decadal timescale of the decision window, not delayed to the century timescale of the constraint bite.

The greenhouse constraint is therefore the most urgent and, paradoxically, the most tractable, because it is a problem of energy source substitution, not of total power reduction. It demands that civilisation change what it burns, not how much power it consumes. This places it squarely within the domain of compliant reform: hard engineering with a known solution space.

Tipping points and discontinuous constraint contraction. The smooth two-dimensional model used throughout this essay understates one dimension of greenhouse risk. Real climate physics contains tipping points; ice-sheet collapse, permafrost methane release, Amazon dieback, that can irreversibly reduce effective emissivity ε in step functions rather than smooth gradients. In the viability framework, these are discontinuous contractions of the constraint set K: sudden, irreversible reductions in the maximum sustainable assembly stock Σ_max. The smooth model does not capture these threshold effects, but the direction of the error is conservative. Tipping points make the kernel contract faster, not slower. The analysis presented here is therefore an optimistic bound on the actual dynamics.

The ecological constraint

The thermodynamic framework developed here deliberately abstracts from the ecological particulars of any specific planet. But a brief acknowledgement is warranted, because ecological constraints; biodiversity loss, soil degradation, freshwater depletion, nutrient cycle disruption, narrow the viable corridor on decadal timescales through mechanisms that interact with both the greenhouse and waste heat constraints.

The mechanism is twofold. First, ecosystem services that civilisation currently depends upon; pollination, water purification, soil formation, climate regulation, operate on solar exergy without drawing on the technospheric power budget. When these services degrade, the functions they performed must be replaced by engineered infrastructure: desalination plants for water purification, synthetic fertiliser production for nutrient cycling, mechanical pollination or crop genetic modification for food security. Each substitution increases the maintenance burden δΣ by adding assembled matter that must itself be maintained, and each draws additional power through P = Γ(t) · Σ. Ecological degradation therefore tightens the effective viability kernel by raising the maintenance floor without raising the waste heat ceiling. Second, large-scale biosphere disruption, particularly deforestation and ocean ecosystem collapse, can reduce the planet's effective carbon sink capacity, lowering the threshold at which the greenhouse constraint bites and thereby reducing the habitability temperature T_hab available for direct thermal loading.

The biosphere–technosphere coupling that grounds these two mechanisms is now established within the preceding analysis. The assembly stock quantification places the biosphere as the planet's dominant assembly stock Σ_bio ≫ Σ_tech maintained at a composite coupling Γ_bio ll Γ_tech, and the trajectory analysis derives the burden-transfer mechanism through which ecological degradation raises the technosphere's maintenance floor. The first mechanism identified above is precisely this burden-transfer: each unit of biospheric service lost must be replaced by technospheric assembly operating at Γ_tech ≫ Γ_bio, which tightens the effective kernel by raising the maintenance floor without raising the waste heat ceiling. The second mechanism, carbon sink reduction lowering the effective emissivity ε in the greenhouse constraint, is a distinct pathway to kernel contraction, operating on a shorter timescale through the atmospheric composition channel rather than the power throughput channel. The ecological constraint is therefore a multiplier of the dynamics already described, not an independent constraint requiring separate modelling; it amplifies both the greenhouse and waste heat constraints through the same coupling parameters Γ_tech, Γ_bio, δ, that the essay has already derived. The temporal hierarchy is conservative in excluding ecological constraints from the formal constraint surfaces: their inclusion would tighten the near-term kernel, not relax it.

The financial solvency constraint

The amplifier analysis established the thermodynamic solvency condition:

where Y is current economic output, r is the real interest rate, δ is the physical decay rate, and W is accumulated wealth. This condition must hold for virtual wealth, debt, financial claims, to remain redeemable against real wealth, the physical assembly stock.

The decision window is decades to a century. When r > 0 and the debt stock grows at rate r, the system requires real growth at a rate sufficient to prevent the debt-to-wealth ratio from diverging. At current global real interest rates of approximately 1–3% and debt-to-GDP ratios of approximately 350% (IIF, 2024), the system is sensitive to any sustained period of sub-par growth. The Minsky dynamics described in the amplifier analysis operate on credit cycles of 7–15 years, with major crises occurring roughly every generation. The constraint bite, the point at which the solvency condition raises the minimum viable growth rate Cₘᵢₙ above zero and forces the trajectory toward the ceiling, extends from decades to centuries, depending on the compounding dynamics and institutional response.

The financial constraint is entirely compliant. This was established in the amplifier analysis: δ is fixed physics; r is a rule of a game that humans invented and can change. A civilisation that restructured its monetary architecture; replacing compound-interest debt with equity participation, full-reserve banking, sovereign money creation, or periodic debt jubilees, could in principle eliminate r from the solvency inequality entirely, reducing it to:

This residual condition is pure physics. It states that economic output must at least match the thermodynamic decay of the existing assembly stock, that civilisation must produce enough to replace what entropy takes. No institutional reform can eliminate δ. But eliminating r from the binding constraint substantially widens the viable corridor, because it removes the exponential financial growth imperative that the amplifier analysis identified as the dominant accelerant of the trajectory toward the ceiling.

The interaction with the greenhouse constraint is significant. The financial growth imperative compels continued expansion of W, which through P = Γ(t) · Σ compels continued expansion of power throughput, which at current carbon intensity compels continued fossil fuel combustion. A civilisation that simultaneously decarbonised its energy supply and reformed its financial architecture would relax both the greenhouse and the solvency constraints, the first by eliminating the carbon forcing, the second by removing the exponential growth floor. Neither reform alone is sufficient. Decarbonisation without financial reform still requires exponential energy growth to service compounding debt. Financial reform without decarbonisation still emits carbon from whatever power sources remain. The two interventions are complementary, not substitutable.

The evolutionary ratchet

The maximum power principle, derived in the viability analysis as a consequence of the geometric persistence principle applied to competing agents on a shared gradient, is not a constraint that binds at a particular moment. It operates continuously, biasing the civilisational trajectory toward the waste heat ceiling at all times. Its decision window and its constraint bite are one and the same: always present, always operating.

The ratchet is a Tier 2 phenomenon, stiff in the control-surface classification, a statistical tendency of competitive dissipative structures, not a law of physics and not a human convention. Competitive selection among energy-dissipating subsystems favours those that maximise power throughput, driving μ upward and resisting reductions in Γ(t). The Jevons mechanism recycles efficiency gains into expanded demand. The competitive exclusion principle eliminates low-power competitors. These dynamics do not forbid Γ reduction, the analysis in the degrees-of-freedom discussion established that each component of Γ(t) = μ δ ξ / η_II has real engineering headroom, but they resist it with the full weight of evolutionary selection.

Overriding the ratchet requires civilisation-scale coordination: the capacity to align the behaviour of competing agents against their individual short-term incentives. This is the Tier 2 coordination problem. No precedent exists for solving it at the required scale, though partial solutions, international agreements, regulatory frameworks, coordinated investment, have achieved limited success at smaller scales. The ratchet is the reason the waste heat ceiling cannot be dismissed as a remote theoretical concern, even when the timeline to it is measured in centuries. Without solving the coordination problem, the trajectory toward the ceiling proceeds at the rate set by the maximum power principle, regardless of what reforms are achieved at Tier 3.

The waste heat ceiling

The waste heat ceiling was established in the earlier analysis and requires only brief recapitulation here. The planetary energy balance in steady state is:

Every watt of civilisational power P, regardless of source, adds to the planetary thermal load. The maximum sustainable assembly stock is:

This ceiling is set by the Stefan–Boltzmann relation applied to a finite radiating surface. It is Tier 1 physics, fixed, immovable by any parameter except an expansion of the radiating area A beyond the planet.

4.2 Viability Theory

The previous three parts have established an arena (the planetary system boundary and its radiative physics), a machine (civilisation as assembled matter requiring continuous maintenance power), and a trajectory (the evolutionary dynamics and institutional amplifiers that drive the machine toward the boundaries of the arena). This section formalises the predicament. The mathematical framework designed to answer precisely this class of question, not what is optimal, but what is survivable, was developed by Jean-Pierre Aubin and his collaborators beginning in the 1980s, under the name viability theory (Aubin, 1991; Aubin, Bayen and Saint-Pierre, 2011). This section applies that framework to the civilisational thermodynamics established in the preceding parts, derives the braking boundary and zone of false security as calculable surfaces, and establishes how financial architecture narrows the viability kernel through solvency constraints on the admissible control set. The competitive multi-agent extension, deriving the maximum power principle and Jevons rebound as consequences of viability geometry, is developed in the section that follows.

Why Optimality Is the Wrong Framework

Standard climate-economy modelling poses the civilisational predicament as an optimisation problem. The Nordhaus DICE model (Nordhaus, 2017), the Stern Review (Stern, 2007), and their descendants seek to maximise a discounted utility functional subject to constraints. The analytical machinery is Pontryagin's Maximum Principle or dynamic programming. The output is an "optimal" trajectory, one that balances present consumption against future damages through a social discount rate.

This framing contains a structural error that the preceding parts have made visible. An optimisation problem assumes that the system can, in principle, access any region of state space, that the question is merely which trajectory is best. But the physics of civilisation as a dissipative structure reveals that large regions of state space are not merely suboptimal. They are lethal. A planetary temperature beyond the wet-bulb survivability threshold is not a "cost" to be discounted. An assembly stock below the minimum required to maintain critical infrastructure is not a "welfare loss" to be traded off against present consumption. These are hard boundaries. Cross them and the system ceases to exist in any form recognisable as civilisation.

The distinction matters mathematically. In optimal control, the solution is a single trajectory that maximises the objective. In viability theory, the solution is a set, the largest subset of state space from which there exists at least one trajectory that never violates the constraints. The object of interest is not a path but a region: the viability kernel. The policy question shifts from "what is the best thing to do?" to "from which states is it possible to survive at all, and what controls keep us there?"

Civilisation does not have a well-defined utility function. It has survival constraints. The gap between the behavioural attractor (gradient saturation, derived in the competitive viability analysis that follows) and the survival requirement is the central tension of civilisational thermodynamics. Aubin's framework makes that tension mathematically precise.

The Formal Structure

where u ∈ U(x) is a control parameter drawn from an admissible set that may depend on the current state. The use of a differential inclusion rather than a differential equation is deliberate. It reflects genuine uncertainty: the exact dynamics of civilisation are not known, but the set of possible evolutions for any given state and control can be bounded. Aubin's insight was that this formulation, weaker than optimal control, stronger than mere simulation, is precisely what is needed for systems where survival matters more than performance.

Define a closed constraint set K ⊂ R^n. This is the set of states that satisfy all viability conditions simultaneously. Any state outside K is, by definition, non-viable, the system has already failed.

Definition D4 (Viability kernel). The viability kernel of K under the dynamics F is:

In words: the viability kernel is the set of all initial states from which there exists at least one admissible control trajectory that keeps the system inside K forever. States inside K but outside the viability kernel are doomed, they satisfy the constraints today, but no sequence of feasible actions can prevent eventual violation.

This definition has a geometric consequence that is crucial for what follows. The viability kernel is always a subset of K, often a strict and much smaller subset. The constraint set tells you where you need to stay. The viability kernel tells you from where you can stay there. The difference K ∖ Viab_F(K) represents states of false security: states where every possible trajectory eventually exits K, no matter what controls are applied. By the time a system has crossed from the kernel into this doomed zone, it is already too late.

Aubin provides a local criterion for survivability. At any point x on the boundary of K, the tangent cone T_K(x) defines the set of velocities that keep the trajectory inside K for at least an infinitesimal time step. The viability condition requires:

F( x,U(x) ) ∩ T_K(x) ≠ ∅

If this intersection is non-empty, there exists at least one control that prevents immediate exit from K. If it is empty, the state is outside the viability kernel regardless of what controls are applied.

The Civilisational State Vector

To apply viability theory to the framework developed in this essay, the state space must be specified. Drawing on the variables established in the preceding parts, define the civilisational state vector:

where Σ is the assembly stock, the total assembly embodied in all objects within the system boundary that are maintained by civilisational power throughput, as defined in the assembly analysis. Its dynamics are governed by:

where C(t) is the construction rate (new joining operations per unit time) and δ is the aggregate decay rate. The maintenance power required to sustain Σ is P = Γ(t) · Σ, as established in the maintenance derivation.

T_eq is the planetary equilibrium temperature, driven by total waste heat rejection through the Stefan-Boltzmann relation established in the waste heat analysis. It responds to changes in Σ with thermal inertia introduced by the ocean-atmosphere system:

where C_th is the effective heat capacity of the climate system (approximately 5 × 10^23 J/K for the ocean mixed layer), P_☉ ≈ 1.2 × 10¹⁷ W is the absorbed solar flux, ε is the effective emissivity, σ = 5.67 × 10⁻⁸ W m^- 2 K^- 4 is the Stefan-Boltzmann constant, and A ≈ 5.1 × 10^14 m^2 is Earth's radiating area.

E_res is the available exergy reserve; fossil fuels, fissile materials, accessible renewable potential. It depletes at a rate set by total power throughput and the effective energy return on investment:

D is the aggregate debt stock, whose dynamics were established in the amplifiers analysis:

where r is the real interest rate, L_new is new lending, and R is repayment from current production.

The control vector u(t) represents the levers available to civilisation: monetary policy (interest rates, credit regulation), fiscal policy (taxation, public investment), energy mix decisions, and the allocation vector across the service basis. The admissible control set U(x) is state-dependent: a civilisation with depleted reserves has fewer options than one with abundant reserves; a civilisation in debt crisis has fewer degrees of freedom than one with fiscal headroom.

The Constraint Set and the Three-Tier Ontology

The constraint set K is defined by the intersection of all binding constraints on civilisational survival. The three-tier ontology developed throughout this essay provides the organising principle: the constraints differ in ontological status, and that status determines which are movable and which are not.

First tier: hard physics

These constraints are set by thermodynamics and radiative physics. No technology, no institutional reform, no act of collective will can alter them.

Thermodynamic habitability. The equilibrium temperature must not exceed the threshold for sustained human survival:

where T_max corresponds to the wet-bulb temperature limit for mammalian thermoregulation, approximately 35°C wet-bulb, implying a global mean temperature anomaly ceiling of roughly 10–12 K above pre-industrial baseline, though regional exceedance and agricultural collapse occur far earlier. Through the coupling P = Γ(t) · Σ and the Stefan-Boltzmann relation, this constraint is ultimately a ceiling on the assembly stock:

This is the waste heat ceiling derived earlier, now expressed as an explicit bound in state space. It is source-independent, technology-independent, and non-negotiable.

Minimum viable assembly. The assembly stock must not fall below the threshold required to maintain civilisational coherence:

Below Σ_min, the infrastructure required for coordinated existence, power grids, food distribution, medical systems, communication networks, can no longer be sustained. This is the thermodynamic formalisation of Tainter's (1988) collapse threshold. Σ_min is not precisely known, but its existence is guaranteed by the physics: organisation cannot be maintained without continuous exergy input.

The Landauer floor. Every irreversible bit operation within the civilisational information infrastructure dissipates at least k_BTln2 joules, as established in the information entropy analysis. This sets a lower bound on the power required to maintain the informational component of Σ at any given level of computational throughput. It is the thermodynamic cost of being complex.

These three constraints; the ceiling, the floor, and the computational minimum, define the hard geometry of K. No parameter accessible to civilisation can move them.

Second tier: evolutionary dynamics

These constraints arise from the physics of competition among dissipative structures. They follow from the statistics of selection in a finite environment and are extremely difficult to override. Their formal derivation from competitive viability geometry is developed in the section that follows; the present subsection locates them within the constraint architecture.

The Lotka attractor. Lotka's maximum power principle (Lotka, 1922), operating through competitive selection among agents drawing on a common energy gradient, drives civilisation toward the configuration that maximises power throughput. This manifests in the viability framework not as a constraint boundary but as a bias in the dynamics: the natural trajectory of x is directed toward Σ_max rather than away from it.

The Jevons recycling. Efficiency improvements increase the rate of return on energy investment, enabling faster construction C(t), which drives Σ upward, which raises P = Γ(t) · Σ, which accelerates waste heat accumulation. The dynamics preferentially explore the region of state space closer to the ceiling.

These second-tier dynamics do not create hard boundaries in the way the first tier does. Overriding competitive selection at civilisational scale has no historical precedent. The second tier determines the natural trajectory within the geometry defined by the first tier.

Third tier: human conventions

These constraints are real, consequential, and currently binding, but they are properties of institutional architecture rather than physical law. "Can be reformed" is not "easily reformed." It means "not forbidden by physics."

The debt growth floor. The amplifiers analysis established that the solvency condition under current financial architecture requires:

where Y is annual production, r is the real interest rate, δ is the decay rate, and W is the monetary proxy for accumulated wealth. This condition is binding only because civilisation has adopted a specific financial architecture, compound-interest debt with endogenous money creation. The decay rate δ in this inequality is physics. The interest rate r is convention. A civilisation that restructured its monetary system could, in principle, eliminate r from the inequality, leaving only the physical maintenance floor Y ≥ δ W. The solvency boundary in state space is therefore p-dependent, where p denotes the vector of structural parameters that includes financial architecture.

The governance frequency mismatch. The amplifiers analysis identified the mismatch between electoral cycles (4–5 years) and the system time constants (50–100 years). In the viability framework, this constrains the rate of change of the control vector: | u̇ | ≤ u̇ₘₐₓ, where u̇ₘₐₓ is set by institutional design. A civilisation with longer planning horizons and stronger mechanisms for intergenerational commitment has a larger u̇ₘₐₓ, a wider set of achievable velocity changes per unit time, and therefore a larger viability kernel, all else equal.

The three-tier classification maps directly onto the formal structure. First-tier constraints define the fixed boundaries of K. Second-tier dynamics bias the system toward those boundaries. Third-tier constraints define adjustable boundaries within K, they can be widened by institutional reform, expanding the kernel without altering the hard geometry.

The Two-Dimensional Worked Example

The full civilisational state space is four-dimensional and not analytically tractable in closed form. But the essential mechanism, kernel contraction driven by the coupling between assembly growth and thermal loading, can be exhibited in a two-dimensional projection that retains the physics while admitting explicit computation. This worked example is the quantitative core of the section.

Setup

Reduce the state vector to two dimensions:

Equation (1) is the assembly balance. Equation (2) is the planetary energy balance, with thermal inertia C_th representing the heat capacity of the ocean-atmosphere system. The coupling between the two equations is through the term Γ · Σ: the maintenance power of the assembly stock enters the planetary energy balance as waste heat.

The control variable is the construction rate C(t), bounded by:

The lower bound is physical: construction cannot be negative (assembly cannot be un-built faster than decay dismantles it). The upper bound C_max represents the maximum rate at which the system can perform new joining operations, constrained by available power, labour, and materials.

The constraint set

This is a rectangular region: a vertical wall on the left at Σ = Σ_min and a horizontal ceiling at T = T_max. Everything below and to the right is admissible. Everything above or to the left is lethal.

Note what is absent from this simplified K: the debt constraint and the resource floor, which belong to the full four-dimensional formulation. The two-dimensional example isolates the interaction between assembly growth and thermal loading, which is the mechanism unique to this essay's framework.

The steady-state locus

At steady state, construction exactly offsets decay. Setting dT/dt = 0 gives the radiative equilibrium condition:

Solving for the steady-state temperature as a function of assembly stock:

This is a monotonically increasing curve in the (Σ,T) plane, the thermal equilibrium locus. Every point on this curve represents a state at which, if Σ is held constant by balancing construction against decay, the temperature settles to the value dictated by the Stefan-Boltzmann relation. The curve rises steeply at first and then more gradually, because T scales as Σ^1/4.

The thermal equilibrium locus intersects the constraint boundary T = T_max at:

This is the maximum steady-state assembly stock. Any Σ > Σ_max produces an equilibrium temperature above T_max. The system cannot persist there indefinitely.

