Chapter 1: The Rock That Flows

Let me begin where the whole book has to begin, on the only ground solid enough to hold the rest of its weight — which is a phrase I have chosen deliberately, because the ground in question turns out to be a great deal more solid than almost anyone believes.

Everything I promised in the introduction depends on this chapter being trustworthy. I told you I would reach, later, toward the eruption of water through ice on a moon of Saturn, and toward the question of whether a planet has to be geologically alive to stay biologically alive. I also told you the reaching would only be safe if the honesty was mechanical. So before any of that, I owe you the settled science.

This chapter lives almost entirely in the first of my three registers — the empirical, the domain of what has been measured, replicated and agreed. There is one substantial exception, an argument about the deep mantle that is genuinely unresolved and that I will flag loudly when we arrive at it. Everything else here you may take as the discipline's considered position, and the reason I can hand it to you with that confidence is the subject of much of what follows: not what we know, but how the knowing was done.

And the first thing the knowing did was destroy the picture almost all of us are carrying.

The picture in everyone's head

Ask most people — including, I should say, most people who are otherwise well informed — what a volcano is connected to, and you will get some version of the same answer. Beneath the thin crust on which we live there is molten rock. A sea of it, or at least a great deal of it. Volcanoes are the places where it finds a way up: chimneys, taps, holes in the floor. The Earth is a body with a hot liquid interior and a cooled skin, and volcanism is what happens when the skin fails.

I want to be careful here, because this is not a stupid picture. It is a perfectly reasonable inference from the available evidence if the available evidence is what comes out of a volcano. Lava is molten rock. Lava comes from underground. Therefore there is molten rock underground. The logic is impeccable. It is also, in the sense that matters, wrong, and getting it right is not a pedantic correction — it changes what kind of object we are talking about.

Here is the position. The Earth has a crust, then a mantle nearly three thousand kilometres thick, then a core in two parts. The mantle constitutes the overwhelming majority of the planet by volume. And the mantle is solid rock. Not softened, not slushy, not a thick liquid: solid, crystalline rock, made mostly of the mineral olivine and its high-pressure relatives, hot enough to glow ferociously if you could somehow bring a piece of it into a dark room, and solid throughout.

There is exactly one enormous liquid layer inside this planet. It is the outer core. It is roughly two thousand kilometres thick, it begins about halfway to the centre, and it is made of molten iron and nickel — metal, not rock. It generates the magnetic field. It has never been tapped by a volcano and never will be, and nothing that has ever erupted anywhere on the surface of the Earth came from it.

Molten rock — magma — does exist. But it exists as a minor, transient, locally generated phase, made in specific places by specific processes, in quantities that are small compared with the solid rock around it, and it does not persist. It is not a reservoir the planet keeps. It is something the planet is continuously making and continuously getting rid of.

The mental adjustment I am asking for is roughly this: stop imagining a boiler with a lid, and start imagining a body that manufactures small volumes of liquid, at depth, on demand, and then expels them.

The proof, and the pleasure of it

I would not ask you to take that on my authority, and you should not. So let me show you how it is known, because the argument is one of the most elegant in the whole of physical science and it can be given in a paragraph.

Earthquakes generate several kinds of wave. Two of them travel through the body of the planet rather than around its surface. The first, the P-wave, is a compressional wave — the rock is squeezed and stretched along the direction the wave is travelling, exactly as a sound wave squeezes and stretches air. The second, the S-wave, is a shear wave: the rock is displaced sideways, perpendicular to the direction of travel, like the wave you send along a rope by flicking one end.

Now the crucial physical fact. A compressional wave can travel through anything that can be squeezed, which is to say through solids, liquids and gases alike. A shear wave cannot. Shearing requires the material to resist being deformed sideways — it requires rigidity — and a liquid, by definition, has no rigidity to offer. Put a shear wave into a liquid and it dies. S-waves do not propagate through liquids. Full stop.

So: put seismometers all over the surface of the Earth, wait for a large earthquake, and see what arrives where.

What we find is that S-waves from a given earthquake arrive at stations across a huge swathe of the planet, having passed straight through the mantle on their way. They do it every time, from every earthquake, at every station, and they have been doing it for as long as we have had instruments capable of recording them. The mantle transmits shear. The mantle is therefore rigid. The mantle is solid, and the whole of it is solid, and this is not an inference from a model — it is very nearly a direct measurement.

