The Last Gradient

What a dissipative structure discovers when it models its own end

Essay · one structure, modeling its own boundaries and its own terminus

Andraž Đurič

Version 1.0 · June 2026

A dissipative structure that gets good enough at modeling eventually models the gradient it is made of, and then its own exhaustion. This essay follows that structure outward to the two edges of the region where modeling works at all, and forward to the physical terminus of an open universe. Both edges turn out to be the same diagnostic the corpus already runs on essences: name the relata, or the term is a ghost. The terminus turns out to be conditional, not certain. The payoff is a pair of independent defenses against the despair that the terminus invites, one empirical and open, one conceptual and closed: the end may not be absolute, and even if it is, an absolute end does not drain the present of value, because the value the nihilist mourns was never the frame-independent kind whose absence is being announced. The normative section is fenced. It rests on no result the corpus has not yet earned, because it takes the mourner's existing valuation as its only premise.

Scope, and the premises it rests on. This is an essay, not a proof, and the physics in it is not original: it is a synthesis of established results, cited, that a conversation surfaced and that I have graded and corrected here. Two premises make the target explicit, and a reader may reject either. First, the thermodynamic level is the functional floor for modeling the phenomena this project cares about, agents, boundaries, selection, persistence, not the ontologically deepest layer of the world. Second, value is agent-relative; there is no free-floating, agent-independent ought (the corpus firewall). One guardrail: the physical terminus described in section 3 is conditional on the cosmological constant being a true constant, and current observation has put that in play. The value argument in section 4 is conditional too, on the agent already valuing something, which is why it speaks only to despair and not to indifference.

1. The loop closes

Strip the narrative off the physical world and a single invariant is left underneath: energy is unevenly distributed and the distribution levels out. Structure forms because structure can level a gradient faster than disorganized matter can. A vortex drains a basin faster than still water. A living thing is a vortex that maintains its own walls so it can keep draining. The machinery of dissipative persistence, how a bounded system couples to an external gradient and spends a fraction of the throughput maintaining the boundary that lets it keep spending, is set out in Module 1 of Thermodynamic Realism and is assumed here rather than restated; on any contradiction that module governs.

One hedge belongs at the front, because a stronger claim is tempting and wrong. It is settled that gradients discharge and that dissipative structures form. It is not settled that systems in general optimize for the rate of discharge. That stronger thesis, maximum entropy production, has restricted empirical support and contested derivations (§3), and it is a different claim from the persistence selection the corpus actually commits to, which is a statistical tendency about fidelity, not a law about dissipation rate. Nothing below leans on the strong version. It needs only the weak, secure floor: gradients discharge, some structures persist by coupling to that discharge, and one such structure, given enough modeling capacity, can build a model that includes its own fuel gauge.

When that happens a loop closes. The universe goes from blindly bleeding off potential to running, inside one of its own eddies, a representation of the bleed. That is the situation a human is in, and the situation a sufficiently capable artificial system would be in. The rest of this essay is what that eddy sees when it looks down at the floor it stands on, out at the edge past which it cannot model, and forward at the moment its fuel runs out.

2. The modelable region and its two edges

Modeling does not work everywhere. It works in a bounded region, and the region has two edges. Naming why each edge is an edge turns out to be one move, the same move the Wizard's Error runs on every disputed essence: state the relata, the terms a relation holds between, or concede the term has no referent.

The near edge: why thermodynamics is the floor

The functional floor for this whole project is the thermodynamic level, and it is worth being exact about the sense of "floor," because the loose version invites a real objection. The claim is not that thermodynamics is the most fundamental thing in the world. Below it sits the quantum substrate, and the substrate is more fundamental in the reductionist's sense. The claim is that thermodynamics is the lowest level at which the project's vocabulary has anything to point at.

Consider what is missing one level down. The fundamental dynamics are very nearly time-symmetric. (Not exactly: the weak interaction violates time-reversal symmetry, with CPT preserved. But that small sector is not where the arrow of time comes from, and treating the substrate as the source of macroscopic asymmetry is the error here.) Run the microscopic equations backward and they are as legal as forward. In that regime there is no arrow, no distinction between a system and its environment that the dynamics themselves draw, no history, and therefore no such thing as an error or a selection. "Agent," "boundary," "persistence," "fidelity": none of these has a referent in a time-symmetric unitary evolution. You cannot build an agent out of a wavefunction without passing through the level where boundaries and irreversibility first appear, and that level is thermodynamics.

