1 millimetre a year: a ball of iron about as hot as the Sun's surface is freezing under your feet

Part IA ball of iron crystals

Chapter 1One millimetre a year

Under the crust lies the rocky mantle. At 3,480 km from the centre it gives way to the outer core, a shell of liquid iron alloy, and inside that, from about 1,220 km in radius down to the centre, is the solid inner core.¹ Its surface is about 5,150 km below our feet.

The headline, about a millimetre a year, is how fast the inner core grows as the liquid freezes onto it. It is the field's standard description: Lauren Waszek, Jessica Irving and Arwen Deuss wrote in 2011 that the inner core grows "at rates of about 1 mm per year", citing J. A. Jacobs's 1953 paper rather than deriving the figure.^2,3 It is an order of magnitude from cooling models, not a measurement (Chapters 5 and 6).

"Crystal" in the title means crystalline. The inner core is a ball of iron crystals, lined up well enough that earthquake waves cross it faster along the Earth's axis than across it. A 1995 model by Lars Stixrude and Ronald Cohen treated it as "a nearly perfectly aligned aggregate" of crystals, close to one giant crystal;⁴ later work describes many aligned grains, in layers that differ (Chapter 4).

The rest of the story is rotation. The inner core is a solid body inside a liquid, held to the mantle by gravity and to the liquid by magnetic forces, but not rigidly, so it can turn fractionally faster or slower than the mantle. Every statement on this page about its rotation is relative to the mantle: "super-rotation" means faster, so that it creeps eastward, and "sub-rotation" slower. Both bodies turn once a day. Most readings of earthquake waves agree that the inner core ran slightly ahead for decades; two of them, by Yang and Song and by Wang, Vidale and colleagues, put the turn around 2008 to 2010, after which it drifted slightly behind.^5,6

Inge Lehmann found the inner core in 1936, and Song and Paul Richards first reported in 1996 that it turns at a slightly different rate from the mantle. Three later lines of work recur here: Vidale and colleagues (2025) on a changing surface, Hrvoje Tkalčić (2024) on how regular the swing is, and Huifeng Zhang and Mathieu Dumberry (2026) on the inner core's pull on the length of the day.

Chapter 2A core inside the core

In 1936 Inge Lehmann, a Danish seismologist, published a paper titled "P′". The liquid core bends earthquake waves so strongly that it leaves a shadow on the far side of the Earth, and faint P waves, the compressional waves of earthquakes, were arriving inside that shadow. Lehmann worked out that the core must contain a separate inner part that sends some of the waves back into the shadow.⁷

That the inner part is solid took longer to show. In 1971 Adam Dziewonski and Freeman Gilbert found that the Earth's normal modes, the whole planet ringing like a bell after a great earthquake, require an inner core with rigidity,⁸ and in 2000 Arwen Deuss and colleagues reported shear waves that had crossed it.⁹ Shear waves cannot travel through a liquid.

The standard numbers come from the Preliminary Reference Earth Model (PREM) of Dziewonski and Don Anderson, published in 1981.¹ It puts the inner-core boundary 1,221.5 km from the centre and the top of the liquid core at 3,480 km. With the Earth's mean radius of 6,371 km, the inner core's surface is 5,149.5 km down. This page rounds to 1,220 km in radius, about 2,440 km across, and 5,150 km down.

PREM	Just inside the inner-core boundary	At the centre
Depth below the surface	5,149.5 km	6,371 km
Density	12.76 g/cm³	13.09 g/cm³
Pressure	328.85 GPa	363.85 GPa
In sea-level atmospheres	3.25 million	3.59 million

Values from PREM; the atmospheres are this page's arithmetic, dividing by 101,325 Pa. The liquid just outside the boundary has a density of 12.17 g/cm³.

Inside, P waves travel at 11.0–11.3 km/s and shear waves at 3.5–3.7 km/s.¹ Adding up PREM's density shell by shell gives the inner core's mass:

M_ic = ∫₀^R 4πr²ρ(r) dr ≈ 9.84 × 10²² kgρ(r) from PREM, R = 1,221.5 km. The mean density is 12.89 g/cm³ (this page's arithmetic).

That is 1.34 times the mass of the Moon, 7.346 × 10²² kg, and 1.65% of the Earth's mass, in 0.70% of the Earth's volume (this page's arithmetic from PREM). The inner core's radius is 70% of the Moon's, 1,737.4 km,¹⁰ and slightly larger than Pluto's, 1,188.3 ± 1.6 km from New Horizons images.¹¹

Nobody has seen it, and no image of it exists. What is known comes from seismic waves, gravity and the length of the day, read alongside laboratory experiments and calculations on iron and the magnetic record in old rocks.

Chapter 3About as hot as the Sun's surface, and solid

The inner core's surface is where the liquid core is freezing, so its temperature is the freezing point of the core's alloy at about 329 gigapascals. In 2013 Simone Anzellini and colleagues melted a speck of iron squeezed between two diamond tips and heated by laser (a diamond-anvil cell), watching it by fast X-ray diffraction, and extrapolated from 200 gigapascals to a melting point for pure iron of 6,230 ± 500 K at that pressure.¹² Lighter elements lower the freezing point by several hundred to more than a thousand kelvin, and the table of core properties compiled by Deguen and colleagues puts the temperature at the inner core's surface at roughly 5,000 to 6,000 K.¹³ A 2025 simulation finds about 5,000 K for a core with 10 mol% carbon (one atom in ten), against 5,300–5,900 K in earlier work on iron–oxygen alloys.¹⁴ The exact value is uncertain by several hundred kelvin.

The International Astronomical Union's nominal temperature for the Sun's visible surface is 5,772 K.¹⁵ So the inner core is about as hot as the surface of the Sun, and far cooler than the Sun's interior.

What keeps it solid is pressure, which raises iron's melting point. J. A. Jacobs proposed in 1953 that the inner core is iron frozen out of the liquid core, growing as the Earth cools.³ In the core the melting point rises with depth faster than the temperature does, so as the core cooled, the first place to fall below its melting point was the centre:

T(r) < T_m(r) ⇒ solid at radius rT: the temperature of the convecting core; T_m: the alloy's melting point at the local pressure. Both fall outwards from the centre, the melting point faster, so as the core cools they first cross at r = 0, and the crossing, the inner-core boundary, then moves outwards.

The ball is mostly iron, with some nickel (about 5% by weight for the core as a whole, in the usual estimate),¹⁴ and a few per cent of lighter elements, since it is slightly less dense than pure iron would be.¹⁶ Which light elements, and how much of each, is not settled. Its iron is thought to be hexagonal close-packed (hcp), with a body-centred cubic (bcc) phase proposed for parts of it.^16,17 Simulations by He and colleagues (2022) suggest that hydrogen, oxygen or carbon in it could be superionic, diffusing almost like a liquid through the iron lattice.¹⁸ That is an interpretation, and the inner core is not liquid.

Ice on a pond is a fair picture, up to a point. A pond freezes from the top down; the core freezes from the centre out, because the Earth's cooling makes it freeze and pressure decides that the centre goes first. And this ice is at about 5,500 K.

Chapter 4What "crystal" means here

In 1983 Georges Poupinet and colleagues found that P waves crossing the inner core close to the Earth's axis arrive earlier than waves crossing it near the plane of the equator.¹⁹ In 1986 Andrea Morelli, Adam Dziewonski and John Woodhouse explained the difference by anisotropy:²⁰ the inner core is faster pole to pole, by about 1% on average in their study.

Later work found anisotropy of up to a few per cent, stronger in the western hemisphere than the eastern, and an upper layer, roughly the top 60–80 km, with little or none. Arwen Deuss's 2014 review describes an isotropic skin, with the same speed in every direction, over an aligned interior; the skin's thickness is an interpretation.²¹

A single crystal of hcp iron is itself anisotropic, but crystals pointing every which way average that out. For the whole ball to prefer one direction, its crystals must on average be lined up. In 1995 Stixrude and Cohen calculated the elasticity of hcp iron at inner-core pressure from first principles and matched the travel times by treating the inner core as "a nearly perfectly aligned aggregate of hcp crystals".⁴ That model, close to one giant crystal, is the origin of the single-crystal inner core of many headlines, and it is history now. The standard explanation is the alignment of many crystals, varying with depth and between hemispheres.^17,21

There may be a layer within. In 2002 Miaki Ishii and Adam Dziewonski found the anisotropy changing character within about 300 km of the centre, an "innermost inner core".²² In 2023 Thanh-Son Phạm and Hrvoje Tkalčić, with waves that cross the Earth's centre several times, described "a ~650-km thick innermost ball" (their figures draw it with a radius of about 650 km), in which P waves are about 4% slower at about 50° from the axis.²³ The 2025 review by Wilson and colleagues gives 250–700 km for its radius.¹⁷ The bulk anisotropy is established; the innermost inner core, and its size, are interpretations.