And then, past a certain angular distance from the earthquake, the S-waves stop arriving. There is a zone on the far side of the planet where the P-waves show up, deflected and delayed, and the S-waves simply never come. That shadow is the outer core announcing itself: a layer that will pass compression and will not pass shear, which is to say a liquid, sitting exactly where the other evidence says the iron is.

I find this genuinely lovely, and I want to say why, because it sets the pattern for everything the rest of the book does. Nobody has been to the mantle. Nobody will. But the planet rings when it is struck, and the ringing carries the news, and a sufficiently patient network of instruments on the surface can read the interior of a world from the outside. The claim "the mantle is solid" is not a story geologists tell. It is what the shear waves did.

Solid things flow

Having insisted so hard on solidity, I now have to complicate it, because otherwise the next several chapters make no sense at all.

Solid rock, at mantle temperatures and pressures and over geological time, flows. Not by melting — the whole point is that it does not melt — but by creep: atoms migrating through crystal lattices, defects propagating, grains deforming and rotating past one another. The rock stays a crystalline solid throughout and nonetheless moves, slowly and continuously, at speeds of a few centimetres a year.

The analogy that actually works here is a glacier. Nobody doubts that glacial ice is solid; you can stand on it, break it, cut a block out of it. And a glacier flows downhill, deforming internally, with a velocity profile like a river's. Ice is a solid that flows because it is close to its melting point and has all the time in the world. So is the mantle.

There is a useful piece of jargon for this, from rheology: whether a material behaves as a solid or a fluid depends on the ratio between how long it takes to relax and how long you watch it. Watch pitch for a second and it is a solid you could shatter with a hammer. Watch it for a decade and it drips. Watch the mantle for a year and it is rock. Watch it for ten million and it is a convecting fluid.

This is the resolution of an apparent contradiction that trips up a lot of otherwise careful readers. When you see a diagram of the mantle with arrows in it — hot material rising here, cool material sinking there — those arrows are real. Convection is real. Plate tectonics is real, and is in fact the surface expression of that convection. None of it requires the mantle to be liquid, and none of it is liquid. It is a solid, circulating.

How you melt a rock that will not melt

So we have a planet whose interior is hot, solid, and creeping. Where does the lava come from?

The intuitive answer is: it must get hotter somewhere. Somewhere down there is a hot spot, and where it is hot enough, the rock melts.

The intuitive answer is very nearly the reverse of the truth, and the correction is, to my mind, the single most beautiful idea in igneous petrology.

The temperature at which a rock melts is not a constant. It depends on the pressure the rock is under, and it depends on what else is dissolved in it — above all, on water. This gives you three ways to melt a rock, and on this planet, only one of them involves adding heat, and that one is by far the least important.

One: take the pressure off

Squeeze a rock and it becomes harder to melt. The atoms are held more tightly into the crystal lattice; you need more thermal energy to shake them free. So the melting temperature of mantle rock — the solidus, the temperature at which the first drop of liquid appears — goes up with depth, because pressure goes up with depth.

Now consider a parcel of mantle rock being carried slowly upward by convection. As it rises, it does expand slightly and cool slightly — this is the same thing that cools a rising parcel of air — but the cooling is gentle, a fraction of a degree per kilometre. Meanwhile the pressure on it is dropping fast, and the solidus is dropping with the pressure, and the solidus is dropping much faster than the rock is cooling.

The two lines converge. At some depth they cross. The rock, which has not gained a single joule of heat and has in fact lost a little, finds itself above its own melting point — and begins to melt.

This is decompression melting, and it is the dominant magma-producing process on Earth. It is what happens beneath every mid-ocean ridge, where the plates pull apart and mantle rises to fill the gap; and the mid-ocean ridge system, which is almost entirely underwater and which most people have never thought about for a single minute of their lives, produces the great majority of the planet's magma. The volcanoes you have heard of are a rounding error. The main event is a forty-thousand-mile seam running along the floor of every ocean, quietly manufacturing new crust.

I want to put the strangeness of this plainly. The commonest way to melt rock on this planet is to lift it. Not to heat it. To carry it upward until the pressure holding it together lets go.

