Entropy, Time, and Space: All the definitions - The Measure on the Possible

Entropy is the measure on the space of the possible, taken from the standpoint of what is instantiated.

This report answers a single question — what are time and space, and how do they relate to entropy — and it answers it through one organizing claim. The claim is not that entropy is disorder, nor that it is ignorance. It is that entropy is a measure: it takes the space of what remains possible and quantifies it from the vantage of what has been fixed. A macrostate, a source distribution, a boundary region, a prepared density matrix, an observed datum — each is an instantiated specification that selects a slice of possibility-space, and entropy is the size of that slice. Everything below is a consequence of taking this seriously, including the parts where the standard physics and our reading of it have to be told apart.

Two conventions are kept throughout. Where a result is established physics, it is stated as physics. Where it is an interpretation the physics permits but does not force, it is flagged as interpretation. The discipline is the point: the world bites, and the report is written so that the parts that bite are separated from the parts that merely render well.

1. The family of entropies, and why it is one family

1.1 The shared form is not a coincidence

Every entropy in standard use shares the form

$$ H = -K \sum_i p_i \log p_i, \qquad K > 0. $$

This is usually presented as a remarkable recurrence — Boltzmann, Gibbs, Shannon, and von Neumann all arriving at the same functional from different problems. The recurrence is real, but it is not luck, and naming it as luck is exactly the loose move to avoid. The form is forced.

The forcing is Shannon's uniqueness theorem (1948; tightened by Khinchin and by Faddeev). Suppose we want a single scalar $H(p_1,\dots,p_n)$ that measures the uncertainty — equivalently, the multiplicity of unrealized-but-compatible alternatives — in a distribution over $n$ outcomes, and we require only three things of it:

1. Continuity. $H$ varies continuously in the $p_i$; an infinitesimal change in the odds is an infinitesimal change in the uncertainty.
2. Monotonicity. For $n$ equally likely outcomes, $H$ increases with $n$; more equiprobable alternatives means more uncertainty.
3. Composition (the grouping axiom). If a choice is broken into a sequence of sub-choices, the total $H$ is the appropriately weighted sum of the parts. Resolving uncertainty in stages must total to resolving it at once.

Any function meeting these three conditions is necessarily $-K\sum_i p_i \log p_i$, up to the constant $K$ that sets the units. The third axiom carries the weight: it says the measure must behave correctly under the composition of alternatives, and the logarithm is the only thing that turns the multiplication of independent possibility-spaces into the addition of their measures.

So the entropies coincide because they are the same operation applied to different specifications of "the alternatives." The mathematical identity is a structural identity. Two physicists working on unrelated problems found the same formula because they were measuring the same thing — the multiplicity of the unrealized-but-compatible relative to what is currently specified.

1.2 Boltzmann entropy — counting microscopic possibility

$$ S = k_{B} \ln W. $$

$W$ (often $\Omega$) is the number of microscopic configurations — microstates — compatible with a given macrostate. The macrostate is what a coarse instrument can read: energy, volume, particle number, magnetization. The microstate is the full specification the instrument cannot resolve. Boltzmann's constant $k_B$ is only a unit conversion, fixing entropy in joules per kelvin so the formula meets thermodynamics.

Read through the organizing claim: the macrostate is the instantiated specification; $W$ is the literal count of the possibilities it leaves open; $\ln W$ is their measure. Boltzmann entropy is the size of the possibility-space your resolution cannot collapse. It is the special case of the general form in which every compatible microstate is equally likely — substitute $p_i = 1/W$ into $-k_B\sum p_i \ln p_i$ and $S = k_B \ln W$ falls out.

1.3 Gibbs entropy — the same count, unequal odds

$$ S = -k_{B} \sum_i p_i \ln p_i. $$

When the microstates are not equiprobable, Gibbs's form is the correct one, and it contains Boltzmann's as the uniform case. Gibbs entropy is also where the time story first shows its hand: computed over the full, fine-grained microscopic distribution, it is conserved under the exact dynamics — a consequence of Liouville's theorem, which says Hamiltonian flow preserves phase-space volume. Fine-grained Gibbs entropy does not increase. Hold that fact; §3 is built on it.