It is tempting to read such layers as tree rings, each recording the time it froze. But no rings have been seen: the "layers" are changes in crystal alignment, inferred from the speeds of waves.

Chapter 5Freezing from the centre out

As the core cools, liquid alloy freezes onto the inner core. The classical model of that growth, by Bruce Buffett, Herbert Huppert, John Lister and Andrew Woods in 1992, has the radius growing roughly as the square root of time:^13,24

r(t) = R √(t/τ), V₀ = R/2τ = ½ × R/τR = 1,221.5 km is today's radius, τ the inner core's age and t the time since it began to freeze. Growth is fastest when the ball is small; at t = 0 the law gives no finite speed. Today's rate V₀ is half the lifetime mean R/τ.

So the age sets today's rate. Renaud Deguen, Thierry Alboussière and Daniel Brito, using the same law, give about 2 × 10⁻¹¹ m/s, 0.6 mm a year, for an inner core about a billion years old, and 0.19–0.63 mm a year across their parameters.¹³

Age τ	Lifetime mean	Today	Iron frozen per second today	Latent heat today
0.5 Gyr	2.44 mm/yr	1.22 mm/yr	9,300 t	5.6–11 TW
0.55 Gyr	2.22 mm/yr	1.11 mm/yr	8,400 t	5.1–10 TW
0.7 Gyr	1.75 mm/yr	0.87 mm/yr	6,600 t	4.0–7.9 TW
1.0 Gyr	1.22 mm/yr	0.61 mm/yr	4,600 t	2.8–5.6 TW
1.5 Gyr	0.81 mm/yr	0.41 mm/yr	3,100 t	1.9–3.7 TW
2.0 Gyr	0.61 mm/yr	0.31 mm/yr	2,300 t	1.4–2.8 TW
3.4 Gyr	0.36 mm/yr	0.18 mm/yr	1,400 t	0.82–1.6 TW

This page's arithmetic from the square-root law and the formulas below, with a latent heat of 600–1,200 kJ/kg. Latent heat only.

So "about a millimetre a year" is the right size for today's growth: for the most-cited ages, half a billion to a billion and a half years, today's rate is 0.4–1.2 mm a year, and for 2 billion years or more, a few tenths. It is also roughly the lifetime average of an inner core about a billion years old (1.22 mm a year at 1.0 billion years, 1.0 at 1.2 billion), whose rate today is half that. Today's rate also fixes how much iron freezes each second, and how much heat that releases:

ṁ = 4πR² ρ V₀, Q_L = ṁ Lṁ: the mass frozen per second; Q_L: the latent heat it releases; L: latent heat per kilogram. Area 4πR² = 1.875 × 10¹³ m²; ρ = 12,763.6 kg/m³ just inside the boundary (PREM); a year of 31,557,600 s. At 1 mm a year, ṁ ≈ 7,600 tonnes a second, and with L = 600–1,200 kJ/kg, Q_L ≈ 4.5–9.1 TW (this page's arithmetic).

At a millimetre a year, about 7,600 tonnes of iron turn solid every second. The 2001 core model of Labrosse and colleagues (Chapter 6) gives a latent heat of 620 kJ/kg, its entropy of crystallisation, 118 J kg⁻¹ K⁻¹, times its solidification temperature at the centre, 5,270 K,²⁵ inside the published range of 600–1,200 kJ/kg, which is uncertain by a factor of two.¹³ That makes a few terawatts: 4.5–9 TW per millimetre a year, and about 1–11 TW across the ages in the table and the range of latent heat (this page's arithmetic, latent heat only). The heat helps power the convection that makes the Earth's magnetic field, though not alone: freezing also expels light elements into the liquid, and that lighter liquid rises and stirs it too.²⁶ A natural nuclear reactor at the centre of the Earth, another proposed source of heat, has been looked for and not found (Nº 13); its heat would be a different quantity, and this page does not compare the two.

A millimetre a year is about 8 cm in a lifetime of 80 years and about 9 cm since Lehmann found the inner core in 1936 (this page's arithmetic). The growth need not be even. Marc Monnereau and colleagues proposed in 2010 a "lopsided growth", one hemisphere freezing while the other melts,²⁷ and local changes in the surface of about a kilometre in a decade have been inferred (Chapter 11).²⁸ Those are interpretations about places; the millimetre a year is a global mean, never measured directly.

Chapter 6How old is the inner core?

In 2001 Stéphane Labrosse, Jean-Paul Poirier and Jean-Louis Le Mouël balanced the core's energy budget: the heat the core must lose to grow the inner core, against what the mantle can carry away. Without radioactive heating in the core, they found, the inner core cannot be older than 2.5 billion years, and "a more acceptable value would be around 1 ± 0.5 Ga".²⁵

The weak point is thermal conductivity. Heat conducted along the core's temperature gradient does nothing to stir the liquid, so a core that conducts well must cool faster to keep the magnetic field going, and has a younger inner core. On the high side, Monica Pozzo and colleagues calculated in 2012 that liquid iron at core conditions conducts two to three times better than earlier estimates, implying 15–16 TW conducted at the top of the core.²⁶ In 2016 Kenji Ohta and colleagues measured iron's electrical resistivity at core conditions (the electrons that carry current also carry heat) and concluded that it implies "a young inner core less than 0.7 billion years old".²⁹ Labrosse's 2015 model gives a similar bound if the core loses more heat at its top than conduction alone can carry,³⁰ and in 2020 Zhang and colleagues found a conductivity of about 100 ± 10 W m⁻¹ K⁻¹ (watts per metre per kelvin) near the core–mantle boundary.³¹

On the low side, Zuzana Konôpková and colleagues measured the thermal conductivity of solid iron directly in 2016 and placed that of the Earth's core at 18–44 W m⁻¹ K⁻¹, a value that "allows for a solid inner core as old as the dynamo", an upper limit on the age rather than a date.³² In 2020 Hsieh and colleagues found about 20 W m⁻¹ K⁻¹ for an outer core in which 15% of the atoms are silicon, with an inner core that "could be older than two billion-years".³³ The dynamo, the churning liquid outer core that makes the magnetic field, is at least 3.4 billion years old: its strength is recorded in rocks 3.4–3.45 billion years old.³⁴ The 2025 review by Wilson and colleagues treats the question as open.¹⁷

Palaeomagnetism, the field's record in magnetised rocks, is read both ways. Andrew Biggin and colleagues read an increase in the field's strength and variability 1.0–1.5 billion years ago as the inner core's birth (2015).³⁵ Richard Bono and colleagues found an ultra-low field about 565 million years ago, pointing to a later start (2019).³⁶ Zhou and colleagues found the field five times stronger by about 532 million years ago and placed the inner core's birth at about 550 million years ago, with the inner core at half its present radius, "where seismic anisotropy changes", by about 450 million years ago (2022).³⁷ Ohta and colleagues say the increase about 1.3 billion years ago "may not be related" to the inner core's birth.²⁹

Nobody knows how the first crystal formed. A liquid must be cooled some way below its melting point, supercooled, before a seed crystal can survive. For pure iron at the centre of the Earth, Ludovic Huguet and colleagues put the need at about 1,000 K (2018).³⁸ The 2025 review puts it at no less than 450 K, against at most 420 K available and more probably under 100 K.¹⁷ Marine Lasbleis and colleagues reach the 100 K limit separately, from the melt trapped in the inner core: the supercooling "cannot have exceeded 100 K".³⁹ That gap is the nucleation paradox.

The proposed ways out lower the need. In one, a seed first forms as metastable bcc iron, a structure that is not the stable one at those pressures but forms more easily, and then turns into hcp (Yang Sun and colleagues, 2022).⁴⁰ With carbon, 10 mol% brings the need down to the 420 K maximum "within uncertainty", and an extrapolation to 15.2 mol% (roughly 4% by weight, the most that may have entered the core) gives "only 266 K" (Wilson and colleagues, 2025; an author correction does not change the result).¹⁴ An alloy with 20% nickel, about four times the 5% usually assumed, needs about 250 K, in a study by Zhai and colleagues (Geophysical Research Letters, October 2026).⁴¹ The carbon and nickel figures are still well above the probable 100 K. They shrink the problem; none has settled it.