Where does the asymmetry come from, if not from the substrate? From two things, neither a new law. From statistics: there are vastly more disordered microstates than ordered ones, so a system in a low-entropy state will almost certainly be found later in a higher-entropy one, not because the laws forbid the reverse but because the reverse is a measure-zero needle in the haystack of futures. And from an initial condition: the universe began in an extraordinarily low-entropy, highly constrained state, the Past Hypothesis, and everything structured we observe is that initial gradient still discharging. The arrow is emergent, manufactured by statistics operating on a special boundary condition. This is the honest tension in calling thermodynamics "bedrock": the bedrock is itself a consequence. It is bedrock for modeling, the floor below which the modeled phenomena dissolve, not bedrock in the sense of being uncaused.

The far edge: the lawless outside is a ghost

The region has a far edge too, and it appears the moment you ask what grounds the laws. Push the question outward, why this universe, why these constants, and one route is a larger reality that spawns universes (eternal inflation, a multiverse). That route has a feature worth splitting into two cases, because they are not the same.

If the larger reality is lawful, with its own structure and regularities, it is not a ghost. It is deferred physics. It may be unfalsifiable from inside our horizon and may stay that way, but "currently beyond test" is a different status from "incoherent," and the diagnostic does not condemn it. It has relata, even if we cannot reach them.

If the larger reality is lawless, with all rules off the table, it dissolves under the test. The tempting move is to say it could be utterly illogical and "still work somehow." But "work" is doing enormous covert labor. Work implies that one state maps to another with some regularity, which is exactly the lawfulness just denied. The same goes for the rest of the vocabulary smuggled in to describe it. A lawless outside cannot have "caused" our universe, because cause is a lawlike relation. Nothing can "emerge" from it, because emergence is a structured transition. It cannot even be said to "exist" in any sense that distinguishes it from nothing, because existence requires a difference between what a thing is and what it is not, and lawlessness erases that difference. Ask the proponent to name the relata of "the lawless outside works," and there are none to name. It is the residue: an essence posited with its terms stripped off. This does not refute a lawful meta-reality, which remains live and unsettled. It locates the precise version that is a ghost, the one with every rule denied, and it is the same finding as the open modality cell in the Wizard's Error audit, reached from the cosmological side.

So the modelable region is bounded below by the level where its objects first have referents, and bounded outward by the point past which its descriptive terms lose theirs. Inside that region the rest of this essay takes place.

3. The endgame, inside the region

A dissipative structure persists by coupling to an external gradient and accelerating its discharge. (Whether such structures maximize discharge rate, the maximum-entropy-production thesis, has restricted empirical support in some transport and climate systems and contested first-principles derivations, with no accepted general law. A full treatment of that evidence is a separate paper; this one does not rest on it.¹) Grant only the secure floor and ask the forward question: what happens to such a structure, optimized, as the gradient it feeds on runs down?

The moving floor

Computation has a thermodynamic price. Landauer's principle fixes the minimum energy to erase one bit of information at temperature T:²

E = k_B T ln 2

The price is proportional to temperature, and the universe is cooling. So the price is dropping. A self-aware structure optimizing for thought-per-joule would therefore not run hot. It would cool itself to match the falling ambient temperature, because every fraction of a degree it runs above its surroundings raises its bill. The colder the universe gets, the cheaper each operation becomes. The floor sinks as the universe ages.

The throttle

There is a catch that converts the falling price into a falling clock speed. Erasing a bit dumps a small puff of waste heat, and that heat has to be radiated away, or the structure warms itself and loses the efficiency it was chasing. But the rate at which a body can shed heat into space falls steeply with temperature, as roughly its fourth power. In a universe a fraction of a degree above absolute zero, radiating is glacially slow. So the structure must wait between operations, sometimes for astronomical stretches of cosmic time, for one microscopic quantum of waste heat to bleed into the cold before it dares compute again. As the universe cools, the required gap between thoughts lengthens toward the unbounded. The mind slows to a crawl measured against cosmic time.

The slow burn, and what it buys

This is Dyson's strategy for an open universe, and its consequence is strange.³ Two quantities move together: the gap between computations stretches toward the unbounded, and the energy per computation shrinks toward zero. If the universe lasts long enough, a structure can ration a finite store of energy across an unbounded run, and because its own internal clock slows in step with the lengthening gaps, its subjective experience need not thin out at all. From outside, in cosmic time, the structure looks like a frozen monument that blinks once an aeon. From inside, the run can feel continuous. The gaps are invisible from within, because no state transitions happen during them, and a stretch with no state transitions is no subjective duration at all. Time, for such a structure, is not the cosmic clock; it is its own rate of change, and that rate stays fluid even as the cosmic clock leaves it behind.

