The Core Is Leaking: Ruthenium-100, Hawaiian Lava, and Gold From the Center of the Earth

Nine Parts Per Million

Start with the rock.

November 1959. Lava floods Kīlauea Iki, a pit crater on the flank of Kīlauea, until the pond stands more than a hundred metres deep and the fountains feeding it throw molten rock 580 metres above the rim. The lake crusts over. It cools for decades. Drill crews from the United States Geological Survey go back into it again and again, because a body of lava with a known start date and a known cooling history is as close as volcanology ever gets to a controlled experiment.

A chip of that rock ends up, decades later, in a clean lab in Göttingen with a filtered air supply. Someone dissolves it in acid. Someone spends weeks on column chemistry pulling out the ruthenium, a silvery and absurdly stubborn platinum-group metal that makes up roughly five parts per billion of the mantle. Someone loads the purified result into a mass spectrometer and waits while the ion beam settles. Then someone runs it again, and again, and again. One run cannot see the thing they are hunting.

The question is narrow. Does this rock carry slightly too much ruthenium-100, measured against its neighbour isotope ruthenium-101, compared with every other rock anybody has ever put through a mass spectrometer?

It does. By about nine parts per million.

Nine parts per million is nine atoms in a million. Count out a million ruthenium atoms one at a time, keeping a tally of which isotope each one is, and you would have to reach the end of the count before noticing that nine of them had landed in the wrong column. A number that size usually is noise.

But nothing that happens inside a planet, at any depth or any temperature the planet can reach, makes ruthenium-100. Radioactive decay does not make it. Melting and weathering do not make it, and neither does a slab sitting in a subduction zone for an age, cooked and squeezed and cooked again. Its abundance was fixed in dying stars before the Sun existed, and the one way to change it in a rock is to mix in material from somewhere that started out with a different mixture, made under a different set of stars. Exactly one reservoir on Earth is rich enough in ruthenium, and isolated enough from the rocks above it, to carry a different mixture of isotopes into a lava flow. Only the one.

The core was supposed to be finished talking to us 4.5 billion years ago. Nine parts per million is a rumour that it never stopped.

The Sealed Room

To understand why a nine-part-per-million excess is startling, you need the standard story, and the standard story is one of the most confidently taught narratives in all of Earth science. So a wobble in it repays attention.

The assembling Earth ran hot enough to be substantially molten over much of its volume, and in a molten planet iron does the one thing iron reliably does under gravity. It sinks. Over something like the first thirty to fifty million years of Solar System history, metallic iron and nickel drained downward through the silicate magma and pooled at the centre, forming a core that today reaches from 2,891 kilometres depth to the middle of the planet and holds about 32% of Earth's mass [12]. The draining metal took a great deal of other chemistry down with it. Elements that prefer metal to rock, which geochemists call siderophile, meaning iron-loving, were scrubbed out of the silicate portion of the planet with an efficiency that is hard to overstate. Gold, platinum, ruthenium, iridium, osmium, rhenium, tungsten. Nearly all of it went down with the iron and stayed.

The numbers are genuinely lopsided. Estimates of core composition put ruthenium at roughly 4,000 parts per billion and gold at roughly 500 parts per billion, while the silicate Earth above is left with about 5 ppb ruthenium and 1.7 ppb gold [10, 11]. Around 99% of Earth's precious-metal inventory now sits where no drill will ever reach. Three million atmospheres of pressure down, and somewhere near 4,000 °C. Gravity does the holding.

And then the seal. The seal rests on thermodynamics, not on tradition. Core and lowermost mantle are in chemical contact only across a thin boundary layer, solid-state diffusion through that layer is glacially slow, convection above it never reaches down to dredge liquid metal, and the metal is denser than the rock by a factor of about two. Whatever mechanical mixing you care to dream up at that boundary, gravity has four and a half billion years in which to undo it, and gravity is patient.

The textbook conclusion followed. A conclusion, note, rather than a hedge dressed up as one. The core has been chemically isolated since the moment it finished forming. On that account the gold in your ring has two perfectly ordinary sources: the small residue left behind in the mantle when the metal drained away, and a "late veneer" of meteoritic material that arrived after the core had already closed, which is why the mantle has any platinum-group metals at all [4, 5].