If the core did supercool, its first few hundred kilometres may have frozen quickly. The 2025 review says that region is "at least as large as the innermost inner core (250–700 km radius)",¹⁷ and gives no consistent speed for it, so this page gives none. Figure 1's "Fast start" is this page's sketch, not a published model: a first region 250 or 700 km in radius, the two ends of that range, frozen at an unknown speed, then the same law,

r² = r₀² + (R² − r₀²) t/τ, V₀ = (R² − r₀²)/(2Rτ)Today's rate is then (1 − r₀²/R²)/2 of the lifetime mean, whatever the age: 0.479 for r₀ = 250 km and 0.336 for 700 km, against a half classically; at τ = 1 billion years, 0.59 and 0.41 mm a year (this page's sketch and arithmetic).

Figure 1 turns an age into today's growth, iron and heat. Pick a published age or set your own, then press Grow it to watch the ball freeze.

Figure 1🎲 Try it

The growing crystal

Set its age

Age of the inner core

Look back

Your age (years)Your age

Today's rate, R/2τGrowing today

Lifetime mean, R/τAverage since it began

Radius thenSize then (centre to edge)

Iron frozen per second, todayIron frozen each second now

Latent heat today, L = 600–1,200 kJ/kgHeat given off now

Growth in your 30 years, at today's rateGrown in your lifetime

The classical growth law r = R√(t/τ) (Buffett et al. 1992) with R = 1,221.5 km (PREM) and the age τ from the slider. The law gives no finite speed at the very start, so the look-back stops 10 million years after it. Iron and heat use Chapter 5's formulas, with L = 600–1,200 kJ/kg (Deguen et al. 2007), latent heat only. Presets are published ages, some of them bounds: Zhou et al. 2022; Bono et al. 2019; Ohta et al. 2016; Labrosse et al. 2001; Biggin et al. 2015; and the oldest age the low-conductivity measurement of Konôpková et al. 2016 allows, as old as the magnetic field (at least 3.4 Gyr, Tarduno et al. 2010): an upper limit on the inner core's age, not an estimate. 'Fast start' is this page's sketch of a supercooled beginning (the box above). This page's model and arithmetic.Slide the age, or pick a published age (some are upper limits). A younger inner core must be growing faster now; an old one grows only a few tenths of a millimetre a year. Press Grow it to watch it freeze: from nothing in the classical picture, fast when it was small and slower since, or, with Fast start (this page's sketch, not a published model), from a first ball that froze quickly at a speed nobody knows. Put in your age to see how much it has grown in your lifetime.

Every published age is a claim about today's growth and heat. The Earth itself is 4.54 billion years old (Nº 13); every estimate here puts the inner core's birth well after the planet's.

Part IIA stethoscope at 5,000 kilometres

Chapter 7Earthquakes that repeat

Several waves from one earthquake reach the far side of the Earth by different roads through the core. PKIKP, also written PKP(DF), crosses the mantle (P), the outer core (K) and the inner core (I); PKiKP reflects off the inner core's surface; PKP(BC) and PKP(AB) turn in the outer core above it and never enter it.^42,43 They share almost the same source and mantle path, so comparing PKIKP with a wave that does not cross the inner core leaves what happened inside. What is measured is the gap between the two arrivals, and how that gap changes from one earthquake to the next:

δt = t_PKP(BC) − t_PKP(DF), Δ(δt) = δt_later − δt_earlierIts change between two earthquakes on the same path is the signal Song and Richards measured in 1996 (Chapter 8).

The workhorse path runs from the South Sandwich Islands, in the South Atlantic, to Alaska. Rays reaching the Eielson array (ILAR), about 150° away (distance measured as the angle at the Earth's centre; the far side is 180°), go deep into the inner core; rays reaching Yellowknife (YKA) in Canada, 135–137° away, graze its top.^6,43 PKIKP takes about 20 minutes to reach the far side of the Earth, on the standard ak135 travel-time tables.⁴⁴

No two earthquakes are alike, so seismologists wait for one to repeat: the same patch of fault breaking years later and sending out nearly identical waves, a doublet. If the waves that do not cross the inner core match and the one that crossed it has changed, the change happened inside. In 2005 Jian Zhang and colleagues used 18 doublets up to 35 years apart, recorded at up to 58 stations in and near Alaska, and confirmed that the inner-core waves change with time.⁴⁵ The changes are tiny: "a change of about 0.01 s per year", as Vidale and colleagues put it in 2000, reaching tenths of a second over decades.^43,46

In 2024 Wang, Vidale and colleagues used 121 earthquakes in 42 locations, 1991–2023, forming 143 pairs. Of the earthquakes, 109 came from published lists of repeaters, chosen by a cross-correlation (a match score, defined below) of 0.90 or 0.95 or more on the arrivals, or phases, that do not cross the inner core; the 12 added from 2021 to January 2023 passed a median above 0.95 at both arrays over 15 seconds.⁶ A 2025 paper extended the set to 168 pairs, 1991–2024.⁴³ The 2024 study then asked whether a later earthquake's PKIKP matches an earlier one's. If it does, the inner core is back where it was at the earlier date, relative to the mantle. The match is scored by the cross-correlation of the two seismograms:

CC(s) = Σ E(t) L(t − s) / √[Σ E² · Σ L²]E, L: the earlier and later seismograms in a window; s: a trial shift. CC = 1 is a perfect match. The repeater test applies only to the windows that do not cross the inner core.

A doctor with a stethoscope listens for a change in a sound that repeats. Here the repeating sound is an earthquake on the same patch of fault, and the change is in the wave that crossed the inner core. The picture breaks in one place: nobody chooses when the next beat comes. Seismologists wait, sometimes decades, for a fault to break again. Figure 2 makes the comparison with seismograms drawn by this page: slide the later quake until its inner-core wave lines up, and watch the other window.

Figure 2🎲 Try it

The stethoscope

Slide the later quake

Repeater test: outer-core window, CC at 0 ms (0.90–0.95 or more in the 2024 study)Same patch of fault? Waves that don't go through the inner core (1 = a perfect match)

Inner-core window: CC at your shiftInner-core wave: how well they match (1 = a perfect match)

Inner-core window: size, later ÷ earlierInner-core wave: size now (1 = the same)

Your shiftYou moved it

What changed: best-fit shiftHow far it moved

Synthetic seismograms, illustrative only: Ricker wavelets (a standard test pulse) at 1.5 Hz with weak noise, built to mimic the published pairs, not copies (the real traces are in Wang et al. 2024 and Vidale et al. 2025). Each pair is two earthquakes on the same patch of fault near the South Sandwich Islands, recorded at Eielson, Alaska (rays deep in the inner core) or Yellowknife, Canada (rays grazing its top). The later trace is aligned on the phases that do not cross the inner core (PKiKP and PKP), and their correlation is Chapter 7's repeater test, over shorter windows than the 2024 study's 15 seconds. The inner-core window is compared at your shift; the best fit maximises its correlation within ±300 ms. Shift sizes follow the published scale, about 0.01 s a year and tenths of a second over decades (Vidale et al. 2000; 2025). Pairs: a large change between the 1990s and after 2010; matches 2003–2023 and 2006–2013, read as the inner core back in the same position relative to the mantle; and 2004–2017, a change at Yellowknife but not Eielson that rotation cannot explain (Vidale et al. 2025). This page's model.Two earthquakes, years apart, from the same patch of fault. The waves that don't go through the inner core look nearly identical, so it really is a repeat. Slide the later quake until its inner-core wave lines up with the earlier one: the shift you need is the change inside the Earth. Try the matching pairs too: when the waves match again, the inner core is back where it was. These wiggles are drawn by this page, not real recordings.

In the first pair no single shift lines up both windows: the fingerprint of a change inside. The two matching pairs show the 2024 kind of result: years apart, no change beyond the noise. The last pair belongs to Chapter 11.

