Lux: Picture a film reel, Hex.
Lux: You thread it into the projector. Press play.
Lux: A ball rolls off a table. Hits the floor. Bounces twice. Stops.
Hex: Okay, Lux. Normal day.
Lux: Now run the film backward.
Lux: The ball leaps off the floor, bounces upward, and lands on the table.
Lux: [beat]
Lux: Some films look just as natural in reverse.
Lux: Others scream "wrong direction."
Lux: Today we're going to put a number on that scream.
Hex: And I'm going to argue that number is lying to us.
Lux: [laughs softly]
Lux: That's the debate. Let's go.
Hex: Here's my position.
Hex: [leaning forward]
Hex: If the microscopic rules are reversible—
Hex: if every transition that can happen forward can also happen backward—
Hex: then the arrow of time is just a statistical accident.
Hex: A trick of our coarse perspective.
Lux: That's a real position. Serious people hold it.
Lux: But the framework says something sharper.
Lux: And to see it, we need to build up from paths.
Lux: A path is a sequence of states the system visits over time.
Lux: [counting on fingers]
Lux: State at time zero. State at time one. State at time two. All the way up to time T.
Lux: Think of it as the individual frames of your film reel.
Hex: So a path is one particular film?
Lux: One particular film. And the forward path law
Lux: tells you how likely that particular film is.
Lux: It's simple: start with your initial distribution—
Lux: the probability of being in each state at time zero—
Lux: then multiply by the transition probability at each step.
Lux: Initial distribution times transition, times transition, times transition.
Lux: One multiplication per frame.
Hex: Same mu-times-P from last episode—just chained together.
Lux: Exactly. Chained T times.
Lux: Now.
Lux: [beat]
Lux: Time reversal.
Lux: Take your film and literally flip the reel.
Lux: Frame T becomes frame zero. Frame zero becomes frame T.
Lux: The path runs backward.
Hex: And you can compute the probability of the backward path too.
Lux: You can. And here's the key object.
Lux: [leaning in]
Lux: Take the KL divergence—
Lux: that's the relative entropy—
Lux: between the forward path law and the reversed path law.
Lux: The framework calls this Sigma (SIG-mah) sub T.
Lux: The emergence calculus uses this number as its formal arrow of time.
Hex: So Sigma-T is a single number that says—
Lux: How much directional evidence is in the path statistics.
Lux: If Sigma-T is zero, the film looks the same forward and backward.
Lux: No arrow. Perfect symmetry.
Lux: If Sigma-T is large, the film obviously runs one way.
Lux: Strong arrow.
Hex: Okay. I see the definition.
Hex: But here's my challenge.
Hex: [pointed]
Hex: What if the arrow only appears because we're watching through blurry glasses?
Hex: We coarse-grain. We lump states together.
Hex: Maybe that lumping creates an apparent asymmetry
Hex: that wasn't there at the microscopic level.
Lux: That's the right question.
Lux: And the answer is a theorem.
Lux: [beat]
Lux: The data processing inequality for path reversal asymmetry.
Hex: The DPI again?
Lux: The DPI.
Lux: Think of coarse-graining as a security camera.
Lux: [gentle]
Lux: You have high-resolution footage and a low-resolution camera pointed at the same scene.
Lux: The theorem says:
Lux: whatever asymmetry the low-res camera sees
Lux: is less than or equal to the asymmetry in the high-res footage.
Hex: Less than or equal.
Lux: Always. Mathematically guaranteed.
Lux: The coarse view can miss asymmetry—
Lux: the grainy footage might not catch the break-in—
Lux: but it cannot invent asymmetry that wasn't there.
Hex: So no false positives.
Lux: No false positives.
Lux: If the microscopic system has zero arrow—Sigma-T equals zero—
Lux: then every coarse observation also sees zero.
Lux: You can get false negatives—hidden irreversibility—
Lux: but you cannot get a fake arrow.
Hex: [thoughtful]
Hex: Okay. I concede the point.
Hex: The arrow I see through my coarse lens is real.
Hex: I might be missing some of it. But I'm not imagining it.
Lux: That's the theorem.