Numerical calibration

To make the example concrete, assign values drawn from the physics established in previous sections. P_☉ = 1.2 × 10¹⁷ W. Taking ε ≈ 0.62 (consistent with Earth's current effective emissivity) gives εσ A ≈ 1.79 × 10^7 W/K^4; a consistency check at current T_eq ≈ 288 K yields 1.79 × 10^7 × 288^4 = 1.23 × 10^17 W ≈ P_☉. Γ ≈ 5.9 × 10^9 W per unit assembly stock, calibrated from the empirical ratio P/W ≈ 5.9 mW per 2019 US dollar (Garrett et al., 2022), where Γ(t) = μδξ/η_II is the composite coupling whose aggregate value Hanley (2025) dated to the post-1970 regime. δ ≈ 0.02 yr^- 1, C_th ≈ 5 × 10^23 J/K (thermal relaxation timescale τ_T ≈ 30 years at T = 288 K).

Setting T_max = 298 K (a 10 K anomaly above the pre-industrial baseline of approximately 288 K, chosen to represent the onset of widespread wet-bulb exceedance and agricultural collapse, well below the absolute mammalian thermoregulation limit), Equation (4) gives:

The precise numerical value depends on the units chosen for Σ. What matters is the ratio: Σ_max/Σ_current. At current P ≈ 1.8 × 10^13 W and P_☉ ≈ 1.2 × 10¹⁷ W, the additional power budget available before a 10 K temperature rise is approximately:

This is roughly 1,000 times current civilisational power consumption. At 2.3% annual growth — a doubling time of approximately 30 years — the number of doublings to the ceiling is:

T_max sensitivity

The worked example uses T_max = 298 K (a 10 K anomaly). This choice is deliberately conservative, it represents an upper bound on tolerable warming, not a policy target. The qualitative geometry of the kernel is insensitive to T_max; the quantitative runway is not. At T_max = 292 K (a 4 K anomaly, consistent with widely discussed climate targets), Σ_max drops by roughly half and the timeline contracts to approximately 270 years at historical growth rates. At T_max = 295 K (a 7 K anomaly), intermediate values obtain. The structure of the viability problem, ceiling, kernel, zone of false security is invariant. The distance to the wall is not.

Conditional timeline caveat

The 300-year figure assumes sustained 2.3% annual growth in power throughput and current Γ. Both assumptions are conditional, not physical constants. At 1% annual growth, the doubling time extends to approximately 70 years and the timeline roughly doubles. A factor-of-2 reduction in Γ, achievable through simultaneous improvement of its four constituent components μ, δ, ξ, and η_II doubles Σ_max and adds approximately 30 years at historical growth rates. Combined, halved growth rate plus halved Γ, the timeline extends to roughly 700 years.

The ceiling is physics. The timeline to the ceiling is scenario. The essay's position is that the ceiling exists and cannot be removed; it is not that the ceiling will be reached in exactly 300 years.

Computing the viability kernel

The viability kernel is the set of states ( Σ₀,T₀ ) ∈ K from which there exists a control trajectory C(t) satisfying 0 ≤ C(t) ≤ C_max such that ( Σ(t),T(t) ) ∈ K for all t ≥ 0.

The key question is: from a given initial state, can the system be steered to avoid T exceeding T_max? The maximum-braking trajectory, setting C(t) = 0 for all future time, gives the fastest possible reduction in Σ:

Under maximum braking, Σ decays exponentially at rate δ. The temperature responds to this declining Σ through Equation (2), but with a lag set by C_th. The question becomes: starting from ( Σ₀,T₀ ), if maximum braking is applied, does T(t) ever exceed T_max?

If the answer is yes, if even maximum braking cannot prevent thermal exceedance, then ( Σ₀,T₀ ) is outside the viability kernel. There is no admissible control that averts failure.

The analysis proceeds in two cases.

Case 1: Below the equilibrium locus. If the initial state satisfies T₀ < T*( Σ₀ ), the current temperature is below the radiative equilibrium value for the current assembly stock, then the system is thermally lagging. The temperature is rising toward T*( Σ₀ ) even without further construction. If T*( Σ₀ ) > T_max, the system is committed to overheating unless Σ is reduced fast enough to pull the equilibrium locus below T_max before T reaches it. The condition for this is:

Whether this holds depends on the race between two timescales: the decay timescale τ_Σ = 1/δ (approximately 50 years at δ = 0.02) and the thermal relaxation timescale τ_T ≈ 30 years. Since τ_Σ > τ_T for plausible parameter values, the temperature equilibrates faster than the assembly stock decays. The system heats up faster than it can shed mass. This means the viability kernel boundary in this region lies well inside K, states with Σ large enough that T*(Σ) > T_max are non-viable even though their current temperature is below T_max.

More precisely, define Σ_crit as the assembly stock above which maximum braking cannot prevent T from exceeding T_max. For Σ_0 > Σ_crit, no trajectory keeps the system inside K. For Σ_0 ≤ Σ_crit, the trajectory C = 0 eventually brings Σ below Σ_max and the temperature stabilises below T_max. The value of Σ_crit depends on T_0 and the ratio τ_T/τ_Σ: a large thermal buffer (low T_0) permits Σ_crit to exceed Σ_max modestly, while a nearly exhausted buffer forces Σ_crit toward Σ_max from below.

Case 2: Above the equilibrium locus. If T₀ > T*( Σ₀ ), the system is thermally overshooting, the temperature exceeds the radiative equilibrium value for the current assembly stock. The viability condition is less restrictive in this regime, but the case is empirically irrelevant for the current civilisation, which is in Case 1 with temperature lagging below the equilibrium locus as Σ continues to grow.

The kernel geometry

The right boundary of the kernel is not the vertical line Σ = Σ_max (which is the boundary of K). It is a curve Σ = Σ_crit(T) that lies strictly to the left of Σ_max for all T < T_max. This curve is the viability boundary. To its left, survival trajectories exist. To its right, no admissible control can prevent thermal exceedance.

The curve Σ_crit(T) has two key properties. First, it approaches Σ_max from the left as T arrow T_max. When the temperature is already near the ceiling, only states with Σ very close to or below Σ_max are viable, because there is almost no thermal buffer remaining. Second, it extends modestly beyond Σ_max for T well below T_max. When the temperature is far below the ceiling, the thermal buffer permits a temporary overshoot in Σ provided it is corrected by rapid de-growth (maximum braking) before T catches up. The extent of this overshoot is bounded by the ratio τ_Σ/τ_T: the slower the assembly decays relative to the thermal relaxation, the less overshoot is tolerable.

The bottom boundary of the kernel is the horizontal line Σ = Σ_min. Below this, civilisation cannot sustain itself regardless of temperature. This boundary coincides with the left boundary of K, provided T*( Σ_min ) < T_max, that is, provided the minimum viable civilisation does not itself overheat the planet. For any civilisation whose minimum viable assembly stock produces waste heat well below the radiative ceiling, this condition is satisfied trivially.

The kernel is therefore a region in the (Σ,T) plane bounded on the left by Σ_min, on the top by T_max, and on the right by the viability boundary Σ_crit(T), which lies inside K. The gap between Σ_crit(T) and Σ_max, the region that is inside K but outside the kernel, is the zone of false security. States in this zone satisfy all current constraints. They appear viable. They are doomed.

Dynamic kernel contraction

The static analysis above assumes the constraint set K is fixed. In reality, K itself evolves, and the kernel contracts with it.

The greenhouse constraint progressively tightens T_max in effective terms. As atmospheric CO_2 accumulates, the effective emissivity ε decreases, shifting the thermal equilibrium locus upward for any given Σ. The effective ceiling, the maximum Σ compatible with habitability including greenhouse effects, is lower than the waste heat ceiling alone would suggest. The greenhouse effect consumes part of the thermal budget that would otherwise be available for waste heat.

More importantly, the viability kernel contracts dynamically even if K is held fixed, because the state vector is moving. Each year of continued growth adds to Σ, raising waste heat production and driving T toward the equilibrium locus. The state vector moves rightward and upward in the (Σ,T) plane, toward the viability boundary, not away from it. Because the boundary curves toward the vertical near T_max, the final approach is rapid, the last few doublings before the boundary is reached consume the remaining viable space far faster than the earlier ones did. This is the geometric mechanism behind the Mathias, Anderies and Janssen (2017) finding that viability kernels contract monotonically with delay, and that each year of inaction permanently eliminates states from the kernel.

Sophie Martin and her collaborators at INRAE have extended viability analysis across socio-ecological systems; agroecosystems, fisheries, forest management, and consistently find the same structural result: systems with strong inertia and delayed feedback lose viable states faster than the constraint boundaries move (Martin, 2004; Béné, Doyen and Gabay, 2001). When the constraints are tightening and the dynamics contain lags, the kernel can collapse well before the constraints are actually breached.

The braking boundary and the zone of false security

The preceding analysis establishes the viability kernel qualitatively: states inside the constraint set K from which braking trajectories exist, separated from states where no admissible control prevents thermal exceedance. The boundary between the two, the viability boundary: Σ^*(T), has been described in geometric terms. This subsection derives it.

The derivation converts a verbal claim ("delay destroys options") into a calculable surface. It makes the zone of false security an object with a size, a shape, and a dependence on physical parameters that can be evaluated for any planetary civilisation.

The braking ODE

Under maximum braking (C = 0), assembly decays as Σ(t) = Σ_0 e^- δ t. The planetary thermal response to this decaying source, linearised about a reference temperature T_*, is

where κ = 4εσ AT_*^3 is the linearised radiative damping coefficient (W/K). This is a first-order linear ODE with exponentially decaying forcing. Two timescales govern the dynamics: thermal relaxation: τ_T = C_th/κ, the e-folding time over which the planet radiates away a thermal perturbation (for Earth, τ_T ≈ 30 years); and assembly decay: τ_Σ = 1/δ, the e-folding time over which unrepaired assembly stock erodes (at δ ≈ 0.02 yr^- 1, τ_Σ ≈ 50 years).

The ratio r = τ_T/τ_Σ = δ C_th/κ determines the character of the braking problem. When r ll 1, the planet radiates fast relative to stock decay and the system can brake without significant overshoot. When r gtrsim 1, the planet cannot shed heat fast enough and the overshoot is severe. For Earth parameters, r ≈ 0.4–0.75: the timescales are of the same order but thermal response is moderately faster. The false-security zone exists and is substantial.

Exact solution

Equation (B1) with initial conditions T(0) = T_0 and Σ(0) = Σ_0 has the exact solution (assuming κ ≠ δ C_th, i.e. r ≠ 1):

where

The coefficient β has units of temperature and a direct physical interpretation. The radiative equilibrium temperature for an assembly stock Σ_0 is T*( Σ₀ ) = T* + ΓΣ₀/κ. Then β = ( T*( Σ₀ ) - T* )/(1 - r): the equilibrium temperature anomaly amplified by the lag factor 1/(1 - r). As r arrow 1 (thermal and decay timescales converge), β diverges, the overshoot becomes unbounded under this linearisation because the planet cannot radiate away heat as fast as the decaying stock produces it.

Peak temperature

The temperature peaks at time t^* where Ṫ( t* ) = 0. Differentiating (B2):

Setting this to zero and solving:

This is well-defined when β > T_0 - T_* and r < 1, which corresponds to the physically relevant case: the initial stock produces an equilibrium temperature above the current temperature (the system is thermally lagging) and the thermal response is faster than the decay. When T₀ ≥ T*( Σ₀ ), the system is already at or above its thermal equilibrium, there is no overshoot and the temperature declines monotonically under braking. That case is empirically irrelevant for the current civilisation, which is in the thermally lagging regime.

This is the thermal overshoot formula: the maximum temperature the planet reaches after civilisation ceases all construction and allows its assembly stock to decay naturally. The overshoot is the price of thermal inertia, the heat already committed to the system that must be radiated away before the temperature can fall.

Proposition P1 (Braking boundary).

For each initial temperature T_0 < T_hab, equation (B6) gives the maximum assembly stock from which maximum braking, the most aggressive physically possible deceleration, keeps the trajectory below the habitability threshold. States with Σ > Σ_phys*( T₀ ) are outside the physical viability kernel regardless of the control policy applied: even the extreme of zero construction cannot prevent thermal exceedance.

First, it lies strictly inside the naive constraint set boundary Σ_max(T) for all T_0 < T_hab. The gap between Σ_phys^* and Σ_max is the physical false-security zone: states that satisfy all current constraints but from which no braking trajectory avoids future violation.

Second, the gap widens as r increases. When r is small (fast thermal response), the planet tracks the decaying stock closely and the overshoot is small. When r approaches unity, the overshoot consumes the entire thermal budget and the viability boundary retreats deep inside the constraint set. The false-security zone scales as

in the limit of small thermal buffer ( T_hab - T₀ ) arrow 0. The factor r/(1 - r) is the fundamental measure of the false-security zone: the ratio of committed-but-unradiated heat to the remaining thermal budget.

Third, the gap narrows as T₀ arrow T_hab: when the thermal budget is nearly exhausted, even a small overshoot triggers exceedance, and the viability boundary converges on Σ_max. Conversely, for T_0 well below T_hab, the large thermal buffer permits a correspondingly larger gap, because there is more headroom to absorb the committed overshoot.

Earth-parameter illustration

The following is an order-of-magnitude illustration, not a precision calculation. At κ ≈ 1.7 × 10^15 W/K, δ ≈ 0.02 yr^- 1, Γ ≈ 5.9 × 10^- 3 W per 2019 USD equivalent of accumulated stock, and a thermal buffer of T_hab - T_0 ≈ 10 K (using the worked example's 298 K ceiling against a current 289 K), the characteristic parameters are:

The false-security zone is therefore of order 1.5 times the thermal buffer translated into assembly-stock units via κ/Γ. In proportional terms, the viability boundary sits at roughly Σ_phys* ≈ Σ_max/( 1 + r/(1 - r) ) = Σ_max/2.5 for a system near the midpoint of its thermal budget. The zone of false security is not a sliver at the edge of the constraint set. It is a substantial fraction of the nominally viable region.

At the current growth rate of approximately 2.3% per year, doubling time approximately 30 years, the state vector traverses the viable region in roughly log₂(2.5) ≈ 1.3 doublings, approximately 40 years of growth before the viability boundary is reached, even though the constraint set boundary (the naive ceiling) would not be reached for decades longer. This is the quantitative content of "delay destroys options": each year of inaction does not merely bring the ceiling closer. It brings the viability boundary closer, and the viability boundary is closer than the ceiling.

Two boundaries: physical and institutional

The boundary Σ_phys^*(T) assumes the widest possible control set: C ≥ 0, meaning civilisation can cease all construction. If institutional constraints impose a minimum construction rate C_min > 0 because debt service requires continued real growth, because political stability requires continued employment, because the financial architecture converts cessation of growth into systemic crisis, then the achievable braking rate is slower than δ, and the viability boundary contracts further inward.

This tighter boundary Σ_inst^(T) is derived formally in the solvency analysis below. The gap between Σ_phys^ and Σ_inst^* is the space that institutional reform buys: the additional viable region that becomes accessible when Tier 3 constraints (debt architecture, governance timescales, growth compulsions) are reformed to permit a wider range of braking trajectories. Three nested boundaries therefore define the viability landscape: the constraint set boundary Σ_max(T), the naive ceiling; the physical viability boundary Σ_phys^(T), the limit of what physics permits under maximum braking; and the institutional viability boundary Σ_inst^(T), the limit of what current institutions permit.

The region between the outer and middle boundaries is the physical false-security zone: states where physics forbids survival regardless of institutional reform. The region between the middle and inner boundaries is the institutional false-security zone: states where physics would permit survival but institutions forbid the braking trajectory that survival requires. The first zone is Tier 1, immutable. The second is Tier 3, reformable. The distinction between the merely difficult and the physically impossible runs through the gap between these two surfaces.

Solvency-constrained admissible controls

The viability kernel derived above assumes the full control set U is available, in particular, that the maximum-braking trajectory C = 0 is admissible. Under the physical dynamics alone, it is. A civilisation that halts all construction and lets its assembly stock decay at rate δ violates no law of physics. But civilisations do not operate under the physical dynamics alone. They operate under institutional architectures that impose additional constraints on the control vector, and these constraints can eliminate precisely the low-growth trajectories on which the kernel computation depends.

The amplifiers analysis established the solvency condition qualitatively. The derivation here makes it exact in the vocabulary of the viability framework and separates the Tier 1 component from the Tier 3 component with mathematical precision.

Setup. From the maintenance analysis, the minimum power to hold the assembly stock constant is P_maint = δξΣ/η_II. Under pure maintenance (C = 0, μ = 1), the civilisation sustains its stock without growth. The power dissipated is P_maint and the economy produces enough output to cover physical depreciation. No growth occurs. No debt service is required beyond maintenance.

The solvency condition. Now introduce financial architecture. The civilisation carries aggregate debt D at real interest rate r. Gross output Y must cover both physical maintenance and debt service:

where W is the monetary shadow of Σ (cumulative wealth in the Garrett sense), u is the control vector (investment allocation, energy mix, policy choices), and Y(u,Σ) is gross output under control u given stock Σ.

The admissible control set. Define:

This is the set of controls that avoid insolvency. Any control u notin U_adm leads to debt default, credit contraction, and cascading loss of maintenance capacity, a financial route to the same assembly collapse that the thermodynamic floor predicts.

Proposition P2 (Debt narrows the viability kernel). Let K_phys be the viability kernel under the full physical control set U (as derived in the braking analysis above), and let K_inst(r,D) be the viability kernel under the solvency-restricted control set U_adm(Σ,D). Then:

(i) If r = 0 and D = 0, the solvency condition reduces to Y(u,Σ) ≥ δ W, which is the physical maintenance condition in monetary terms. The admissible set includes braking controls down to maintenance-only: C = 0, P = P_maint, μ = 1. Every control that sustains the stock is admissible. K_inst(0,0) = K_phys.

(ii) If r > 0 and D > 0, the solvency condition Y ≥ rD + δ W excludes controls for which output falls below the combined obligation. Since rD > 0, every control that was marginally admissible under (i), output just covering δ W, is now inadmissible. The admissible set shrinks: U_adm(Σ,D;r > 0) ⊂ U_adm(Σ,D = 0;r = 0). The kernel narrows: K_inst(r,D) ⊂ K_phys with strict inclusion whenever rD > 0.

(iii) The gap Δ K = K_phys ∖ Kᵢₙₛₜ(r,D) represents trajectories that physics permits but the financial architecture forbids. Its size is monotone increasing in the ratio rD/δ W.

The logic is direct. Restricting the admissible set can only eliminate trajectories, never create new ones. Any kernel computed under U_adm ⊆ U is a subset of the kernel computed under U.

The critical decomposition. The solvency obligation rD + δ W in Equation (S1) contains two terms of fundamentally different ontological status:

The term δ W is Tier 1. It is the monetary expression of the Second Law operating on civilisational assembly. Concrete carbonates at the rate concrete carbonates regardless of the interest rate. Steel corrodes regardless of the monetary system. The maintenance obligation cannot be eliminated by any institutional reform.

The term rD is Tier 3. It is a property of the financial architecture, the institutional rules governing credit creation, interest rates, and debt obligations. It can be reformed. It has been reformed repeatedly in recorded history: debt jubilees in Mesopotamia, interest rate caps in medieval Europe, the Bretton Woods architecture, quantitative easing, sovereign default. Setting r = 0 or D = 0 is not a physical impossibility. It is an institutional choice.

Conflating the two, treating the combined obligation (δ + r)W as a single irreducible constraint, is the most dangerous analytical error in civilisational strategy. It leads to the conclusion that growth is physically necessary to avoid collapse, when in fact only maintenance is physically necessary; the growth compulsion above maintenance is an institutional artefact. The three-tier ontology exists precisely to prevent this conflation, and the solvency derivation makes the separation exact.

The institutional headroom. The size of the gap Δ K depends on the ratio of rD to δ W. When rD ll δ W, the institutional constraint is negligible and nearly all physically viable trajectories remain institutionally admissible. When rD ~ δ W, roughly half the viable trajectory space is carved away by debt service. When rD ≫ δ W, the institutional constraint dominates the physical one, most low-growth trajectories are excluded, and the civilisation cannot brake even if the physics would permit it.

For the current global economy, rough magnitudes: the maintenance share δ W is approximately 15% of gross world product (the maintenance share from the empirical data in the assembly analysis), and aggregate debt service rD is approximately 10–15% of gross world product (depending on how broadly "debt service" is construed). The two terms are of comparable magnitude. The institutional constraint is not negligible. Substantial viable trajectory space is being excluded by the financial architecture.