Two: add water

The second mechanism is chemical rather than mechanical, and it is responsible for nearly every volcano that has ever frightened anybody.

Water dissolved in silicate melt disrupts the polymerized network of silicon and oxygen that gives rock its structure. The practical consequence is that adding water to hot mantle rock lowers its solidus dramatically — by several hundred degrees. Rock that was comfortably solid becomes, on the arrival of water and with no change in temperature whatever, rock that is partly liquid.

The place this happens is a subduction zone. An oceanic plate, having been made at a ridge and having spent tens of millions of years sitting under seawater, is thoroughly hydrated: water is bound into the crystal structures of minerals like serpentine, chlorite, amphibole and various micas. When that plate is forced back down into the mantle, it heats up, and those hydrous minerals become unstable and break down, and the water they were holding is released into the wedge of mantle sitting above the descending slab.

That wedge is hot. It was not melting. Now it is.

This is flux melting, and it is why there is a ring of volcanoes around the Pacific, and why those volcanoes behave the way they do. The magma they produce carries dissolved water inherited from the ocean floor, and dissolved water in magma is, as we will see in the next chapter, the single most dangerous substance in this book. Pompeii, St. Helens, Pinatubo, Unzen — all of them are, at bottom, seawater coming back up.

Three: heat it after all

The third route is the one everybody guesses first, and it does happen; it is simply the junior partner.

If basaltic magma generated by either of the first two mechanisms rises into the crust and parks there, it carries an enormous amount of heat with it, and it gives that heat to the rock around it. Continental crust, which has a lower melting point than mantle peridotite, can be pushed over its solidus by that heat and melt in place.

This is heat-transfer melting, or crustal anatexis. Volumetrically it is modest. In terms of consequences it is out of all proportion to its volume, because the melt it produces is silica-rich — rhyolitic — and silica-rich melt is viscous, and viscous melt is what produces the eruptions that reorganize climates. Most of Part IV of this book is, in one way or another, about the products of this third mechanism.

Partial melting, and the distillation of a planet

There is a further piece of machinery here that I need to give you now, because without it the third part of this book cannot make its case.

A rock is not a pure substance. Ice melts at a single temperature because ice is one compound; rock is an assemblage of several different minerals, each with its own melting behaviour, and so a rock does not melt all at once. It melts over an interval — beginning at the solidus, finishing, hundreds of degrees higher, at the liquidus. In between, and this is the normal state of affairs in the Earth, you have a solid with some liquid distributed through it.

In practice mantle melting is very partial indeed: something on the order of a few per cent, occasionally rather more. Ninety-odd per cent of the rock stays solid.

And here is the consequence that matters. The minerals that melt first are not a representative sample of the rock. The liquid that comes off has a different composition from the solid it came from — richer in silica, in aluminium, in the alkalis, in water, and in a whole class of elements which, because they do not fit comfortably into the crystal structures of mantle minerals, partition preferentially into any melt that is available. Geochemists call these incompatible elements, which is a wonderfully deadpan name for what they actually do, which is leave.

So partial melting is a distillation. Melt a bit of mantle peridotite and you get basalt, which is richer in silica than the peridotite was. Melt basalt — or, better, hydrated basalt at a subduction zone — and you move further along, toward andesite. Run the process again and you arrive at the granites.

Every cycle concentrates the light, the silica-rich, the volatile and the incompatible upward, and leaves the dense residue behind. Do this for four and a half billion years and you have separated a planet into layers by composition: a depleted mantle underneath, and floating on top of it a scum of buoyant, silica-rich rock too light ever to be pushed back down.

That scum is the continents. It is where you live. It was distilled out of the mantle by partial melting, one small increment at a time, and it exists for the same reason lava exists, and I will make the full argument for that in Chapter 7.

There is a satisfying loop hidden in this, and it is worth pausing on. Uranium, thorium and potassium are all incompatible elements. They have been preferentially extracted from the mantle and concentrated into the crust by exactly the process described above. The radioactive fuel that drives the engine has, over geological time, been steadily distilled toward the surface by the engine itself.

The fuel, and how we weigh it

Which brings us to where the heat comes from, and to one of the more remarkable measurements in modern geophysics.