Hex: Why does it work, though? Intuitively?
Lux: The key insight is geometric.
Lux: Reversing a path and then coarse-graining
Lux: gives the same result as coarse-graining and then reversing.
Lux: The two operations commute.
Lux: And once you have that, the rest follows from a general property of KL divergence:
Lux: pushing through any map can only shrink it.
Hex: Wait—really?
Hex: That commutation is doing all the work?
Lux: All of it.
Lux: Flip then blur equals blur then flip.
Lux: That one fact plus the contraction of KL gives you the entire theorem.
Hex: Here's something I didn't expect.
Hex: [surprised]
Hex: This definition—Sigma-T—
Hex: it doesn't require the system to be in equilibrium?
Lux: It does not.
Lux: You can start from any initial distribution.
Lux: The definition works. The DPI works.
Lux: Stationarity—being in equilibrium—
Lux: only enters when you want to interpret Sigma-T as an entropy production rate.
Hex: So the arrow exists even far from equilibrium.
Lux: Even far from equilibrium.
Lux: The number might be infinite in pathological cases—
Lux: if some reverse paths have zero probability—
Lux: but the definition is always there.
Lux: Let's make it concrete.
Lux: [counting on fingers]
Lux: Take our three marbles from episode six. Red, green, blue.
Lux: The kernel that breaks detailed balance—red mostly stays red,
Lux: green mostly goes to blue, blue mostly goes to red.
Lux: Now pick a specific path: red, green, blue, red, blue.
Lux: Five frames.
Hex: Okay. And the forward probability?
Lux: Probability of starting at red—from your initial distribution—
Lux: times the transition probability from red to green,
Lux: times green to blue, times blue to red, times red to blue.
Lux: Multiply them all together.
Hex: And the reversed path is: blue, red, blue, green, red.
Lux: Right. And you compute its probability the same way.
Lux: The ratio of forward to backward
Lux: tells you how much this particular path "prefers" the forward direction.
Lux: Sum the log-ratios over all possible paths, weighted by their forward probabilities—
Lux: that's Sigma-T.
Hex: So Sigma-T is a weighted vote across all films.
Lux: [beat]
Lux: A weighted vote on which direction is more natural.
Lux: And the DPI says: putting on blurry glasses can only reduce the vote margin.
Lux: One more connection.
Lux: [thoughtful]
Lux: In the Six Birds framework,
Lux: one of the companion papers defines time as a closure artifact.
Lux: Three ingredients: an ordering of events, a measure of change,
Lux: and an irreversible record that makes "before versus after" matter.
Lux: Sigma-T is the measure of change and the irreversible record
Lux: rolled into one number.
Hex: And the DPI holds in quantum too?
Lux: It does. The quantum version uses quantum relative entropy
Lux: instead of classical KL divergence.
Lux: And the channels—quantum operations—play the role of coarse-graining maps.
Lux: Same structural guarantee: a quantum channel cannot create distinguishability.
Hex: So it's the same principle across classical and quantum.
Lux: Same principle. Different mathematical clothing.
Lux: Let's land this.
Lux: [beat]
Lux: Three things.
Lux: One: the arrow of time is Sigma-T—
Lux: the KL divergence between the forward path law and its time reversal.
Lux: It measures how much directional evidence is baked into the path statistics.
Lux: Two: the data processing inequality guarantees no false positives.
Lux: Coarse-graining can hide the arrow but cannot create one.
Lux: Three: stationarity is not required.
Lux: Sigma-T is defined and the DPI holds for any initial distribution.
Lux: Equilibrium only matters when you want an entropy production rate.
Hex: So what's the test?
Lux: The test is: compute Sigma-T.
Lux: If it's zero, the process has no arrow at any resolution.
Lux: If it's positive, the arrow is real—and any coarse-graining
Lux: can only give you a lower bound on how strong it is.
Hex: Fair.
Hex: [beat]
Hex: Next time—we zoom out.
Hex: Theory packages. How the framework bundles state, dynamics, and audits
Hex: into a single portable object you can carry across domains.
Lux: The toolkit gets its carrying case.