Grouping and Maxwell’s Demon
Neil D. Lawrence
2026-11-03
FW26, William Gates Building
Define This Week
Questions; MacKay’s balance; die grouping \(H=H(w)+\sum_i w_i H(p_i)\)
Gaussian vs Maxwell: variable, grouping, and measure \(m\)
The demon: compartments and two ledgers
Worksheet 3: where can the entropy sit?
This Session
Do not open Szilard or Landauer in this slot.
After This Lecture
Worksheet 2 due today, before the quiz
Worksheet 3 released (due 17 November, with Quiz 3)
Quiz 2
Paper; no notes, no network, no LLMs
Opening block (about fifteen minutes as delivered)
Worksheet 2 in as you sit
As Taught
Harry’s speed question first. Compartments last, as Worksheet 3.
Grid and concentric rings: why speed peaks away from zero
Variable, grouping, measure: Jaynes’ \(m\) is only the last
Demon: three (maybe four) compartments; number ledger versus energy ledger
After the break, the shell figure and the compartment schematic are what this session is for. MacKay strategy stays deferred. Stop at the fork; Landauer waits.
Questions, a Balance, and a Die
Composition wrapper only. Through-line: the game is the idealised yes/no world; the pans and then the cube are where mechanism spoils it. Stop MacKay after two-versus-three readings if the clock is tight. Protect three-faces-left as the known-ball echo, then the \(1\) –\(3\) / \(4\) –\(6\) cut versus odd/even as different groupings, then the extensive/intensive sentence. Do not include the Gaussian–Maxwell measure files here — lectures that want both include this file, then that one.
Bits as Balanced Questions
A fair yes/no answer distinguishes two equally likely cases — one bit .
\(N\) equally likely objects \(\Rightarrow\) worst case \(\lceil\log_2 N\rceil\) bits
Twenty questions reach about a million hypotheses (\(2^{20}\) )
Unequal priors: best questions still split probability as evenly as possible
The same idea, with a different instrument, comes next — MacKay’s balance.
MacKay’s Balance and Balls
State the puzzle first. Do not yet say that the instrument is ternary. Keep the mechanism (what we put on the pans) on a different slide from the maths (how many bits that use can return).
Twelve Balls, One Odd
MacKay’s weighing problem, Exercise 4.1 (MacKay, 2003) : twelve balls; exactly one is odd; it may be heavier or lighter.
A two-pan balance and twelve identical balls.
Instrument: a two-pan balance
Task: identify the odd ball and whether it is heavy or light
State the puzzle. Do not list the readings of the balance, and do not walk through which balls go on which pan.
How Should We Proceed?
We have a finite list of hypotheses and a physical instrument. How should we think?
What, exactly, do we need to name?
What would count as a useful use of the balance?
When would we know we were done?
Leave this live. Do not say left, right, or level, and do not say three. After they have named the hypotheses, try the two-outcome story next — as if this were the information game again.
Let’s Try Two Outcomes
Treat the balance as the information game: left heavy or right heavy only.
Two readings \(\Rightarrow\) at most \(\log_2 2\) bits per use
Mechanism next: what we actually put on the pans
Maths stays on its own slides — the two accounts can disagree
Say “log two bits” — one bit if the two readings are equally likely. Do not yet mention a third reading. Board 6 vs 6 as the move people reach for from twenty questions.
Maths: Two Readings
Maths. Start from \(\log_2 24\) . Each use returns at most \(\log_2 2\) .
\[
\log_2 24 \approx 4.58,\qquad 2^4 = 16 < 24 \le 32 = 2^5
\]
Four uses cannot name 24 hypotheses
Five uses are not ruled out, if each use really returns one bit
That “if” is the mechanism, not entropy /information maths
Mechanism: 6 vs 6
Mechanism. Put six against six.