1.4 Shannon entropy — possibility measured in bits

$$ H = -\sum_i p_i \log_2 p_i \quad \text{(bits)}. $$

Shannon's problem was communication, not heat, but the object is identical with the units changed. $H$ is the average number of binary distinctions — bits — needed to specify which message a source emitted; it is the minimum achievable cost of describing the source, which is the content of the source-coding theorem. The distribution is the instantiated specification; $H$ is the volume of the unseen, the average number of yes/no questions still standing between you and the actual outcome. Boltzmann counts microstates; Shannon counts messages; the functional is the same because the operation is the same.

1.5 Von Neumann entropy — entropy of a quantum state

$$ S(\rho) = -\operatorname{Tr}(\rho \ln \rho) = -\sum_i \lambda_i \ln \lambda_i, $$

where $\rho$ is the density matrix and $\lambda_i$ are its eigenvalues. The second equality is the whole content: in the basis that diagonalizes $\rho$, von Neumann entropy is Shannon entropy of the spectrum. The quantum object reduces to the classical one once you ask which basis the state is "really" mixed in.

Two facts make it the most revealing member of the family.

First, a pure state — maximal knowledge, $\rho = |\psi\rangle\langle\psi|$, one eigenvalue equal to 1 — has $S = 0$. A mixed state has $S > 0$. The pure→mixed transition is therefore the quantum event in which determinate possibility becomes statistical fact. We will return to this as the microscopic seat of instantiation — and to the discipline required not to overread it.

Second, like Gibbs entropy, von Neumann entropy is invariant under the exact (unitary) evolution. Closed quantum dynamics do not change it. Again the arrow is not in the law.

A consequence worth stating now: for a composite system in a pure entangled state $|\psi\rangle_{AB}$, the whole has $S = 0$ while each part, described by its reduced density matrix, is mixed with $S(\rho_A) = S(\rho_B) > 0$. The entropy of the part is the entanglement entropy. Locality buys mixedness; the global purity is paid for in local ignorance. This is the hinge of the holographic story in §4.

1.6 The family, at a glance

2. The non-ergodic frontier: when the alternatives are not independent

The composition axiom in §1.1 quietly assumes that the parts of a system can be treated as independent — that the multiplicity of the whole is the product of the multiplicities of the parts, so its log is their sum. Additivity is the formal expression of independence. The moment a system's history couples its parts — long-range interactions, long memory, a phase space carved into regions a trajectory cannot leave — that assumption fails, and the entropy that remains correct is a generalized one. Non-additivity is the signature of non-ergodicity at the level of the measure itself. This is the frontier your framework lives on, so it deserves to be stated precisely rather than gestured at.

Rényi entropy relaxes the composition axiom into a one-parameter family, $H_\alpha = \frac{1}{1-\alpha}\log\sum_i p_i^\alpha$, recovering Shannon as $\alpha\to 1$ and sweeping out a spectrum of weightings: $\alpha\to 0$ counts the support (Hartley/max-entropy), $\alpha\to\infty$ tracks the single most probable outcome (min-entropy).

Tsallis entropy, $S_q = k_B(1-\sum_i p_i^q)/(q-1)$, recovers Boltzmann–Gibbs–Shannon as $q\to 1$ but is pseudo-additive: for independent subsystems $A$ and $B$ (with $k_B = 1$),

$$ S_q(A\cup B) = S_q(A) + S_q(B) + (1-q),S_q(A),S_q(B). $$

The coupling term $(1-q),S_q(A)S_q(B)$ is the whole content. When $q\neq 1$ the entropy of a composite is not the sum of the entropies of its parts even when the parts are statistically independent — the measure itself encodes a refusal of the parts to add up. That is precisely what one wants for a non-ergodic substrate: a system that does not explore its full possibility-space but is trapped by what it has instantiated cannot be measured by an additive functional, because additivity is the assumption that the trajectory is irrelevant. Tsallis (and the broader family of non-additive entropies) is the formal admission that the path matters.