Chapter 8A degree a year, then a tenth of that

The differential rotation was predicted before it was seen. In 1995 Gary Glatzmaier and Paul Roberts's computer model of the geodynamo had the inner core super-rotating, dragged eastward relative to the mantle by an eastward flow in the outer core that pulls on it through the magnetic field, a mechanism their December 1996 paper calls "analogous to a synchronous motor".⁴⁷ In July 1996 Song and Richards reported that on the South Sandwich Islands–Alaska path the separation of PKP(BC) and PKP(DF) had changed by about 0.3 s over about 30 years (the figures are Richards's later summary). They read this as rotation "on the order of 1° per year faster than the daily rotation of the mantle and crust".⁴² Su, Dziewonski and Jeanloz found about 3° a year the same year.⁴⁸

Then the estimates came down. Free oscillations (the normal modes of Chapter 2), which average over the whole inner core, gave near zero.⁴⁹ Waves scattered inside the inner core from two Soviet nuclear tests at Novaya Zemlya, in 1971 and 1974, gave about 0.15° a year in 2000;⁴⁶ Vidale's 2019 re-analysis, "halving our rotation rate estimate", gave 0.07° a year,⁵⁰ and Wang and Vidale's of 2022 gave 0.10° a year about the Earth's axis, 0.29° in all over 1971–1974.⁵¹ Doublets gave more: 0.3–0.5° a year in 2005,⁴⁵ and in 2013 a "shuffling" mean of 0.25–0.48° a year over 1961–2007, with decadal swings of order 1° a year.⁵² By 2024 the estimates had narrowed, in Wang and colleagues' summary, to 0.05–0.15° a year.⁶ Each is that study's interpretation; none is a single "true" rate.

None of these numbers is the inner core's spin. The inner core and the mantle each turn once a day. Counted against the stars, that is 366.25 turns in a year of 365.25 days: the Earth's trip round the Sun uses up one turn a year, so there is one more turn than there are days (this page's arithmetic). Only the tiny difference between the inner core's turning and the mantle's is in question:

Δφ = ∫ Δω dt, s = (2πR/360°) Δφ = 21.3 km per degreeΔω: the inner core's rotation rate minus the mantle's; Δφ: the angle gained on the mantle; s: the slide of a point on the inner core's equator. The daily spin itself carries that point round at ΩR ≈ 89 m/s (this page's arithmetic).

At 0.1° a year, a point on the inner core's equator slides about 2.1 km a year relative to the mantle, some two million times faster than the inner core grows. A full lap relative to the mantle would take about 3,600 years, or 2,400–7,200 years across 0.05–0.15° a year, if the rate held (this page's arithmetic). The record says it does not.

The long-term average may be tiny, or not. From frozen-in structure in the inner core that shifts eastward with depth, Waszek, Irving and Deuss inferred an "extremely slow steady inner core super-rotation of 0.1°–1° per million years".² Zhang and Dumberry's 2026 analysis implies a steady 0.12° a year instead, close to three full turns relative to the mantle in 9,000 years.⁵³ The two refereed answers, each one model's inference, are a factor of 10⁵ to 10⁶ apart.

Picture two runners lapping a track once a day, one edging a fraction of a stride ahead, then dropping back. Neither stops, and the one behind is not running backwards.

Chapter 9A pause, and an argument about clocks

In January 2023 Yi Yang and Xiaodong Song reported that paths whose inner-core waves had changed for decades had barely changed over the previous decade: the rotation relative to the mantle had paused, around 2009–2010 (the paper divides its doublets at 2009). They read the record since 1964 as part of an oscillation of about seven decades, with an earlier turning point in the early 1970s, matching multidecadal changes in the length of the day and in the magnetic field.⁵ It is one group's model, not a discovered cycle.

This page sketches it. Yang and Song's angle, as plotted by Zhang and Dumberry, runs from about −3.7° in 1972 to about +0.95° in 2010, about 4.7° from trough to peak.⁵³ A sinusoid, the regular swing of a pendulum, turning at those dates is

Δω(t) = A sin[2π(t − 1972)/P], Δφ(t) = φ_mid − (AP/2π) cos[2π(t − 1972)/P]This page's sketch, not Yang and Song's model. φ_mid: the angle midway between trough and peak. Turning points 38 years apart give P ≈ 76 years; half the swing, 2.35°, gives a peak rate A ≈ 0.19° a year (0.21° a year for P = 70).

"About 70 years" and 76 are the same claim at the precision the source gives. At A = 0.20° a year (Figure 3's default) the sketch's mean rate from 1991 to 2010 is 0.127° a year, the same as Yang and Song's twin-station measurement for 1991–2010, which is disputed (below).⁵⁴ But the published curve falls only about 0.25–0.3° by 2019–2021, while a symmetric sinusoid falls about 0.6° by 2019 (this page's arithmetic): the motion is lopsided, as Chapter 10 shows another way.

Chapter 10Back the way it came

On 12 June 2024 Wang, Vidale, Pang, Koper and Wang reported in Nature which earlier earthquake each later one matched, for 121 repeating earthquakes between 1991 and 2023. The matches say that the inner core moved forward relative to the mantle from 2003 to 2008, then from 2008 to 2023 went "two to three times more slowly back through the same path": later waves repeated earlier ones in reverse order.⁶ The paper's Methods section puts the turn at 2008.5 ± 0.18 and the return about 2.5 times slower. The method fixes the order of the positions, so it gives the reversal and the ratio of the speeds, not the rate; that the first leg was eastward comes, in the authors' words, "together with previous studies". This page writes the fit as:

Δφ = v(t − 2003) before t_r, then falls at v/r; t₂ = t_r + r(t_r − t₁)t_r ≈ 2008.5, the turn; r ≈ 2.5, forward speed ÷ backward speed; v is not measured. A position held at t₁ before the turn recurs at t₂.

With 2008.5 and 2.5, the 2003 position recurs in about 2022.3 and the 2006 position in about 2014.8. The 2025 paper reports matches between 2003 and 2023 and between 2006 and 2013.⁴³ Taken as exact, those two pairs would need a turn near 2007.6 and a ratio near 3.3 (this page's arithmetic), so the matches are good to a year or two. The 2024 authors add that 2005–2015 "is more difficult to resolve" and interpret the rotation there as slowing before the lead turns into a lag.⁶

The asymmetry is the point. The motion is "more complicated than a symmetric function such as a sinusoid", the authors write, and will need new models.⁶ A symmetric swing turning in 2010 brings the 2003 position back in 2017, not 2023. Its authors call it "the most definitive evidence so far" that the inner core moves relative to the rest of the Earth, and it says nothing about a reversed spin: the inner core turned once a day throughout. Figure 3 sets six readings side by side, with a dial that keeps both bodies turning: switch between them and press Play, and the last tile finds the year each position comes round again.

Figure 3🎲 Try it

Ahead, then behind

Switch the reading

Peak rate AFastest rate of the swing

Year

Angle since 2003, relative to the mantleAhead (+) or behind (−) its 2003 spot

Rate relative to the mantleGaining (+) or losing (−) each year

Slid at its equator since 2003How far its equator slid

Own-spin effect on the day (this page's estimate)Change to the day (ms: thousandths of a second), from its own spin alone (this page's estimate)

Same position as inSame spot as in

Six readings of the inner core's rotation relative to the mantle, as angle (top) and rate (bottom), zero at each model's 2003 position. Song and Richards's 1996 steady rate is drawn only to 2000; Zhang et al.'s 2005 rate is steady. 'About 70 years' is this page's sketch of Yang and Song's 2023 model, Chapter 9's sinusoid (period 76 years, peak rate A). The 20–30-year swing uses Tkalčić's 2024 periods at the same peak rate; he gives no phase, so this page turns it at 2010, as the other readings do. The six-year oscillation is Wang and Vidale's 2022 model (0.5° peak to peak, less if part of the outer core moves with it; the 1971–74 tests give 0.29°, which the authors count as a match within uncertainties), which their 2024 data do not detect as the main motion. The backtracking line is Wang et al.'s 2024 fit (turn at 2008.5, return 2.5 times slower); its rate is not measured and is set here to 0.1° a year. Shaded years extrapolate the models; they are not published forecasts (the 1996 and 2005 rates are shaded after publication). Matches between years are approximate (± 1 year; ± 2 in the poorly resolved 2005–2015). Marks: Vidale 2019 and Wang and Vidale 2022 (1971–74); Yang and Song 2022 (1991–2010, disputed); Wu et al. 2026; reported matches 2003–2023 and 2006–2013 (Vidale et al. 2025). The 1971–74 and 1991–2010 marks give rates, as slopes; their height on the angle panel is not a measurement. The day tile assumes the inner core hands over only its own spin (Chapter 12): 0.054 ms per 0.1° a year. As of October 2026. This page's model.How far the inner core is ahead of (up) or behind (down) the spot it held in 2003, measured against the mantle, and below that whether it is gaining or losing each year, under six readings of the waves from earthquakes (and, for one, nuclear tests). Two curves are this page's sketches: the 70-year one, and the 20–30-year one, whose timing this page chose. The dial shows both turning together; only the gap between them changes, magnified so you can see it. Switch readings and press Play. The recent readings have it swinging between ahead and behind; they disagree on how fast and how regularly. With Data marks on, the two sloping lines show how fast a study found it turning, not where it was. Shaded years are guesses that carry each reading forward, not predictions.