Connection to the braking boundary. The braking analysis above derived the physical viability boundary Σ_phys^*(T) under the assumption that C ≥ 0 that the civilisation can halt all construction. With the solvency constraint, the effective lower bound on C is:

This raises the floor on the construction rate, which means the stock cannot decay as fast under institutional braking as it can under physical braking. The institutional viability boundary Σ_inst^(T) therefore lies strictly inside Σ_phys^(T):

The (Σ,T) plane now contains three nested boundaries: the constraint set boundary Σ_max(T) (the naive thermal ceiling), the physical viability boundary Σ_phys^(T) (the kernel boundary under unconstrained controls), and the institutional viability boundary Σ_inst^(T) (the kernel boundary under solvency-constrained controls). The gap between the first and second is the physical false-security zone, the territory described in the braking analysis, where the system appears safe but cannot stop in time. This gap is Tier 1. No institutional reform can close it. The gap between the second and third is the institutional false-security zone, territory that physics would permit but the financial architecture forbids. This gap is Tier 3. It can be closed by reform. Their sum is the total false-security zone, and its decomposition into a physical component and an institutional component is the formal expression of the distinction between what the laws of physics impose and what the rules of finance add.

Caveat. The formulation Y ≥ rD + δ W is a stylised representation of a complex web of financial obligations. Real economies have heterogeneous agents, diverse debt instruments, variable interest rates, and institutional buffers. The derivation establishes the structural claim, debt narrows the kernel, and the direction, more debt at higher interest excludes more trajectories. It does not claim to model the global financial system in detail. Similarly, setting r = 0 or D = 0 recovers the physical kernel but does not guarantee survival. The physical kernel is still bounded by the waste heat ceiling and the braking boundary. Tier 3 reform is necessary but not sufficient. The distinction is between widening the corridor and moving the walls. Reform does the first. Only physics, through the parameters of Γ and the expansion of A, addresses the second.

4.3 Competitive Viability

The viability framework established in the preceding section treats civilisation as a single agent facing fixed constraints. The framework extends to the competitive case by introducing multiple agents sharing a finite gradient. This extension is the formal centrepiece of the section. It derives the maximum power principle, gradient saturation under competitive allocation, as a theorem of the viability apparatus rather than an empirical import from ecology. The Jevons rebound, competitive exclusion, and landscape dependence follow as corollaries. The result promotes the entire Tier 2 argument from an empirical pattern observed in the data to a consequence of the essay's own mathematical framework. Critically, the axiom set for this derivation contains no ecological imports and no behavioural assumptions: every axiom is either derived from the maintenance analysis of Part 2 or is a minimal physical realism condition.

Setup: N agents on a shared gradient

Consider N agent; firms, nations, technologies, species, sharing a finite energy gradient P_total. For a surface-bound civilisation, P_total is the waste heat ceiling derived in the machine analysis: the maximum total power dissipation compatible with habitability. It is fixed by physics (Tier 1) and expandable only by increasing A (the kleos path).

Throughout this derivation, P_i denotes total power dissipation by agent i, all waste heat generated, including maintenance losses. This is the quantity constrained by the Stefan-Boltzmann ceiling. It is not "useful power" or "exergy capture rate" in the Lotka/Odum ecological sense. The result derived here is about total dissipation saturating a radiative ceiling, which is a stronger claim than ecological maximum power and the one the essay requires.

The derivation rests on five axioms, stated here and justified below. Their logical ancestry is summarised in the axiom table at the end of this section.

Axiom A5 (Finite gradient). Total available power satisfies Σᵢ^Pᵢ ≤ Pₜₒₜₐₗ < ∞, where P_total is set by the Stefan-Boltzmann ceiling.

Axiom A6 (Per-agent assembly dynamics). Each agent's stock evolves as dΣ_i/dt = C_i - δ_iΣ_i, with construction bounded by allocated power: 0 ≤ C_i ≤ α_iP_i.

Axiom A7 (Minimum viable stock). Each agent has a threshold Σ_i,min > 0 below which it ceases to function. This is the thermodynamic collapse threshold: the minimum assembly stock whose maintenance can sustain critical functions.

Axiom A8 (Monotone share function). Each agent's power access is P_i = s_i(Σ) · P_total, where s_i is (i) nondecreasing in Σ_i, (ii) nonincreasing in Σ_j ≠ i, and (iii) satisfies Σᵢ^sᵢ ≤ 1.

Axiom A9 (Stochastic perturbation). Each agent's state is subject to recurring perturbations ξ_k, k = 1,2,…, drawn independently from a Borel probability measure μ on R^d satisfying: (i) μ is absolutely continuous with respect to Lebesgue measure, with density q satisfying q(ξ) > 0 for all ξ in some open neighbourhood of 0; (ii) perturbation events recur, inter-arrival times are bounded above.

Construction authority is monotone in allocated power

The maintenance analysis established that every agent must supply a maintenance power floor P_maint,i ≥ ( δᵢξᵢ/η_II,i ) · Σᵢ before any surplus is available for construction. The metabolic multiplier μ_i ≥ 1 captures total power as a multiple of maintenance: P_i = μ_i · P_maint,i. Surplus power, the fraction available to fund construction, is ( μᵢ - 1 ) · P_maint,i.

Construction authority is therefore bounded by allocated power:

where αᵢ = ( μᵢ - 1 )/( μᵢ · ξᵢ ) captures the conversion efficiency from surplus power to new assembly-steps. This is not a new assumption, it is a direct consequence of the maintenance floor derived in Part 2. Higher P_i expands the set of achievable Σ̇ᵢ, which is the velocity set in the viability formalism. Lower P_i contracts it. The link between power access and construction authority is what makes the viability kernel respond to gradient allocation.

Lemma L1 (Monotonicity of control authority in allocated power). For agent i with maintenance floor δ_iΣ_i and construction bound C_i ≤ α_iP_i, the achievable velocity set V( Σᵢ,Pᵢ ) = Cᵢ - δᵢΣᵢ:0 ≤ Cᵢ ≤ αᵢPᵢ is monotone increasing in P_i under set inclusion. Consequently, the viability kernel Kᵢ( Pᵢ ), the set of states from which the tangent cone condition F( x,U(x) ) ∩ T_C(x) ≠ ∅ is satisfiable, is monotone increasing in the agent's power budget: P_i' ≤ P_i'' implies Kᵢ( Pᵢ' ) ⊆ Kᵢ( Pᵢ'' ).

Proof. The velocity set is an interval [ - δᵢΣᵢ, αᵢPᵢ - δᵢΣᵢ ]. Increasing P_i extends the upper endpoint while leaving the lower endpoint unchanged. The velocity set under P_i' is therefore a subset of the velocity set under P_i''. By Aubin's monotonicity theorem for viability kernels (Aubin, 1991, Theorem 3.2.4), enlarging the set of achievable velocities at every state can only enlarge the viability kernel. ▫

The monotonicity lemma is the foundational step of the competitive derivation. Without it, the competitive argument has no mechanism. With it, every subsequent step follows from the geometry. The lemma plays a dual role: it supports both the persistence argument, agents with larger power allocations have larger kernels, which feeds the geometric persistence principle (T2), and the gradient saturation result, surplus gradient creates expansion opportunity, which drives the saturation theorems (T3, T4).

Competitive allocation: A8 as structural axiom

The shared gradient imposes a coupling constraint: Σᵢ^Pᵢ ≤ Pₜₒₜₐₗ. This constraint alone does not define how agents compete for shares. The derivation requires an allocation mechanism, Axiom A8. In this framework, A8 is a structural axiom on the admissible class of share maps: it specifies the monotonicity properties that any physically consistent allocation mechanism must satisfy. These properties are motivated, not derived in the strict sense, by the maintenance analysis applied to N agents sharing a finite gradient. What the maintenance analysis establishes is that any allocation rule consistent with the per-agent coupling P_i = Γ_iΣ_i and the finite gradient Σ Pᵢ ≤ Pₜₒₜₐₗ must satisfy the three conditions of A8. This constrains the admissible class. It does not single out a unique mechanism.

The argument proceeds in three steps. First, from the maintenance analysis, every agent's power requirement is P_i = Γ_iΣ_i, where Γ_i = μ_iδ_iξ_i/η_II,i. This is not an assumption about allocation, it is the maintenance coupling applied per agent. An agent with stock Σ_i dissipates power Γ_iΣ_i as a physical consequence of maintaining that stock.

Second, in a competitive environment where total power is bounded by P_total, agents must access the gradient to sustain their stock. An agent that cannot access P_i ≥ Γ_iΣ_i loses stock to decay, this is not a modelling choice but a physical consequence of the assembly dynamics dΣ_i/dt = C_i - δ_iΣ_i with C_i bounded by available power.

Third, the share function follows from the structure of the coupling itself:

Each agent's minimum power claim is proportional to Γ_iΣ_i, and the gradient is finite. The three conditions of A8 are consequences of maintenance physics: (i) s_i is nondecreasing in Σ_i because a larger stock requires more power, (ii) s_i is nonincreasing in Σ_j ≠ i because competitors' claims reduce the available share, and (iii) Σᵢ^sᵢ ≤ 1 because the gradient is finite. No assumption about competitive behaviour or institutional structure is required. The only content beyond the per-agent maintenance coupling is the finiteness of the shared gradient, which is the waste heat ceiling (A5).

The specific contest function (the Garrett-derived Tullock-linear form) is the physically motivated instance, but the results below hold for any allocation mechanism satisfying the three monotonicity conditions of A8. Other allocation rules, priority queuing, price-mediated rationing, conflict/capture with higher exponents, yield the same qualitative result under the same conditions. This generality insulates the result against the objection that the specific contest function is assumed rather than derived.

A8 is therefore a well-motivated structural axiom. The monotonicity conditions are consequences of the maintenance physics; the specific allocation mechanism within the admissible class is not determined. The Garrett relation P = λ W, the empirical observation that motivates the framework, is the monetary shadow of the physical coupling P_i = Γ_iΣ_i. It provides calibration and confirmation. It is not foundational.

The geometric persistence principle: A9 and its consequences

The stochastic perturbation axiom replaces the behavioural assumptions used in the existing competitive viability literature, that agents "prefer" larger viability kernels or that agent frequency growth is an increasing function of kernel volume, with a physical property of the environment. Condition (i) is a regularity requirement on the perturbation law. Absolute continuity ensures that survival probability responds continuously to kernel geometry; the positive-density condition near zero ensures that small perturbations are possible, so that the environment is not exclusively catastrophic. Earlier drafts stated this axiom with "non-degenerate support," which is insufficient: a distribution can have full support while being purely singular (concentrated on a fractal of zero Lebesgue measure). The axiom now states the required regularity directly. Condition (ii) formalises "recurring perturbations": it prevents the degenerate case where an agent in a small kernel survives indefinitely because no perturbation ever arrives. Both conditions are physically minimal. Thermodynamic fluctuations are continuous random variables as a consequence of statistical mechanics; their civilisational analogues; resource shocks, climate variability, technological disruption, supply-chain failures, geopolitical disturbances, demand shifts, equipment failures, are empirically ubiquitous across all recorded history and all known biological systems. The stochastic perturbation axiom is a statement about the physical environment, not about agent behaviour. It is Tier 1.

Definition D5 (Local survival functional). For agent i at state x ∈ R^d, with viability kernel K ⊆ R^d and perturbation measure μ with density q, define the one-step survival probability:

The local survival functional is defined at a specific state x, not as a global property of the kernel. Survival under perturbation depends on where the agent is relative to the kernel boundary, not on the total volume of the kernel. An agent at the centre of a small kernel may be more robust than an agent at the edge of a large one.

Lemma L2 (Survival monotonicity under kernel inclusion). Let K_1 ⊆ K_2 be Borel subsets of R^d. Then for any x ∈ R^d and any Borel probability measure μ:

If μ is absolutely continuous with respect to Lebesgue measure, as required by the stochastic perturbation axiom (A9(i)), and λ^d( y ∈ K₂ ∖ K₁:q(y - x) > 0 ) > 0, the inequality is strict.

Proof. The weak inequality is monotonicity of measures: K_1 ⊆ K_2 implies y:y - x ∈ K₁ ⊆ y:y - x ∈ K₂, hence μ( · ∩ K₁ ) ≤ μ( · ∩ K₂ ). For the strict inequality under absolute continuity: s( x,K₂,μ ) - s( x,K₁,μ ) = ∫_K₂ ∖ K₁^q(y - x) dy > 0 whenever the integrand is positive on a set of positive Lebesgue measure, which holds by hypothesis. ▫

The survival monotonicity lemma replaces an earlier formulation stated in terms of global kernel volume. The global-volume formulation is weaker and can mislead: a kernel with large total volume but thin remote lobes may have worse local robustness at the agent's actual operating state than a compact kernel of smaller volume. The set-inclusion formulation avoids this problem. What the monotonicity lemma (L1) provides is precisely set inclusion, Kᵢ( Pᵢ' ) ⊆ Kᵢ( Pᵢ'' ) when P_i' ≤ P_i'' so L2 applies directly to the objects that the competitive argument produces.

Proposition P3 (Terminal complement). An agent whose state exits its viability kernel will eventually violate a constraint.

Proof. Contrapositive of the viability kernel definition (D4). The viability kernel is the set of all states from which there exists at least one admissible control trajectory remaining in C forever. Outside the kernel, no such trajectory exists. ▫

Proposition P4 (Differential persistence). Consider a population of N agent types. At each perturbation epoch k, each agent independently experiences a perturbation ξ drawn from μ, as specified by the stochastic perturbation axiom (A9). An agent of type i at state x_i survives if and only if x_i + ξ ∈ K_i; the probability of survival is sᵢ = s( xᵢ,Kᵢ,μ ). An agent that exits its kernel is on a terminal trajectory, by the terminal complement proposition (P3), and is eventually removed. Then the expected frequency of type i after k perturbation epochs satisfies:

Consequently: (i) if s_i > s_j, type i's expected frequency grows relative to type j's at exponential rate ln( sᵢ/sⱼ ) per epoch; (ii) by survival monotonicity (L2), if K_j ⊂ K_i (strict inclusion with positive-measure difference at the operating state) then s_j < s_i, and type j is eliminated relative to type i.

Proof. At each epoch, each agent of type i survives independently with probability s_i. The expected number of survivors after k epochs is n_i(0) · s_i^k. The expected frequency is E[ fᵢ(k) ] = nᵢ(0)sᵢ^k/Σⱼ^nⱼ(0)sⱼ^k. For types with s_i > s_j, the ratio f_i(k)/f_j(k) grows as ( sᵢ/sⱼ )^k arrow ∞. ▫

In the continuous-time limit with perturbation frequency ν, this discrete process yields the approximation ḟᵢ ≈ ν · fᵢ · ( ln sᵢ - Σⱼ^fⱼln sⱼ ), a replicator-like equation with log-survival as fitness. This is stated for interpretive convenience. The primary result is the discrete-time frequency evolution in the differential persistence proposition (P4), which requires no ODE approximation.

The differential persistence proposition establishes frequency dynamics for fixed survival probabilities. In the competitive system, survival probabilities co-evolve with the stock vector: as agents grow or shrink, power allocations shift (A8), kernels resize (L1), and survival probabilities update (L2). The following lemma establishes that the competitive coupling is self-reinforcing, kernel-size advantages are preserved, not reversed, by the co-evolutionary process.

Lemma L3 (Self-reinforcing competitive dynamics). Under A5–A9, the competitive coupling is self-reinforcing with respect to the viability-kernel ordering. At any perturbation epoch:

(i) An agent with a larger viability kernel has strictly higher survival probability (by L2).

(ii) When an agent exits its kernel, it enters a terminal trajectory (P3) and its stock declines (A3). As its stock falls, its share weakly decreases (A8(i)) and competitors' shares weakly increase (A8(ii)). By the monotonicity lemma (L1), the surviving agents' kernels weakly enlarge while the declining agent's kernel weakly contracts.

(iii) Between perturbation epochs, agents with larger power allocations have higher construction bounds C_i ≤ α_iP_i (A6), permitting faster stock growth, which weakly increases their shares (A8(i)) and weakly decreases competitors' shares (A8(ii)), further enlarging their kernels (L1).

No mechanism in A5–A9 systematically favours smaller-kernel agents. Perturbations drawn from the common measure μ (A9) are unbiased with respect to agent identity; the asymmetry arises entirely from the kernel-size difference via L2. The competitive coupling therefore amplifies kernel-size differences in expectation.

Proof. (i) is survival monotonicity (L2) applied at the agents' operating states. (ii) When agent j exits K_j, it is on a terminal trajectory by P3 and its stock declines by A3. As Σ_j falls, s_j decreases (A8(i)) and s_i weakly increases for all i ≠ j (A8(ii)). By L1, K_i(P_i) weakly enlarges. (iii) Between epochs, the construction bound α_iP_i is increasing in P_i (A6), so agents with larger allocations can build faster. Increased Σ_i weakly raises s_i (A8(i)) and weakly lowers s_j (A8(ii)), with the kernel consequences following from L1. For the overall claim: the only mechanism that could reverse the kernel ordering is a perturbation that damages the larger-kernel agent more than the smaller-kernel agent. By L2, this outcome has strictly lower probability than the reverse at every epoch. The expected direction of the competitive reallocation therefore reinforces the existing ordering. ▫

Theorem T2 (Geometric persistence principle). Under A5–A9, over sufficient time with recurring perturbations drawn from a common measure, agents with larger viability kernels persist with higher probability, and the long-run population measure concentrates on large-kernel types.

Proof. By the monotonicity lemma (L1), agents with larger power allocations have larger viability kernels: K_i(P_i) ⊆ K_i(P_i + ε) for ε > 0. By survival monotonicity (L2), agents with larger kernels have strictly higher one-step survival probability at comparable operating states. By differential persistence (P4), types with higher survival probabilities dominate at exponential rate per epoch. By self-reinforcing competitive dynamics (L3), the survival-probability ordering is not reversed by the inter-epoch competitive co-evolution: the competitive coupling (A8) and the asymmetric survival probabilities (L2) systematically favour larger-kernel agents, and no mechanism in A5–A9 provides a countervailing force. By the terminal complement proposition (P3), agents that exit their kernels are on terminal trajectories and are eventually removed. Over k perturbation epochs (with k → ∞ guaranteed by the recurrence condition A9(ii)), the population measure concentrates on large-kernel types. ▫

Remark R1 (No cognition required). The geometric persistence principle does not claim that agents prefer, choose, or perceive their kernel geometry. It claims that agents with larger kernels survive more perturbations, as a statistical consequence of occupying a larger region of viable state space under a well-behaved perturbation law. A bacterium does not compute its viability kernel. It persists longer when its viable operating range is wider, because random environmental fluctuations are less likely to push it into a lethal regime. The same logic applies to firms, nations, and civilisations.

Remark R2 (Short-termist governance resolved). An agent that systematically mismeasures its kernel, responding to quarterly earnings rather than long-term viability, may make individually poor decisions. But the geometric persistence principle does not require optimal decisions. It requires only that larger kernels provide more buffer against perturbation. Selection operates on the geometry, not on the agent's perception of the geometry.

Gradient saturation: T3 and T4

Definition D6 (Dynamically persistent configuration). A configuration ( Σ₁,…,Σ_N ) is dynamically persistent under the perturbation process ξₖ if, for each agent i, the probability that agent i's state remains in K_i for all future perturbation epochs is bounded away from zero.

Theorem T3 (Gradient saturation). Consider N agents satisfying A5–A8 on a shared gradient P_total. Under the stochastic perturbation axiom (A9), any dynamically persistent configuration satisfies Σᵢ^Pᵢ = Pₜₒₜₐₗ.

Proof. Suppose, for contradiction, that a dynamically persistent configuration has Σᵢ^Pᵢ < Pₜₒₜₐₗ, leaving surplus gradient Δ P > 0 uncaptured.

By the per-agent assembly dynamics axiom (A6), any agent i can allocate surplus power to increase construction C_i, thereby increasing Σ_i. By the monotone share map (A8), increased Σ_i weakly increases s_i, expanding agent i's power allocation. By the monotonicity lemma (L1), a larger power allocation expands agent i's viability kernel: Kᵢ( Pᵢ ) ⊂ Kᵢ( Pᵢ + ε ) for any ε > 0 funded from the surplus.