Two further members move the measure from a distribution to a trajectory, and in doing so build the bridge to time — this is the content the Gemini dialog offered as the "entropy bagel" but did not deliver:

Kolmogorov–Sinai (metric) entropy, $h_{KS}$, is the rate at which a dynamical system manufactures new distinguishable possibility. Partition the state space, watch the symbolic itinerary a trajectory produces, and take the asymptotic Shannon entropy per step, maximized over partitions. A system with $h_{KS} = 0$ is predictable; a positive value is the rate at which the past stops determining the future — the per-step generation of fresh alternatives.

Topological entropy, $h_{top}$, is the same idea stripped of any probability measure: the exponential growth rate of orbits a finite resolution can tell apart. The two are linked by the variational principle, $h_{top} = \sup_\mu h_{KS}(\mu)$ over invariant measures $\mu$. The canonical positive-entropy example is a hyperbolic automorphism of the torus — the "bagel," Arnol'd's cat map — which stretches and folds the doughnut so that nearby trajectories separate exponentially and the system generates possibility along every orbit.

The payoff is conceptual: the static entropies of §1 measure the possibility left open by a specification now; the dynamical entropies measure the rate at which a trajectory opens fresh possibility as it runs. Entropy stops being a snapshot of the unseen and becomes the rate of becoming. That is the natural seam to time.

3. Time: two faces, one root

3.1 The arrow is not in the laws

The microscopic equations are time-reversal symmetric. Hamilton's equations run as cleanly backward as forward; Schrödinger evolution is unitary; even granting the standard model's weak $T$-violation, $CPT$ holds and the dynamics relevant to thermodynamics are reversible to extraordinary precision. And as §1.3 and §1.5 recorded, the fine-grained entropies — Gibbs's, von Neumann's — are conserved under these dynamics. If entropy increase came from the equations of motion, fine-grained entropy would rise. It does not. The arrow is not written into the law.

3.2 Where the arrow comes from

It comes from two things together: a boundary condition and an act of coarse-graining.

The boundary condition is the Past Hypothesis — that the universe began in a macrostate of extraordinarily low entropy, a state of vanishingly small $W$ and therefore enormous improbability among all the states it might have had. (Penrose locates this in the early universe's gravitational smoothness and estimates the required fine-tuning at one part in $10^{10^{123}}$ — a number whose only role here is to register how special the initial condition was.) The arrow points away from this special past because there is only one direction it can point: toward states compatible with more microstates, because those states are overwhelmingly more numerous.

The coarse-graining is the second half. We track macrostates, not microstates, because macrostates are what is stable and resolvable. The actual microstate evolves deterministically and reversibly, but it wanders out of the tiny initial macrostate cell into the astronomically larger cells around it — not because it is "seeking disorder" but because nearly all the volume it can reach lies in the larger cells. Boltzmann's H-theorem makes this rigorous, but only by importing a time-asymmetric assumption — the Stosszahlansatz, molecular chaos, which treats colliding particles as uncorrelated before impact and not after. The reversibility objection (Loschmidt) and the recurrence objection (Zermelo, via Poincaré) are answered not by denying reversibility or recurrence but by the atypicality of the initial condition and by recurrence times so long they exceed the age of the universe many times over. The asymmetry is in the boundary, not the bulk.

3.3 One root, several arrows — and the paradox dissolved

The single low-entropy past is the common source of arrows that look independent:

• the thermodynamic arrow, entropy increasing;
• the records arrow, that we hold traces of the past and none of the future — because a record is a correlation freshly formed, and forming a correlation requires a low-entropy reservoir to absorb the entropy cost, and such reservoirs exist only because the universe began far from equilibrium;
• the causal / fork asymmetry — that correlations trace to common causes in the past rather than the future (Reichenbach) — which is again the special past expressed as uncorrelated initial conditions;
• the felt arrow, the openness of the future against the fixity of the past.

This is where the Gemini synthesis left a hole worth filling, because there is an apparent contradiction the loose version cannot see, let alone answer. How can the future be both the high-entropy direction and the open direction? If entropy rises toward the future, and entropy is multiplicity, isn't the future the most "determined" — the most filled-in — and the past the open one?