Set side by side, the four readings published from 2022 on all change sign, and agree on little else; the steady rates of 1996 and 2005 never do. The 1996 rate, kept up, would have moved the inner core 30° in 30 years.

Chapter 11Turning, or changing shape?

The alternative to rotation is a changing surface. In 2006 Lianxing Wen studied a South Sandwich Islands doublet of 1 December 1993 and 6 September 2003, recorded at three stations in Russia and Kyrgyzstan: the reflection off the inner core, PKiKP, arrived earlier in 2003, by "at least from 39 to 70 milliseconds" in his words. He inferred that the radius had "enlarged locally beneath middle Africa by 0.98 to 1.75 kilometers", and left two causes open: "either a differential motion of the inner core", if its surface is bumpy, "or a rapid growth of the inner core by this amount".²⁸ A kilometre in a decade is about a hundred thousand times the mean growth rate (this page's arithmetic).

Wen's group argues that the surface alone explains the changes: localised, episodic changes over 1993–2013, including a drop in the surface of 5.59 km beneath Zimbabwe or Kenya, and one PKiKP change within 8–85 days that rotation could produce only at 8.6° a year or more relative to the mantle, so that rotation "cannot provide a consistent or reasonable interpretation".⁶¹ In September 2026 Xin Zhang and Wen, mapping the surface beneath eastern South America over 2003–2024 with one cluster of repeating earthquakes, found irregular episodes of "change, quiescence and recovery" and explained them without any rotation.⁶¹ Song's group argues the opposite: the changes come mostly from PKIKP and begin before PKiKP arrives, which rules out the surface as the only source, and the small PKiKP shifts reported since 2006, Wen's among them, in their view likely come from station clock errors and instrument changes.⁶² The two groups have exchanged comments and replies, most recently in 2024.^58,62

In February 2025 Vidale and colleagues found changes between 2004 and 2008 at Yellowknife, whose rays graze the top of the inner core, but not at Eielson, whose rays go deeper. Rotation would change both. Since only the grazing rays changed, the authors place the changes in the shallow inner core; they were gone by 2008–2009. The conclusion is "tentatively both" rotation and local change. The local change is most likely viscous deformation, the surface slowly flowing, pulled by dense regions of the mantle or dragged by the outer core's flow; the shallow inner core's viscosity, its resistance to flow, is "very roughly 10¹⁶ to 10¹⁷ Pa·s" at most.⁴³ Figure 2's last pair is a sketch of this kind of change.

A yolk in a spinning egg can turn slightly out of step with its shell. But a yolk is loose, while the inner core is held by gravity and magnetic forces, which is why it leads or lags by only fractions of a degree a year; and a yolk keeps its shape.

Where the argument stands (this page's summary). Everyone in this debate agrees the inner-core waves change with time. They disagree on how much is rotation and how much is a changing surface, on the cycle (6, 20–30 or about 70 years, or no regular cycle), and on the exact year the lead became a lag. The newest papers cited here still divide: Wu, Song and Yang (January 2026) favour a slow swing, Vidale and colleagues (2025) rotation with surface change, and Zhang and Wen (September 2026) surface change with no rotation at all.

Part IIIThe core and the day

Chapter 12The length of the day

If the inner core matters at the surface, it is through the length of the day (LOD):

LOD − 86,400 s = d(TAI − UT1)/dtt in days. TAI: International Atomic Time; UT1: the Earth's rotation angle as a time, measured since the 1980s by very-long-baseline interferometry (VLBI) and before that by optical astrometry. This page takes annual means from the monthly UT1 − UTC series EOP 20 C04 of the International Earth Rotation and Reference Systems Service (IERS), the array Nº 9 uses, with the leap seconds added back to turn UTC, the civil time scale, into atomic time.

On that record the day was about 3.1 ms longer than 86,400 s in 1972, 2.4 ms in 1993, 0.3 ms in 2003 and 1.3 ms in 2016, and slightly shorter from 2020 to 2025, −0.25 ms in 2022: a fall of 3.37 ms from 1972 to 2022 (this page's arithmetic from IERS data).⁶³ One millisecond a day is 11.6 parts per billion of the Earth's spin (Nº 16). The same record lies behind the decision that Nº 9 describes, to end leap seconds from 20 May 2027.

The long lengthening, 2.3 ms per century from tides alone and 1.8 ms per century observed over 2,700 years (Nº 9, Nº 16), has nothing to do with the core. The swings over decades come mainly from the core and the mantle exchanging spin.⁶⁴ Which torque, or twisting force, carries it is not settled: coupling at the core–mantle boundary, or, as Buffett proposed in 1996 and a 2026 study suggests from the data, the inner core's pull (Chapter 13).⁶⁵

On top sits a six-year oscillation. Richard Holme and Olivier de Viron characterised a 5.9-year oscillation of about 0.13 ms (0.127 ms in the authors' posted manuscript).⁶⁴ Two explanations compete. Buffett showed in another 1996 paper that gravitational coupling between the mantle and a slightly lopsided inner core could appear in the day, and Jon Mound and Buffett (2006) and Hao Ding and Benjamin Chao (2018) read the six-year signal that way;^65,66 Nicolas Gillet and colleagues found in 2010 that "a torsional wave recurring every six years" in the liquid outer core "accounts well for both the phase and the amplitude of the six-year signal", without the inner core.⁶⁷ In 2025 Mathieu Dumberry, in a theory paper, found that in a magnetic field of a few millitesla, the strength expected in the core, the inner core's gravitational oscillation merges into the outer core's torsional waves, so that the six-year signal "cannot be interpreted as the signature" of it.⁶⁸ The 2026 lever study finds the pull "inefficient at periods shorter than a few decades".⁵³ The latest theory therefore favours the torsional waves; no observation has settled it.

If the inner core handed over only its own spin to the mantle, the day would change by

ΔLOD = 86,400 s × (C_ic/C_m) × Δω/ΩC_ic = 5.86 × 10³⁴ kg m², the inner core's moment of inertia, from PREM; C_m = 7.12 × 10³⁷ kg m², that of the mantle and crust; ratio 8.2 × 10⁻⁴; Ω = 7.2921151467064 × 10⁻⁵ rad/s, 131,850° a year. Δω = 0.1° a year gives 0.054 ms, 0.62 parts per billion (this page's estimate).

That is about 0.11 ms at the 0.2° a year peak of this page's sketch of the 2023 reading, and 0.22 ms from the top of its swing to the bottom. The 3.37 ms fall of 1972–2022 is about 60 times the effect of 0.1° a year and about 16 times that whole swing. Carried by the inner core's own spin alone, the six-year oscillation would need about ±0.24° a year, ±0.22° in angle (this page's estimate), close to Wang and Vidale's ±0.25°;⁵⁶ if the liquid in the cylinder that touches the inner core's equator turned with it, the rate needed would fall two- or threefold.⁶ VLBI measures changes this small; Nº 16 shows how a ring laser feels the day change, and why VLBI keeps the long record.

Figure 4 draws the day since 1972 with the own-spin curve on the same axis, so you can see how small it is. Move the year to one you remember, then switch to the six-year oscillation, or to the lever of Chapter 13.

Figure 4🎲 Try it

The length of the day

Find your year

Year

Day in 1990, minus 86,400 sYour year's day was longer by

Spin slower by

Own-spin swing, peak to peak (its own spin handed over alone)If it only handed over its own spin: the whole swing, top to bottom

Observed 1972–2022 ÷ own-spin swingThe real swing is bigger by

The length of day minus 86,400 s, annual means 1972–2025, computed by this page from IERS EOP 20 C04 UT1 − UTC (Nº 9's monthly series, leap seconds restored). 'Its own spin' adds the change the inner core would make by handing only its own angular momentum to the mantle (C_ic/C_m = 8.2 × 10⁻⁴; 0.054 ms per 0.1° a year) while following Figure 3's sketch. 'The six-year oscillation' shows the series, with its slow changes filtered out by this page, and Holme and de Viron's 5.9-year oscillation (about 0.13 ms; 0.127 ms in the authors' posted manuscript) at this page's fitted phase; its two explanations are in Chapter 12. 'As a lever' follows Zhang and Dumberry 2026, who suggest the multidecadal swings are driven mainly by the inner core's gravitational pull on the mantle, the spin supplied by the outer core: the curve is smoothed by this page (Gaussian, half power at about 30 years) with the observed 1.8 ms per century removed, but not the atmospheric and oceanic parts the paper removes; the sliders give ΔLOD = T t LOD/(C_mΩ) for the time-varying part of the pull, which peaks at about 2 × 10¹⁷ N m in their fit (their Fig. 1b; the steady part is balanced, as the note under the slider says), treating that part as the whole net torque on the mantle. In the paper the torques at the core–mantle boundary resist it, so this is an upper-end estimate; their fit also takes Yang and Song's history as input (Chapter 13). Data to September 2026; as of October 2026. This page's arithmetic.How much longer than 86,400 seconds the average day has been, year by year, against atomic clocks. Move to your birth year. Switch views: on its own spin the inner core explains only a sliver of the changes; working as a lever, as a 2026 study suggests, its pull could move the day by milliseconds. Nobody has settled which is right.