Nine parts per million of ruthenium-100, measured in a rock you can pick up and turn over in your hand, is a crack in that room.

How to Weigh One Atom Against Its Twin

Follow the procedure. It matters, because the entire result lives or dies on whether the measurement is real, and the measurement is at the edge of what instruments can do.

Step one: define your units honestly. You never report an isotope ratio raw, but always as a deviation from a standard, because the absolute number carries an instrument bias that no amount of care will remove. For ruthenium the convention is epsilon notation, parts per ten thousand:

So \(\varepsilon^{100}\mathrm{Ru} = +0.09\) means the sample is 0.09 parts in ten thousand rich in ruthenium-100, which is nine parts per million. Hold on to that conversion. Both units are in use, often in the same paragraph of the same paper, and the conversion between them is the single most common place in this field to lose a factor of a hundred.

Step two: dissolve, and separate. Ruthenium is present in the rock at five parts per billion, which is five atoms in every thousand million. To measure its isotope ratios to a precision of one part in ten thousand you first have to get it away from everything else in the rock, because the mass spectrometer cannot tell ruthenium-100 from molybdenum-100 by mass alone. Weeks of column chemistry go into each sample, in a lab where the blank, meaning the ruthenium contamination already sitting in your own reagents, has to come in smaller than the signal you are hunting.

Step three: accept that one measurement is not enough. Here is the number that made us respect this paper. Across 72 measurements of the reference material OREAS 684, the external reproducibility quoted for \(\varepsilon^{100}\mathrm{Ru}\) comes to ±0.13 (2σ) [1], while the signal being chased is +0.09. The signal-to-noise ratio on a single run is 0.69. The noise is bigger than the thing. One measurement, taken alone, cannot see this at all.

What can see it is the mean of many. The standard error on a mean shrinks as \(1/\sqrt{n}\), so with enough repeats a ±0.13 single-run scatter collapses to a ±0.03 uncertainty on the population mean. Averaging down to ±0.03 is legitimate. The averaging does fix what kind of claim this is, and the distinction is worth being pedantic about. The excess belongs to a population of Hawaiian rocks, not to any one of them, and it was earned by repetition rather than seen in a single spectrum. Nobody photographed a core atom.

Two Clocks, Set Independently

One isotope anomaly is a curiosity. Two anomalies in different elements, measured by different methods and pointing at the same answer, is an argument.

Ruthenium is the sensitive instrument. The core is roughly 800 times richer in ruthenium than the mantle, so a trace of core metal completely dominates the ruthenium isotope budget of whatever it is mixed into, however little of it there is. A 0.07% addition moves the ratio by nine parts per million. The mixing is weighted by concentration rather than by mass, which is the whole trick of the method.

Tungsten is the blunt instrument, and that is precisely its value. The core is only about 36 times richer in tungsten than the mantle, a contrast twenty times weaker than ruthenium's. Tungsten's anomaly is radiogenic, produced by the decay of hafnium-182 to tungsten-182, a clock with a half-life of 8.9 million years that ran down long before the first ocean. Core formation stripped tungsten downward and left hafnium behind, so the mantle went on making tungsten-182 for tens of millions of years after the core had stopped. Against the silicate Earth, the core runs 200 parts per million short in tungsten-182 [9].

So: add core metal to a mantle source, and you should see a ruthenium-100 excess and a tungsten-182 deficit, in a fixed ratio set by two concentration contrasts that were established independently, 4.5 billion years ago, by two different chemical processes.

The prediction can fail, which is what separates this paper from the twenty years of tungsten-only work behind it, where a deficit could always be explained some other way [2, 3, 7]. Tungsten deficits in these rocks have been known since 2017, and they have been argued over ever since. A rival explanation has stood beside them the whole time (§8). Ruthenium excesses had not been reported in hotspot lavas at all. Finding both together, in the proportion the arithmetic demands, is new.