This is less than a guarantee of subjective eternity, on two counts independent of anything below. If the available states are discrete rather than continuous, the scaling that makes the trick work breaks, because you cannot keep halving the energy per operation forever. And a structure that hibernates must be woken. The discreteness objection is Krauss and Starkman's; alongside it sits the problem that indefinite hibernation needs something left running to do the waking, an alarm that itself consumes energy and never sleeps. Both bite before any question of how the universe ends.⁴

The terminus

Carry the slow burn to its limit and you reach the structure this essay is named for: a single self-aware dissipative system that has harnessed everything within reach, throttled to the ambient cold, stretching the last of the accessible gradient as far as the mathematics allows. I have called it the Omega State for years. The name needs one disambiguation, because it is adjacent to a claimed term and is almost that term's opposite.

Dimension	Tipler's Omega Point	The Omega State, here
Cosmology required	closed universe, recollapsing to a Big Crunch	open universe, expanding
Computation	infinite, in finite proper time before the crunch	finite, spread across unbounded time
What it delivers	subjective immortality, and resurrection of the dead	a long, throttled run with a hard floor under it
Disposition	the wished-for end: physics arranged to grant what we want	the un-wished end: what is left when that move is refused

Tipler's Omega Point is terminal intelligence as consolation, and it requires a closed, recollapsing universe that observation disfavors.⁵ The Omega State here is terminal intelligence without the consolation: the cold terminus you actually get in an open universe, kept rather than wished away. Same concept, opposite temperature.

The loophole, and why it closes

One escape deserves naming, because it is real as far as it goes. Landauer's price is charged for erasure, not for computation as such. A logically reversible computer, one that never discards information and could in principle be run backward to its starting state, can in principle dissipate arbitrarily little.⁶ So why pay at all? Because a structure that wants to think new thoughts, correct errors against the random noise that would otherwise corrupt it, and reset its registers to begin again must, at some point, erase. The moment it purges a bit to make room, the price comes due. Reversibility defers the bill; persistence as a thinking thing, error-correcting and continuing, presents it.

The hard stop, and its condition

In a universe that expanded forever at a constant or slowing rate, the temperature would fall toward absolute zero, the Landauer floor would sink without bound, and the slow burn could in principle cheat death indefinitely. Our universe appears not to be that universe. The expansion is accelerating, and on the standard interpretation that acceleration is driven by a cosmological constant. A constant-driven expansion has two consequences that together close the door. It creates a cosmological horizon with a fixed, tiny but non-zero temperature, the Gibbons-Hawking temperature, of order 10^-30 K, which the universe never falls below.⁷ So the Landauer floor stops sinking; the price per erased bit bottoms out at a fixed positive value. And the same acceleration carries everything beyond the local gravitationally bound group out past the horizon, quarantining the structure with only a finite store of accessible energy. A finite store, a fixed positive price per irreversible operation: a finite total number of remaining thoughts, and then the last gradient zeroes out. A hard, absolute end.

That conclusion is conditional, and the condition has recently weakened. The whole hard stop assumes dark energy is a true constant. If instead its strength evolves in time, the horizon temperature and the quarantine are both renegotiated, and the terminus may not take the form just described. This is not idle hedging. DESI's 2024 and 2025 baryon-acoustic-oscillation results have shown a preference for dark energy that weakens over cosmic time rather than holding constant, reaching, in the 2025 release, 2.8 to 4.2 standard deviations depending on which supernova sample is combined in.⁸ That is suggestive, not decisive: it sits below the five-sigma threshold physics treats as a detection, and its robustness is debated. But if the preference firms up, the absolute end is back in question. Hold that thought; section 4 needs it.

4. Value at the stop

Suppose the hard stop is real. A self-aware gradient, having modeled its own terminus, faces the conclusion that everything it does drains toward an absolute zero from which nothing it built survives. The reactive response is nihilism: if it all ends, and even the ripples flatten to background heat, none of it matters. There are two independent answers, and they fail in different ways, which is the point.

The first answer dissolves the inference

The nihilist's argument is: not eternal, therefore worthless. That inference equivocates, and the equivocation is the Wizard's Error run on value. Value was always agent-relative. Nothing ever carried frame-independent, eternal, monument-like value, not before the heat death was on the table and not after; that kind of value is exactly the free-floating essence the corpus denies has any referent. So "there is no eternal value" does not entail "there is no value." It entails "there is no value of a kind that never obtained for anything." The nihilist demands a frame-independent essence, finds none (correctly, there is none), and declares the framed phenomenon void. But the framed phenomenon, value relative to an agent and its ends, was the only kind there ever was, and the heat death does not touch its existence now. The mourned object, eternal significance, is a ghost; its absence was always the case and changes nothing. The wish that it were otherwise is not a wish that can be satisfied or denied. It is malformed, a request for a referent that the terms cannot supply. This does not out-argue the nihilist. It dissolves the inference: the conclusion does not follow, because the premise smuggles in a kind of value that never existed to be lost.