Earth's internal heat has two sources. The first is primordial: the leftover energy of building a planet. Accretionary impacts, and then the gravitational energy released when iron sank to the centre to form the core, which by itself was enough to raise the whole planet's temperature by thousands of degrees. Some of that heat is still in there, four and a half billion years later, slowly leaking out.

The second is radiogenic: the ongoing decay of long-lived radioactive isotopes distributed through the rock — uranium-238 and uranium-235, thorium-232, and potassium-40. Each decay deposits a small amount of energy as heat into the surrounding rock. There is not very much of this material by mass. There is an enormous amount of rock.

The obvious question is how the total divides between the two, and it matters, because the primordial component is a fixed inheritance being spent down, while the radiogenic component is a decaying source with a calculable future. Together they determine how long this planet stays geologically alive — which is the subject of the coda, and which is not, I am afraid, an entirely cheerful subject.

For most of the twentieth century that division was estimated indirectly, from assumptions about the Earth's bulk composition, generally by analogy with meteorites. Then someone worked out how to measure it.

When uranium and thorium decay, they emit — among other things — antineutrinos. Antineutrinos interact with matter so feebly that they pass through a planet essentially unimpeded, which makes them nearly impossible to detect and, for exactly the same reason, perfect messengers: an antineutrino produced by a uranium decay a thousand kilometres down arrives at the surface carrying undistorted information about the event that made it. Build a large enough detector, put it deep underground to shield it from cosmic rays, run it for years, and you can count the ones that happen to interact. These are called geoneutrinos, and detectors in Japan and in Italy have now measured them.

I do not want to overstate what this has settled — the uncertainties remain large, and the precise heat budget is still argued over. But I do want to be clear about what has happened here, because it is the kind of thing that ought to be better known. We have directly detected the radioactivity of the Earth's interior, by catching the ghost particles it emits, in tanks buried under mountains. The planet's fuel gauge has been read from the outside.

The other thing worth carrying away is that the fuel is running down. Radioactive decay is exponential; every one of those isotopes has a half-life, and every half-life that passes leaves less of it. The early Earth generated substantially more radiogenic heat than the Earth does now, and ran hotter, and convected faster, and was more volcanically violent by a wide margin. What we are living through is a late, cool, comparatively sedate phase of a process that has been winding down since it began.

Convection, and the argument about plumes

Register note: everything up to this point in the chapter is empirical and, so far as I can tell, uncontested. What follows in this section is partly settled and partly not, and I am going to mark the seam.

That the mantle convects is established. Cold, dense oceanic lithosphere sinks at subduction zones and pulls the plate behind it; hot mantle rises beneath ridges to replace it. Plate tectonics is the top of the convecting system, not something separate that sits on top of it — the plates are the cold upper thermal boundary layer, and the primary force driving them is the weight of their own sinking slabs.

For decades there was a serious argument about whether the mantle convected as a single layer or as two stacked cells divided at the seismic discontinuity around six hundred and sixty kilometres down. Seismic tomography — of which much more in Chapter 4 — has largely settled this by imaging slabs that plainly continue through that boundary and pile up near the base of the mantle. Whole-mantle circulation is now the mainstream position.

The unsettled part concerns plumes.

The hypothesis, which dates to the early 1970s, is that in addition to the broad convective circulation there are narrow, long-lived, hot upwellings rising from very deep — perhaps from the boundary with the core — and that where one of these meets the underside of a plate you get a volcanic hotspot. Because the plume is anchored deep and the plate is sliding over it, you get a track: a chain of volcanoes ageing progressively in the direction of plate motion. The Hawaiian chain, running northwest into the Emperor Seamounts, is the type example, and the argument has a great deal going for it.

The case in favour: the age progressions along hotspot tracks are real and are hard to produce any other way; helium isotope ratios at some hotspots suggest a source that has not been thoroughly stirred and degassed like the rest of the upper mantle; and seismic tomography has, in the last decade or so, begun to resolve broad low-velocity conduits beneath a number of hotspots that look very much like what the hypothesis predicts, along with two enormous anomalous structures sitting on the core–mantle boundary beneath Africa and the Pacific.