The pans will tip — the odd ball is on one side
Follow the down pan, as if the odd ball were heavier
Six balls remain under that story
Ask what we actually learned. Under “it is heavier,” 12 balls become 6, one bit toward which ball. We do not know the direction. The down pan is six maybe-heavies; the up pan is six maybe-lights. All twelve balls are still in play. Maths next.
Maths: What 6 vs 6 Bought
Maths. Score the result of the action on the 24 hypotheses.
24 \(\to\) 12 remaining hypotheses: exactly \(\log_2 2\) bits
Zero balls eliminated — each ball is still possibly odd
The bit went into a coupling of group and direction
Mechanism: 3 vs 3
Mechanism. Keep going as a binary search of the balls: three against three.
Under the heavier-side story, three balls remain
Same move, smaller pile
We are still not asking about light versus heavy
Are there only two possibilities with 3 vs 3?
Mechanism: Three Left
Mechanism. Three balls remain. We cannot put one and a half on a pan.
2 vs 2 needs a fourth ball
Borrow one of known normal mass
Then 1 vs 1 on whatever the 2 vs 2 leaves
Make them say the known ball out loud. Ask: how many bits do we get from 2 vs 2 now? Do not calculate before they see that the split of three remaining balls cannot be even.
Aha: It Balances Three Ways
The pans also sit level . Left heavy, right heavy, or balance.
The instrument was never a yes/no question
That was a mechanism choice that threw away a symbol
Let them say the three readings. Then go straight to the new budget. Do not start the 4 vs 4 tree.
Maths: Three Readings
Maths. Each use now returns at most \(\log_2 3\) .
\[
\log_2 3 \approx 1.58,\qquad 3^2 = 9 < 24 \le 27 = 3^3
\]
Two uses cannot name 24 hypotheses
Three uses can, if each use really returns \(\log_2 3\)
Same “if” as before: mechanism must realise the split
Mechanism Must Collect \(\log_2 3\)
Mechanism. To earn \(\log_2 3\) , all three readings must be possible and near equal.
First weighing: 4 vs 4, not 6 vs 6 — so level can happen
A known ball, or emptying a reading, is again a tax
Optimal play keeps the three readings as near equiprobable as possible
Doable in Three — Strategy Deferred
The counting argument shows three weighings must be enough in principle.
We do not give the sequence of groupings today
Info gained and mechanism are separate ledgers — score both
Same moral as twenty questions: split the prior evenly, now three ways
Stop here. Invite them to try Exercise 4.1 before reading MacKay’s solution figure. Optional aside: 1–6 vs 7–12 can never balance, so it yields only one bit and is already suboptimal once we know there are three readings .
Coarse-Graining a Die
A fair die has \(H(X)=\log_2 6\approx 2.58\) bits.
Stage the question: does every yes/no cost one bit?
First cut: odd or even?
Second cut: is it a six?
Same total either way — different prices for the first answer
Board both partitions. Odd/even is balanced so exactly 1 bit. ``Is it six?’’ is not one bit — students often assume every binary question buys a bit. That is the MacKay moral applied to grouping.
Odd or Even: Exactly One Bit
Groups \(\{1,3,5\}\) and \(\{2,4,6\}\) are equiprobable.
\[
H(G_{\mathrm{parity}})=1 \text{ bit},
\qquad
H(X\mid G_{\mathrm{parity}})=\log_2 3 \approx 1.585 \text{ bits}
\]
Group label: fair coin — one bit
Inside either group: three equally likely faces
Total: \(1+\log_2 3=\log_2 6\)
“Is It a Six?”: Not One Bit
A yes/no question is one bit only when the two answers are equiprobable.
\[
w=\Bigl(\tfrac16,\tfrac56\Bigr),
\qquad
H(G_{\mathrm{six}})=h_2\Bigl(\tfrac16\Bigr)\approx 0.65 \text{ bits}
\]
Yes (\(X=6\) ): no residual uncertainty
No: five faces left, \(H=\log_2 5\)
Remaining: \(\tfrac56\log_2 5\approx 1.93\) bits
Total again \(\approx 0.65+1.93=2.58=\log_2 6\)
Same Total, Different First Cut
However you slice the questions, you pay \(\log_2 6\) in total.