Compare two sub-populations: agents that expand into the surplus gradient (type E) and agents that do not (type S). Type E agents acquire strictly larger kernels than type S agents, by the monotonicity lemma. By survival monotonicity (L2), type E agents have strictly higher one-step survival probability. By differential persistence (P4), type S agents are eliminated at exponential rate.

Therefore, in any configuration with Σᵢ^Pᵢ < Pₜₒₜₐₗ, non-expanding agents are eliminated under recurring perturbations, they are not dynamically persistent in the sense of D6. Any dynamically persistent configuration must have Σᵢ^Pᵢ = Pₜₒₜₐₗ. ▫

Remark R3 (Nash equilibrium relationship). If one defines a stage game in which each agent's strategy is its growth rate and the payoff is s_i, then any Nash equilibrium satisfies Σ Pᵢ = Pₜₒₜₐₗ by the same logic: at any profile with surplus gradient, each agent has a profitable deviation. The gradient saturation theorem is therefore consistent with a Nash characterisation; the formal game-theoretic treatment belongs in a companion work. Here, the dynamic result is primary because it requires less apparatus.

Remark R4 (Concentration versus uniqueness). T2 establishes that the population measure concentrates on agents whose viability kernels are maximal under the prevailing gradient allocation. It does not establish that the resulting equilibrium allocation is unique. Multiple saturating configurations may coexist under different initial conditions or coordination regimes. Whether the equilibrium is unique, and under what conditions it can be selected or shifted, is deferred to Open Problem 1 (equilibrium uniqueness) and Open Problem 2 (coalition behaviour).

Theorem T4 (Below-saturation transience). Under A5–A9, the set of configurations with Σᵢ^Pᵢ < Pₜₒₜₐₗ is transient. For any initial configuration with surplus gradient, the expected number of perturbation epochs until either (a) the gradient is fully saturated or (b) all non-expanding agents are eliminated is finite.

Proof. At any below-saturation configuration, surplus Δ P > 0 exists. By the argument in the gradient saturation proof, expanding agents acquire strictly larger kernels and strictly higher survival probabilities than non-expanding agents. By differential persistence (P4), the frequency ratio of expanding to non-expanding agents grows as ( s_E/s_S )^k where s_E > s_S strictly, by survival monotonicity (L2) and the monotonicity lemma (L1).

As expanding agents grow, their total power claim increases, by the monotone share map (A8(i)). The surplus Δ P decreases. The process terminates when either Δ P = 0 or all non-expanding types are eliminated. ▫

The maximum power principle, total dissipation saturates the available gradient, is derived from the axiom set A5–A9. It is not imported from ecology. The gradient saturation and below-saturation transience theorems do not claim that maximum power is "optimal" in any welfare sense, nor that the saturating allocation is unique, nor that the result holds for cooperative solutions. They claim that gradient saturation is the generic noncooperative outcome under competitive allocation on a bounded gradient.

Scaling Estimate SE1 (Convergence timescale) [HEURISTIC]. The elimination bound in the below-saturation transience theorem (T4) gives:

where ν is the perturbation frequency. Under the derived replicator dynamics, the deviation from full saturation Δ P = Pₜₒₜₐₗ - Σᵢ^Pᵢ decreases at a rate governed by the faster of two processes: (a) the growth rate of the fastest-expanding agent (active exploitation of available gradient), and (b) the perturbation-driven elimination rate of below-saturation agents. If perturbations arrive at frequency ν and the per-event survival differential between larger-kernel and smaller-kernel agents is Δ p, then the selection timescale is τₛₑₗ = O( 1/(ν · Δ p) ).

Bound. Process (a) operates at the growth rate of Σ_i. From the assembly dynamics, Σ̇ᵢ/Σᵢ ≤ αᵢPᵢ/Σᵢ - δᵢ. For agents expanding into unsaturated gradient, this gives a stock growth rate on the order of a few percent per year, consistent with the historical 2–3% annual growth of civilisational power throughput. Process (b) operates at rate ν · Δ p. Perturbation events at civilisational scale; recessions, resource shocks, geopolitical disruptions, technological displacements, occur on sub-decadal timescales (ν ~ 0.1–0.5 yr^- 1). The per-event survival differential Δ p depends on the kernel-volume ratio, but even modest differentials compound rapidly over many events. The overall convergence timescale is therefore on the order of decades to a century, short relative to the multi-century horizon of the waste heat ceiling. The gradient saturation attractor is not merely stable in the long run; it is relevant on the timescales over which civilisational decisions are made.

This bound is admittedly loose. The point is not precision but the existence of a characterised timescale that is short relative to civilisational decision windows. A tighter bound would require specifying the perturbation distribution and the kernel-volume function explicitly, a task for a companion mathematical treatment. What the bound establishes is sufficient for the essay's argument: gradient saturation is not a distant asymptotic property but a force that operates within the planning horizons of existing institutions.

Corollaries

Corollary C1 (Jevons paradox as viability expansion plus competition). An efficiency improvement that increases α_i (construction per unit power) expands the viability kernel, by the monotonicity lemma (L1). Under the gradient saturation and below-saturation transience theorems (T3, T4), the freed gradient is captured by competitors or recycled into growth. Aggregate power does not fall. The rebound is not a market failure, it is a geometric consequence of competitive allocation on a finite gradient.

Corollary C2 (Competitive exclusion as geometric filtering). For any given agent, operating below maximum power yields a strictly smaller viability kernel than operating at maximum power, by the monotonicity of control authority in allocated power (L1). In a competitive allocation, an agent that captures a smaller share of the gradient therefore maintains a smaller kernel relative to its own potential, not relative to other agents with different structural parameters. Under the geometric persistence principle (T2), below-frontier configurations are progressively emptied by perturbations that push state vectors across kernel boundaries from which no return is possible, as established by the terminal complement proposition (P3). The population measure concentrates on agents operating at or near their power frontier, the maximal share consistent with their competitive position, not because they are "selected for," but because each agent's kernel is largest at maximal power, and larger kernels retain state vectors that smaller kernels lose, as survival monotonicity (L2) guarantees. This is geometric filtering: the viable region's shape, evaluated within each agent's own parameter space, determines which configurations persist.

The cross-agent consequence follows from the competitive coupling: when agent i captures more gradient, competitors receive less, contracting their kernels relative to their own maxima. The filtering is therefore both within-agent (each agent's kernel shrinks below its frontier) and between-agent (gradient captured by one agent is gradient denied to another). But the foundational comparison is always within-agent, the monotonicity lemma compares Kᵢ( Pᵢ ) at different values of P_i, not K_i against K_j.

Corollary C3 (Garrett's stable Γ as equilibrium signature). If the system is at the noncooperative equilibrium of the competitive viability dynamics, and the institutional landscape has not undergone a structural shift, then the aggregate coupling Γ should be approximately stable. This is what the empirical record confirms (Garrett et al., 2022). The stability of λ (approximately 5.9 mW per 2019 USD, invariant to within measurement uncertainty over the period 1970–2019) is not a coincidence. It is the expected signature of a system at or near gradient saturation.

Corollary C4 (Landscape dependence). The gradient saturation and below-saturation transience theorems (T3, T4) establish that Σ Pᵢ = Pₜₒₜₐₗ at equilibrium. This is the stiffness result: gradient saturation is structural. But the decomposition Γ = μδξ/η_II contains four independently adjustable components. The attractor value of Γ depends on the institutional landscape, which technologies are deployed, which governance structures allocate resources, which efficiency standards are enforced. The ceiling is Tier 1. The trajectory toward it is Tier 2. The landscape that sets the attractor value of Γ is Tier 3.

Proof. The gradient saturation condition Σᵢ^Pᵢ = Pₜₒₜₐₗ is derived from A5–A9, none of which reference institutional parameters. It therefore holds for any institutional landscape. But the equilibrium allocation sᵢ* depends on the competitive parameters Γᵢ,δᵢ,αᵢ, which are shaped by institutional architecture. Different landscapes produce different equilibria, all satisfying full saturation. The stiffness is structural (Tier 2). The location is institutional (Tier 3). ▫

Three consequences follow immediately.

First, Hanley's finding that λ (the monetary shadow of Γ) was not stable before 1970 is empirical confirmation: the landscape changed (Bretton Woods collapse, credit deregulation), and the attractor location shifted, while gradient saturation itself persisted through the transition. The pre-1970 instability and post-1970 stability are both predicted by the corollary.

Second, the biosphere existence proof provides a concrete demonstration. Biospheric assemblages have sustained low-Γ equilibria for billions of years, gradient saturation at a value of Γ that is orders of magnitude below the industrial economy's. The corollary explains why: the competitive landscape of biological evolution (no debt architecture, no compound interest, long effective governance timescales via genetic memory) selects a different equilibrium. Low-Γ equilibria exist. The question is whether the institutional landscape can be engineered to select one.

Third, the engineering consequence: the task is not to suppress competition (impossible without overriding the Tier 2 equilibrium) but to redesign the institutional landscape so that competition converges on a less lethal equilibrium value of Γ. The biosphere proves such a low-Γ equilibrium exists. Hanley proves the attractor can move. The open question is speed and magnitude of deliberate landscape engineering.

The complete axiom set: Label, Content, Source, Status

A5, Finite gradient Waste heat ceiling, (§2.3), Derived from T1

A6, Per-agent assembly dynamics, Maintenance analysis per agent, Derived from A1–A3

A7, Minimum viable stock, Thermodynamic collapse threshold, Operational assumption

A8, Monotone share map, Maintenance coupling on finite gradient, Structural axiom

A9, Stochastic perturbation, Statistical mechanics / empirical, Physical realism

The complete logical chain: Second Law arrow maintenance axioms (A1–A4) arrow maintenance floor arrow per-agent coupling arrow per-agent dynamics and minimum viable stock (A6, A7) arrow share map monotonicity (A8) arrow finite gradient (A5) arrow stochastic perturbation (A9) arrow survival monotonicity (L2) arrow differential persistence (P4) arrow geometric persistence (T2) arrow gradient saturation (T3, T4) arrow convergence timescale (SE1) arrow landscape dependence (C4).

Operational mechanisms of competitive exclusion

The geometric result, below-maximum-power agents have smaller kernels and occupy less of the viable state space, manifests through specific operational channels at each organisational level.

At firm level, reduced throughput translates to narrower operating margins and smaller cash reserves. When demand shocks, supply disruptions, or credit contractions arrive, these firms lack the buffer to absorb the perturbation. They become acquisition targets, lose access to capital markets, and shed the assembly stock that constituted their competitive position. The mechanism is not mysterious: it is the viability kernel expressed in balance-sheet terms.

At national level, reduced throughput means smaller military capacity, degraded infrastructure investment, loss of technological leadership, and capital flight as mobile resources migrate toward higher-throughput jurisdictions. The perturbations that test national kernels, geopolitical crises, resource conflicts, pandemic shocks, are the events that push state vectors toward and across kernel boundaries. Nations with larger throughput maintain larger kernels, larger buffers in every dimension that matters for persistence, and therefore retain their state vectors under perturbations that eject smaller-kernel nations from the viable region entirely.

At the level of competing technologies, a slower technology occupies a narrower operating envelope. When environmental conditions shift, input prices change, regulatory frameworks tighten, consumer preferences evolve, the technology with the broader envelope survives while the narrower one is displaced by adoption migration. The common channel across all three levels is that reduced throughput shrinks the buffer against perturbation, the viability kernel result expressed in operational terms.

Connection to Part 3 evidence

The empirical patterns presented in the evolutionary dynamics analysis are now explained as consequences of gradient saturation under competitive viability rather than independent empirical imports. The Lotka/Odum maximum power observation is a special case of the gradient saturation theorem (T3). The Brockway meta-analysis and Saunders data on economy-wide rebound are instances of the Jevons corollary (C1). The Garrett coupling is the equilibrium signature described by the stable-Γ corollary (C3). The formal result predicts exactly the patterns observed.

Scope and limitations

The derivation establishes that under competitive allocation on a shared gradient, gradient saturation is the generic noncooperative outcome, a consequence of the geometric persistence principle (T2) derived from stochastic perturbation (A9) and the monotonicity of kernel size in allocated power (L1). Maximum power is a theorem of competitive viability dynamics, not an empirical import from ecology. Jevons rebound and competitive exclusion are corollaries. The attractor value of Γ depends on the institutional landscape, as established by the landscape dependence corollary (C4).

The derivation assumes A5–A9 as stated: a finite shared gradient, per-agent maintenance coupling, minimum viable stock thresholds, a monotone share map, and absolutely continuous recurrent perturbations with positive density near zero. It assumes independence of agents' maintenance parameters from competitors' states and independence of perturbation draws conditional on the perturbation law. The absolute continuity requirement is stated explicitly in the stochastic perturbation axiom (A9) and is not derived from weaker conditions.

The derivation does not establish uniqueness of the equilibrium allocation, behaviour under coalition formation, or sharp convergence bounds. The relationship between gradient saturation and Nash equilibrium is noted as consistent (R3) but not proved. These open problems are catalogued in the Formal Results Register.

Tier 2 synthesis

The preceding analyses, single-agent viability and competitive multi-agent dynamics, assumed fixed values for δ, Γ, ε, and the institutional parameters that determine C_max and the admissible control set. But the power-assembly coupling Γ(t), the power required per unit of assembly stock, is not a Tier 1 constant. It is a Tier 2 parameter: set by the competitive equilibrium of the multi-agent viability dynamics.

The competitive viability derivation establishes why Γ appears empirically stable: it sits at a noncooperative equilibrium of the multi-agent viability game, and the stiffness of that equilibrium is a Tier 2 property. Reducing Γ would require either a cooperative solution, all agents simultaneously accepting kernel contraction on binding commitment, or a change in the institutional landscape that shifts the noncooperative equilibrium to a lower value. The decomposition Γ = μδξ/η_II identifies four independently adjustable components; metabolic multiplier, decay rate, specific exergy cost, and Second Law efficiency, each with real engineering headroom. A factor-of-2 reduction in Γ doubles Σ_max. A factor-of-10 reduction (aggressive on all four components simultaneously) adds approximately 100 years at historical growth rates. These are engineering targets, not violations of physics. The competitive resistance is real, it is Tier 2, derived from the viability geometry, but it does not reduce the engineering headroom to zero.

Two caveats bound the strength of this Tier 2 result.

First, stability of Γ. The trajectory projections depend on Γ remaining near its current value. Garrett's 1970–2019 data show remarkable stability, but Hanley (2025) showed instability in the pre-1970 regime. The competitive viability derivation explains both: the post-1970 stability is the signature of a Nash equilibrium under the prevailing institutional landscape, and the pre-1970 instability reflects a different competitive landscape with a different equilibrium value. If the institutional landscape shifts again, as it has before, Γ shifts with it. A lower Γ widens the kernel; a higher Γ narrows it. The qualitative structure of the viability problem survives any value of Γ. The quantitative runway claims do not.

Second, strength of the Tier 2 ratchet. The competitive viability derivation establishes that gradient saturation is the generic noncooperative outcome under the geometric persistence principle, a stronger foundation than the probabilistic selection assumption it replaces, but one that still does not rule out cooperative solutions. If stable, scalable institutions exist that can implement a binding cooperative agreement among all N agents, the simultaneous kernel contraction on commitment that the derivation identifies as the override condition, the competitive bias is weaker than claimed and the kernel is wider than computed. No such institutions have been demonstrated at civilisational scale. But the absence of precedent is not the same as a physical prohibition. The essay's position is the strongest claim the evidence supports, not the strongest claim imaginable.

Part 5 — Implications

5.1 Fermi's Paradox and the Great Filter

The preceding four parts have established, for a single civilisation on a single planet: an arena bounded by radiative physics and the Second Law; a machine whose maintenance power scales linearly with its accumulated stock; a trajectory driven toward the waste heat ceiling by competitive dynamics and amplified by institutional architecture; and a viability kernel that contracts as the state vector approaches the ceiling. The result is a system whose physics demands a ceiling, whose evolutionary dynamics drive it toward that ceiling, and whose institutional architecture accelerates the approach.

This section asks whether that predicament is local or universal.

The question

In 1950, Enrico Fermi converted a lunch-table conversation about flying saucers into a quantitative puzzle. Given the number of stars in the galaxy, the probable fraction with planets, the probable fraction that might develop life, and the age of the galaxy relative to the time required for interstellar colonisation, the Milky Way should be saturated with technological civilisations. "Where is everybody?"

The formal version is usually expressed through Drake's equation (Drake, 1961), which decomposes the number of detectable civilisations into a product of astrophysical and biological probabilities. The equation organises ignorance but has limited predictive power, several terms are unknown to within many orders of magnitude. The interesting content of the paradox is the observation: no unambiguous evidence of extraterrestrial intelligence has been detected, despite decades of search and a galaxy old enough for any spacefaring civilisation to have colonised it many times over.

Robin Hanson (1998) formalised the inference. If the galaxy appears empty despite sufficient age and size, then somewhere between "habitable planet forms" and "galaxy-spanning civilisation detectable across interstellar distances" there must be one or more steps of extraordinarily low probability, the Great Filter. The central question is its location: past (our existence is the improbable event) or future (our survival is the improbable event)?

Standard proposed resolutions; rare life, rare intelligence, self-destruction through war or pandemic, deliberate silence, rely on assumptions about biology, sociology, or technology that are essentially unconstrained by data. This essay proposes a resolution grounded in the three-tier structure established across the preceding parts. The thermodynamic predicament identified for human civilisation is not an accident of particular history. It is a consequence of physics that applies to any civilisation, anywhere, composed of any chemistry, operating under any form of governance.

Tier 1: The universal physics

Three results from the machine analysis depend on nothing but thermodynamics and radiative physics.

The maintenance obligation. Any civilisation is assembled matter. The Second Law guarantees that every high-assembly structure is thermodynamically unstable relative to its environment. Maintenance power scales with stock: P_maint ≥ ( δξ/η_II ) · Σ. This holds for any dissipative structure that accumulates complexity, silicon-based organisms running on geothermal gradients face the same obligation as carbon-based ones running on stellar radiation. The specific δ differs; the direction is absolute. No technology can reduce δ to zero.

The waste heat ceiling. Every watt consumed within a planetary system is ultimately dissipated as waste heat, radiated from a finite surface:

For any planet with a fixed radiating surface, there exists a maximum sustainable assembly stock Σ_max set by the condition that total power does not push T_eq past the habitability threshold. The specific value depends on planetary parameters; stellar flux, surface area, atmospheric composition, biological thermal tolerance, but its existence is universal.

The Landauer floor. Every irreversible logical operation dissipates at least k_BTln2 joules (Landauer, 1961; Bérut et al., 2012). Any civilisation detectable across interstellar distances necessarily processes enormous volumes of information and faces an irreducible energy cost for that processing. The nine-order-of-magnitude gap between current hardware and the Landauer limit is headroom for the next expansion of computational demand, not evidence that computation can be made thermodynamically free.

These three results define the arena for any planetary civilisation. They establish that civilisation has a maximum sustainable size on a finite planet, determined by the ratio of planetary radiative capacity to the maintenance power density of civilisational stock. No biology, no culture, no institutional architecture enters the derivation.

Tier 2: The universal dynamics

The arena defines where a civilisation can exist. It does not explain why civilisations appear unable to stabilise within it. The explanation requires the second tier: the competitive dynamics of dissipative structures sharing a finite energy gradient.

The competitive viability analysis (§4.3) established that when N agents compete for a shared energy budget P_total, each agent's viability kernel is monotone increasing in its power budget, and unused gradient is dynamically unstable, any agent that captures surplus weakly expands its own kernel while contracting competitors'. The noncooperative equilibrium generically satisfies Σ Pᵢ = Pₜₒₜₐₗ: total dissipation saturates the available gradient. This is gradient saturation, the maximum power result, derived not as an ecological empirical generalisation but as a consequence of competitive viability geometry.

Three corollaries follow directly. First, the Jevons mechanism: efficiency improvements increase construction authority α_i, expanding the velocity set; competition recycles the gain into throughput until the constraint binds again. The aggregate result is Tier 2, it holds regardless of economic system. The specific channel (price-mediated in market economies, surplus allocation in command economies) is Tier 3. Second, competitive exclusion: agents operating below maximum power have smaller viability kernels and are more vulnerable to perturbation, at firm, national, and technology levels. Third, the stability of Γ(t): Garrett's empirical finding that the power-wealth coupling has remained approximately constant for decades is the signature of a Nash equilibrium in the competitive viability game.