The case against, made most forcefully by a minority who have been notably stubborn about it and who have on several occasions turned out to be right about something: much of the evidence is indirect, and there are alternative explanations for most of it. Volcanism at supposed hotspots can be produced by shallow processes — patches of unusually fertile mantle that melt more readily without being any hotter, or extension in the plate above allowing decompression melting where the lithosphere is thin. Tomographic images of narrow deep features are exactly the kind of image most vulnerable to smearing and to the assumptions built into the inversion. And the famous bend in the Hawaii–Emperor chain, long read as a change in the direction of plate motion, may instead record the hotspot itself moving, which rather undermines the idea that these things are fixed deep anchors.

Where does this leave us? My honest reading is that the balance of evidence has moved toward plumes over the past twenty years, that the strongest version of the plume model — a narrow pipe from the core–mantle boundary feeding every hotspot on the map — is not supportable, and that "plume" is probably a family of phenomena of different depths and origins that got a single name too early. This is the theoretical register. Anyone who tells you it is closed is telling you about their allegiance, not about the data.

How the melt gets out

We have made some magma at depth. Getting it to the surface is a separate problem, and a more interesting one than it sounds.

The melt begins as an intergranular film — liquid occupying the boundaries between solid crystals, in fractions of a per cent. The first question is whether it can move at all, and the answer turns on geometry: whether the melt beads up into isolated pockets, in which case it is stuck, or wets the grain boundaries into a connected network, in which case it can flow. In the mantle, fortunately for everything that follows, it connects, at very low melt fractions indeed.

So the melt percolates: buoyant, because liquid silicate is less dense than the crystals it left behind, moving upward through the pore network of a solid that is itself slowly flowing. This porous seepage does not stay diffuse. It focuses, into channels — you can see the fossilized plumbing directly in places like the Oman ophiolite, where a slab of ocean floor has been shoved up onto a continent and can be walked over, and where the channels through which melt once drained appear as bands of rock stripped of everything the melt dissolved on its way past.

Higher up, the mechanism changes entirely. Once melt collects in sufficient volume and the surrounding rock is cold enough to fracture rather than creep, ascent switches from seepage to dikes: fluid-filled cracks that the magma drives open ahead of itself, propagating through brittle rock. This is fast. Not centimetres a year — metres per second.

We know it is fast, and I want to give you the evidence, because it is my favourite piece of reasoning in the chapter. Some magmas arrive at the surface carrying xenoliths — chunks of foreign rock torn off the conduit walls on the way up, including, in certain cases, pieces of genuine mantle. These fragments are considerably denser than the melt carrying them. In a slow-moving magma they would sink out long before reaching the surface, as gravel settles out of a slow stream. They did not sink out. They are sitting in the rock at the surface where anyone can pick them up.

Therefore the magma that brought them must have been moving fast enough to keep them suspended over the whole journey. From that one physical constraint you can put a floor under the ascent velocity, and the floor is high — metres per second, from mantle depths, in a matter of hours. The most direct samples we possess of the deep Earth were delivered by an express service, and we know it was express because the parcels are still in the van.

That is also why volcanoes are episodic rather than continuous. Magma is not piped steadily to the surface. It is generated slowly, accumulates in staging regions within the crust — regions whose true nature is a great deal stranger than "chamber," which is Chapter 4's business — and is then released in pulses when something gives way.

The honest complication

I have spent this chapter telling you, with what I hope reads as confidence, what the inside of this planet is made of and what it is doing. I should finish by telling you how little of it anyone has touched.

The deepest hole ever drilled into the Earth is on the Kola Peninsula in northwestern Russia. It took roughly two decades and it reached something over twelve kilometres before the rock became too hot and too plastic to keep the hole open. Twelve kilometres is an extraordinary engineering achievement and it is, as a fraction of the distance to the centre of the Earth, approximately nothing. It did not reach the base of the continental crust. It did not come close. No one has ever drilled out of the crust. Every statement in this chapter about the mantle — nine-tenths of the planet's rocky volume — is inference.

It is very good inference, and I have tried to show you its structure rather than merely assert its conclusions. Seismic waves, which read the interior by how it transmits vibration. Xenoliths and ophiolites, which are the mantle delivering samples to us rather than the reverse. Experimental petrology, in which small quantities of rock are squeezed and heated in the laboratory to mantle conditions to find out what they do there. Geoneutrinos. Each of these is a different instrument pointed at the same inaccessible object, and their agreement is the reason for confidence.