On its own spin, the inner core is a footnote to the day; but it need not act alone.

Chapter 13A lever made of iron

In September 2026 Huifeng Zhang and Mathieu Dumberry combined Yang and Song's rotation history with outer-core flows derived from the magnetic field's slow changes, for 1964–2019, and asked what drives the day's multidecadal changes, those with periods over 30 years. Their curve falls about 3 ms, from +1.4 ms in the early 1970s to −1.7 ms in 2019, after removing the effects of the atmosphere, the oceans and the long trends and applying a 30-year filter. Their results "suggest that the multidecadal LOD changes are driven primarily by the gravitational torque" of the inner core, "resisted by electromagnetic and topographic torques" at the core–mantle boundary.⁵³

Buffett proposed the mechanism in 1996: the liquid core's electromagnetic pull turns the inner core, and gravity hands the spin on to the mantle over decades.⁶⁵ In the 2026 fit, flows inside the tangent cylinder, the cylinder of liquid that touches the inner core's equator, drag the inner core round. The inner core's shape, with 11–29 m of relief, is turned out of line with the mantle's denser and lighter regions (its mass anomalies) by a further "about 1°", on top of a permanent offset (below), "sufficient to provide the necessary gravitational torque on the mantle":

Γ_g = Γ α, Γ ≈ 1.3–1.5 × 10¹⁹ N m per radianα: the misalignment; Γ: a strength factor, not the torque (95% range 0.6–4.2 × 10¹⁹). The time-varying part of the torque peaks at about 2 × 10¹⁷ N m in their Fig. 1b (this page's check: at 1.3 × 10¹⁹ N m per radian, each degree of misalignment is worth 2.3 × 10¹⁷ N m).

Beneath the swings sits a permanent misalignment of 1.19°, whose steady torque, about 2.9 × 10¹⁷ N m by this page's estimate, balances the mean torque at the core–mantle boundary and does not change the day; the bulges flow back into line (relax) in 8–10 years at best fit (95% range 2–31).⁵³ Why can a lever do what the inner core's own spin cannot? This page's estimate:

ΔL = C_mΩ × ΔLOD/LOD ≈ 1.8 × 10²⁶ N m s for 3 ms, ΔLOD = T t × LOD/(C_mΩ)ΔL: the angular momentum the mantle must gain or lose to change the day by ΔLOD. A torque T on the mantle (here the time-varying part of Γ_g) held for a time t moves T t of it; LOD = 86,400 s.

To supply that from its own spin, the inner core would have to change its rate by about 5.6° a year, far beyond the 0.05–0.15° a year of the 2024 summary. A lever needs no such change. Held for a decade, 2 × 10¹⁷ N m moves 6.3 × 10²⁵ N m s and changes the day by 1.05 ms, an upper end, since the core–mantle torques resist the pull; the change in the day depends only on the torque and how long it acts. The spin comes from the outer core, which has about 155 times the inner core's moment of inertia, so its rate need change by only 0.036° a year to supply all 3 ms. Figure 4's lever view turns this arithmetic into two sliders, for the pull and its duration.

The caveats are the paper's. Its torque histories are "tied to the accuracy of the inner core rotation and core flow models". It takes Yang and Song's history as input, so it supports that history only if the model is right. And the abstract's starting point, that "the precise nature of the core–mantle torque remains unknown", still holds for the shorter swings: below a few decades the pull is inefficient, the paper's modelled boundary torques "do not contain large decadal fluctuations" either, and the paper says its models of the boundary coupling may "need refining". A commentary published with the paper in Nature (a News & Views) states the result more firmly;⁶⁹ this page keeps the paper's own word, "suggest".

So the inner core may move the day as a lever, not a flywheel: an idea from 1996, now backed by one refereed study of the data, a month old as of October 2026.

Chapter 14Out of step, not stopped

The inner core turns once a day, like the mantle and like you. Everything this page says about its rotation is a difference of a fraction of a degree a year: a lead that, in the 2023 and 2024 readings, became a lag around 2008 to 2010. Nº 16 ends with the Paris pendulum on the same question, "relative to what?"; here the answer is the mantle.

What the story does not mean:

Stopped spinning: It turns once a day throughout; only its lead on the mantle changed sign (Chapter 10).
Spinning backwards: A lag is not a reversed spin, any more than a runner a stride behind runs backwards (Chapter 8).
Slowing the Earth: The day's long lengthening is tidal; the core swings it both ways by milliseconds (Chapters 12 and 13).
A proven 70-year cycle: One group's model, against 20–30 years, six years and an asymmetric backtrack (Chapters 9 and 10).
Exactly a millimetre a year: An order of magnitude: 0.4–1.2 mm a year today for the most-cited ages (Chapter 5).
One giant crystal: A ball of many aligned iron crystals (Chapter 4).
Hotter than the Sun: About as hot as the Sun's surface, far cooler than its interior (Chapter 3).

Where things stand, as of 25 October 2026: the inner-core waves change with time; the 2024 matches are, in their authors' view, the strongest evidence that the lead became a lag; how much is rotation and how much surface, the period, the inner core's age and the lever are all argued.

This page's view

The headline number is the least argued thing on the page and the least measured: nobody disputes its order of magnitude, and nobody has measured it. "About a millimetre a year" is the right size for today's growth, 0.4–1.2 mm a year for the most-cited ages, and the lifetime average of a ball about a billion years old; the dispute over its age moves today's rate by a factor of a few.

The rotation is the opposite: measured, with difficulty, and argued in public. The 2024 matches are a different kind of evidence: a run of waveforms that come back in reverse order is harder to explain by local change than one waveform that changes, though Zhang and Wen (2026) describe a surface that changes and recovers with no rotation at all. What the matches say is modest: relative to the mantle, the inner core went one way for a few years and came back more slowly. It never stopped turning with the rest of us.

Glossary

Inner-core boundary, core–mantle boundary: The inner-core boundary, 1,221.5 km from the centre, and the core–mantle boundary, 3,480 km from it.
PREM: The Preliminary Reference Earth Model (1981): density, pressure and wave speeds at every depth, averaged over the sphere.
P and S waves: Compressional (P) and shear (S) seismic waves. Shear waves cannot cross a liquid.
PKIKP, PKiKP, PKP: PKIKP (PKP(DF)) crosses the inner core; PKiKP reflects off it; PKP(BC) and PKP(AB) turn above it.
Coda: The later wiggles after a wave's first arrival; for PKIKP, waves scattered inside the inner core.
Normal modes: The whole Earth ringing at fixed frequencies after a great earthquake.
Seismic anisotropy: Wave speed depending on direction; in the inner core, faster pole to pole.
hcp, bcc iron: Two ways of stacking iron atoms: in hexagonal layers (hcp), or in cubes with one atom at the centre (bcc).
Superionic: A solid in which light atoms diffuse almost like a liquid through a fixed lattice.
Innermost inner core: A central region, 250–700 km in radius, whose crystals seem aligned differently.
Latent heat: The heat released when a liquid freezes: 600–1,200 kJ/kg for iron at the inner-core boundary.
Geodynamo: The convecting liquid outer core that generates the Earth's magnetic field.
Thermal conductivity: How well a material carries heat; disputed for core iron, and with it the inner core's age.
Supercooling, nucleation paradox: Cooling below the melting point before crystals form; pure iron needs more than the core probably had.
Palaeomagnetism: The ancient magnetic field's record in rocks.
Doublet: Two earthquakes years apart on the same patch of fault, with nearly identical waveforms.
Cross-correlation: How well two waveforms match at a given shift, from −1 to 1.
Differential rotation; super-, sub-rotation: The inner core turning slightly faster or slower than the mantle; both turn once a day.
LOD, UT1: The length of the day against atomic time; UT1 is the Earth's rotation angle as a time.
VLBI: Very-long-baseline interferometry: radio telescopes timing quasars, which keeps the long UT1 record.
Torque: A twisting force, in newton metres (N m). Held for a time, it changes a body's spin: torque × time is the angular momentum moved.
Moment of inertia, angular momentum: A body's resistance to changes in its spin, and the spin it carries.
Gravitational coupling: The pull between the inner core's uneven shape and the mantle's mass anomalies.
Torsional waves: Oscillations in which cylinders of outer-core liquid twist about the axis, held by the magnetic field.

