ATOM 10
Shannon vs Thermodynamic Entropy
Block 3 · Information, Stochasticity & Frontier Formalisms
one word, two quantities · Landauer · Maxwell's demon

The most important untangling in the field

"Entropy increases with age." But which entropy? This single word names two genuinely different quantities — and silently swapping one for the other is the most common, most consequential error in physics-of-aging writing. Here is how to keep them straight.

THERMODYNAMIC joules / kelvin heat · energy SHANNON bits uncertainty Landauer same math, different units — not automatically the same thing
two boxes, one fragile bridge
01 / 06
Level: for physicists — the rigorous distinction and its bridge Level: for gerontologists — the clarification that prevents bad claims Level: for everyone — the deepest "gotcha" in the field

The same equation, twice

There is a famous near-coincidence at the root of all the confusion: the formula for thermodynamic entropy and the formula for information entropy look almost identical. That resemblance is real — and it has misled people for seventy years.

One "entropy" lives in physics: it's about heat and energy spreading out, measured in physical units. The other lives in information theory: it's about uncertainty and surprise in a message, measured in bits. They share a name and a near-identical equation — but a coin flip's "entropy" and a hot cup of coffee's "entropy" are not obviously the same kind of thing. The whole field gets into trouble when people forget that.

This atom is the field's essential hygiene. Thermodynamic entropy (Clausius/Boltzmann/Gibbs) quantifies thermal-energy dispersal, in J/K. Shannon entropy quantifies uncertainty in a probability distribution, in bits. Their formulas share a form, which has tempted decades of writers to treat "rising entropy with age" as one claim. When a paper invokes entropy, you must always ask which one — and whether the bridge between them has actually been built or merely assumed.

The structural identity is exact in form: Gibbs S = −kBΣpi ln pi vs Shannon H = −Σpi log pi. Shannon explicitly took the cue from statistical-mechanical entropy. But identical functional form over a probability distribution does not entail identical physical content: one carries units of J/K and refers to a thermal phase-space measure; the other is dimensionless (bits) and refers to an information source. Whether they are the same quantity, related quantities, or merely analogous is the precise question this atom resolves — and it is decisive for every entropy-based claim in atoms 7 and 9.

02 / 06

Two quantities, side by side

Lay them out plainly. They answer different questions, in different units, about different things — even though the algebra rhymes.

Thermodynamic

  • About heat & energy dispersal
  • Units: joules per kelvin (J/K)
  • Clausius, Boltzmann, Gibbs
  • Tied to temperature & physical states
  • Governed by the second law

Shannon (informational)

  • About uncertainty in a message
  • Units: bits (dimensionless)
  • Shannon, 1948
  • Tied to a probability distribution
  • No thermometer required

The right-hand "entropy" doesn't need a temperature at all — you can compute the Shannon entropy of a deck of cards or a string of text without any heat involved. The left-hand one is inseparable from heat and energy. So when someone measures the "entropy" of your DNA methylation pattern (Atom 9), they're almost always computing the right-hand, informational kind — even if they then talk as if it were the physical kind.

This is exactly why the proxies in atom 7 (methylation entropy, ECG-interval irregularity) and the whole information theory of atom 9 are informational entropy — computed from probability distributions over states, in bits, with no calorimetry. They are scientifically useful and predictive. The error is the silent slide from "this informational entropy rises with age" to "therefore thermodynamic entropy / the second law is driving aging." That inference requires a bridge, not a shared word.

Operationally: thermodynamic S requires a physical phase-space measure and a temperature; Shannon H requires only a distribution over distinguishable outcomes. The methylation-entropy and dysregulation clocks compute H over epigenetic/physiological state distributions — informational, dimensionless, calorimeter-free. They have genuine predictive validity, but they do not, on their own, license thermodynamic conclusions. Conflating them is a units error dressed as a physical insight.

03 / 06

The bridge: Maxwell's demon and Landauer

And yet — the two are not unrelated. There is a real, hard-won bridge between them, built over a century by way of a famous thought experiment and a deep principle about erasing information.

Imagine a tiny "demon" sorting fast and slow molecules to create useful heat from nothing — seemingly breaking the second law. The resolution, after a century of debate: the demon has to store and then erase information to keep sorting, and erasing information has an unavoidable physical cost in heat. That is the bridge: at the moment you erase a bit of information, the two entropies finally touch.

The link runs through Maxwell's demon and Landauer's principle. Szilard and later Landauer showed the demon cannot violate the second law because erasing one bit of information dissipates at least kBT ln2 of heat to the environment. This is the rigorous point of contact: information processing has a thermodynamic cost. It connects the informational and thermal worlds — but only for the specific act of logically irreversible erasure.

LANDAUER'S BOUNDQerase ≥ kBT ln2 per bit logically irreversible erasure of one bit dissipates at least k₋T·ln2 of heat

Szilard's engine and Landauer's principle exorcise Maxwell's demon: the demon's measurement–feedback can extract up to kBT ln2 per bit, but resetting (erasing) its memory costs at least that much, preserving ΔSuniv ≥ 0. This is the genuine bridge — but it bridges via logically irreversible operations on physical memory, not via a blanket identity. The conversion factor kB simply carries bits to J/K. The bridge is real, narrow, and conditional.

