Introduction: The Deep Field and the Ledger

Introduction — The Deep Field and the Ledger

> Everything flows. > — attributed to Heraclitus

> Information is physical. > — Rolf Landauer

The most expensive photograph ever taken

In December 1995, Robert Williams, then Director of the Space Telescope Science Institute, did something with his discretionary time that a number of his colleagues considered a waste of a national instrument. He pointed the Hubble Space Telescope at nothing.

The patch he chose sat near the handle of the Big Dipper, and he chose it precisely because it was empty — no bright foreground stars, no nearby galaxies, nothing catalogued, nothing interesting. It covered a piece of sky about the size of a grain of sand held at arm's length. Then he left the shutter open, in one form or another, for ten days. Three hundred and forty-two separate exposures, stacked.

What came back has been reproduced so many times that it has lost its capacity to frighten anyone. Roughly three thousand galaxies, crowded into a window of nothing, each one a hundred billion stars, most of them so distant that their light left before the Earth existed.

The usual lesson drawn from this image is abundance. Look how full the universe is. Look how small you are. It is a fine lesson and I have no argument with it, but it is not the lesson of this book, and I think it slightly misses what the photograph actually is.

Any single one of those 342 exposures showed nothing at all. Blank frame. The photons were arriving the entire time — they had been arriving for billions of years, and they arrive still, on the back of your hand, right now, as you read this. Nothing was hidden. What was missing was a system willing to hold still long enough, and to spend enough, to keep them.

And even then, the image you have seen is not the light. It is a compressed record of the light: ten days of arriving photons reduced to a fixed array of numbers, most of the incoming signal discarded as noise, the surviving fraction stretched and colour-mapped so that a human retina — itself a lossy instrument with a bad dynamic range and a blind spot — can make something of it.

The Deep Field is not a picture of how much is out there. It is a picture of what it costs to keep anything at all.

That transaction is the subject of this book.

The law under the floor

Here is a fact that most people find either trivial or unbelievable when they first meet it, and which is neither.

Forgetting has a price, and the price is heat.

In 1961, an IBM physicist named Rolf Landauer worked out that erasing a single bit of information — genuinely destroying it, collapsing two possible states into one — requires the release of a minimum quantity of energy into the surrounding environment. The number is kT ln 2: at room temperature, around three sextillionths of a joule. Absurdly small. Also unavoidable. It is not an engineering limitation to be designed around; it is a floor set by thermodynamics, and it applies to any physical system that stores and clears states, which is to say all of them, including the one reading this sentence.

For fifty years this remained a beautiful piece of theory. Then in 2012, a group led by Antoine Bérut measured it — a single microscopic particle in a double-well optical trap, erased over and over, the heat counted `[VERIFIED]`.

Landauer's own summary of what he had found was four words long: information is physical.

I want to be careful here, because this is the point where books of this kind usually start writing cheques the physics will not honour. Landauer's principle does not say the universe is a computer. It does not say your thoughts are made of bits, or that consciousness is computation, or that reality is code. It says something narrower and stranger: that memory and forgetting are not abstractions floating above the physical world. They are transactions in it, and they show up on the same ledger as friction and combustion.

There is a second fact, and it comes from the least forgiving object we know of.

In the early 1970s, Jacob Bekenstein and then Stephen Hawking established that a black hole has an entropy, and that this entropy is proportional not to its volume but to the area of its horizon `[VERIFIED]`. The implication has been rattling around theoretical physics ever since. If the maximum information you can pack into a region of space scales with the surface you could wrap around it rather than the space inside it, then information has a maximum density, and that density is set by a boundary.

Put the two together and you have the floor this entire book stands on:

Keeping information costs energy. Losing it costs energy. And there is a hard ceiling on how much any bounded thing can hold.

Everything else in these pages is an attempt to take that seriously.

The thesis

If those constraints are real, then something follows that sounds obvious and turns out not to be.

Nothing keeps everything.