Bench Notes: We Did the Arithmetic Ourselves

We began, as everyone does, by averaging two numbers that had no business being averaged. Of course we did. If the core is at \(\varepsilon = +0.25\) and the mantle is at \(0.00\), and the measured rock is at \(+0.09\), then the rock is 36% core, right? We wrote it on the whiteboard. Then somebody said it out loud: a third of Hawaiʻi is liquid iron?

Isotope ratios do not mix by mass. They mix weighted by how much of the element in question each component brings to the mixture. A component that is 800 times richer in ruthenium gets 800 times the vote. Written properly, with \(f\) the mass fraction of core, \(C\) the ruthenium concentrations and \(\varepsilon\) the compositions:

and rearranged to give the thing we actually want:

A second-year student can check that algebra by hand. It drives every number in this article. Plug in \(\varepsilon_{\text{mix}}=0.09\), \(\varepsilon_c=0.25\), \(\varepsilon_m=0\), \(C_c=4000\) ppb, \(C_m=5\) ppb and you get \(f = 7.03\times10^{-4}\). Not 36%. 0.070%. One part in 1,423.

Then the honest part. Working out what we did not know took longer than working out the equation. Four of the five inputs are uncertain, and \(f\) is a ratio of sums rather than a simple product, so the uncertainty that comes out the far end is skewed. Adding percentages in quadrature would have given the wrong shape. We ran 200,000 seeded Monte Carlo draws instead, and the seed sits in the linked script, so anyone can rerun this and land on our digits. The measured excess went in as a normal distribution about \(+0.09\). The core composition went in as a uniform distribution over \(+0.15\) to \(+0.35\), because we genuinely do not know it better than "somewhere in that range." The core ruthenium concentration carried ±25%, the mantle one ±20%.

The answer: median \(f = 7.1\times10^{-4}\), or 0.071% core by mass. The 68% interval runs 0.044–0.129%, the 95% interval 0.029–0.240%. No draws had to be thrown out as unphysical. The paper's own modelling puts an upper bound below 0.25% for bulk core entrainment [1]. Our median sits an order of magnitude under that ceiling, and our 95% upper edge nearly touches it, which is about what you would hope for from a model this crude. Another way to hold 0.071%: weigh out a kilogram of the Hawaiian source rock and the core's share of it is seven tenths of a gram.

Then we overreached. We took our ruthenium-derived \(f\) and pushed it through the same mixing equation using tungsten's numbers (core 470 ppb W at μ¹⁸²W = −220 ± 30 ppm, mantle 13 ppb at 0) with no refitting of any kind. The prediction came out at −5.5 ppm, 68% interval −9.9 to −3.4. Measured Hawaiian values run about −5 to −12 [2].

Nobody in the room expected that to work. That was the moment the paper stopped being a headline and became a result.

The Ledger

Plain numbers now. No rhetoric. Every quantity below is an input we chose or an output the script printed; the provenance column says which, and the last column says how much we trust it.

The gold flux is one multiplication:

volume flux times density times core fraction times gold concentration in the core. Nothing clever.

Read line R5 again. Three hundred thousand tonnes of core metal a year is a cube of iron 33.8 metres on a side, arriving under Hawaiʻi every twelve months, and again every twelve months after that. Now look at R6, and notice how small the gold number is when you set it beside the iron. 159 kg. Gold runs at only 500 parts per billion even in the core, so that whole block of iron carries a lump of gold roughly twenty centimetres on a side.

Line R10 is the one that keeps the story honest. Even in the leakiest rocks measured, roughly 83% of the gold is ordinary mantle gold. The core is a minority shareholder.

Things We Kept Asking Each Other

The Steelman for "No"

Suppose you are the reviewer who wants it wrong. The job is to find a version of the world in which every measurement in the paper is correct, the chemistry is clean, and the conclusion is still wrong. Being contrary is easy and worth nothing. Here is the best case we can build against it.