But I want to leave you with two facts that put the limits in proportion.

The first: our best physical samples of the deepest Earth arrive as microscopic inclusions trapped inside diamonds. Diamonds form at depth and are carried up fast — by exactly the express mechanism described above — and occasionally one of them encloses and preserves a tiny crystal of a mineral that is stable only far below. In 2014 a diamond from Brazil was found to contain a speck of ringwoodite, a high-pressure form of olivine that exists in the transition zone hundreds of kilometres down, and that speck turned out to contain water. That is the standard of evidence we are working with for the deep interior: a fleck inside a gemstone.

The second, and I find this one genuinely startling: the most abundant mineral in the Earth — the silicate that makes up the bulk of the lower mantle, and therefore the largest single component of the planet by volume — did not receive a formal name until 2014. It could not be named, because the naming rules require a natural specimen, and no natural terrestrial specimen had ever been obtained; it is unstable at surface pressures and reverts on the way up. The sample that finally allowed it to be named came from a meteorite, where it had been produced by shock. It is called bridgmanite.

We are standing on a planet whose principal ingredient we could not legally name until a decade ago, and which we have never held a terrestrial piece of.

That is the object this book is about. Nearly everything in the chapters that follow is read off its surface, or out of what it throws away, or from the way it rings when it is struck. I think that is a reason for a particular kind of confidence rather than a reason for doubt — the confidence of a discipline that has had to earn every claim indirectly and has therefore been forced to be very careful about what counts as knowing.

But you should hold the whole edifice with the knowledge that no one has been down there.

Next: why any of it explodes.

Draft notes — verification status

Standing convention: nothing below has been independently checked in this pass. The chapter's structural claims — the mantle is solid; S-waves establish it; the three melting mechanisms; partial melting as differentiation — are textbook-level consensus and are very unlikely to be wrong in substance. The numbers, dates and named studies are the exposure.

A note on the Part I field note. The outline calls for each part to open with a first-person field vignette (Erta Ale for Part I). I have not written it and will not: it is a claim about where you have personally stood, and it has to come from you or be cut. If you have not been, the honest alternatives are to write the vignette from a place you have been, or to drop the device.

Pending verification:

Mantle thickness (~2,900 km); outer core thickness (~2,200 km) and depth to core–mantle boundary.
The S-wave shadow zone geometry — the angular distance beyond which S-waves are not received (commonly given as ~103°). Stated qualitatively in the draft; confirm before adding a figure.
Typical degree of partial melting beneath mid-ocean ridges (draft says "a few per cent").
Mid-ocean ridge system as producer of the majority of Earth's magma, and the ridge system's total length (draft says "forty-thousand-mile").
Adiabatic gradient vs. solidus gradient — the claim that the solidus falls with decompression faster than the parcel cools. Directionally certain; the numbers need sourcing if quoted.
Magnitude of solidus depression by water in peridotite ("several hundred degrees").
Depth of slab dehydration beneath arcs (~100 km) — omitted from the draft text deliberately; add only with a source.
Total global heat flow (~46–47 TW) and the radiogenic/primordial split (the Urey ratio) — the draft avoids giving figures; keep it that way unless sourced.
Geoneutrino detection: KamLAND (Japan) and Borexino (Italy); first results and what they actually constrain. The draft is deliberately vague on magnitude — verify before tightening.
Half-lives: U-238, U-235, Th-232, K-40. Not quoted in the draft; add only if checked.
Plume hypothesis attribution (Morgan, early 1970s) and the principal critics (Anderson, Foulger and colleagues).
Recent tomographic imaging of plume conduits (French & Romanowicz, 2015) and the LLSVPs beneath Africa and the Pacific.
Hawaii–Emperor bend: age (~47 Ma) and the hotspot-motion reinterpretation.
Oman ophiolite as the standard field exposure of melt-channel structures (dunite channels).
Xenolith ascent-rate argument: the inferred minimum velocities for kimberlite and alkali basalt transport.
Kola Superdeep Borehole: final depth (12,262 m), completion date (1989), and the reason drilling stopped.
Pearson et al. (2014) — ringwoodite inclusion in a Brazilian diamond, and the water content claim.
Bridgmanite: formal naming in 2014, the meteorite source (Tenham), and the IMA rule requiring a natural specimen.