Not the Deep Field, which threw away most of ten days of light to produce one array of numbers. Not the star, which converts a cloud of gas into a point of fusion and radiates the difference away forever. Not the genome, which carries a working summary of four billion years and not a single one of the individual lives that wrote it. Not the memory, which retains the shape of a Tuesday in your childhood and has permanently deleted the other seven thousand. Not the civilisation, which keeps its laws and loses its languages.

Every persistent structure in the universe is, in this sense, the same kind of thing: a policy about what to discard.

This is the reversal the book is built on. We tend to treat structure and loss as opposites — order over here, entropy over there, the one holding out against the other. That framing is not wrong so much as it is badly cut. Structure is not the opposite of loss. Structure is a particular, disciplined way of losing. The star is not resisting entropy; it is producing enormous quantities of it, and the light and the heavy elements are what falls out of the process. The memory is not preserving your childhood; it is destroying almost all of it in a specific pattern, and the pattern is what you experience as having a past.

The question a structure answers is never how do I keep this? It is always: given that most of this is going, what do I keep, and what will I pay?

On the word "recursion," and what I am refusing

The title needs defending, because the obvious reading of it is one I want to shut down before we go any further.

Nearly every book that puts the atom next to the solar system, or the neuron next to the cosmic web, is making a version of the ancient claim: as above, so below. The universe is a single pattern repeating itself at every scale. The small is a copy of the large. Find the source code and you find everything.

It is a gorgeous idea. It is also, as usually stated, unfalsifiable — and I have spent a long time trying not to write sentences that cannot be wrong.

The problem is not that the resemblances are fake. Several of them are real and measurable. The problem is the causal story smuggled in underneath: that the resemblance exists because something is running the same program twice. That claim has no mechanism, no evidence, and no way to fail. Once you accept it, every subsequent similarity becomes confirmation, and you have stopped doing anything except collecting rhymes.

So here is the version of recursion this book will actually defend:

When the constraints are identical, the solutions converge.

A star, a galaxy, a genome, a nervous system, and an institution all face the same problem: finite energy, finite bandwidth, a maximum information density, and an entropy gradient that runs in one direction and cannot be argued with. Nobody arranged for them to be alike. They are alike because a shared constraint has a limited number of good answers, and each of them, independently, found one.

This is convergence, not repetition. It is the same reason a shark, an ichthyosaur, and a dolphin arrived at the same body — not because there is a fish template written into the fabric of the cosmos, but because water is unforgiving in a very specific way, and there are only so many shapes that get through it.

That is a smaller claim than as above, so below. It is also one that could be shown to be false, which is the entire point.

The three registers

Every claim in this book is tagged, and I would rather you learn the system in the first ten pages than discover halfway through that I have been quietly changing gears.

Established `[VERIFIED]` `[SOURCED]` — measured, replicated, and not seriously in dispute. Landauer's limit. Hydrostatic equilibrium. The r-process. The Deep Field. When I say the periodic table was written by dying stars, I am in this register.

Inferred `[INTERPRETATION]` `[BOUNDARY]` — well-supported models built on top of measurements, including some that the field is currently arguing about. Dark matter is here: the discrepancy is measured, the halo models work beautifully, and nobody has caught the particle. So is the holographic principle. So is the current status of dark energy, which changed twice while I was writing Chapter Eleven.

Analogical `[ILLUSTRATIVE]` — every mapping from physics to a human life. All of it. When I say that trauma behaves like an event horizon, I am making a claim about your intuitions, not about your neurons. The physics is doing rhetorical work, not evidential work, and I will say so each time.

The line between the second and third registers is where books like this go to die. I have tried to draw it in ink.

Four slots

There is a specific way an idea like compression rots. It starts meaning something precise, and then, because it is useful, it is stretched to cover one more case, and then another, until it means anything that happens to anything, at which point it explains everything and predicts nothing.

The defence is a fixed schema. Every chapter in this book takes one class of object and fills in the same four boxes, from real physics, before it is permitted to say a single word about human beings.

Input. What information enters the system? Operator. What lawful process throws part of it away? Invariant. What is retained, and in what form? Cost. What is paid — in energy, in time, in irreversibility?