Odd/even first: pay \(1\) , then \(\log_2 3\)
Six-or-not first: pay \(\approx 0.65\) , then \(\approx 1.93\)
Coarse-graining keeps only the group term and throws away the rest
Partitions That Do Not Overlap
Coarse-grain into groups that do not leak into each other.
Face \(d\) ; partition index \(i\)
\(p(d\mid i)\) is not a general conditional (once \(i\) is known, the support is restricted)
Divide so the partitions do not overlap: we do not want things to leak in ways we are not modelling
Here \(d\) is the die face and \(i\) is the partition: low/high or \(1\) –\(2\) versus \(3\) –\(6\) . Outside the active cell the probabilities are zero. Soft mixtures (several \(i\) active for one \(d\) ) come next.
Why Non-Overlap Matters
The log of a sum is a headache. The log of a product is trivial. Non-overlap lets us use the product.
\[
p(d)=\sum_i p(d\mid i)\,p(i)
\]
Product rule and sum rule give the marginal (that much is ordinary)
For each face \(d\) , only one \(i\) is valid
The other terms in the sum are zero, so the log-sum collapses
If \(d\in\{1,2\}\) then only \(i=1\) contributes; if \(d\in\{3,4,5,6\}\) then only \(i=2\) . The partitions are on or off: that is the trick.
Overlapping Components Are a Different Model
A mixture of Gaussians (or a mixture of experts) lets components overlap. Then the trick fails.
Soft mixture: several \(i\) can be active for the same \(d\)
The sum inside the log stays a sum
Coarse-graining as used here is the exclusive case (break the world down, and throw the within-cell uncertainty away)
A mixture of Gaussians or mixture of experts is the overlapping contrast. We do not need EM here. Then back to the die identity.
Grouping Composition
Uncertainty splits into stages: which group, then where inside the group.
\[
H(X)=H(w)+\sum_i w_i\, H(p_i)
\]
\(w_i\) : probability of group \(i\)
\(p_i\) : distribution inside group \(i\)
First term: cost of the coarse label; second: average residual detail
This is the chain rule for a deterministic grouping \(G=g(X)\) : \(H(X)=H(G)+H(X\mid G)\) . Khinchin’s point is only that consistency under staged questions characterises Shannon \(H\) ; no proof today, and no mutual information yet.
Reverse Direction
Learning the face determines the group for free. What is left to find after the group?
\(H(G\mid X)=0\) : the coarse label is redundant given \(X\)
Cost of the coarse question: \(H(G)\)
Still to find after the group: \(H(X\mid G)\)
For exclusive partitions \(H(G\mid X)=0\) (today’s assumption). Soft / overlapping mixtures leave \(H(G\mid X)>0\) ; that derivation waits for week 5, once we treat the leftover term with mutual information.
Why the Total Is Invariant
Entropy is the uncertainty that is consistent no matter how you decompose a question into stages.
Any partition: \(H(w)+\sum_i w_i H(p_i)\) returns the same \(H(X)\)
That consistency is special to the weighted logarithm form
Grouping Is a Choice of Abstraction
Keeping \(\{1,2,3\}\) versus \(\{4,5,6\}\) and refusing the face is the same move as a macrostate.
The group label is what we keep; \(\sum_i w_i H(p_i)\) is what we ignore
Macroscopic variables: extensive \(N\) ,\(V\) ,\(E\) ,\(S\) ; intensive \(T\) ,\(P\)
We cannot follow Laplace’s curves, so we keep a few numbers
Callback only — extensive versus intensive was week 2. Do not re-derive. Do not open MaxEnt, Lagrange, or the demon. The next include, if you want it, is the Gaussian-Maxwell measure arc. Units: the die was bits; those figures are nats. Same \(H\) , different \(\log\) base.