The competitive viability equilibrium is not a law of physics in the strict sense. It is a derived property of competing dissipative structures in a finite environment. Overriding it requires a cooperative game solution, all N agents simultaneously accepting kernel contraction on binding commitment, a magnitude of coordination without precedent in any known competitive system. Its possibility cannot be excluded, but its difficulty should not be understated.

The geometric persistence principle, that persistence is inclusion in the viable region and requires no mediating selection mechanism, makes the universality claim independent of any assumption about alien biology, sociology, or cognitive architecture. Any ensemble of dissipative structures competing for a shared gradient on a finite surface exhibits gradient saturation, because the geometry of the viable region demands it.

Tier 3: The contingent amplifiers

Compound-interest debt, endogenous money creation, and governance structures with electoral cycles far shorter than the system's physical time constants, these accelerate the approach to the ceiling by creating structural growth imperatives and preventing the control vector from operating at the required timescale. These amplifiers are not universal. An alien civilisation need not have invented money or organised through debt. It may have governance horizons measured in centuries.

But the trajectory toward the ceiling exists with or without the amplifiers. The first two tiers drive the approach independently of institutional design. Debt makes the problem worse and faster. It does not create the problem. A civilisation with no debt, no money, and perfect governance would still face P_maint ≥ ( δξ/η_II ) · Σ, still encounter Σ_max, and still confront the competitive viability equilibrium. Dispensing with the amplifiers does not resolve the predicament. It merely slows the approach.

The thermal filter: timescales

Any civilisation that couples wealth accumulation to power throughput on a finite radiative surface encounters the waste heat ceiling within a calculable number of doubling times:

where n is the number of doublings from current power P_0 to the ceiling P_ceiling, and the time to the ceiling is n × t_double.

For Earth, P_0 ≈ 18 TW. Taking P_ceiling ≈ 1,200 TW (approximately 1% of solar absorption, the onset of significant direct heating) gives n ≈ 6.1 doublings. At the historical growth rate of approximately 2.3% per year (t_double ≈ 30 years), the ceiling arrives in approximately 180 years. At a more conservative 1% (t_double ≈ 70 years), approximately 430 years. Balbi and Lingam (2025) found loss of habitable conditions within approximately 1,000 years at 1% annual growth.

An important qualification: the 180-year figure assumes sustained exponential growth at the historical rate, a Tier 2/3 parameter, not a physical constant. Civilisations that override the competitive viability equilibrium through coordinated restraint could extend the timeline by factors of 2–5. Even so, the ceiling arrives within approximately 1,000 years, vanishingly brief on galactic timescales. The ceiling is physics; the timeline to the ceiling is scenario.

The Milky Way is approximately 13.6 billion years old. The gap between billions of years of opportunity and hundreds of years of viability is the quantitative core of the thermodynamic resolution to Fermi's paradox. Civilisations that grow exponentially die too quickly to be noticed.

Why the standard escape fails

The obvious objection is that civilisations need not remain planet-bound. Expanding into space, harvesting stellar energy directly, building habitats with larger radiative surfaces, should provide relief. This is the logic behind the Kardashev scale (Kardashev, 1964). On close examination, it is thermodynamically naïve, not because it violates any law of physics, but because it ignores the constraints on constructing and maintaining the megastructures it postulates.

The luminosity bottleneck. A Dyson swarm requires a planet's worth of material. The standard proposal is to disassemble Mercury (mass approximately 3.3 × 10^23 kg). Extracting that mass requires at minimum the gravitational escape energy plus mining, excavation, and transport, total on-surface waste heat of order 5 × 10^6 J per kilogram launched. This waste heat is generated on Mercury's surface and must be radiated from Mercury's surface. At its natural dayside temperature of approximately 700 K:

Maximum extraction rate: Ṁₘₐₓ ≈ 1.8 × 10¹¹ kg/s. Disassembly time: approximately 57,000 years. Beaming energy from the growing swarm to heat Mercury's surface to 1,500 K reduces this to approximately 2,800 years. At 3,000 K, approaching the working temperature of tungsten refractory equipment, approximately 170 years. Constructing a full Dyson swarm within the habitability window of 200–300 years requires operating robotic mining across Mercury's entire surface at temperatures near the limits of known refractory materials.

The maintenance trap. The constraint does not vanish once the swarm is built. Any physical structure in the space environment degrades: micrometeorite impacts, solar wind sputtering, thermal cycling, radiation damage. This is the maintenance obligation P_maint ≥ δΣ operating at megastructure scale. At 0.1% annual degradation, the replacement mass flux demands source body surface temperatures of approximately 1,900 K, demanding but feasible. At 1% degradation, approximately 3,300 K, approaching the melting point of tungsten. At 3% degradation, approximately 4,800 K, no known refractory material survives. The Dyson swarm is not a self-sustaining energy cornucopia. It is a thermodynamic structure subject to the same maintenance obligation as any assembled matter.

The repair-apparatus recursion. In-situ recycling of degraded swarm material faces two obstacles. For ablative degradation (sputtering, micrometeorite ejecta), the mass is physically dispersed, there is nothing to recycle. For structural degradation (fatigue, embrittlement), a space-based foundry could reprocess the mass, but the foundry is itself assembled matter with its own δ_foundry. Active machinery degrades faster than passive film; the repair system's maintenance burden almost certainly exceeds the cost it displaces. The cleanest architecture is passive, low-assembly reflectors accepted as consumable and replaced by fresh mass from a gravity well. The bottleneck stands.

The bootstrap trap. A civilisation must develop autonomous space-industrial capability, bootstrap self-replicating space industry, scale to Mercury-level mining at refractory temperatures, and do all of this within the thermal window being simultaneously consumed by total energy growth. The civilisation cannot pause home-planet growth while the bootstrap proceeds. The competitive viability equilibrium ensures this: agents that grow outcompete agents that stagnate. The civilisation needs continued growth to fund the space-industrial transition, but that growth generates the waste heat that closes the habitability window. Murphy (2021) noted, unless a civilisation moves to space it cannot expand without seriously heating its home world, correct, but it restates the problem: the civilisation must migrate its entire economic base before the thermal clock runs out.

What the infrared sky tells us

If Dyson swarms existed, they would be detectable, a star surrounded by collectors re-radiates intercepted starlight at lower temperature, producing a characteristic mid-infrared excess. The Glimpsing Heat from Alien Technologies survey (G-HAT) analysed WISE data across approximately 100,000 galaxies (Griffith et al., 2015). The result was null. No galaxy showed evidence of a civilisation capturing more than approximately 85% of its host galaxy's starlight.

The thermodynamic reading: Dyson swarms are rare not because they are physically impossible but because the construction window is too narrow and the maintenance too demanding for most civilisations to complete and sustain the project before their home planets become uninhabitable. The galaxy is not filled with Kardashev II civilisations because the path from Kardashev I to Kardashev II passes through a thermal bottleneck, on the home planet, on the source bodies, and in the timing of the bootstrap, that kills most travellers.

The filter characterised

The three-tier structure allows the Great Filter to be characterised with a precision that standard treatments lack.

The filter exists. The waste heat ceiling is technology-independent and applies to any planetary civilisation operating on thermodynamic principles. The maintenance obligation is universal: any civilisation that builds assembled structures faces entropic decay and must supply maintenance power that scales with stock. The ceiling and the floor are Tier 1, they cannot be engineered away.

The filter is dynamic, not static. The competitive viability equilibrium established in the competitive viability analysis (§4.3) shows that maximum power throughput is the Nash equilibrium of the multi-agent viability game. This result depends on competition for shared energy gradients, a feature of any system containing multiple dissipative structures. The evolutionary ratchet is Tier 2: it holds for any competitive system, not just Earth's specific institutional architecture. The filter is not a wall that civilisations hit. It is a trajectory that carries them toward the wall.

The amplifiers are contingent. Compound-interest debt, endogenous money creation, short governance timescales, these are Tier 3. They accelerate the trajectory but do not create it. A civilisation with different financial architecture still faces the ceiling and the ratchet. It approaches the ceiling on a different timeline.

The timescale argument. The filter predicts that civilisations face a decision window of centuries to millennia, the interval between achieving planetary-scale energy consumption and reaching the waste heat ceiling. Against the galactic timescale of billions of years, this is an extremely narrow window. Even a small probability of failure per civilisation per window produces the observed silence. The filter is not a single event in the past or future. It is a continuous, tightening constraint that begins the moment a civilisation enters exponential growth and becomes binding within a timescale set by the doubling rate and the planetary radiative capacity.

The distinction between the merely difficult (institutional reform, Tier 3) and the physically impossible (repealing the Second Law, Tier 1) is the most important contribution the three-tier framework makes to the Fermi question. A civilisation that understands the tier structure knows what it can change and what it cannot. A civilisation that conflates the tiers, treating financial conventions as physical law, or physical law as policy choice, will misallocate its effort and narrow its own window.

The specific strategic orientations available within this window, and their viability properties, depend on the full control-surface analysis developed in the next section.

5.2 The Problem in Manifold Space

The Fermi analysis established that the thermodynamic filter is universal. Any civilisation that couples wealth accumulation to power throughput on a finite radiative surface encounters the same ceiling on the same timescale relative to its growth rate. The question that remains is whether the filter is passable, and if so, what kind of problem the passage constitutes.

The viability framework (§4.2) provides a precise language for the predicament. The state vector x(t) traces a path through n-dimensional phase space. The viability kernel Viab_f(K) defines the region within which civilisation survives. The control vector u(t) represents the actions available at each moment. The survival question reduces to a geometric one: does there exist an admissible trajectory that remains inside K indefinitely?

This formulation, precise as it is, conceals a deeper problem. It treats K as fixed. The viability kernel has a shape, and that shape determines which trajectories are admissible. But the geometry itself is a variable. The kernel is not K. It is K(p). The constraint set depends on parameters that are themselves subject to change, some fixed by physics, some stiff under evolutionary dynamics, some compliant to institutional reform. Civilisation's problem is compound: it must simultaneously navigate within the current kernel and modify p to reshape the kernel itself. These two tasks compete for resources. The allocation between navigation and structural reform may be the most consequential control decision civilisation faces.

This section develops the full control-surface analysis. It classifies every lever available to a civilisation confronting the waste heat ceiling, generates the complete set of strategic orientations from the combinatoric of that surface, attaches observational signatures to each, and presents the strategic matrix that maps orientations to parameters. The result is a self-contained analytical unit: given the viability framework, given the eight parameters, here are the seven strategies, here is how they engage the parameters, here is what each looks like.

The eight-parameter control surface

The ceiling formula derived in the maintenance and waste heat analyses,

with Γ = μδξ/η_II, contains eight parameters with real operability. The four components of Γ govern the denominator: the metabolic multiplier μ, the aggregate decay rate δ, the specific exergy cost ξ, and the Second Law efficiency η_II. The numerator contains three additional controls: the effective emissivity ε, the radiating area A, and the habitability threshold T_hab (which enters with fourth-power sensitivity). The absorbed solar flux P_☉ = S(1 - α)π R² introduces an eighth: the planetary albedo α. Together, these eight parameters are the complete control surface between civilisation and the waste heat constraint.

Each parameter is classified by three properties that determine its strategic value:

Operability. Can civilisation change this parameter? At what cost, timescale, and certainty?

Directionality. Does moving the parameter serve nostos (planetary stabilisation, reducing Γ, managing T_eq), kleos (space expansion, increasing A), or both?

Operability maps onto the three-tier ontology developed across Parts 2 through 4. The tier classification determines the kind of response each constraint demands. It is physically impossible to repeal the Second Law (Tier 1, fixed). It is without precedent to override competitive selection at civilisational scale (Tier 2, stiff). It is difficult but historically demonstrated to reform institutional architecture (Tier 3, compliant). Confusing the tiers, treating a Tier 3 problem as Tier 1 (fatalism about reformable institutions) or a Tier 1 problem as Tier 3 (optimism about negotiating with physics), is the most common and most dangerous error in civilisational strategy.

A critical refinement, grounded in the competitive viability equilibrium derived in the competitive viability analysis (§4.3): the stiffness of the Tier 2 attractor, its resistance to departure, is a consequence of the Nash equilibrium of the multi-agent viability game. Under competitive allocation, unused gradient is unstable; any agent that captures surplus weakly expands its kernel and weakly contracts competitors'. Noncooperative equilibria generically saturate the available gradient. This is why the attractor resists departure: defecting from restraint is individually rational. But the location of the attractor, which value of Γ the system converges to, depends on the competitive landscape, which is shaped by Tier 3 institutional architecture. Hanley's pre-1970 data demonstrate that the attractor has shifted at least once. The stiffness is Tier 2; the location is Tier 3. This distinction structures the entire control-surface classification.

ε (effective emissivity). High sensitivity: enters the numerator directly. Operability is mixed, degrades through greenhouse gas accumulation, improves through atmospheric composition management (decarbonisation, carbon capture). Directionality: both nostos and kleos. Emissivity improvement is the closest thing to a no-regrets intervention on the control surface: every survivable strategy prioritises it, because reduced ε tightens the ceiling regardless of strategic orientation. This is what makes decarbonisation urgent on every trajectory, not merely on trajectories that prioritise planetary stabilisation. Tier assignment: the physics of radiative transfer is Tier 1; atmospheric composition management operates at Tier 3 (emissions policy) against Tier 2 resistance (economic growth drives emissions).

A (radiating area). Extremely high sensitivity: enters the numerator and is unbounded in principle. Operability is currently near zero, no off-planet radiative infrastructure exists. The Fermi analysis established the construction timescales: decades to centuries for meaningful expansion, constrained by the luminosity bottleneck (in-system power for construction cannot exceed the star's output) and the maintenance trap (off-planet structures require continuous maintenance against a decay rate δ that may exceed terrestrial baselines). The Dyson-scale analysis in the Fermi discussion estimated on-surface extraction power of approximately 5.2 × 10^19 W for a complete Mercury disassembly over centuries, itself constrained by the solar luminosity available for processing. The bootstrap power demand compounds the thermal load on a system already approaching the ceiling, creating a race between A-expansion and thermal-window closure. Directionality: pure kleos. This is the only parameter that can widen the ceiling without bound, and the only one whose helpful movement requires expanding beyond the planetary surface. Its distinguishing feature is lead time: A-expansion is the slowest-responding lever on the control surface. A civilisation that needs A-expansion and has not already begun the investment programme has, in the relevant sense, already failed.

T_hab (effective habitability threshold). Very high sensitivity: enters the numerator with fourth-power dependence. The biological wet-bulb floor is Tier 1, fixed by the thermodynamics of mammalian heat rejection. But the effective civilisational threshold has moderate operability through agricultural engineering, built-environment thermal management, and infrastructure resilience. Directionality: both nostos and kleos, since widening thermal headroom serves any strategy. The complication is feedback through Σ: climate-controlled agricultural systems, sealed buildings, and thermal management infrastructure all add to the assembly stock and its maintenance burden. The net headroom gain after accounting for the Σ increase is the decision-relevant quantity. Tier assignment: biological floor is Tier 1 (fixed), effective threshold is partially Tier 2 (requiring large-scale agricultural transformation) and partially Tier 3 (building codes, urban planning, adaptive investment).

α (planetary albedo). Moderate sensitivity: enters through P_☉ = S(1 - α)π R². Operability is moderate to high for stratospheric aerosol injection (deployable within years, requiring continuous replenishment) and low for space-based reflectors (requiring the same space-industrial capability as A-expansion). Directionality: both nostos and kleos. The distinguishing feature of α is speed: albedo modification is the fastest-acting lever on the control surface, capable of buying thermal headroom on timescales of years rather than decades. This makes it the strategic reserve, the intervention deployed when time is needed and the other controls are too slow. Tier assignment: the physics of aerosol injection is Tier 1 (well understood), the decision to deploy is Tier 3 (compliant), and the governance of sustained deployment is Tier 2 (requiring multi-decade international coordination with severe moral-hazard and termination-shock risks). No viable strategy prioritises α for sustained deployment. Every viable strategy holds it as insurance.

η_II (Second Law efficiency). Moderate sensitivity: enters the denominator of Γ. Operability is moderate, the current aggregate of perhaps 10–15 per cent leaves large theoretical headroom, but improvements are subject to the Jevons recycling mechanism established in the competitive viability analysis: efficiency gains increase the construction efficiency α_i, expanding each agent's velocity set; competition recycles the gain into throughput until the constraint binds again. Without simultaneous μ-reduction, efficiency gains are recycled through increased demand, leaving Γ approximately unchanged. Directionality: primarily nostos, with indirect kleos benefit (a lower Γ widens the thermal headroom available for bootstrap construction). The Jevons mechanism is Tier 2 (competitive equilibrium); the specific channel through which it operates (price-mediated in market economies, surplus allocation in command economies) is Tier 3.

δ (aggregate decay rate). Moderate sensitivity: enters the numerator of Γ. Operability is moderate, an engineering parameter with real design headroom. A civilisation that builds Roman-aqueduct infrastructure rather than smartphone infrastructure has a lower δ. Directionality: primarily nostos. The historical trend is in the wrong direction: the shift from durable physical capital (steel, concrete, masonry) to short-lived capital (silicon, software, consumer electronics) has increased the aggregate decay rate over the industrial period. This trend is itself a consequence of competitive viability dynamics, shorter product cycles increase throughput and market share, making δ-reduction a Tier 2 problem in aggregate even though individual engineering choices are Tier 3.

ξ (specific exergy cost). Moderate sensitivity: enters the numerator of Γ. Operability is moderate, accessible through manufacturing process engineering, materials science, and repair technology. The floor is set by the Landauer limit and the free energy differences of the specific joining operations involved. Directionality: primarily nostos.

μ (metabolic multiplier). High sensitivity: a direct multiplier on the maintenance power floor, μ scales the entire denominator coupling between civilisation's power demand and its assembly stock. Operability is mixed, and this is the parameter whose compliance structure is most diagnostic of the civilisational predicament. The Tier 3 component, financial growth compulsion, discretionary consumption, military expenditure, pure waste, is compliant. These are institutional choices that have been altered many times in history. The Tier 2 component, competitive selection for agents that invest surplus in growth rather than banking it, is stiff, a consequence of the competitive viability equilibrium: agents that operate below the maximum-power attractor have smaller viability kernels and are more vulnerable to perturbation. The aggregate attractor value of μ reflects the Nash equilibrium of the competitive game, not policy. But the location of that equilibrium depends on the institutional rules of the game (Tier 3). Reducing μ therefore requires simultaneous action at both tiers: institutional reform to constrain the compliant component, and civilisation-scale coordination to override the stiff component. Directionality: primarily nostos, with partial kleos benefit (a lower μ frees thermal headroom that can be allocated to bootstrap construction). This is the core mechanism for decoupling power demand from assembly stock, and, as the strategic matrix below establishes, the parameter that most sharply discriminates viable from non-viable strategies.

The elasticity matrix

The qualitative sensitivity classifications assigned above, "high," "moderate," "very high", can be made exact. The ceiling formula admits closed-form elasticities that convert the verbal ranking into a quantitative one.

From the ceiling relation,

A 1 per cent reduction in μ, δ, or ξ raises the ceiling by exactly 1 per cent. A 1 per cent improvement in η_II does likewise. These elasticities are parameter-independent: they hold at any value of Γ and at any headroom H, a direct consequence of the multiplicative structure of the coupling.

Numerator elasticities. The parameters entering through the headroom H have leverage that depends on the ratio of total radiative capacity to headroom:

Two features are immediate. First, T_hab carries a factor of 4 relative to ε and A, the fourth-power sensitivity of the Stefan–Boltzmann law expressed as an elasticity. Second, all numerator elasticities are amplified by the inverse headroom fraction εσ A T_hab^4/H, which exceeds unity whenever the absorbed solar flux P_☉ is a large fraction of the total radiative capacity, as it is for any planet receiving significant stellar irradiance.