For a molecular cloud collapsing into a star: the input is the positional and thermal microstate of a cloud light-years across; the operator is gravitational collapse; the invariant is one bound object, a mass, a composition, an angular momentum; the cost is a hundred thousand years of radiated heat and the permanent loss of everything else the cloud used to be.

If a chapter cannot fill all four boxes, the object does not belong in the book. This is a leash, and it is on me, not on you.

How this could be wrong

Two ways, and I would like them stated plainly at the front where they can be used against me.

One. If a persistent structure exists that retains all of its input information at no thermodynamic cost, the floor is gone and the book collapses into decoration.

Two. If the resemblances across scales turn out to be a curated collection — if I have chosen the six objects that rhyme and quietly declined to mention the sixty that don't — then there is no convergence, only an author with a filing system. The test is whether the shared constraint can be identified independently of the resemblance it supposedly produces. Where I can do that, I will show the work. Where I can't, I will say so and mark the passage `[ILLUSTRATIVE]`, and you should read it as literature.

The question underneath

I have spent a long time mapping particular territories. The body, and the metabolism it runs on without ever having designed it. The psyche, and the solar system as an interface for thinking about it. The long argument of evolution. And the machine we assembled out of our own written record, which turned out to be a mirror.

I did not know, for most of that time, that they were the same question.

They are, in this sense: each of them is about a system that must decide what to keep. A body deciding which proteins to fold and which to break down. A mind deciding which of the day to consolidate overnight and which to let go. A lineage deciding which mutations survive contact with the world. A model deciding which of ten trillion words of human writing to compress into a set of weights, and what is lost in the compression.

This book is what happens when you ask that question at the largest available scale and discover that the universe has been answering it for thirteen point eight billion years, in public, with the receipts still visible in the sky.

The route runs like this.

Part One establishes the rules of the channel — the first great erasure at recombination, which is why the sky has a wall in it, and then Landauer's ledger, which is why erasure is never free.

Part Two is the foundry: how any structure gets made at all. Molecular clouds and the threshold problem, including the honest and unpopular fact that some clouds are simply not massive enough to ignite. The main sequence and the physics of staying lit for ten billion years. And then the supernova — the only place in nature where information gets written to permanently durable media, at the cost of destroying the medium that wrote it.

Part Three is the archive: the neutron star, which throws away everything a star was in order to keep a single number and broadcast it into the dark with more fidelity than anything else in the sky, and the black hole, where the question of whether loss can ever be total becomes an open problem in physics rather than a figure of speech.

Part Four is the network — the invisible scaffolding that holds galaxies together and the fact that we have measured it without catching it, and then galactic cannibalism, where growth by absorption turns out to require digestion, and where I will show you a prediction about our own galaxy's fate that has been publicly revised twice in the last two years.

Part Five is what is left. How to read a sky that is almost entirely empty of evidence — the discipline of inferring a planet from a fraction of a percent of dimming, and of publishing your own corrections when you get it wrong. And then the last erasure: the heat death, the evaporation, and the strange discovery that in the deep future the universe will lose observational access to its own beginning. A cosmos that forgets its origin. The thesis, stated by the thing itself.

The stake

I should say what I think is actually at issue, because it is not astronomy.

You are going to keep some of your life and lose the rest, and the ratio is not close. The forgetting is not a malfunction, and it is not a moral failure, and it cannot be fixed by discipline or by an app or by writing more of it down — because the constraint is not psychological. It is thermodynamic. Every structure that has ever persisted has faced it, and every one of them solved it the same way: by paying for a small, specific invariant and letting the rest go.

The only real question available to you is which invariant, and whether you chose it or inherited it.

That is a harder and better question than how do I remember more. It is the one the sky has been demonstrating, at enormous scale and with total indifference, for the whole of cosmic history.

Ten days of light. Three thousand galaxies. One image.

Everything else discarded.

Let's begin.

THE CHANNEL

The rules everything else obeys