Objection one: the highly siderophile elements do not add up. The strongest objection, and it comes from inside the paper itself. If you dump 0.07% core metal into a mantle source, you are dumping in a lot of platinum, iridium, osmium and rhenium along with the ruthenium, because the core is enormously enriched in all of them. The most isotopically anomalous samples should therefore be visibly enriched in the highly siderophile elements as well. Simple entrainment overpredicts by a factor of 2.5 in the most depleted samples [1]. The authors' answer is that magmatic processes obscure the source-level signal on its way up. Sulfide saturation and crystal fractionation both sit between a source and a lava flow, as does the manner in which the melt was drawn off in the first place. That answer is plausible. It also, unavoidably, explains away an inconvenient prediction.

Objection two: tungsten has a rival, and it has had one since 2017. Negative μ¹⁸²W does not require the core. An early silicate reservoir with a low hafnium/tungsten ratio (a magma-ocean cumulate that crystallised in the first tens of millions of years and was never remixed afterwards) would also be deficient in tungsten-182, because it never got its share of the hafnium-182 that was still decaying [6, 7]. Worse for the core hypothesis: metallic iron droplets produced by disproportionation during deep magma-ocean crystallisation would carry moderate tungsten but very low highly siderophile element abundances, which fits observation better than bulk core does [6]. That model has been on the table for years. Nobody has killed it.

Objection three: the sample set is small. Forrest Horton, a geochemist at Woods Hole not involved in the work, called the sample size "a little tentative" while acknowledging the analytical rigour behind it, and he is right on both counts [19]. The number of Hawaiian samples carrying a resolvable ruthenium excess is not large, and the signal is defined by a population mean rather than by individually significant measurements.

Objection four is ours, not the literature's. Our own model could be badly wrong and we would never know. Run the sensitivity. If the core's true \(\varepsilon^{100}\mathrm{Ru}\) is +0.15 rather than +0.25, our \(f\) jumps from 0.070% to 0.187%, and every gold number downstream jumps with it by that same factor of 2.7. We have no independent handle on that input at all. Nobody does.

Now the rebuttal, because a steelman that is never answered is just a hedge. Objection two explains tungsten and says nothing about ruthenium. Crustal recycling and silicate differentiation contribute negligible ruthenium. You cannot make a ruthenium-100 excess by rearranging silicates [1]. Objection one is a quantitative discrepancy of a factor of 2.5 in a trace-element budget notorious for being scrambled during melting, not a contradiction in the isotope systematics. More samples will answer objection three, which makes it the good kind of objection to have. Objection four is real, and it is why every number of ours in the table carries a range.

What remains is an inference that is strong, not settled.

Forty Years of Almost

The idea did not arrive in 2025. It arrived in the 1980s, then sat. Four decades in the literature, never quite dying and never quite proving itself. An idea that arrives ahead of its instruments usually waits like that.

The first hints were osmium. In the 1980s and 1990s, workers measuring osmium isotopes in Hawaiian and Siberian lavas found ratios that were hard to make from ordinary mantle and that could, with some squinting, be produced by adding a little outer core. The debate that followed ran long and technical, across two decades of conference sessions, and settled nothing. By 2005 Brandon and Walker could publish a review titled, simply, The debate over core–mantle interaction, which is what you call a paper when the field has been arguing for twenty years without resolution [8]. The trouble was always that osmium had rivals available to it. Recycled oceanic crust could mimic the signal, and the rivals kept winning on parsimony.

Then helium. Ocean-island basalts from Hawaiʻi, Iceland and Samoa carry unusually high ³He/⁴He ratios, a marker of a primordial reservoir that has never been degassed. Whatever the plumes are tapping, it is old and it has been left alone. The helium did not point at the core specifically. It did establish that plumes reach something ancient.

Tungsten arrived in 2017. Mundl and colleagues found that ocean-island basalts with the highest ³He/⁴He also had the lowest μ¹⁸²W, down to 18 parts per million below the ambient mantle [2]. Two independent primordial markers, correlated. That was genuinely exciting, and it launched a decade of work on tungsten in deep-plume lavas [6, 7]. The magma-ocean rival sat right beside it the whole time, explaining the same data with no core at all.

Which brings us to the quotation that best captures where the field actually stood, in the words of the person who eventually moved it. "About 40 years ago," the paper's lead author told Nature's news desk, "people first came up with the theory that maybe the core is losing some material into the mantle, but the signals we got so far were really ambiguous" [18].