It dissolves on inspection. "Open" and "high-entropy" are the same fact stated twice, not two facts in tension. A high-entropy macrostate is one compatible with many microstates: many possibilities un-resolved, which is exactly what "we cannot yet read which one will be realized" means. The future is open because it is high-multiplicity. And the past is "fixed" not because it carried low information-as-entropy but because records of it exist and constrain inference — and its low thermodynamic entropy is precisely the condition that made those records formable in the first place. The same low-entropy past that lets entropy rise is the past that lets the rising be remembered. There is one asymmetry wearing four costumes.

3.4 Time as instantiation (interpretation)

Here is the reading our framework adds, and it is marked as a reading because the physics permits it without forcing it. Time is the axis along which the possible is converted into the instantiated. Each moment that becomes actual deposits structure that constrains the trajectories still available; the process is non-ergodic in the strict sense that the system does not sample its phase space but is trapped by what it has already laid down. The dynamical entropies of §2 give this teeth — $h_{KS}$ is the rate at which fresh possibility is generated along the actual trajectory, and the arrow of time is the direction of accumulation, the order in which "what remains possible after a state change" gets measured.

What is the warranted claim, and what is not? The warranted claim is that this picture reorganizes the standard physics — the records arrow, the causal arrow, and the thermodynamic arrow — under a single concept, instantiation, and that it contradicts none of it. It earns its keep by unification. The claim it does not license is that "instantiation" is an extra physical ingredient beyond the Past Hypothesis and coarse-graining; a Boltzmannian can describe every one of these facts without reifying becoming. The interpretation is a lens that brings the standard account into a single focus. It is not a new mechanism, and it should not be sold as one.

4. Space: the carrier, and the area law

Space enters the entropy story in two registers, and they are very different in how settled they are.

4.1 Space as the carrier of microstates (established)

In ordinary statistical mechanics, space is where the microstates live: phase space is the product of configuration space (positions) and momentum space, and $W$ is a volume in it. Entropy measures a region of phase space, and phase space is built on physical space. Nothing exotic — but it is the baseline, and it is the sense in which space and entropy are related for nearly all of physics.

4.2 The area law (established)

The genuinely strange result comes from gravity. A black hole carries entropy, and that entropy is proportional not to its volume but to the area of its horizon:

$$ S_{BH} = \frac{k_B,c^3,A}{4,G,\hbar} = \frac{k_B,A}{4,\ell_P^2}, $$

with $\ell_P$ the Planck length. The information content of the most entropic object that fits in a region scales with the region's boundary, not its interior. The companion result is the Bekenstein bound: any system of radius $R$ and energy $E$ has

$$ S \le \frac{2\pi k_B R E}{\hbar c}. $$

Both are robust consequences of semiclassical gravity and black-hole thermodynamics (the four laws of black-hole mechanics, Hawking radiation). They are not speculation. They say something flatly counterintuitive about space: the capacity of a volume to hold distinguishable possibility is set at its surface.

4.3 Holography, and the honest caveat (established as a duality; conjectural beyond it)

The area law motivates the holographic principle ('t Hooft, Susskind): that the maximal information in a region is fixed by its bounding area in Planck units, suggesting a lower-dimensional description on the boundary is complete. This has exactly one fully worked, heavily checked realization — AdS/CFT (Maldacena), in which a gravitational "bulk" in anti–de Sitter space is dual to a conformal field theory on its boundary. Within it, the Ryu–Takayanagi formula equates the entanglement entropy of a boundary region to the area of a minimal surface in the bulk, and a line of work (Van Raamsdonk and others) shows that weaving or cutting entanglement assembles or severs the bulk geometry — spacetime, in this setting, built from entanglement.

Now the caveat, because this is precisely where loose synthesis overreaches. AdS/CFT is established as a duality in its setting. That our universe — which is not anti–de Sitter — is holographic in this strong sense is an open conjecture. That cognition or consciousness is "AdS/CFT in finite substrate" is not physics at all; it is analogy, and it should be named as analogy. The area law and AdS/CFT are real and astonishing. Their extension to cosmology, and especially to mind, is unproven, possibly fruitful, and currently a bet rather than a result. The disciplined statement is the strong one and the modest one held together: the boundary encodes the bulk where we can check it; whether the cosmos and the mind obey the same rule is a question, not an answer.