04 / 06

Why the bridge is narrower than people think

It's tempting to use Landauer to declare information entropy and thermodynamic entropy simply equal. That overreaches — and a careful course flags exactly where the bridge does and doesn't hold.

The tempting shortcut: "information entropy and physical entropy are the same, Landauer proved it." Not quite. Landauer ties them together only for a specific operation (erasing information), and even then the real physical cost depends on the actual hardware, not just the logic. The two remain distinct quantities that touch at one well-defined point.

So the honest picture is: two different things, with one narrow, real bridge between them. Useful — but you can't walk across it for free, and you can't use it to declare them identical everywhere.

The careful caveat: the naive identification ΔS = −kBΔH is, in the words of the physics-of-information literature, "very misleading," because actual dissipation depends on the physical implementation (the Hamiltonian), not on the logical operation alone. Some authors even contest the unconditional validity of Landauer's principle. So information and thermodynamic entropy are related, at a specific point, under specific conditions — not interchangeable. For aging research this is the crucial license check.

The identity ΔS = −kBΔH is misleading because it suggests dissipation depends only on the logical map; in reality it depends on the physical Hamiltonian realizing the gate (Landauer's bound is a minimum, approached only by specific quasi-static, irreversible phase-space expansions). The unconditional generality of Landauer's principle has itself been challenged (e.g. Norton). Net: the bridge is a conditional inequality for logically irreversible erasure, not a universal equation of state. Treat informational and thermodynamic entropy as distinct quantities linked by a narrow, implementation-dependent bound.

05 / 06

What this means for a theory of aging

Now the payoff. This distinction is not pedantry — it directly determines which aging claims are well-posed and which are quietly broken.

When you read "aging is rising entropy," ask one question: which entropy, and was the bridge actually built? If a study measures the information-disorder of cells and predicts age — great, that's real and useful. But if it then claims this is the second law of thermodynamics dooming us, that's a leap that needs the bridge. Often the bridge is just assumed. Now you can spot it.

Concretely: the information theory of aging (atom 9) is built on Shannon entropy of epigenetic states — legitimate and predictive as informational entropy. The thermodynamic schools (atoms 7–8) deal in thermal σ. These are different claims. They may connect — restoring epigenetic information (reprogramming) presumably has a Landauer-type thermodynamic cost — but that connection must be demonstrated, not assumed from the shared word "entropy." This is the single most important consistency check when evaluating any gerophysical claim.

The distinction sharpens every claim in the course. Methylation-entropy clocks and the ITOA (atom 9) quantify Shannon H; entropy-production and dissipation accounts (atoms 7–8) quantify thermodynamic σ. A well-posed gerophysical theory must state which it uses and, if it crosses between them, supply the Landauer-type bridge explicitly (e.g. the thermodynamic cost of epigenetic restoration). Claims that aging "increases entropy" without specifying which, or that silently equate rising H with rising σ, are not yet falsifiable physics. This is the litmus test revisited in atom 20.

06 / 06

What to take from this atom

"Entropy" names two distinct quantities: thermodynamic (heat dispersal, J/K, second law) and Shannon (uncertainty, bits, information). Their formulas rhyme, but they are not automatically the same. The one real bridge — Landauer's principle (erasing a bit costs ≥ kBT ln2) — is narrow, conditional, and implementation-dependent.

The rule for the rest of the course, and for reading any paper: when you meet "entropy," ask which one, and whether a bridge was built or assumed. Most aging "entropy" measures are informational; most thermodynamic theories are thermal; equating them silently is the field's signature error. You can now catch it.

Next (atom 11): stochastic thermodynamics — Jarzynski, Crooks, Seifert — exact thermodynamic statements at the single-molecule, fluctuating scale.

Next (atom 11): how modern physics makes thermodynamics exact even for tiny, noisy molecular systems — the rigorous floor under molecular aging.

Up next: the cutting edge — how physics now handles the constant random jiggling of single molecules, where life actually happens.

Check your understanding

3 questions

Answers are scored instantly, with an explanation. Nothing is submitted anywhere — the quiz lives only in your browser.

01What is the key difference between thermodynamic entropy and Shannon entropy?
They share a near-identical formula but answer different questions in different units. Thermodynamic entropy (J/K) concerns thermal-energy dispersal and the second law; Shannon entropy (bits) concerns uncertainty in a distribution and needs no thermometer. Same algebra, different physical content.
02What does Landauer's principle establish, and why does it matter here?
Landauer's principle: logically irreversible erasure of one bit costs at least k_B·T·ln2 of dissipated heat. This is the genuine point of contact between the two entropies (and what exorcises Maxwell's demon) — but it's a conditional, implementation-dependent bound, not a blanket identity.
03A study measures rising "entropy" of DNA methylation patterns with age and concludes "the second law of thermodynamics drives aging." What's the problem?
Methylation entropy is computed from a probability distribution over states — informational entropy, in effect. It can be real and predictive, but concluding the thermodynamic second law is the driver requires explicitly building the bridge (e.g. the Landauer cost), not relying on the shared word "entropy." That silent slide is the field's signature error.

Ask about this topicoptional · your key

Want to dig deeper or ask in your own words? Connect your own API key and you can ask questions about this atom. The key is for you alone: the course works fully without it, through the slides and quiz above.

Your key is stored only in this browser tab and goes straight to the provider. This site never saves it or sends it to any server of its own.