Gaussian, Maxwell–Boltzmann, and the Missing Measure
Composition wrapper only. Children carry the figures. As taught: open from Harry’s speed-peak question and the anthropomorphic bridge, then histogram arithmetic, then the velocity-space shell before the 3D Maxwell pair. Stop before MaxEnt / Lagrange / the die.
Entropy of a Gaussian Velocity Histogram
Equal-width bins of one Cartesian component. Discrete \(H\) is a sum over masses.
Midpoint rule: \(p_i\approx f(x_i)\Delta\) on equal-\(\Delta v\) bins of \(v_x\sim\mathscr{N}\left(0,\sigma^2\right)\)
\(H=-\sum_i p_i \log p_i\) depends only on those masses
Analytic \(h=\tfrac12\log(2\pi e \sigma^2)\) ; the Riemann partition tracks \(h-\log\Delta\)
As taught: open from Harry’s ``why isn’t the speed peak at zero?’’ and the anthropomorphic bridge, then this Riemann picture. Board \(p_i\approx f\Delta\) once. The figure is the piecewise-constant midpoint approximation, not a sample histogram. The two numbers should agree; \(h\) itself is smaller by \(\log\Delta\) (and is negative if \(\Delta\) is large). Nats throughout. The shell-geometry figure follows the 1D fold.
Histograms as Riemann Sums
A histogram replaces \(f(x)\) by a piecewise-constant function: height \(f(x_i)=f_i\) on each bin of width \(\Delta\) .
Bin mass \(p_i = \int_{\text{bin } i} f(x)\,\text{d}x \approx f(x_i)\Delta\) (midpoint rule)
\(\sum_i g(x_i)\Delta \to \int g(x)\,\text{d}x\) as \(\Delta\to 0\)
Used twice: \(\sum_i f_i\Delta \approx 1\) and \(-\sum_i f_i\log f_i\,\Delta \approx h\)
So \(H \approx h - \log\Delta\) : the \(-\log\Delta\) comes from the \(\Delta\) inside the log
Draw the Gaussian, overlay rectangles of width \(\Delta\) and height \(f(x_i)\) , shade one rectangle and label its area \(p_i\) . Then split \(\log(f_i\Delta)\) on the board. Point out that the second sum is not a Riemann sum of anything finite as \(\Delta\to 0\) , which is why \(H\) diverges while \(h\) doesn’t.
Histograms as Riemann Sums
Equal-width midpoint Riemann approximation of a Gaussian velocity component. Discrete \(H\) matches \(h-\log\Delta\) ; \(h\) itself is the remainder after the bin-width term is subtracted.
One Dimension: Velocity Versus Modulus
Same one-dimensional sample: \(v\) peaks at zero; folding to \(|v|\) keeps the mode at zero.
Left: \(v\sim\mathcal{N}(0,\sigma^2)\) — peak at \(0\)
Right: \(s=|v|\) — half-normal; support \([0,\infty)\) , mode still at \(0\)
Folding alone does not push the peak away from zero
Side by side. Point at both modes: both sit at zero. The student’s ``why isn’t the peak at zero?’’ question is about speed in three dimensions, not about taking a modulus. Board \(f(s)=\sqrt{2/(\pi\sigma^2)}\,e^{-s^2/2\sigma^2}\) briefly if needed.
One dimension. Left: a Gaussian velocity component peaks at zero. Right: the same sample folded to \(|v|\) is half-normal — the mode is still at zero.
Velocity Space: Grid and Speed Rings
Same sample in the \((v_x,v_y)\) plane. Equal-\(\Delta v\) bins are a grid; equal-\(\Delta s\) bins are concentric rings.