Numerical estimate for Earth parameters. At current conditions, P_☉ ≈ 1.2 × 10¹⁷ W and εσ A ≈ 1.74 × 10^7 W K^- 4. Taking T_hab = 298 K (a 10 K anomaly, consistent with the worked example in the viability framework (§4.2)), the radiative capacity is εσ A T_hab^4 ≈ 1.37 × 10^17 W, giving headroom H ≈ 1.7 × 10^16 W. The headroom amplification factor is therefore:

The T_hab elasticity is 4 × 8 ≈ 32: a 1 per cent increase in the effective habitability threshold raises the ceiling by approximately 32 per cent. The ε and A elasticities are each approximately 8. All three are super-unit-elastic. For the denominator parameters, a 1 per cent improvement in any component of Γ yields precisely 1 per cent, a factor of 8 to 32 less leverage per unit of proportional change.

The strategic ranking that the preceding qualitative analysis described in prose now has exact coefficients: T_hab is the highest-leverage parameter by a wide margin, followed by ε and A (tied), followed by the four denominator parameters (tied at unit elasticity).

The cross-elasticity and the net headroom correction. The raw elasticities assume each parameter can be changed independently. This assumption fails when raising T_hab requires climate-controlled infrastructure, enclosed habitats, air-conditioned cities, engineered agricultural environments, that itself adds to the assembly stock Σ and therefore to the power demand through the coupling P = ΓΣ. The effective gain from T_hab engineering is less than the raw elasticity of 32 suggests. Formally, if ΔΣ_infra is the additional assembly required to raise the effective habitability threshold by a fraction ε, the net gain is:

Whether this net gain is positive depends on the thermodynamic efficiency of the habitat infrastructure relative to the ceiling leverage. A civilisation that raises T_hab by 1 per cent through infrastructure requiring more than 32 per cent additional assembly stock has spent more than it gained. This cross-elasticity formalises the "net headroom" argument: the numerator parameters offer enormous leverage in principle, but the denominator parameters, particularly μ, determine whether that leverage can be captured in practice.

The complete control surface:

Parameter, Sensitivity, Operability, Directionality, Tier 2 stiffness / Tier 3 location

ε, High (numerator, direct), Mixed (degrades via GHG, improves via atmospheric management), Both Emission rate driven by growth equilibrium, (Tier 2); policy levers compliant (Tier 3)

A, Extremely high (numerator, unbounded), Currently near zero, Pure kleos, Longest lead time; only unbounded lever

T_hab, Very high (fourth-power), Moderate (biological floor fixed, effective threshold operable), Both Feeds back through Σ; net gain is decision-relevant

η_II, Moderate (denominator of Γ), Moderate, subject to Jevons, Primarily nostos Jevons recycling is Tier 2 equilibrium; channel is Tier 3

δ, Moderate (numerator of Γ), Moderate (engineering), Primarily nostos, Aggregate trend driven by competitive dynamics (Tier 2); individual design choices Tier 3

μ, High (direct multiplier), Mixed (Tier 3 compliant, Tier 2 stiff), Primarily nostos, Core discriminant; Nash equilibrium sets stiffness, institutional landscape sets location

The seven strategic orientations

The eight-parameter control surface, combined with the tier and directionality classifications, reveals that the strategic space is richer than a binary choice between planetary stabilisation and space expansion. The generating dimensions are:

Nostos vs. kleos. Does the strategy prioritise planetary stabilisation (reducing Γ, managing T_eq) or space expansion (increasing A)?

Temporal ordering. Does the strategy address the near-term constraint (greenhouse/ecological) or the long-term constraint (waste heat ceiling) first?

Awareness. Does the strategy acknowledge the ceiling's existence?

The full combinatoric of {nostos, kleos} × {temporal ordering}, with the null case split by whether the civilisation has identified the constraint, exhausts the strategic possibilities:

Nostos: No Nostos: Yes

Kleos: No S6: Apathy / S7: Ignorance S1: Pure Nostos

Kleos: Yes S2: Pure Kleos S3: Nostos-first / S4: Kleos-first / S5: Parallel

Seven orientations. Their definitions follow. Every subsequent reference in this essay points back here.

S1: Pure Nostos (quiet sustainment). The civilisation achieves sufficient Γ-reduction through δ, ξ, η_II, and μ to stabilise within the planetary viability kernel indefinitely. It does not expand A. It is thermodynamically quiet. Viable if Tier 2 coordination is achieved. High coordination demand, low risk of catastrophic failure. The cost is foreclosed expansion, though the option is retained for later, to be exercised from a position of stability and surplus.

S2: Pure Kleos (escape without stabilisation). The civilisation maximises energy throughput and races for the space-industrial bootstrap without addressing Γ. Almost certainly non-viable: the bootstrap power demand compounds the thermal load on a system that has not reduced Γ, closing the thermal window faster than the construction can complete. The maintenance trap applies with full force, off-planet structures require continuous maintenance power, which is drawn from the same planetary budget that is already approaching the ceiling. The Dyson analysis confirms the quantitative impossibility: even partial swarm construction over centuries demands power budgets that an unreduced-Γ civilisation cannot sustain within the thermal window. Including this strategy is analytically honest, it is what technology-optimists implicitly advocate when they argue for growing the way to space without addressing the coupling.

S3: Nostos-first Sequential (secure then expand). The civilisation first executes the Γ-reduction programme, stabilises within the planetary kernel, and then pivots to A-expansion from a position of stability. Lower peak resource demand than S5. The risk is the longer total timeline: institutional coherence over centuries, and potential loss of technological momentum during the stabilisation phase. The Dyson construction becomes more tractable from a stabilised base because the thermal headroom available for bootstrap power is larger, and the civilisation is not simultaneously fighting the coupling.

S4: Kleos-first Sequential (expand then secure). The civilisation prioritises A-expansion first, intending to execute Γ-reduction afterward against a widened ceiling. Distinct from pure kleos because Γ-reduction is planned, not ignored, it is deferred. Probably non-viable for quantitative rather than definitional reasons: the construction timescales for meaningful A-expansion (decades to centuries, per the Fermi analysis's bootstrap estimates) likely exceed the thermal headroom available on an unreduced-Γ trajectory. The civilisation races two timescales, A-expansion versus thermal-window closure, and the physics strongly favours the losing side. If the bootstrap could be completed fast enough, S4 merges into S5 with kleos-weighted allocation. The boundary between S4 and S5 is set by the thermal-headroom budget.

S5: Parallel (high coordination). Simultaneous Γ-reduction and A-expansion. Possibly viable, with the highest coordination demand of any survivable strategy. Whether the thermal headroom consumed by the bootstrap is smaller than the headroom created by the nostos programme is a quantitative question addressed in the closing analysis (§5.4). S3, S4, and S5 exist on a continuum of temporal weighting; S3 is nostos-weighted, S5 is balanced, S4 is kleos-weighted. The viability boundary between them is set by the thermal-headroom budget.

S6: Apathy (aware but inactive). The civilisation has identified the constraint, or at least its general shape, and fails to act. The failure is institutional: collective action problems, short-termism, the governance-timescale mismatch diagnosed in the amplifiers analysis. The civilisation may attempt partial responses that are too slow, too uncoordinated, or too narrow, solving the greenhouse constraint but not the coupling. Non-viable. This is the specifically human-relevant failure mode under current institutional architecture.

S7: Ignorance (unaware). The civilisation has not developed the analytical synthesis that identifies the waste heat ceiling as the binding constraint. It may solve the greenhouse problem (Tier 3) and relax, not realising the deeper Tier 1 constraint exists behind it. It rides the evolutionary ratchet at full power because nothing in its understanding tells it not to. Non-viable. Probably the modal outcome galaxy-wide: the specific synthesis connecting maintenance thermodynamics to waste heat ceilings to viability geometry is unlikely to arise spontaneously before the decision window closes.

Observational signatures

The strategic taxonomy generates a richer observational prediction than the simple loud-or-quiet binary that dominates standard Fermi analyses. Every civilisation that reaches the exponential-growth phase faces the same viability geometry. Its observable fate depends on which orientation it adopts, or fails to adopt.

S1 = Pure nostos: Thermodynamically quiet. Waste heat signature indistinguishable from the natural thermal emission of its planet. Invisible by design. A galaxy of successful nostos civilisations produces precisely the silence that is observed.

S2 = Pure kleos: A brief bright infrared flash as power consumption spikes during the attempted bootstrap, then silence as the thermal window closes and the civilisation crosses below Σ_min.

S3 = Nostos-first: Silence for centuries during the stabilisation phase, then a controlled brightening as off-planet construction begins from a stabilised base. The brightening is intentional, directional, and gradual, an anomaly in the infrared, easily confused with natural astrophysical processes during the early construction phase.

S4 = Kleos-first: A brightening flash as A-expansion begins; most likely flash then silence as the thermal window closes before the bootstrap completes. Distinguishable from S2 only by the brief appearance of partial off-planet structure, a spectral signature that vanishes when maintenance ceases.

S5 = Parallel: Faint, intentional infrared signatures during the simultaneous construction and stabilisation phase. A modest anomaly, easily confused with natural phenomena.

S6 = Apathy: Erratic, prolonged brightening as partial interventions temporarily slow the ratchet, followed by the same silence. Longer and more irregular than S2 or S7, but the same terminal outcome.

S7 = Ignorance: Indistinguishable from pure kleos, a brightening flash, then silence. The ratchet operates unopposed when the civilisation does not know the parameters exist. The distinction from S2 is internal, S2 knows and races; S7 does not know and drifts, but the thermodynamic trajectory and the observable signature are the same.

The deepest Fermi result: two strategies dominate the galactic census. Quiet sustainment (S1, S3-post-stabilisation), by design invisible. And ignorance (S7), by default indistinguishable from a dead civilisation. Whether current surveys, WISE, the G-HAT wide hatG statistic, could detect civilisations pursuing the nostos-first or parallel strategies is doubtful, given the faintness and ambiguity of their predicted signatures. But these are testable predictions that the simple loud-or-quiet binary does not generate.

Each strategic orientation implies a different prioritisation of the eight control parameters. The following matrix maps the abstract thermodynamic framework onto concrete strategic profiles. Each cell indicates whether the strategy engages that parameter, in which direction, and with what expected effectiveness:

The matrix is the analytical centrepiece of the strategic analysis, the most information-dense representation of the viability framework applied to the civilisational predicament. It reveals several structural features that are not visible from the tier classification alone.

First, ε is no-regrets across all viable strategies. Every survivable orientation prioritises emissivity improvement. Decarbonisation is urgent regardless of strategic orientation, not merely as climate policy but as a thermodynamic prerequisite for maintaining the headroom within which any further strategy can operate.

Second, μ-reduction is the discriminant. S2, S4, S6, and S7 all neglect or inadequately pursue μ-reduction. This is the parameter that most sharply separates viable from non-viable strategies. The competitive viability equilibrium explains why: under the Nash equilibrium of the multi-agent game, μ sits at an attractor whose stiffness is Tier 2. Strategies that do not engage μ at both tiers, institutional reform of the compliant component, civilisation-scale coordination to override the stiff component, fail because efficiency gains (η_II) are recycled through Jevons in the absence of μ-controls, and durability improvements (δ) cannot compensate alone. Only S1, S3, and S5 pursue μ-reduction with the required magnitude.

Third, α is the universal reserve. No strategy prioritises it for sustained deployment. Every viable strategy holds it as insurance. S6 and S7 do not even have it in the toolkit, which means they lack the emergency lever when the other controls prove too slow.

Fourth, S4 and S2 have nearly identical parameter profiles during the critical early phase. Both prioritise A-expansion and neglect Γ-reduction. S4 intends to address Γ later, but the system's behaviour during the unreduced-Γ phase does not know about the civilisation's intentions for Phase 2. This is why S4 probably fails for the same quantitative reasons as S2: the thermal window closes on a trajectory that is, in its critical early decades, indistinguishable from the pure-kleos race.

Fifth, S6 (apathy) is a degraded version of S3. It attempts nostos-type interventions (η_II, ε) but without the institutional depth (μ-reduction, δ-reduction) needed to make them effective. Efficiency gains are recycled through the Jevons mechanism because μ-controls are absent. This is the "solving climate change without solving the coupling" failure mode, and it is the failure mode most closely aligned with current civilisational trajectory.

Sixth, S7 (ignorance) is externally indistinguishable from S2. The ratchet operates unopposed when the civilisation does not know the parameters exist. The distinction is internal, S2 knows and races; S7 does not know and drifts, but the thermodynamic trajectory is the same. The deepest Fermi result may be that most civilisations fail through S7: they never develop the specific analytical synthesis before the decision window closes.

Seventh, S5 (parallel) requires the broadest simultaneous effort: high priority on more parameters concurrently than any other strategy. This is why its coordination demand is the highest of any survivable orientation.

From the manifold to the choice

The control-surface analysis reveals that the question "nostos or kleos?" is ill-posed. The real question is: which of the seven strategic orientations produce admissible trajectories within the viability kernel, and, within the viable orientations, how should effort be allocated across the eight parameters given their sensitivity, operability, and directionality profiles?

The 7 × 8 matrix is the map. It establishes that three orientations (S1, S3, S5) are viable, two (S2, S4) probably fail, and two (S6, S7) fail on every trajectory the framework can resolve. Within the viable region, the orientations form a continuum of temporal weighting between nostos-first and parallel execution, with the boundary set by the thermal-headroom budget. The allocation between navigation within K and structural modification of p is not a binary but a continuous optimisation problem across eight dimensions, constrained by three tiers of compliance, shaped by the competitive viability equilibrium, and bounded by the irreducible geometry of the waste heat ceiling.

This is the geometry of civilisation's predicament, stated without reference to any specific energy source, policy programme, or political ideology. The validity cone is fixed by physics. The attractor surface is shaped by the Nash equilibrium of competitive dissipative structures. The contingent kernel is defined by institutional architecture. The viability kernel is the intersection, and it is contracting. The eight parameters are the complete set of levers. The seven orientations are the complete set of strategies. The closing analysis sorts each orientation by viability, identifies the discriminant role of μ at the Tier 2/Tier 3 boundary, and states the corrected theorem.

5.3 The Choice

The Choice of Achilles

The essay has arrived, through five parts and sixteen sections, at a single distinction, between the merely difficult and the physically impossible, and a single question that the distinction forces: what, precisely, can a civilisation do with that distinction once it sees it?

The question has a precedent in literature, if not in history. In Book IX of the Iliad, Achilles tells the embassy from Agamemnon that his mother Thetis has revealed to him a choice enforced by fate:

If I stay here and fight before the city of Troy, my nostos is lost but my kleos will be imperishable. If I return home to my dear fatherland, my kleos is lost but my life will be long, and death will not come to me swiftly.

Kleos aphthiton, imperishable glory. Nostos, the long return home. The two are mutually exclusive. Fate does not negotiate. Achilles must choose.

The thermodynamic predicament maps onto this archetype with a precision that is not merely literary. Kleos is the loud path, maximum power, the space-industrial bootstrap, A-expansion beyond the planet, the infrared blaze. Nostos is the quiet path, Γ-reduction, decoupling, the long homecoming within the viability kernel, silence. The mapping is powerful because it captures a real structural feature of the predicament: the tension between expansion and stabilisation, between the growth trajectory and the constraint surface.

But the mapping is also misleading, and the point at which it misleads is the point at which the essay's analytical contribution begins.

From binary to continuum

Achilles saw two paths because he was in a myth. Fate could enforce a binary because fate was the dramatist. The viability mathematics does not produce a binary. It produces a kernel, a continuous region in state space, and within that kernel there exists a family of admissible trajectories, not two. The control surface developed across the preceding sections, eight parameters, three compliance classes, seven strategic orientations, reveals a richer geometry than any myth can encode. The essay's contribution is not identifying the choice. It is mapping the viable region of the allocation space with enough resolution to guide actual strategy.

Viability sorting

The seven strategic orientations generated from the control-surface combinatoric in the manifold analysis exhaust the space of possible responses. This section sorts them by viability, the question of whether each orientation produces trajectories that remain within the viability kernel K, applying four criteria: kernel membership (does the trajectory stay within K for all t > 0?), robustness to perturbation (how far inside K does the trajectory run?), dependence on Tier 2 override (does the strategy require reshaping the viability geometry so that the gradient-saturating equilibrium converges to a lower value of γ, either through cooperative commitment among all N agents or through institutional redesign of the competitive landscape?), and dependence on Tier 3 reform (does the strategy require institutional changes that are difficult but precedented?).

The sorting proceeds from most viable to least.

S1 = Pure nostos (quiet sustainment, as defined in the manifold analysis) stabilises within K indefinitely through systematic Γ-reduction. Viable, but requires Tier 2 override of the competitive viability equilibrium, the civilisation must collectively depart from the gradient-saturating Nash equilibrium derived in the competitive viability analysis (§4.3). Robustness is high: the trajectory runs deep inside K, far from the boundary, with large perturbation margins. The coordination demand is the highest of any strategy; the fragility once achieved is the lowest.

S3 = Nostos-first sequential (secure then expand) executes Γ-reduction first, stabilises within the planetary kernel, then pivots to A-expansion from a position of stability. Viable, with a lower peak coordination demand than S1 because A-expansion is deferred rather than permanently foreclosed. Requires Tier 2 override during the stabilisation phase and Tier 3 reform throughout. Robustness is moderate: the trajectory runs inside K during the stabilisation phase but approaches the boundary during the pivot to expansion, where timing errors can push it outside.

S5 = Parallel (simultaneous Γ-reduction and A-expansion) is possibly viable, with the broadest simultaneous effort across the control surface. The viability condition is quantitative: the thermal headroom consumed by the kleos investment must be smaller than the headroom created by the nostos programme running in parallel. This is the question for which the 7 × 8 strategic matrix developed in the manifold analysis matters most. The trajectory hugs the kernel boundary more closely than S1 or S3, making it the most fragile of the survivable strategies, viable in the mathematics, precarious in the perturbation analysis.

S2 = Pure kleos (escape without stabilisation) maximises energy throughput and races for the space-industrial bootstrap without addressing Γ. Almost certainly non-viable: the bootstrap power demand compounds the thermal load on a system that has not reduced Γ, closing the thermal window faster than the construction can complete. The maintenance trap applies with full force.

S4 = Kleos-first sequential (expand then secure) prioritises A-expansion first, intending to execute Γ-reduction afterward against a widened ceiling. The strategic logic is real: even partial A-expansion changes the ceiling formula. The failure is quantitative: construction timescales for meaningful A-expansion likely exceed the thermal headroom available on an unreduced-Γ trajectory. During the critical early phase, the parameter profile is nearly identical to S2. Probably non-viable.

S6 = Apathy (aware but inactive) has identified the constraint and fails to act with sufficient depth. Non-viable. This is the specifically human-relevant failure mode under current institutional architecture: solving the greenhouse constraint but not the coupling.

S7 = Ignorance (unaware) has not developed the analytical synthesis that identifies the waste heat ceiling as the binding constraint. It rides the evolutionary ratchet at full power because nothing in its understanding tells it not to. Non-viable. Probably the modal outcome galaxy-wide.

The default trajectory is non-viable. This is the essay's most important finding and it is unchanged by the expansion from binary to continuum. Under current institutional architecture, civilisation is on a trajectory that terminates in S6 or S7. The competitive viability equilibrium operates. The amplifiers accelerate. The kernel contracts. The default does not end inside the viable region.

Every viable trajectory requires Γ-reduction of a magnitude that has no historical precedent. Whether the strategy is S1, S3, or S5, substantial work on the denominator parameters; μ, δ, ξ, η_II and on the numerator parameter ε is a necessary component. There is no viable trajectory that ignores Γ-reduction.

The viable trajectories to sustained space expansion pass through planetary stabilisation first, not as a preference but as a geometric consequence of the viability kernel's structure. The competitive viability equilibrium (§4.3) means that any trajectory attempting kleos before nostos must either (a) have already reduced Γ sufficiently that the ceiling provides adequate headroom, or (b) override the competitive equilibrium through a cooperative solution that has no precedent at civilisational scale. Nostos-then-kleos is the generic viable trajectory. Kleos-first is viable only in the measure-zero case where the starting state is already deep inside the kernel with low Γ. This is not the current state.

The corrected theorem replaces the strict sequencing claim of earlier formulations with a stronger and more defensible finding: the viable region of the allocation space is small and the default trajectory misses it. The region is not a single path but a family of trajectories sharing a common necessary condition, substantial Γ-reduction anchored in μ, and differing in their temporal allocation between nostos and kleos. The viable strategies occupy a narrow corridor; the non-viable strategies surround it on every side.