Sources

A. M. Dziewonski and D. L. Anderson, "Preliminary reference Earth model," Phys. Earth Planet. Inter. 25, 297–356 (1981), doi:10.1016/0031-9201(81)90046-7.
L. Waszek, J. Irving and A. Deuss, "Reconciling the hemispherical structure of Earth's inner core with its super-rotation," Nat. Geosci. 4, 264–267 (2011), doi:10.1038/ngeo1083.
J. A. Jacobs, "The Earth's inner core," Nature 172, 297–298 (1953), doi:10.1038/172297a0.
L. Stixrude and R. E. Cohen, "High-pressure elasticity of iron and anisotropy of Earth's inner core," Science 267, 1972–1975 (1995), doi:10.1126/science.267.5206.1972.
Y. Yang and X. Song, "Multidecadal variation of the Earth's inner-core rotation," Nat. Geosci. 16, 182–187 (2023), doi:10.1038/s41561-022-01112-z; Author Correction, Nat. Geosci. 16, 189 (2023), doi:10.1038/s41561-023-01136-z.
W. Wang et al., "Inner core backtracking by seismic waveform change reversals," Nature 631, 340–343 (2024), doi:10.1038/s41586-024-07536-4.
I. Lehmann, "P′," Publications du Bureau Central Séismologique International, Série A, Travaux Scientifiques 14, 87–115 (1936) (a bureau publication, without a DOI).
A. M. Dziewonski and F. Gilbert, "Solidity of the inner core of the Earth inferred from normal mode observations," Nature 234, 465–466 (1971), doi:10.1038/234465a0.
A. Deuss et al., "The observation of inner core shear waves," Geophys. J. Int. 142, 67–73 (2000), doi:10.1046/j.1365-246x.2000.00147.x.
NASA Space Science Data Coordinated Archive, "Moon fact sheet," nssdc.gsfc.nasa.gov.
F. Nimmo et al., "Mean radius and shape of Pluto and Charon from New Horizons images," Icarus 287, 12–29 (2017), doi:10.1016/j.icarus.2016.06.027.
S. Anzellini et al., "Melting of iron at Earth's inner core boundary based on fast X-ray diffraction," Science 340, 464–466 (2013), doi:10.1126/science.1233514.
R. Deguen, T. Alboussière and D. Brito, "On the existence and structure of a mush at the inner core boundary of the Earth," Phys. Earth Planet. Inter. 164, 36–49 (2007), doi:10.1016/j.pepi.2007.05.003, arXiv:0801.2897.
A. J. Wilson et al., "Constraining Earth's core composition from inner core nucleation," Nat. Commun. 16, 7685 (2025), doi:10.1038/s41467-025-62841-4; Author Correction, Nat. Commun. 16, 10318 (2025), doi:10.1038/s41467-025-66094-z.
A. Prša et al., "Nominal values for selected solar and planetary quantities: IAU 2015 Resolution B3," Astron. J. 152, 41 (2016), doi:10.3847/0004-6256/152/2/41.
H. Tkalčić, "Complex inner core of the Earth: the last frontier of global seismology," Rev. Geophys. 53, 59–94 (2015), doi:10.1002/2014RG000469 (review).
A. J. Wilson et al., "The formation and evolution of the Earth's inner core," Nat. Rev. Earth Environ. 6, 140–154 (2025), doi:10.1038/s43017-024-00639-6 (review).
Y. He et al., "Superionic iron alloys and their seismic velocities in Earth's inner core," Nature 602, 258–262 (2022), doi:10.1038/s41586-021-04361-x.
G. Poupinet, R. Pillet and A. Souriau, "Possible heterogeneity of the Earth's core deduced from PKIKP travel times," Nature 305, 204–206 (1983), doi:10.1038/305204a0.
A. Morelli, A. M. Dziewonski and J. H. Woodhouse, "Anisotropy of the inner core inferred from PKIKP travel times," Geophys. Res. Lett. 13, 1545–1548 (1986), doi:10.1029/GL013i013p01545.
A. Deuss, "Heterogeneity and anisotropy of Earth's inner core," Annu. Rev. Earth Planet. Sci. 42, 103–126 (2014), doi:10.1146/annurev-earth-060313-054658 (review).
M. Ishii and A. M. Dziewoński, "The innermost inner core of the earth: evidence for a change in anisotropic behavior at the radius of about 300 km," Proc. Natl Acad. Sci. USA 99, 14026–14030 (2002), doi:10.1073/pnas.172508499.
T.-S. Phạm and H. Tkalčić, "Up-to-fivefold reverberating waves through the Earth's center and distinctly anisotropic innermost inner core," Nat. Commun. 14, 754 (2023), doi:10.1038/s41467-023-36074-2.
B. A. Buffett et al., "Analytical model for solidification of the Earth's core," Nature 356, 329–331 (1992), doi:10.1038/356329a0.
S. Labrosse, J.-P. Poirier and J.-L. Le Mouël, "The age of the inner core," Earth Planet. Sci. Lett. 190, 111–123 (2001), doi:10.1016/S0012-821X(01)00387-9.
M. Pozzo et al., "Thermal and electrical conductivity of iron at Earth's core conditions," Nature 485, 355–358 (2012), doi:10.1038/nature11031.
M. Monnereau et al., "Lopsided growth of Earth's inner core," Science 328, 1014–1017 (2010), doi:10.1126/science.1186212.
L. Wen, "Localized temporal change of the Earth's inner core boundary," Science 314, 967–970 (2006), doi:10.1126/science.1131692.
K. Ohta et al., "Experimental determination of the electrical resistivity of iron at Earth's core conditions," Nature 534, 95–98 (2016), doi:10.1038/nature17957.
S. Labrosse, "Thermal evolution of the core with a high thermal conductivity," Phys. Earth Planet. Inter. 247, 36–55 (2015), doi:10.1016/j.pepi.2015.02.002.
Y. Zhang et al., "Reconciliation of experiments and theory on transport properties of iron and the geodynamo," Phys. Rev. Lett. 125, 078501 (2020), doi:10.1103/PhysRevLett.125.078501.
Z. Konôpková et al., "Direct measurement of thermal conductivity in solid iron at planetary core conditions," Nature 534, 99–101 (2016), doi:10.1038/nature18009.
W.-P. Hsieh et al., "Low thermal conductivity of iron-silicon alloys at Earth's core conditions with implications for the geodynamo," Nat. Commun. 11, 3332 (2020), doi:10.1038/s41467-020-17106-7.
J. A. Tarduno et al., "Geodynamo, solar wind, and magnetopause 3.4 to 3.45 billion years ago," Science 327, 1238–1240 (2010), doi:10.1126/science.1183445.
A. J. Biggin et al., "Palaeomagnetic field intensity variations suggest Mesoproterozoic inner-core nucleation," Nature 526, 245–248 (2015), doi:10.1038/nature15523.
R. K. Bono et al., "Young inner core inferred from Ediacaran ultra-low geomagnetic field intensity," Nat. Geosci. 12, 143–147 (2019), doi:10.1038/s41561-018-0288-0.
T. Zhou et al., "Early Cambrian renewal of the geodynamo and the origin of inner core structure," Nat. Commun. 13, 4161 (2022), doi:10.1038/s41467-022-31677-7.
L. Huguet et al., "Earth's inner core nucleation paradox," Earth Planet. Sci. Lett. 487, 9–20 (2018), doi:10.1016/j.epsl.2018.01.018.
M. Lasbleis, M. Kervazo and G. Choblet, "The fate of liquids trapped during the Earth's inner core growth," Geophys. Res. Lett. 47, e2019GL085654 (2020), doi:10.1029/2019GL085654, arXiv:1912.12258.
Y. Sun et al., "Two-step nucleation of the Earth's inner core," Proc. Natl Acad. Sci. USA 119, e2113059119 (2022), doi:10.1073/pnas.2113059119.
J. Zhai, L. Wei, C. Gao and Y. Sun, "Quantifying the effects of nickel on Earth's inner-core nucleation," Geophys. Res. Lett. 53, e2026GL124720 (2026), doi:10.1029/2026GL124720; preprint arXiv:2608.27139.
X. Song and P. G. Richards, "Seismological evidence for differential rotation of the Earth's inner core," Nature 382, 221–224 (1996), doi:10.1038/382221a0.