Left: isotropic cloud in two Cartesian components
Equal-\(\Delta v\) grid: each square has the same area
Concentric rings: shell circumference grows as \(s\) — more cells at larger speed
If a student asks why speed peaks away from zero, this figure is the answer before the 3D histograms. A ring at larger \(s\) holds more grid cells, so even an isotropic Gaussian piles probability away from the origin once you bin by speed. Two dimensions is enough for the geometry; three dimensions replaces rings by spherical shells (\(s^2\) ).
Two Cartesian velocity components. The pale grid is equal-\(\Delta v\) area. The red circles are equal-\(\Delta s\) shells: circumference grows with \(s\) , so a speed histogram weights larger speeds more heavily. Three dimensions replaces rings by spherical shells of area \(4\pi s^2\) .
Three Dimensions: Component Versus Speed
Same 3D sample: one Cartesian component peaks at zero; speed peaks away from zero.
Left: \(v_x\sim\mathcal{N}(0,\sigma^2)\) — peak at \(0\)
Right: \(s=|\mathbf{v}|\) — Maxwell; factor \(s^2\) from shell volume
The nonzero mode is geometry, not folding
Side by side with the 1D pair and the 2D grid-and-rings figure still in mind. Left mode at zero; right mode at \(\sigma\sqrt{2}\) . Point at the \(s^2\) : that is the spherical shell in velocity space — the three-dimensional version of the concentric rings just drawn. Absolute value alone did not move the peak; the Jacobian does.
Three dimensions. Left: one Cartesian component still peaks at zero. Right: speed peaks at \(\sigma\sqrt{2}\) because of the \(s^2\) shell volume — not because of taking a modulus.
Pushed Bins Keep the Same Entropy
The same cells, written on the speed axis: masses unchanged, widths change.
Equal-volume shells in 3D velocity: equal width in \(u=s^3\)
Push the edges to \(s=u^{1/3}\) ; \(\Delta s_i\) now grows then shrinks
\(H\) is a functional of the masses — it does not see the widths
Left panel: probability bars, equal visual width, height \(p_i\) . Right: density \(p_i/\Delta s_i\) with the true speed widths. Same \(H\) on both. This is a 1–1 relabelling of a partition of speed (volume coordinate \(\leftrightarrow\) speed). Folding \(v\mapsto|v|\) , or grouping Cartesian cubes into shells, merges cells and drops \(H\) . That is not this figure.
Equal-volume shells in 3D velocity, written two ways. Left: midpoint masses \(p_i\) . Right: the same partition as a piecewise-constant density on the speed axis. The bins change shape; \(H\) does not.
The Continuous Limit Drops \(\log\Delta\)
As the bins shrink, discrete \(H\) tracks \(-\log\Delta\) to infinity. Differential \(h\) is what remains.
\(H(X^\Delta)\approx h-\log\Delta\)
\(-\log\Delta\to+\infty\) as \(\Delta\to 0\)
Passing to a density throws that term away
The scatter should sit on the line \(h-\log\Delta\) . The horizontal line at \(h\) is the finite remainder. The move to a continuous entropy is exactly ignoring a term that is driving to \(+\infty\) .
The Continuous Limit Drops \(\log \Delta\)
Discrete entropy of an equal-width Gaussian histogram against \(-\log\Delta\) . The points track \(h-\log\Delta\) ; \(h\) is the finite part. The continuous limit discards a term that diverges.
Entropy of the Maxwell Speed Law
The speed \(s=|\mathbf{v}|\) is a different random variable, with a different \(h\) .
\(f(s)=4\pi s^2(2\pi\sigma^2)^{-3/2}e^{-s^2/2\sigma^2}\)
\(h(s)=\log(\sigma\sqrt{2\pi})+\gamma-\tfrac12\) , not \(\tfrac12\log(2\pi e\sigma^2)\)
Why is this number not the Gaussian’s \(h\) ?
Overlay the Maxwell curve on equal-\(\Delta s\) midpoint Riemann rectangles. Read both numbers. Headline the three-way split: different variable, grouping (direction), measure (Jacobian). Leave the pushforward density for the next snippets.