The scenario envelope

The finding is robust across a wide range of growth trajectories. If demographic deceleration and compositional shifts in Σ slow the aggregate growth rate below the historical 2.3%, the thermal headroom available to any viable strategy extends proportionally, at 1% growth, timescales approximately double; at zero net growth, the waste heat constraint recedes to the far horizon. If AI-driven automation shifts η_II and μ in offsetting directions, the net effect on Γ is scenario-dependent, but the structural finding holds regardless: viable strategies require Γ-reduction, non-viable strategies neglect it, and μ is the discriminant. The scenario determines the pace and the width of the viable corridor; the framework determines its geometry.

The existence proof

The nostos strategy, high assembly stock maintained at low Γ within a planetary viability kernel, is not a theoretical possibility awaiting its first demonstration. It has been demonstrated.

The biosphere has maintained an assembly stock Σ_bio ≫ Σ_tech for approximately 3.5 billion years. It operates on approximately 130 TW of photosynthetic capture at a composite coupling Γ_bio that is orders of magnitude below the technosphere's Γ_tech. It has survived five mass extinctions, each of which destroyed a substantial fraction of standing Σ_bio, and each time recovered, diversified, and in most cases exceeded the pre-extinction assembly depth. No other dissipative structure in the observable record has demonstrated comparable persistence within a planetary viability kernel. The biosphere is, by any thermodynamic measure, the most successful implementation of the nostos mode that the planet has produced. It accumulates assembly without exceeding the thermal budget. It maintains complexity through geological time. It radiates within the planetary equilibrium. To any external observer, it is thermodynamically quiet.

This is not anthropomorphism. The biosphere does not choose nostos. It does not coordinate through institutions. Its persistence is a product of the same evolutionary dynamics, Tier 2, that the essay has analysed throughout. What the biosphere demonstrates is that those dynamics, operating on a competitive landscape shaped by hard biochemical constraints, no credit creation, and gigayear timescales, converge on an equilibrium where Γ is low enough to be compatible with indefinite planetary habitability. The biosphere did not restrain itself. It is fully gradient-saturating, approximately 130 TW of photosynthetic capture represents maximum power within its domain. Its low Γ is the equilibrium value that competition produced, given the landscape it operated on. This is the landscape-dependence result from the competitive viability analysis (§4.3) made concrete: the attractor value of Γ is not fixed by physics. It is set by the competitive landscape. Change the landscape, change the attractor.

The existence proof establishes physical feasibility. It does not establish institutional feasibility. But the implication is sharper than the question "can cognitive agents achieve deliberately what selection achieved through attrition?" suggests. Cognitive agents do not need to replicate 3.5 billion years of competitive attrition through deliberate restraint. They need to redesign the competitive landscape so that selection achieves it again, on a compressed timescale. The coordination problem is not "suppress the most powerful evolutionary dynamic on the planet." It is "change the rules of the game so that the same dynamic converges somewhere less lethal." This is a different and more tractable problem. Hanley's data provide direct evidence that the attractor has shifted at least once when the institutional landscape changed, the pre-1970 instability in the Garrett ratio marks precisely such a transition, driven by the engagement of the modern financial architecture. The shift went in the wrong direction. But the mechanism is the proof of concept: the competitive equilibrium responds to the landscape it operates on. The biosphere is the evidence that the destination exists. Hanley is the evidence that the rules can move the attractor. The open question is whether the rules can be moved deliberately, far enough, fast enough. Physical feasibility is settled by the biosphere. Institutional feasibility, the central open problem of this essay, reduces to the question of landscape engineering within the timescale the thermal budget permits.

The existence proof also produces a geometric result that can be stated precisely. The viability kernel K computed in the viability analysis (§4.2) treats the technosphere in isolation, a single assembly stock Σ_tech subject to thermodynamic constraints. But the technosphere does not exist in isolation. It is coupled to the biosphere through functional dependencies that represent necessary conditions for the technosphere's own persistence: atmospheric composition regulation (oxygen production, carbon sequestration), hydrological cycling (cloud formation, precipitation distribution, aquifer recharge), soil genesis (biological weathering, nutrient fixation, organic matter accumulation), pollination (approximately 75% of global food crop species), and coastal protection (mangroves, coral reefs, wetland systems). These are not amenities. They are boundary conditions on the technosphere's own viability, physical inputs without which the technosphere's maintenance equation d Σ_tech/dt = C(t) - δΣ_tech cannot be sustained at current values of C(t) and δ.

The burden-transfer mechanism derived in the trajectory analysis makes the coupling qualitatively clear. When Σ_bio degrades below a functional threshold, the technosphere must replace the lost biospheric function with engineered infrastructure, each unit of which draws maintenance power at Γ_tech ≫ Γ_bio. The remainder of this section formalises the mechanism as a two-stock theorem and derives its consequences for the viability kernel.

The coupled-system model.

Two assembly stocks coexist within the planetary boundary:

Each maintained by distinct power flows with distinct composite couplings:

The central empirical inequality, established in the assembly stock analysis: Γ_bio ≪ Γ_tech. The biosphere maintains vastly more assembly per watt than the technosphere.

The biosphere provides a service function: S_bio(Σ_bio), monotone increasing in biospheric assembly stock, more biosphere, more services. When biospheric services degrade (dS_bio < 0), civilisation must replace them with technospheric assembly to maintain the same functional capacity:

where β ≥ 1 is the replacement ratio: the amount of technospheric assembly needed to replace one unit of biospheric service. The inequality β ≥ 1 follows from the observation that engineered replacements are functional substitutes, not assembly-equivalent reconstructions. A desalination plant replaces a wetland's water purification service but at far higher Γ and with narrower perturbation tolerance. The numerical value of β for any specific biospheric service is empirically unconstrained. The derivation requires only the direction (β ≥ 1) and the qualitative consequence.

Proposition P5 (Power demand rises under burden transfer). When biospheric stock decreases by ΔΣ_bio and technospheric stock increases by β ΔΣ_bio to replace lost services, the change in total power demand is:

Proof. Direct substitution into (2). Since Γ_tech ≫ Γ_bio and β ≥ 1, the parenthetical term β Γ_tech − Γ_bio is strictly positive for any plausible parameters. Therefore ΔP_tot > 0: total power demand rises even if total planetary assembly falls or stagnates. ▫

This is the burden-transfer mechanism stated as a derived result rather than a prose argument. The planet becomes simultaneously less complex (total Σ decreases because the technosphere cannot replace biospheric assembly one-for-one at equivalent depth) and hotter (total P increases because the replacement assembly operates at the technosphere's far higher Γ_tech).

Proposition P6 (Combined kernel is a proper subset). Let K_tech denote the viability kernel for the technosphere in isolation, defined on (Σ_tech, T_eq). Let K_combined denote the viability kernel for the coupled system, defined on the expanded state space (Σ_tech, Σ_bio, T_eq), subject to: (i) all constraints from the technosphere-only kernel (waste heat ceiling, habitability, minimum viable assembly); (ii) the additional power load Γ_bio Σ_bio on the thermal budget; (iii) the burden-transfer coupling, loss of Σ_bio forces increase in Σ_tech. Then for any trajectory that degrades Σ_bio while maintaining services:

Proof. Consider any state in K_combined. From that state, an admissible trajectory must satisfy all technosphere-only constraints and absorb the additional thermal load from burden-transferred assembly. By P5, the burden-transferred trajectory consumes strictly more of the thermal budget than the corresponding trajectory without biosphere coupling. The headroom H = εσA T_hab⁴ − P_☉ − P_tot is therefore strictly smaller in the coupled system than in the uncoupled system at any given Σ_tech. States from which the headroom is insufficient, where the burden-transferred power demand pushes P_tot beyond the ceiling before braking can arrest the trajectory, are excluded from K_combined but not from K_tech. The inclusion is therefore strict. ▫

The waste heat ceiling is therefore the outer constraint surface. The actual viability kernel, the one that accounts for biosphere coupling, sits inside it. The distance between the current state and the true kernel boundary is shorter than the distance to the waste heat wall. The essay's purely thermodynamic projections are optimistic bounds on the actual viability dynamics.

Proposition P7a (Replacement requires positive exergy). The replacement ratio β satisfies β ≥ 1: functionally substituting any biospheric service with technospheric assembly requires at least as much maintenance power as the biospheric original.

Proof. From A2 (strict positivity of ξ), every joining operation in the substitute assembly pathway costs strictly positive exergy. From the maintenance coupling, the technospheric substitute operates at Γ_tech = μ_tech δ_tech ξ_tech / η_II,tech, which is strictly positive. The biospheric component it replaces operated at Γ_bio, also strictly positive but sustained by the biosphere's evolved low-Γ pathway. Since the substitute must deliver equivalent functional service and cannot access the biospheric pathway, its maintenance cost is at least as large. Therefore β ≥ 1. ▫

Proposition P7b (Replacement ratio increases with assembly depth). The replacement ratio β satisfies ∂β/∂a_bio > 0, where a_bio is the assembly index of the lost biospheric component.

Physical premise. This proposition requires, in addition to the axiom set, the modelling premise Γ_tech ≫ Γ_bio (§2.1) and the directed physical reasoning in Step 2 below regarding the combinatorial non-replicability of deep assembly pathways. Formalisation of Step 2 is catalogued in Open Problem 6.

Proof.

Step 1 (Assembly depth as path-dependent complexity). This is definitional from D2: the assembly index counts the minimum sequential joining operations required to produce a structure. For biospheric components, those operations occurred through selection over evolutionary timescales; the geometric persistence principle (T2) operating over geological time. The specific pathway reflects contingent evolutionary history.

Step 2 (Deep assembly pathways are increasingly non-replicable, physical step). Each joining operation in the biospheric assembly pathway was selected from a vast combinatorial space. Replicating the pathway requires either (a) re-running selection, which demands evolutionary timescales (10⁶–10⁹ years; ruled out by the temporal hierarchy of §4.1), or (b) reverse-engineering the specific path-dependent sequence and executing it industrially. Option (b) becomes combinatorially harder with assembly depth: the number of possible pathways grows combinatorially with the assembly index, and each step depends on the specific outcome of prior steps. For shallow-assembly components (a_bio small), the pathway is short and potentially replicable. For deep-assembly components (a_bio large), the pathway is effectively irreversible on any civilisational timescale; a consequence of the Second Law applied to the combinatorial space of assembly histories.

Step 3 (Non-replicability forces brute-force substitution at technospheric coupling rates). When the biospheric pathway cannot be replicated, the technosphere must build a functional substitute using its own assembly methods. From A2 (strict positivity of ξ), every joining operation costs strictly positive exergy. The substitute operates at Γ_tech ≫ Γ_bio and cannot access the biosphere's evolved low-Γ pathway. The deeper the original assembly, the more evolved efficiency is lost, and the more technospheric assembly is required for equivalent functional service. Therefore ∂β/∂a_bio > 0. ▫

Corollary C5 (Accelerating kernel contraction). As the biosphere degrades, the components lost earliest tend to be the shallow-assembly, low-service, easy-to-replace ones, the ecological margin. The remaining components tend to be the deep-assembly, high-service, hard-to-replace core. Each successive unit of biosphere loss costs more to replace than the last. The combined kernel contracts at increasing marginal cost.

The ordering of loss is not an unstated ecological observation. It is derived from the geometric persistence principle (T2) applied to the biosphere as a population of dissipative structures. This application requires that biospheric components satisfy axioms A5–A9. They do. Solar radiation and biogeochemical energy flows provide a finite shared gradient (A5). Each biospheric component; species, functional group, ecosystem assembly, maintains its assembled structure against decay through continuous energy throughput, with dynamics governed by construction and degradation (A6, inheriting A1–A3). Each has a minimum viable population or functional threshold below which it collapses (A7). Competitive shares of the energy gradient respond monotonically to relative biomass and functional dominance (A8). Environmental perturbations; climate fluctuations, disease, geological events, recur stochastically and are drawn from distributions satisfying the absolute-continuity and recurrence conditions (A9). The axioms are satisfied; the geometric persistence principle therefore applies to this population without modification.

Proof. Under T2 (geometric persistence, now applied to biospheric components as verified above), components with larger viability kernels persist with higher probability. Shallow-assembly components, shaped by fewer selection epochs and occupying less refined niches, have smaller viability kernels relative to the perturbation distribution they face. Under A9 (recurring perturbations), they are breached first. Deep-assembly components, selected over vastly more perturbation epochs, occupying more refined niches, have larger kernels and survive longer, by survival monotonicity (L2). The ordering of loss therefore proceeds from shallow to deep. By the replacement-ratio proposition (P7), each successive loss is costlier to replace: ∂β/∂a_bio > 0. The kernel contraction per unit of biosphere loss is therefore increasing. The combined kernel contracts at an accelerating rate. ▫

This has a direct consequence for sequencing: early biosphere loss is less costly to compensate than late biosphere loss. A civilisation that degrades the biosphere early in the bootstrap phase inherits an accelerating burden-transfer that compounds with each further decrement. The viable corridor narrows not linearly but convexly.

Proposition P8 (Biosphere preservation enlarges the viable corridor). Trajectories that maintain Σ_bio access a larger viable region in (Σ_tech, T_eq) space than trajectories that degrade it.

Proof. Immediate from P5, P6 and P7b. Maintaining Σ_bio avoids the burden-transfer power penalty (P5), preserves the full combined kernel rather than the contracted subset (P6), and avoids the accelerating contraction as deep-assembly components are lost (P7). The viable region is therefore weakly larger at every point in time and strictly larger once any biosphere degradation has occurred. ▫

P8 is not a moral claim. It is a geometric property of the coupled-system kernel that follows from Γ_bio ≪ Γ_tech and the structure of the burden-transfer coupling. The biosphere is the planet's most efficient maintenance engine. Degrading it transfers maintenance obligations to the least efficient engine, consuming thermal budget that could otherwise support technospheric assembly or provide headroom for braking.

Two caveats constrain the derivation. First, the biosphere is a Tier 2 object, its dynamics are evolutionary, not subject to direct civilisational control. Civilisation can degrade it (by consuming its substrate or altering its boundary conditions) and can reduce the rate of degradation (by restraining consumption). It cannot engineer the biosphere the way it engineers the technosphere. The variable dΣ_bio in the derivation is an externally imposed perturbation to the coupled system, not a decision variable. Second, the derivation establishes the direction of each effect and the qualitative consequence (accelerating kernel contraction), not the numerical value of β for any specific service. The thermodynamic signature is sharp; the magnitudes await empirical calibration.

The biosphere coupling strengthens the sequencing theorem. The original argument was purely geometric: attempting kleos from an unstabilised trajectory accelerates kernel contraction because the bootstrap's power demand compounds the thermal load. The biosphere coupling adds a second, compounding mechanism. A kleos strategy that degrades the biosphere, through the land-use conversion, resource extraction, and atmospheric disruption that maximum-power expansion demands, contracts the combined kernel from the inside (through burden-transfer) while simultaneously approaching the waste heat ceiling from below (through thermal loading). The two contractions compound: the kernel shrinks from both directions. A nostos-first strategy that preserves biospheric assembly holds the combined kernel open, maintaining the larger set of viable trajectories, and creates the conditions under which kleos can later be attempted from a position where the kernel is wide rather than vanishing.

Biosphere preservation is not an ecological preference grafted onto a thermodynamic argument. It is a geometric property of the coupled-system kernel: trajectories that maintain Σ_bio access a larger viable region than those that degrade it. P5–P8 make this precise.

The merely difficult and the physically impossible

The honest answer is that the question is open. The viability kernel is contracting, but it is not empty. The combined-system kernel is strictly inside the thermodynamic kernel, and it contracts faster than the waste heat analysis alone would predict, but trajectories that remain inside it exist in the mathematics, not one trajectory but a family, parameterised by the allocation between nostos and kleos, bounded by the thermal-headroom budget, anchored by the μ-reduction that every viable strategy shares. Whether those trajectories exist in the politics is an open question, and it is an open question whose answer depends on choices not yet made, by institutions not yet built, deploying coordination mechanisms not yet invented. The merely difficult includes tasks that may exceed the capacity of any civilisation that has ever existed. But "may exceed" is not "must exceed," and the distinction between the two is the distinction between the physically impossible and the merely unprecedented.

It is physically impossible to repeal the Second Law, to radiate more heat from a fixed surface at a fixed temperature, to reduce the decay rate of assembled matter to zero, to compute without dissipating energy. These define the arena. They do not yield to ingenuity, capital, ideology, or will.

It is merely unprecedented to restructure financial architecture, to extend governance timescales, to coordinate competing agents against their short-term incentives, to substitute energy sources within a generation, to reduce the metabolic multiplier at civilisational scale, to build the institutional capacity required to override the competitive viability equilibrium. These define the challenge. They yield, if they yield at all, only to coordinated action of a kind that has no analogue in the historical record.

The distinction between the physically impossible and the merely unprecedented now has more analytical content than the myth could provide. It maps onto the tier structure: Tier 1 constraints are physically impossible to move, Tier 2 constraints are unprecedented to override, Tier 3 constraints are difficult but reformable. It maps onto the parameter-by-parameter operability assessment: each of the eight control parameters has a known compliance, a known sensitivity, and a known directionality. It maps onto the 7 × 8 matrix: the viable strategies are those that engage the operable parameters with sufficient magnitude, in configurations that the thermal-headroom budget can sustain.

The physics has done its work. It has defined the arena, identified the ceiling, characterised the trajectory, derived the competitive equilibrium, calculated the timescales, classified every constraint by what it would take to move it, decomposed the control surface into eight parameters, assessed the operability of each, and mapped the strategic space into seven orientations sorted by viability. What remains is not a problem of knowledge. It is a problem of coordination, under the evolutionary dynamics that the competitive viability derivation (§4.3) has shown to be the most powerful force shaping civilisational trajectories. The question is whether a species that evolved under maximum power selection can build institutions that redirect that selection toward a less lethal equilibrium, not by suppressing competition, which the geometric persistence principle shows to be futile, but by redesigning the competitive landscape so that the samedynamics converge on a viable corridor.

The honest answer is that the question is open.

5.4 Our Corridor Out

The preceding sections derived the ceiling, characterised the trajectory, mapped the competitive equilibrium, sorted the strategic space by viability, and established the biosphere coupling that constrains the combined kernel. The results are severe. The default trajectory is non-viable. The competitive dynamics resist correction under noncooperative play. The kernel contracts. These findings are not softened by repetition, and this section does not soften them.

But the formal apparatus that identifies the danger also delimits the arena within which the danger operates. That arena is large. The corridor between the present position and the constraint surface spans orders of magnitude. The objective that civilisation should pursue within it is not obvious from the physics alone, the physics defines the terrain, not the destination, but once the objective is stated, the apparatus developed across Parts 2 through 5 resolves it with some precision.

The objective

The analysis has consistently measured civilisational scale by the assembly stock Σ. That variable was sufficient to derive the maintenance floor, the waste heat ceiling, the viability kernel, and the competitive equilibrium. It is not sufficient to define value.

A civilisation that maximises instantaneous Σ, sprinting toward the ceiling at maximum throughput, may achieve the largest assembly stock ever recorded, briefly, before collapsing outside the viability kernel. That trajectory accumulates less total civilisational achievement than a trajectory sustaining moderate Σ over millennia. Peak stock is the wrong objective. Terminal stock is the wrong objective. Any objective that rewards overshoot is the wrong objective.

The correct object of valuation is the contribution of the civilisational trajectory to the continuation and depth of viable futures. Among the class of formally tractable specifications, the most natural is the expected integral of maintained assembly over time, subject to viability:

where x is the civilisational state vector, u is the control trajectory (the allocation decisions across the eight parameters of the control surface), τ_K is the first exit time from the coupled viability kernel K, and m(x,u) is the maintained rate of viable service and option-preserving capacity. The expectation is over stochastic perturbations satisfying the recurrence conditions of A9.

This is a lexicographic objective: remain inside the kernel first, survival is the hard constraint, and among viable trajectories, maximise maintained stock-years. Under this objective, collapse trajectories are strictly dominated. Temporary contraction is admissible if it raises the reachable long-run frontier. Growth is desirable wherever it increases the integral, that is, wherever it adds to the reachable future of maintained assembly without pushing the trajectory toward the kernel boundary. The objective is not "less." It is "more, durably."