Forty years of ambiguous signals. Then an element that nobody had managed to measure precisely enough in hotspot lavas to settle anything.

How Would You Break It?

If this result is going to survive the next decade, it will do so by being attacked well and holding up. Here is what we would want to see, in rough order of how decisive it would be, from the cheapest test to the one that would settle the argument.

Test 1. Measure ruthenium in a hotspot that, on everyone's model of the mantle, should not be tapping the core. The core hypothesis says the signal should track plumes rooted at the core–mantle boundary, and only those. Iceland is deep-rooted, and so are Galápagos and La Réunion, all three plausibly tapping the base of the mantle. Plenty of other volcanic provinces are not. If ε¹⁰⁰Ru excesses turn up in shallow-sourced lavas too, the story collapses. If they stay confined to the deepest-rooted plumes, the case for a leak hardens substantially.

Test 2. Find the correlation, or fail to. If core entrainment is real, ε¹⁰⁰Ru and μ¹⁸²W should be correlated sample by sample, along a mixing line whose slope is fixed by the concentration ratios in Figure 2. A slope, with a value predicted in advance, measured rock by rock. "Both anomalous on average" is weaker, and weaker is where the field sits today. With enough samples the slope becomes a prediction sharp enough to fail. We would like that measurement made more than any other.

Test 3. Attack the highly siderophile element problem head-on. Objection one in §8 is a factor-of-2.5 mismatch. Either somebody demonstrates experimentally that sulfide saturation during Hawaiian melting can hide that much platinum-group enrichment, or the mismatch stands as a genuine mark against simple entrainment and in favour of the exotic alternatives, such as an oxide-rich layer at the top of the outer core, which would fractionate elements differently on the way out [1].

Test 4. Connect the isotopes to the seismology. The ultra-low-velocity zones sitting directly on the core–mantle boundary, and the two continent-sized low-velocity provinces beneath Africa and the Pacific that cover roughly 30% of that boundary [15, 19], are the obvious candidates for where the exchange happens. Deep plumes appear to rise from their edges [16]. Does the strength of the isotopic signal vary with which structure a given plume roots into? Nobody can currently draw that link.

Test 5. Do the physics. How does metal cross a boundary that gravity defends? Two candidate mechanisms are real enough to test: solid-state diffusion of siderophile elements into the lowermost mantle over geological time [14], and morphological instabilities that let molten iron penetrate upward into silicate along grain boundaries under the right wetting conditions [13]. Both are laboratory-scale results, awaiting a demonstration that anything seen in a press scales up to a planet. Until one of them does, the phrase "the core leaks" describes a signature in a mass spectrum, not a process.

What a Ring Is Made Of

Back to the rock, at a different scale.

A plain wedding band holds about four grams of gold, which is a number you can feel in your hand. The plume delivers that much to the crust every thirteen minutes. Call it forty thousand rings a year out of a single hotspot, made of atoms that spent four and a half billion years at the centre of the planet and then five hundred million more riding a column of hot rock upward.

None of that gold is in your ring. Almost certainly not. Ore deposits are built in the crust by hydrothermal fluids working on a mantle that has been stirred for aeons, and our own model says that even in the leakiest Hawaiian basalt only about 17% of the gold is core-derived. The honest statement is smaller and stranger than the headline: some fraction of the gold circulating in the world's economy came, originally, from a reservoir that geology spent a century describing as permanently closed.

And that is the part worth sitting with. The gold is the hook. The boundary is the finding. For decades the boundary was drawn in textbooks as a line: a clean discontinuity, the deepest full stop in Earth science, and nothing was thought to cross it. Nine parts per million turns that full stop into a membrane. Slow, selective, almost perfectly effective, and leaking.

Everything in §5, §6 and Figures 2–4 is the club's own model. Not the paper's analysis. Messling and colleagues did the chemistry, the measurement and the real error budget, which is the hard part. We did the arithmetic a high-school reader can check. Our assumptions are listed so that you can break them.

References