Entropy of the Maxwell Speed Law
Equal-width midpoint Riemann approximation of Maxwell speed. Discrete \(H\) tracks \(h(s)-\log\Delta\) . The Maxwell \(h(s)\) is not the Gaussian \(h(v_x)\) .
Two Discretisations of the Same Speeds
Equal \(\Delta s\) is a different partition from the pushed equal-volume cells. Discrete \(H\) changes.
Same speed law \(f(s)\) , two partitions
Pushed: equal 3D volume (\(u=s^3\) ); uniform: equal \(\Delta s\)
The discretisation is the measure
Side by side: same \(f\) , same \(s_{\max}\) , same number of bins. Only the edge placement changes. Midpoint Riemann rectangles; two values of \(H\) . Uniform-in-\(s\) is Lebesgue on the speed axis; equal-volume is Lebesgue in 3D, written in \(s\) .
The same Maxwell density, two midpoint Riemann partitions, two values of discrete \(H\) . Left: equal 3D volume, pushed to the speed axis. Right: equal \(\Delta s\) . The discretisation is the measure.
Bin Widths Are a Radon-Nikodym Derivative
The ratio of widths is the discrete Jacobian. It converts one \(h\) into the other.
Same cells: \(H\) is shared; \(h\approx H+\sum_i p_i\log\Delta_i\)
\(\Delta s_i/\Delta u_i \to |{\mathrm{d}s}/{\mathrm{d}u}|=1/(3s^2)\)
Continuum name: Radon-Nikodym derivative between the two measures
Board \(h\approx H+\sum p_i\log\Delta_i\) for unequal bins. The difference of the two remainders is \(\sum p_i\log(\Delta s_i/\Delta u_i)\) , which tracks \(\mathbb{E}\log|J|\) . Do not do \(\sigma\) -algebras. The width ratio is the discrete RN derivative.
Bin Widths are a Radon-Nikodym Derivative
Width ratio of the pushed equal-volume cells, against the Jacobian \(1/(3s^2)\) . The average log-ratio converts \(h(u)\) into \(h(s)\) . That ratio is the discrete Radon-Nikodym derivative.
Radon-Nikodym Derivative
Recall RN derivative between measures.
Maxwell’s Demon, Compartments, Two Ledgers
Composition wrapper only. Overview (quote and canvas), then the compartment schematic (three, maybe four), then number versus energy at the gate, then the fork. Canvas handedness: hot left to right, right bulk heats. Flux-balance already includes the ledgers and the fork. Do not include Gaussian–Maxwell, grouping composition, Szilard, Landauer, or Ellis here. Stop before the worksheet.
Maxwell’s Demon
Maxwell’s demon was designed to highlight the statistical nature of the second law of thermodynamics.
Maxwell’s Demon. The demon decides balls are either cold (blue) or hot (red) according to their velocity. Balls are allowed to pass the green membrane from right to left only if they are cold, and from left to right only if they are hot. The displayed entropy is the Shannon entropy of the velocity histogram (a coarse-grained proxy, not full thermodynamic entropy).
A Compartment Model of the Demon
The box as compartments, not as one fluent Maxwellian.
Left bulk: roughly local equilibrium at a cold \(T_L\)
Right bulk: roughly local equilibrium at a hot \(T_R\)
Kinetic layer at the membrane: not in equilibrium
Optional fourth: the gate itself (the velocity rule)
Three compartments, maybe four. The bulks are far enough from the membrane that collisions restore a Maxwellian; the band at the gate is where the sorting rule acts. We do not resolve the demon here: Worksheet 3 presses the particle-flux versus energy-flux bookkeeping.
Three compartments for Worksheet 3: cold left bulk, hot right bulk, and a non-equilibrium kinetic layer at the membrane. The green gate is an optional fourth compartment — the microscopic velocity rule. Number flux and energy flux are the two ledgers that must both close.
Flux Balance Across the Membrane
What Does the Membrane Do?