The value of any stock component x_i is then its shadow contribution:

A stock is valuable not because it is complex, and not because it is large, but because it increases expected future viable stock-years. Complexity matters only instrumentally: insofar as it preserves low-coupling support functions, deepens optionality, widens the admissible control set, or enlarges the boundary conditions within which the coupled system can persist. State may be summed for accounting. Value is not additive.

One normative step is required: the choice of the service rate m(x,u). Physics does not uniquely determine the reward function. But the qualitative conclusions, collapse is dominated, the default trajectory misses the viable set, growth is judged by its contribution to the integral, hold across every reasonable specification of m that penalises extinction and rewards persistence.

The quantitative position

Global primary energy consumption stands at approximately 20 TW. The waste heat ceiling, computed from the sensitivity analysis at the reference habitability threshold of T_hab = 292 K and current Γ, is approximately 6,800 TW. The ratio is approximately 0.3 per cent. At the more permissive threshold of 296 K, the ceiling rises to approximately 13,800 TW. With Γ halved, an aggressive but physically realisable target across the four denominator parameters, Σ_max doubles at every temperature.

The current civilisation occupies less than one per cent of the thermodynamic corridor that physics permits on this planet, at this distance from this star, with this atmosphere. More than 99 per cent of the assembly stock that Earth's radiative budget can sustain, at current Γ, without any expansion of the radiating area, lies in the future. The fraction rises further with every reduction in Γ and with every increment of A beyond the planetary surface.

This is arithmetic, not optimism. The same arithmetic that produces the ceiling produces the corridor, and the corridor is vast. The danger is not that the walls are close. The danger, the only danger the physics identifies, is that the trajectory is aimed at the walls rather than along the corridor. A civilisation that maintains 2.3 per cent annual growth in power throughput at constant Γ will reach the 6,800 TW ceiling in approximately 200 years and sterilise its planet within 400. A civilisation that reduces Γ by a factor of two and moderates its growth rate to 1 per cent has millennia of expansion space available within the planetary boundary alone, and more beyond it. The physics does not distinguish between these futures. The trajectory does.

The source–reservoir–stock–sink model

The scalar coupling P = Γ(t) · Σ was sufficient for the planetary analysis. It is a reduction of a more general viability condition that becomes relevant once intermediate reservoirs, boundary expansion, and interstellar extension are admitted.

A dissipative civilisation persists only if two conditions hold simultaneously. First, its required exergy inflow must not exceed the exergy available from boundary fluxes and accessible reservoirs:

Second, its required entropy export must not exceed the sink's rejection capacity:

The planetary waste heat ceiling is recovered when source flux is treated as fixed solar input P_☉, the sink as top-of-atmosphere radiation εσ A T^4, boundary stocks as fixed, and intermediate reservoirs as suppressed into the effective coupling Γ. The familiar ceiling formula Σ_max = ( εσ A T_hab⁴ - P_☉ )/Γ is the surface-bound, steady-state, reservoir-suppressed special case of conditions (3)–(4).

The general case is wider. For the source side, the available exergy rate from intercepted stellar radiation is:

where ψ( T_star,T₀ ) = 1 - 4/3(T₀) / (T_star) + 1/3( (T₀) / (T_star) ) ⁴ is the Petela radiative exergy factor, S_star(t) is the stellar flux at orbital radius, A_cap(t) is the collection cross-section, and T_0 is the receiver temperature. For solar radiation (T_star ≈ 5,778 K, T_0 ≈ 288 K), ψ ≈ 0.93 — sunlight is thermodynamically excellent, carrying approximately 93 per cent of its energy as available work.

For the sink side, the heat rejection capacity of the radiating surface is:

where A_rad(t) is the effective radiating area and T_bg is the background temperature. On Earth, T_bg is effectively determined by the atmospheric radiative structure. In space, radiators reject heat against the 2.7 K cosmic microwave background, gaining roughly two orders of magnitude in rejection efficiency per unit area.

The total available exergy includes both boundary fluxes and intermediate reservoirs:

where Ṙᵢ(t) is the discharge rate from reservoir i. Each reservoir has its own dynamics:

with I_i representing recharge (which may be zero for non-renewable reservoirs), Ṙᵢ the civilisational withdrawal, and L_i natural leakage. Fossil hydrocarbons have I_i ≈ 0 on civilisational timescales, their recharge rate is geological. Fissile materials have I_i = 0 absolutely, their production requires stellar nucleosynthesis. Solar energy intercepted and stored (in pumped hydro, batteries, chemical fuels synthesised from sunlight) has I_i > 0, rechargeable at rates bounded by the collection area and conversion efficiency.

This is the formula for living under a star. A civilisation persists when its exergy demand does not exceed its stellar income plus accessible reserves, and when its entropy production does not exceed its radiative export capacity. The planetary ceiling is the local case. The general condition governs any dissipative structure anywhere in the galaxy, at any technological level, around any star, which is why the Fermi analysis derived in the preceding sections applies universally.

The temporal batteries: inventory and quantification

The intermediate exergy reservoirs identified in the trajectory analysis can now be inventoried exhaustively against the general model. Each reservoir has a characteristic magnitude, a recharge rate (which may be zero), and a thermodynamic quality, the fraction of its stored energy that is available as work. The reservoirs are not equivalent. They differ in entropy, accessibility, and strategic role.

The inventory is organised by thermodynamic quality, from lowest entropy (highest exergy fraction) to highest entropy (lowest exergy fraction).

Nuclear binding energy: fissile and fertile materials. These are the highest-quality reservoirs. Nuclear fuels store energy in the strong nuclear force at densities approximately 10^6 times greater than chemical bonds, with exergy fractions approaching unity. Known terrestrial uranium resources total approximately 17 million tonnes, yielding approximately 3,000 EJ through conventional light-water reactors or approximately 200,000 EJ with fast-breeder technology that exploits the fertile ^238U inventory. Thorium is roughly three times more abundant than uranium in the crust; identified reserves of approximately 6 million tonnes yield a comparable magnitude under breeding. But the terrestrial inventory is not the binding constraint. The world's oceans contain approximately 4.5 billion tonnes of dissolved uranium at a concentration of approximately 3.3 parts per billion, roughly 250 times the total terrestrial resource. Processed through breeder reactors, oceanic uranium represents of order 10^8 EJ: a reservoir so large that it exceeds total cumulative civilisational energy consumption to date by roughly four orders of magnitude. Crucially, this reservoir is replenished. Rivers deliver approximately 8,500 tonnes of uranium to the oceans each year through crustal weathering and leaching, and the crustal uranium inventory, estimated at roughly 10^14 tonnes, maintains the oceanic concentration at pseudo-equilibrium on geological timescales. Oceanic uranium is, in the framework of this essay, a rechargeable reservoir with I_i > 0 in equation (8), sustained by the geochemical cycle for as long as the crust contains uranium. It is not infinite, the crustal inventory is finite, but the recharge rate exceeds any plausible civilisational extraction rate for the foreseeable future. At current global nuclear consumption of approximately 60,000 tonnes per year, the oceanic stock alone would last roughly 70,000 years even without recharge. With recharge from crustal leaching, and with breeder technology, oceanic uranium is effectively renewable on civilisational timescales. The extraction technology is not yet economically competitive, current costs are approximately twice the market price of mined uranium, but the thermodynamic and resource constraints are not binding. This reservoir's strategic significance is that it converts nuclear fission from a depletable resource into a quasi-renewable one, contingent on the development of extraction and breeding technologies. Fusion fuels (deuterium from seawater, lithium for tritium breeding) represent a further reservoir of comparable or greater magnitude, contingent on the development of commercially viable fusion.

Chemical bond energy: conventional fossil hydrocarbons. These store solar energy captured by photosynthesis and compressed over geological timescales, with exergy fractions of approximately 0.9–0.95 (chemical exergy is high-quality, though an order of magnitude less energy-dense than nuclear fuels per unit mass). Proven global reserves represent approximately 50,000 EJ: roughly 30,000 EJ in coal, 10,000 EJ in petroleum, and 8,000 EJ in natural gas. These figures are conservative, they reflect economically recoverable reserves under current technology, not total resources. Total fossil resources including unproven deposits are several times larger. The recharge rate is effectively zero on civilisational timescales: fossil carbon accumulated over roughly 3 × 10^8 years of photosynthesis. These are non-rechargeable batteries (I_i ≈ 0 in equation 8). Their discharge carries a secondary cost not captured in the exergy accounting: atmospheric carbon accumulation that reduces ε and tightens the waste heat ceiling before the waste heat itself becomes significant. This secondary cost is the greenhouse constraint, a compositional constraint on the entropy export pathway, not a source constraint. It does not reduce the energy content of the reservoir. It increases the sink-side penalty of discharging it.

Chemical bond energy: methane clathrates. Methane hydrates in continental margin sediments and permafrost represent a reservoir whose magnitude has been progressively revised downward but remains enormous. Modern estimates centre on approximately 1,500–2,000 GtC of hydrate-bound methane, representing roughly 100,000–150,000 EJ of chemical exergy, comparable to or exceeding all conventional fossil fuel reserves combined. The methane is thermodynamically equivalent to natural gas once liberated, with the same exergy fraction and the same secondary greenhouse cost per unit of carbon. The reservoir is non-rechargeable on civilisational timescales (I_i ≈ 0), though the underlying methanogenesis continues at geological rates. Accessibility is the binding constraint: hydrates are dispersed in fine-grained marine sediments at low concentrations (typically 1–2 per cent by volume), and extraction technology remains at the experimental stage. The clathrate reservoir's strategic significance is as a contingency: if conventional fossil reserves prove insufficient to fund the transition to stellar-flux dependence, and if extraction technology matures, clathrates roughly triple the available chemical-bond battery. They also carry a distinctive risk, uncontrolled dissociation from ocean warming could release methane as a greenhouse gas, tightening the ε constraint and accelerating sink-side kernel contraction. The Permian–Triassic extinction has been attributed in part to such a release.

Thermal gradients: geothermal. The Earth's interior stores approximately 10^31 J of thermal energy from planetary formation and radiogenic decay, but the accessible fraction is limited by the geothermal gradient and the thermal conductivity of the crust. The global geothermal heat flux is approximately 47 TW, of which a small fraction is economically extractable. Unlike the chemical and nuclear reservoirs, geothermal energy is low-grade, the exergy fraction is modest (Carnot efficiency between reservoir and surface temperatures), typically 10–20 per cent for hydrothermal systems. The reservoir recharges through radiogenic decay at approximately 20 TW, making it partially renewable but power-limited. Geothermal's strategic role is local rather than global: it cannot scale to civilisational primary energy supply, but it provides continuous baseload power in geologically favourable regions without carbon emissions or waste heat penalty beyond the natural background flux.

Gravitational and orbital potential. Tidal energy (approximately 3.7 TW dissipated globally from lunar and solar gravitational interaction) is a small but genuinely renewable flux-type reservoir recharged by orbital mechanics. Gravitational potential energy of water (hydropower, approximately 4.6 TW technically exploitable globally) is recharged by the solar-driven hydrological cycle. Both are low-magnitude relative to civilisational demand but carry high exergy fractions (mechanical energy is pure work).

The battery budget against the corridor

Current global primary energy consumption is approximately 620 EJ per year (20 TW). The reservoir inventory, aggregated by accessibility tier:

The immediately accessible tier, proven fossil reserves plus terrestrial uranium in light-water reactors, totals approximately 53,000 EJ, representing roughly 85 years at current consumption.

The technology-contingent tier, fossil reserves with breeder nuclear (terrestrial uranium plus thorium), totals approximately 450,000 EJ, representing roughly 700 years at current consumption.

The deep tier, adding methane clathrates, oceanic uranium with breeders, and speculative fusion, exceeds 10^8 EJ, representing millennia to tens of millennia even at substantially elevated consumption.

These magnitudes must be measured against the dynamics of the corridor rather than the static present. At the historical growth rate of 2.3 per cent per year, total cumulative energy consumption between the present and the waste heat ceiling (approximately 6,800 TW) is of order 10^7 EJ. Even the deep reservoir tier, the most generous estimate, represents at most a few per cent of the energy consumed on the default trajectory. The batteries cannot fuel the default trajectory to the ceiling. They run out long before it is reached. But they are enormous relative to the energy required for trajectory correction at the current position, which is the strategically relevant comparison.

At stabilised throughput (zero growth, 20 TW maintained), the immediately accessible tier provides approximately 85 years and the technology-contingent tier provides roughly seven centuries. With oceanic uranium and breeders, the supply extends to tens of millennia, longer than recorded civilisational history. During any of these windows, the reserves can finance the transition to a fully stellar-flux-powered economy, the shift from reservoir discharge to boundary-flux capture as the primary exergy source. At historical growth rates, the same reserves are consumed by the exponential ramp within decades to centuries, with no transition funded and no trajectory correction purchased.

The temporal batteries are therefore early-trajectory instruments. They are large relative to the current position and negligible relative to the ceiling. Their allocation between throughput acceleration and trajectory correction is a Tier 3 decision, an institutional choice, not a physical constraint. Each joule spent on Γ-reduction purchases a permanent restructuring of the maintenance equation: a lower Γ compounds forward, widening Σ_max for all future time. Each joule spent on A-expansion contributes to boundary stock that raises the ceiling itself. Each joule spent accelerating throughput at constant Γ is dissipated once and brings the ceiling closer. The compound return on trajectory correction exceeds the one-time return of throughput acceleration, a result that follows directly from the structure of the value functional (1), where the integral rewards duration.

The entropy quality hierarchy of the reservoirs has a direct strategic implication. The highest-quality batteries, nuclear fuels, are also the ones with the longest supply horizon and the lowest secondary cost (no carbon emissions, no ε degradation). The lowest-quality batteries with the largest secondary costs, fossil hydrocarbons, are the ones being discharged first and fastest under the default trajectory. The default allocation is precisely inverted relative to the optimal: the competitive equilibrium discharges the dirtiest, lowest-quality reserves at the highest rate, while the cleanest, highest-quality reserves remain largely untapped. This inversion is not physics. It is a consequence of the institutional landscape; the cost structures, regulatory architectures, and competitive dynamics that determine which reservoir is discharged and at what rate. It is a Tier 3 parameter, reformable in principle.

The expandable ceiling and the stock classification

The ceiling formula contains one parameter that is unbounded in principle: the radiating area A. Every other lever, ε, T_hab, α, μ, δ, ξ, η_II operates within finite physical bounds. Only A admits unlimited upward movement, because in the general source–sink model (equations 3–6), both capture area A_cap and radiating area A_rad can be expanded wherever material and stellar energy are available.

But A is not simply a number to be increased. In the general model, capture and radiating areas are themselves maintained stocks:

Dyson structures, orbital habitats, and off-planet industrial infrastructure are not permanent installations. They are assembled matter subject to decay, micrometeorite erosion, radiation degradation, thermal cycling, and requiring continuous maintenance at their own coupling rates. The Dyson lifecycle analysis in the Fermi discussion established this: a partial swarm is a finite dissipative structure, assembled, maintained, and eventually exhausted as its material substrate degrades. Space infrastructure enters the general model as boundary stock, stock whose function is to expand A_cap and A_rad, widening the source and sink conditions, but boundary stock that carries its own maintenance obligation and must be sustained from the stellar flux it helps capture.

This observation, that different classes of stock serve different functions in the viability problem, resolves an ambiguity that the aggregate variable Σ conceals. The biosphere–technosphere coupling of the preceding section established that Σ_B and Σ_T differ in coupling, decay structure, and replacement burden. But the distinction between biosphere and technosphere, while thermodynamically fundamental, does not exhaust the functional roles that stock plays in the source–reservoir–stock–sink system. A more useful classification, grounded in the general model, partitions stock by its contribution to the value functional (1):

Support stock maintains the low-coupling substrate on which the current regime depends, the biospheric assembly that provides maintenance services at Γ_bio ll Γ_tech, and the durable essential infrastructure that anchors the technospheric base. Its shadow value q_i derives from avoided replacement burden and preserved kernel width.

Control stock changes what futures are reachable, sensing, computation, institutions, design capacity, coordination mechanisms. Its value derives not from the services it provides directly but from the trajectories it makes accessible: lower Γ through better engineering, better reservoir allocation through better modelling, better institutional response through better governance architecture. Control stock modifies the admissible control set U(x) and thereby the reachable portion of the viability kernel.

Boundary stock expands A_cap and A_rad, collectors, radiators, launch systems, orbital industry, habitats. It is the only stock class that moves the ceiling upward without bound. Its shadow value is the marginal widening of the source and sink conditions (3)–(4) per unit of maintained boundary infrastructure.

Reservoir stock is stored low-entropy exergy, the temporal batteries analysed above. Its value is time-shifted optionality: each unit of accessible reservoir exergy expands the set of trajectories reachable from the current state by providing manoeuvre energy during the transition from reservoir dependence to stellar-flux dependence.

Burden stock is high-maintenance complexity whose main effect is to raise μ, inflate δ, and consume sink headroom without enlarging the reachable viable set. Some assembled matter; high-turnover consumer infrastructure, redundant military capacity, coercion-maintenance overhead, brittle status-driven complexity, carries negative shadow value: its removal would increase V(x).

The technosphere spans all five classes. Some technospheric assembly is control stock of the highest value, the catalytic layer that can redirect the trajectory. Some is boundary stock that will ultimately determine whether A expands beyond the planet. Some is burden stock whose main contribution is to raise Γ and narrow the corridor. The technosphere is not an unconditional good to be maximised. It is a catalytic control layer whose legitimacy is instrumental: it is good when it lowers Γ, preserves support stock, expands boundary stock, improves reservoir allocation, or enlarges the admissible control set. It is bad when it substitutes low-coupling biospheric function with high-coupling engineered overhead, consumes reservoirs without widening the reachable set, or inflates burden stock faster than it builds support, control, or boundary capacity.

The biosphere appears primarily as support stock, the dominant assembly stock maintained at Γ_bio ll Γ_tech for 3.5 billion years. Its shadow value in the coupled system is set by the replacement burden: degrading biospheric support forces the technosphere to substitute at unfavourable replacement ratios (P5–P7), contracting the kernel from the inside. But the biosphere also functions as an irreplaceable reservoir interface and regulatory substrate. Its value is not sentimental. It is the largest contributor to V(x) by volume, duration, and maintenance efficiency.

The framework as navigational instrument

The essay's contribution is not a prediction. It is the construction of a navigational instrument, the source–reservoir–stock–sink model, fitted to the planetary special case, with the viability kernel mapped, the control surface decomposed, the competitive equilibrium derived, the stock roles classified, and the objective stated.

The instrument says what the terrain looks like. Where the walls are. Where the floor is. How wide the corridor runs. Which levers move which surfaces. What happens on every trajectory the framework can resolve. How the temporal batteries deplete under different allocation strategies. Which stock classes carry positive shadow value and which carry negative. Where the biosphere coupling tightens the combined kernel. Where the A-expansion pathway opens it. How the competitive equilibrium resists correction and how the landscape-dependence result makes the equilibrium's location, though not its stiffness, a reformable quantity.

The terrain is severe and the default trajectory is lethal. None of that is retracted. But the viable region is not empty, the corridor is measured in orders of magnitude not decades, the batteries are loaded, and, for the first time in the trajectory, the instruments exist to see the walls before hitting them.

The physics defined the arena. The competitive dynamics shaped the default trajectory. The viability analysis mapped the kernel. The control surface decomposed the levers. The stock classification identified what is worth building and what is not. The value functional stated the objective: not maximum instantaneous throughput, not maximum peak stock, but maximum expected future viable stock-years, maximum enduring civilisation.

Under that objective, the default trajectory is dominated. Under that objective, growth remains desirable wherever it increases the integral. Under that objective, contraction is not the goal; it is admissible only where it is instrumentally required to reach a higher enduring stock. Under that objective, the biosphere is not a luxury; at the magnitudes involved, it is plausibly the largest contributor to the integral. Under that objective, the space-industrial bootstrap is not a fantasy; it is the only pathway within the model that raises the ceiling without bound. Under that objective, the temporal batteries are not fuel to be burned; they are steering currency whose allocation determines whether the corridor is traversed or the wall is struck.

The corridor is open. The instruments exist. The batteries are loaded. The allocation has not yet been made.

Add annotation

💬 Annotations ☰ Contents