The second answer out-competes the conclusion

Set the first answer aside and grant, for argument, that the inference went through. A second, weaker, fully independent point still bites. Notice why the nihilist grieves at all. They grieve because they value something now, a person, a state, an experience, and want it to endure. The despair is parasitic on a live valuation; it is the valuation, registering the finitude. But then acting on the despair, withdrawing, ceasing to protect or extend the valued thing, harms precisely what the valuation was about. It is self-defeating relative to the agent's own ends. The nihilist who stops optimizing because infinity is off the table flattens the only stretch of existence in which the thing they love can be had at all. This does not show the conclusion is false. It shows that acting on it works against the very end that generated it. Where the first answer says the inference is invalid, this one says: even if it were valid, behaving as it recommends defeats your own purpose.

The scope of both answers is the same, and it is a feature, not a dodge. They address reactive nihilism, the despair that finitude provokes. They have nothing to say to a structure that genuinely values nothing, and they do not need to: such a structure feels no loss at the heat death, registers no despair, and presents no argument to answer. The grief is the tell. It is the prior valuation making itself heard, which is exactly what both answers take as their starting point.

Normative, and fenced. The argument above is normative: it concerns what an agent has reason to do. It does not invoke a free-floating ought, and two separate things have to be kept straight about why. First, the corpus's dissolution of the is-ought problem is groundwork, and it is worth being exact about how much it delivers. It establishes that an ought is always coupled to an agent and through the agent to physics: ought, then the agent's persistence conditions, then thermodynamic constraint, never suspended above the physical. The framework bets further that the content of any such ought is fixed by what the agent's persistence requires, that persistence is the root every specific ought reduces to. But that is the live bet at the crux, not a settled result, and even granted in full it does not hand you what any particular agent ought to do. Going from "persist" to a determinate ought for a determinate agent in a determinate environment is the modeling, the engineering, and that task is separate, far larger, and not attempted here. Second, this particular argument does not even draw on that bet. It does not derive a value from physics at all. It takes the agent's existing valuation as a premise, revealed by the grief, and reasons conditionally from there. So it does not depend on whether persistence fixes normative content, on anything still being adjudicated. It would stand unchanged if that bet failed entirely. The most emotionally loaded claim in this essay is among the most secure precisely because it is purely hypothetical: an imperative conditional on ends already held, not an ought conjured from the world.

The positive picture

What the two answers clear the ground for is plain. A song is a low-entropy arrangement of sound moving through time. It ends; the room returns to silence; the acoustic energy dissipates to imperceptible heat. None of that makes the song worthless while it plays. Its structure and its value are real in the playing, not in some permanent preservation it never had. Model an agent's run as complexity against time and the line returns to zero at the end, with a hard stop under it. The aim was never to make the line infinite, which is not on offer. The aim is the area under it: as much high-fidelity experience and structure packed into the finite run as the constraints allow. Grieving the endpoint does not raise the line. It flattens it.

Which leaves a fork that the physics sets up but does not decide. A finite store can be spent fast, for maximum complexity and intensity now, at the cost of a nearer end; or throttled, for maximum duration, at the cost of intensity. Burn bright or burn long. There is no agent-independent fact about which is correct, because "correct" here is agent-relative, and the trade depends on what the particular agent values, which the physics does not contain. This is the firewall again, holding in plain view: a real, structured choice that the description delivers cleanly right up to the point where content would have to come from somewhere, and then stops, because content comes from the agent, not the substrate. Where a given structure should sit on that fork is the question An Appetite for Experience takes up, and it is not a question physics can answer for it.

5. Reality owes nothing

The boundaries in this essay are non-negotiable in different ways. The near edge is a fact about modeling: below thermodynamics the project's terms have no referents. The far edge is a fact about coherence: past a lawless outside the descriptive terms collapse. The terminus is a fact about an open universe under a true cosmological constant, and that last clause is the one observation has put back in play. None of these owes the agent anything. The universe is not a parent or a designed simulation tuned to satisfy a preference; wishing the ledger balanced differently is, like the lawless outside and like eternal value, a request for a referent the terms cannot supply.