The JavaScript demon is a selective membrane : one speed threshold \(v_\star\) , opposite rules on each side.
Left \(\to\) right: pass only if \(v > v_\star\) (hot)
Right \(\to\) left: pass only if \(v \le v_\star\) (cold)
Elastic collisions remix velocities; the membrane only gates crossings
Match the canvas: red = hot, blue = cold, green strip = membrane. Threshold in the sim is \(v_\star=3\) with initial speed \(5\) .
Steady State \(=\) Balanced Fluxes
In a steady state the number of balls leaving left equals the number leaving right — flux balance of membrane crossings.
Assume each side thermalises to a 2D Maxwellian at its own \(T_L\) , \(T_R\)
Particle flux \(L\to R\) : only the hot tail, \(v>v_\star\)
Particle flux \(R\to L\) : only the cold body, \(v\le v_\star\)
Steady state: \(\Phi_L^{\mathrm{hot}}(n_L,T_L)=\Phi_R^{\mathrm{cold}}(n_R,T_R)\)
Local equilibrium is an approximation: collisions fast compared with crossings. Energy flux need not balance under the same Maxwellians — flag that after the calculation.
Left: temperatures on each side of the selective membrane from particle-flux balance at equal populations (simulation energy budget). Right: entropy of the two-Maxwellian state minus the single-temperature equilibrium. The red point is the simulation threshold \(v_\star=3\) .
Temperatures Split; Entropy Falls
For the simulation budget, particle-flux balance at equal \(N\) yields \(T_L\approx 1.9\) , \(T_R\approx 23\) (\(k_B=1\) ) — a large split.
Unconstrained equilibrium: one \(T_0=\langle\mathrm{KE}\rangle\) , entropy \(S_0\)
Membrane steady state: \(S_{\mathrm{ss}}<S_0\) (here \(\Delta S\approx -25\) nats at \(N=40\) )
The gas alone looks like a second-law violation
Live the numbers from the plotcode printout. Canvas entropy is a coarse velocity-bin proxy; this \(S\) is the ideal-gas Shannon entropy of two Maxwellians.
Maxwellians Are Only an Approximation
Particle balance and energy balance cannot both hold for two Maxwellians — the true steady state depletes velocity tails.
Hot crossings carry more energy per ball than cold crossings
Equal number fluxes still leave a net energy flux under Maxwellians
Collisions rebuild the tails; the membrane continually sculpts them
Local-\(T\) picture still predicts the direction and an entropy deficit
Do not oversell the numbers as exact NESS values. The point is the ledger: gas entropy down; cost elsewhere.
Where Did the Entropy Go?
The membrane implements Maxwell’s sorting policy. The gas entropy falls; Clausius is not repealed.
Prescription: the velocity gate (who may cross)
No-go: you cannot harvest that \(\Delta S\) for free in a cycle
Next: Szilard priced measurement; Landauer priced erasure; Bennett (1982) made the erasure resolution
Bridge forward. Ellis locates dissipation at measurement/gating; information ledger locates it at stored outcomes and erasure. Same Clausius, different bookkeeping.
The Demon Sorts Groups
Hot versus cold is a coarse label — this lecture’s grouping, now with a policy.
Group \(G\) : pass / no-pass under the membrane rule (tail vs body)
Within group: collisions try to restore a Maxwellian \(\pi_G\)
Local-equilibrium assumption: mixing inside each side is fast vs crossings
Callback to the die: the demon pays attention to \(H(w)\) over coarse velocity sets and hopes the within-group term stays near equilibrium. The flux-balance calculation above already used that hypothesis.
Lumpability (Pointer)
When do the groups themselves form a Markov process?
If exit rates agree for every microstate in a group \(\Rightarrow\) coarse chain is Markov (lumpable)
Otherwise coarse-graining leaves memory — projection operators (Zwanzig/Mori), not this course’s core
Demon membrane: approximately lumpable only while local equilibrium holds