J. E. Vidale et al., "Annual-scale variability in both the rotation rate and near surface of Earth's inner core," Nat. Geosci. 18, 267–272 (2025), doi:10.1038/s41561-025-01642-2.
B. L. N. Kennett, E. R. Engdahl and R. Buland, "Constraints on seismic velocities in the Earth from traveltimes," Geophys. J. Int. 122, 108–124 (1995), doi:10.1111/j.1365-246x.1995.tb03540.x.
J. Zhang et al., "Inner core differential motion confirmed by earthquake waveform doublets," Science 309, 1357–1360 (2005), doi:10.1126/science.1113193.
J. E. Vidale, D. A. Dodge and P. S. Earle, "Slow differential rotation of the Earth's inner core indicated by temporal changes in scattering," Nature 405, 445–448 (2000), doi:10.1038/35013039.
G. A. Glatzmaier and P. H. Roberts, "A three-dimensional convective dynamo solution with rotating and finitely conducting inner core and mantle," Phys. Earth Planet. Inter. 91, 63–75 (1995), doi:10.1016/0031-9201(95)03049-3; G. A. Glatzmaier and P. H. Roberts, "Rotation and magnetism of Earth's inner core," Science 274, 1887–1891 (1996), doi:10.1126/science.274.5294.1887.
W.-j. Su, A. M. Dziewonski and R. Jeanloz, "Planet within a planet: rotation of the inner core of Earth," Science 274, 1883–1887 (1996), doi:10.1126/science.274.5294.1883.
G. Laske and G. Masters, "Limits on differential rotation of the inner core from an analysis of the Earth's free oscillations," Nature 402, 66–69 (1999), doi:10.1038/47011.
J. E. Vidale, "Very slow rotation of Earth's inner core from 1971 to 1974," Geophys. Res. Lett. 46, 9483–9488 (2019), doi:10.1029/2019GL083774.
W. Wang and J. E. Vidale, "Earth's inner core rotation, 1971 to 1974, illuminated by inner-core scattered waves," Earth Planet. Sci. Lett. 577, 117214 (2022), doi:10.1016/j.epsl.2021.117214.
H. Tkalčić et al., "The shuffling rotation of the Earth's inner core revealed by earthquake doublets," Nat. Geosci. 6, 497–502 (2013), doi:10.1038/ngeo1813.
H. Zhang and M. Dumberry, "Gravitational torque drives multidecadal variations in length of day," Nature 658, 119–124 (2026; online 23 September 2026, open access), doi:10.1038/s41586-026-10999-2.
Y. Yang and X. Song, "Inner core rotation captured by earthquake doublets and twin stations," Geophys. Res. Lett. 49, e2022GL098393 (2022), doi:10.1029/2022GL098393.
H. Tkalčić, "On the inner-core differential-rotation (un)resolvability from earthquake doublets: the traps of data selection," Geophys. Res. Lett. 51, e2023GL107043 (2024), doi:10.1029/2023GL107043.
W. Wang and J. E. Vidale, "Seismological observation of Earth's oscillating inner core," Sci. Adv. 8, eabm9916 (2022), doi:10.1126/sciadv.abm9916.
G. Pang and K. D. Koper, "Excitation of Earth's inner core rotational oscillation during 2001–2003 captured by earthquake doublets," Earth Planet. Sci. Lett. 584, 117504 (2022), doi:10.1016/j.epsl.2022.117504.
D. Tian and L. Wen, "Comment on 'Inner core rotation captured by earthquake doublets and twin stations' by Yang and Song," Geophys. Res. Lett. 50, e2023GL103173 (2023), doi:10.1029/2023GL103173.
K. Wu, X. Song and Y. Yang, "Pattern of inner-core differential rotation from long-term earthquake sequences and USArray network," Geophys. Res. Lett. 53, e2025GL119295 (2026), doi:10.1029/2025GL119295.
H. Lecomte et al., "Gravitational constraints on the Earth's inner core differential rotation," Geophys. Res. Lett. 50, e2023GL104790 (2023), doi:10.1029/2023GL104790.
J. Yao, L. Sun and L. Wen, "Two decades of temporal change of Earth's inner core boundary," J. Geophys. Res. Solid Earth 120, 6263–6283 (2015), doi:10.1002/2015JB012339; J. Yao et al., "Temporal change of seismic Earth's inner core phases: inner core differential rotation or temporal change of inner core surface?," J. Geophys. Res. Solid Earth 124, 6720–6736 (2019), doi:10.1029/2019JB017532; X. Zhang and L. Wen, "Episodic change, quiescence, and recovery of Earth's inner core surface," J. Geophys. Res. Solid Earth 131, e2026JB034595 (2026), doi:10.1029/2026JB034595.
Y. Yang and X. Song, "Origin of temporal changes of inner-core seismic waves," Earth Planet. Sci. Lett. 541, 116267 (2020), doi:10.1016/j.epsl.2020.116267; J. Yao et al., "Comment on 'Origin of temporal changes of inner-core seismic waves' by Yang and Song (2020)," Earth Planet. Sci. Lett. 553, 116640 (2021), doi:10.1016/j.epsl.2020.116640; Y. Yang and X. Song, "Reply to Yao et al.'s comment on 'Origin of temporal changes of inner-core seismic waves'," Earth Planet. Sci. Lett. 553, 116639 (2021), doi:10.1016/j.epsl.2020.116639; Y. Yang, X. Song and A. T. Ringler, "An evaluation of the timing accuracy of global and regional seismic stations and networks," Seismol. Res. Lett. 93, 161–172 (2022; online November 2021), doi:10.1785/0220210232; X. Zhang and L. Wen, Comment on Yang et al. (2021), Seismol. Res. Lett. 96, 595–610 (2025; online 4 September 2024), doi:10.1785/0220230292; Y. Yang, X. Song and A. T. Ringler, Reply to Zhang and Wen, Seismol. Res. Lett. 96, 611–619 (2025; online 4 September 2024), doi:10.1785/0220240004.
IERS Earth Orientation Centre, EOP 20 C04 Earth-orientation series, monthly UT1 − UTC (read from the array embedded in Nº 9, 24 October 2026), iers.org.
R. Holme and O. de Viron, "Characterization and implications of intradecadal variations in length of day," Nature 499, 202–204 (2013), doi:10.1038/nature12282.
B. A. Buffett, "Gravitational oscillations in the length of day," Geophys. Res. Lett. 23, 2279–2282 (1996), doi:10.1029/96GL02083; B. A. Buffett, "A mechanism for decade fluctuations in the length of day," Geophys. Res. Lett. 23, 3803–3806 (1996), doi:10.1029/96GL03571; J. E. Mound and B. A. Buffett, "Detection of a gravitational oscillation in length-of-day," Earth Planet. Sci. Lett. 243, 383–389 (2006), doi:10.1016/j.epsl.2006.01.043.
H. Ding and B. F. Chao, "A 6-year westward rotary motion in the Earth: detection and possible MICG coupling mechanism," Earth Planet. Sci. Lett. 495, 50–55 (2018), doi:10.1016/j.epsl.2018.05.009.
N. Gillet et al., "Fast torsional waves and strong magnetic field within the Earth's core," Nature 465, 74–77 (2010), doi:10.1038/nature09010.
M. Dumberry, "The mantle-inner core gravitational mode of oscillation in a strong magnetic field regime," J. Stud. Earth's Deep Interior 1 (2025; online 26 September 2025), doi:10.46298/jsedi.15735, arXiv:2505.21271.
D. Jault and P. Personnettaz, "Gravitational forces inside Earth drag some decades out — and make others fly by," Nature 658, 37–38 (2026; News & Views), doi:10.1038/d41586-026-02773-1.

1 millimetre a year

A ball at the centre of the Earth

A core inside the core

About as hot as the Sun's surface, and solid

What "crystal" means here

Freezing from the middle out

How old is it?

A stethoscope made of earthquakes

Two runners on one track

A pause, and an argument about clocks

Back the way it came

Turning, or changing shape?

The length of your day

A lever made of iron

Out of step, not stopped

One millimetre a year