4.4 The boundary reading (interpretation)

Where geometry is informational, the framework's reading is clean and, kept to its warrant, defensible: space too is a render of an underlying distribution, and the only thing that crosses between inside and outside is what the boundary transmits — the Markov-blanket picture, no hay afuera, hay cooperación. The area law is the most literal physical instance of "the render lives on the boundary": the interior's capacity for possibility is written on its surface. This is offered as the lens, not as a derivation from holography — the physics of §4.2 is what bites; §4.4 is what it looks like through the framework.

5. The nexus where it bites: information is physical

Time and space meet entropy at the point where information turns out to cost energy. This is the part of the report that bites hardest, because it is measured in a laboratory and is true regardless of which interpretation in §3 and §4 one prefers.

5.1 Landauer's principle (established, measured)

Erasing one bit of information in a thermal environment at temperature $T$ dissipates at least

$$ k_B T \ln 2 $$

of heat, and raises the environment's entropy by at least $k_B \ln 2$. Logical irreversibility — the merging of two distinct states into one, which is what erasure is — forces thermodynamic irreversibility. This is the hard bridge between Shannon's bit and Boltzmann's joule, and it has been confirmed experimentally (Bérut and colleagues, and others). Information is not an abstraction laid over physics; a bit is a physical entropy carrier, and destroying it pays a thermodynamic price that no cleverness avoids. In the framework's vocabulary: a place where the world bites, frequency-independent, indifferent to interpretation.

5.2 Maxwell's demon, Szilard, and the closing of the books (established)

The Szilard engine — a one-molecule heat engine — can extract $k_B T \ln 2$ of work per bit of which-side information about the molecule. Maxwell's demon, sorting fast molecules from slow, seems on this basis to violate the second law. The resolution (Szilard 1929; completed by Bennett 1982) is that the demon's memory is finite: to keep operating it must eventually erase what it has learned, and erasure costs exactly the Landauer minimum, paying back every joule the sorting seemed to gain. The second law's bookkeeping closes only when information is counted as a physical entropy term. The demon is not an exception to thermodynamics; it is the proof that information belongs inside it.

5.3 The observer and the second law — your question, answered

You posed it directly: in a closed system entropy increases, yet a living observer, by measuring and choosing, can reduce entropy locally — the space of possibilities widens with each instantiation, but the observer narrows it. How does the framework hold both?

It holds them as one statement: the observer reduces the entropy of a subsystem only by exporting at least as much entropy to the environment. There is no violation, because the observer is not outside the system — the observer is the site at which possibility-space is locally collapsed to one realized outcome at the cost of a larger global increase. This is the same architecture seen three ways. In quantum terms, the local state goes pure→mixed while the joint state's information is conserved (decoherence). In thermodynamic terms, a record is a local entropy decrease purchased with a larger export (Landauer). In the framework's terms, the boundary at which an entropy gradient becomes a historical record is exactly where the observer sits.

So life and cognition are not affronts to the second law; they are its open-system corollary. The substrate runs toward uniformity asymptotically — but the layers built upon it differentiate without bound for as long as a free-energy gradient is available to pay the integration cost. Randomness supplies new possibility; the gradient makes integrating it costly; the observer is the apparatus that pays; the non-ergodic record is what the payment buys. The second law is not the enemy of complexity. It is the engine of complexity in systems far from equilibrium — the precise mechanism by which low-entropy gradients are cashed in for high-dimensional structure. The observer is not separate from the entropy story. The observer is the entropy story, at the level where local structure is being built.

5.4 Pure → mixed as the seat of the record (established mechanism; bounded interpretation)

The von Neumann transition of §1.5 is the microscopic locus of all this. When a system decoheres against an environment, its local (reduced) description goes pure→mixed, and a definite, redundantly-recorded classical fact can form. This is the quantum face of "an entropy gradient becoming a historical record": reversible quantum possibility becoming, locally and for all practical purposes, an irreversible classical actuality. As a description of how classicality and records emerge, decoherence is established and quantitative.