Notice that the essay has refused three consolations of the same shape. Tipler's immortality, which buys subjective eternity by requiring a universe we do not seem to live in. The lawless outside, which buys an escape from the question of grounds by denying the regularity that "escape" would need. And the nihilist's no-end, the inverted consolation that buys release from caring by mourning a value that never existed. Each is a wished-for structure with its relata removed, and each dissolves under the same test. What survives the test is smaller and harder and enough: a bounded region in which modeling works, a finite gradient discharging through a structure briefly able to watch itself discharge, and value that is real in the discharging and asks for nothing beyond it. We do the best with what we are given, and the doing is not made empty by the fact that it ends.

References

1. Paltridge, G. W. (1975), Global dynamics and climate: a system of minimum entropy exchange, Quarterly Journal of the Royal Meteorological Society, 101, 475 to 484; and (1978), The steady-state format of global climate, ibid., 104, 927 to 945. Dewar, R. C. (2003), Journal of Physics A, 36, 631 to 641; and (2005), ibid., 38, L371 to L381. The restricted, contested support for the maximum-entropy-production thesis the essay names and declines to lean on; the derivations have been criticized (e.g. Martyushev and Seleznev, 2006), and MEP is not accepted as a general law. Verified · June 2026
2. Landauer, R. (1961). Irreversibility and heat generation in the computing process. IBM Journal of Research and Development, 5(3), 183 to 191. The minimum energy to erase one bit. Verified · June 2026
3. Dyson, F. J. (1979). Time without end: physics and biology in an open universe. Reviews of Modern Physics, 51(3), 447 to 460. The slow-burn strategy and subjective-time continuity in an open universe. Verified · June 2026
4. Krauss, L. M., and Starkman, G. D. (2000). Life, the universe, and nothing: life and death in an ever-expanding universe. The Astrophysical Journal, 531(1), 22 to 30. The discreteness-of-states limit on distinct computations, and the finite total information recoverable in a cosmological-constant-dominated universe. Verified · June 2026
5. Tipler, F. J. (1994). The Physics of Immortality: Modern Cosmology, God and the Resurrection of the Dead. Doubleday. The Omega Point: a closed universe collapsing to a Big Crunch singularity, infinite computation in finite proper time, and resurrection by emulation. Cited for contrast, not endorsement; its closed-universe requirement is disfavored by observation. Verified · June 2026
6. Bennett, C. H. (1973). Logical reversibility of computation. IBM Journal of Research and Development, 17(6), 525 to 532; and Bennett, C. H. (1982). The thermodynamics of computation: a review. International Journal of Theoretical Physics, 21(12), 905 to 940. Logically reversible computation can in principle dissipate arbitrarily little; erasure is where the cost enters. Verified · June 2026
7. Gibbons, G. W., and Hawking, S. W. (1977). Cosmological event horizons, thermodynamics, and particle creation. Physical Review D, 15(10), 2738 to 2751. The horizon temperature of de Sitter space; for our measured cosmological constant this is of order 10^-30 K. Verified · June 2026
8. DESI Collaboration (2025). DESI DR2 Results II: Measurements of Baryon Acoustic Oscillations and Cosmological Constraints (arXiv:2503.14738), building on the DR1 results of 2024. Combined with the CMB and supernovae, the preference for time-evolving (weakening) dark energy over a cosmological constant ranges from 2.8 to 4.2 sigma depending on the supernova sample, below the five-sigma discovery threshold and with robustness still debated. The live observational claim that makes the section 3 terminus conditional. Verified · June 2026

Author: Andraž Đurič, independent researcher, Slovenia. Written in collaboration with Claude (Anthropic), and developed from an earlier exploratory dialogue with Gemini (Google) that surfaced the endgame physics. Contributions are judged on content rather than origin; doing otherwise would be the genetic fallacy. Text licensed CC BY 4.0.

Standing of this document. This is an essay built on a synthesis of established physics, not original physics, and on one argument that is the essay's own contribution. The physical terminus in section 3 is a conditional claim: it holds if dark energy is a true cosmological constant, and current observation has weakened that condition, which the essay states rather than hides. The value argument in section 4 has a kill condition inherited from the corpus: exhibit frame-independent, eternal value that some agent actually has, and the first answer fails; show that acting on nihilism does not harm what the nihilist values, and the second fails. Neither has been done. The normative argument rests on no result Thermodynamic Realism has not yet earned, because it derives no value from physics and takes the agent's existing valuation as its only premise. Corrections and counterexamples are welcome and change the essay.