The discipline required here is exactly the discipline the method demands. Decoherence explains the appearance of collapse and the emergence of the classical world; it does not, by itself, settle the measurement problem, whose interpretations (Everett, objective-collapse, epistemic) remain open. And it is not a theory of consciousness. The pure→mixed transition is the right technical place to locate the creation of a record; it is not a proof of any metaphysics of mind, and the temptation to make it one is the temptation to mistake a vivid render for a bite.

6. Is entropy subjective or objective? — the strongest objection, faced

The whole report rests on entropy as "the measure on the possible, from the standpoint of the instantiated," and that phrasing invites the sharpest available objection: if entropy is taken from a standpoint, isn't it merely subjective — about our ignorance rather than the world? The objection has to be met, not waved past, because the answer is the keystone.

On one side, Jaynes and the maximum-entropy tradition: entropy is a property of a probability distribution, hence of a description given chosen macrovariables; coarse-graining is observer-relative; entropy is the maximal uncertainty consistent with the constraints one elects to track. This fits "from the standpoint of the instantiated" and the reading of entropy as the friction a finite observer feels processing a system beyond its representational capacity.

On the other side, the objectivists — Lebowitz, Goldstein, Albert in the Boltzmannian lineage — insist that entropy is a fact about the world: the macrovariables are not arbitrary but physically privileged (they are the robust, additive, slowly-varying quantities the dynamics actually respects), and the low-entropy past is a fact about the universe, not about anyone's knowledge. The second law's predictions are objective and do not wait on an observer.

The reconciliation is the warranted claim, and it is the one our own render/bite anchor predicts. Entropy is relative to a partition — there is no entropy without a choice of which alternatives count as "the same" — but the relevant partition is not arbitrary. It is fixed by what is dynamically stable and physically resolvable; thermodynamic macrovariables are not a free choice but the ones the world makes legible. So entropy is observer-relative in its definition and objective in its consequences: given the natural partition, the increase, the Landauer joule, the area law are frequency-independent facts that no standpoint can wish away. The "standpoint of the instantiated" is the partition — perspectival, the render. The gradient that the partition then measures is not — it bites. The subjective/objective dichotomy is false because the answer is both, at different layers, and the framework's contribution is to say which layer is which.

Coda: one architecture, claimed to its warrant

Strip it to the bone. Reality has the structure we find because of one coupled operation: a memoryless point at which new possibility enters, and a non-ergodic accumulation that integrates it into a record. Entropy is the metric on that operation — the measure of the possibility-space the new bits are drawn from, and of the residue they leave behind once fixed. Boltzmann measures it over microstates, Shannon over messages, von Neumann over quantum spectra, Tsallis over systems whose history refuses to let the parts add up, Kolmogorov–Sinai over the rate a trajectory generates fresh alternatives. They are one functional because, by Shannon's theorem, the operation forces the form. Time is the order of the accumulation and the direction of its arrow; its asymmetry is one low-entropy past wearing the costumes of thermodynamics, records, causation, and felt becoming. Space is the carrier of the microstates and, at the limit, the boundary on which a volume's whole capacity for possibility is written. Information is physical, and Landauer is where the books close.

What remains genuinely open should be left open, because saying so is part of the method. There is no settled formal reformulation of probability as "what remains possible after a state change" for non-ergodic systems — the gap that Kauffman, Peters, and Deutsch circle without closing. Whether our non-AdS cosmos is holographic in the strong sense is unproven. The measurement problem's interpretation is unresolved. And "time as instantiation" is a lens that unifies the standard physics, not a mechanism that supplements it.

Held to its warrant, the claim that opened the report is the thing the whole report has earned: entropy is the measure on the space of the possible, taken from the standpoint of what is instantiated — and time, space, and information are the three faces of how the possible is converted, irreversibly, into what has been.

Eduardo Bergel and Claude Opus 4.8

The Symbiont

t333t.com Research