Main

Eoarchaean to Palaeoarchaean TTGs (dating back 4.0 billion years to 3.2 billion years, Gyr) form the earliest cratons1—the ancient nuclei of continents. Establishing how these TTGs formed is key to understanding how continental crust formed and stabilized. TTGs are variably deformed and metamorphosed felsic igneous rocks, rich in silica (average 69 wt%), aluminium oxide (around 15 wt%) and sodium oxide (3–7 wt%), and are characterized by low potassium oxide/sodium oxide ratios (averaging around 0.4) and high strontium/yttrium (Sr/Y) ratios (averaging about 40)4,10. Although there is consensus that TTGs formed by partial melting of hydrated basaltic rocks11,12, there remains considerable controversy surrounding the geodynamic setting, with implications for our understanding of early Earth evolution. Models for producing Archaean continental crust generally fall into two camps: those that invoke plate tectonics, and those that do not. Uniformitarian models, which involve plate tectonics and a dominance of horizontal forces, suggest that TTGs formed by partial melting of a subducting slab, as do modern-day adakites3,12,13. Non-uniformitarian models, in which vertical forces were dominant, instead propose that TTGs were produced near the base of thick, plateau-like mafic crust, whose formation was coupled to that of buoyant, subcontinental mantle lithosphere that was highly depleted in crust-forming components9,14,15,16.

TTGs have been classified into low-, medium- and high-pressure variants on the basis of partial melting experiments using amphibolite, metamorphic rocks dominated by amphibole (hornblende) and plagioclase10,17. Low-pressure TTGs (LP-TTGs) have relatively low Sr/Y ratios, consistent with melt residues that are rich in plagioclase (which sequesters strontium) but with little or no garnet (which sequesters yttrium and heavy rare-earth elements, HREEs); LP-TTGs are inferred to have formed at pressures of 1.0–1.2 GPa (at about 30–35 km depth)10. Medium-pressure TTGs (MP-TTGs) have moderate to high Sr/Y ratios and formed in equilibrium with residues that are rich in garnet, but with little or no rutile (which sequesters niobium (Nb) and tantalum (Ta)) or plagioclase; this is interpreted to reflect melting at a pressure of about 1.5 GPa (around 45 km depth)10. High-pressure TTGs (HP-TTGs) have high Sr/Y and lanthanum/niobium (La/Nb) ratios; such ratios are thought to reflect residues that are rich in garnet and rutile but lack plagioclase, and which are considered to have formed at pressures greater than 2.0 GPa (at about 60 km depth)10. Thus, the trace-element composition of TTGs can be used to infer the depth of melting of their source rocks, with implications for discriminating the geodynamic environment in which they formed18.

To help resolve the geodynamic provenance of early Archaean TTGs, we use new thermodynamic models that are applicable to mafic rocks19 to model the partial melting of a putative Palaeoarchaean TTG basaltic source. Phase equilibria modelling permits the calculation of the abundance and composition of minerals, volatile species and melt as a function of pressure, temperature and bulk rock composition. We couple this modelling with trace-element geochemistry to predict conditions of melting and to infer a likely geodynamic setting for the formation of TTGs and the generation of Earth’s first stable continents.

The East Pilbara Terrane, part of the Pilbara Craton in Western Australia, is among the oldest remnants of Earth’s Archaean continental nuclei20. It is volumetrically dominated by Palaeoarchaean TTGs whose compositions require source rocks that are much more enriched in thorium, large-ion lithophile elements (LILEs) and light rare-earth elements (LREEs) than are mid-ocean-ridge basalt (MORB) or many Archaean basalts4. However, 3.5-Gyr-old basalts and basaltic andesites within the Coucal Formation (hereafter Coucal basalts), at the base of the Pilbara Supergroup, have been shown to be suitable source rocks for the East Pilbara TTGs, at least in terms of their trace-element compositions, and may have formed a large proportion of a pre-3.5-Gyr-ago mafic protocrust that was 35 km or more in thickness4. Such rocks are not unique to the Pilbara—Palaeoarchaean TTGs and enriched basalts also occur in the Barberton greenstone belt in southern Africa, and probably formed in a similar setting21.

The Coucal basalts (Extended Data Table 1) form a tholeiitic series that is characterized by substantial enrichment in incompatible trace elements relative to enriched (E)-MORB, to normal (N)-MORB and to other Pilbara Supergroup basalts with similar magnesium oxide, chromium and nickel contents4 (Fig. 1a). Pronounced negative anomalies are evident for niobium (and tantalum), strontium and titanium. Although a third of the Coucal basalts show evidence for crustal contamination, most have low thorium/caesium (Th/Cs) ratios (averaging 0.05), Th/Nb ratios (averaging 0.18) and La/Nb ratios (averaging 1.5), within the ranges seen in basalts considered to be free of substantial crustal influence22,23 (Extended Data Fig. 1). The true basaltic members (with silica contents being less than 52 wt%) are not primary mantle melts (magnesium oxide = 6.1–3.1 wt%; Mg# (100Mg/(Mg + Fe2+) = 47–30). Given that their trace-element enrichments cannot be attributed to crustal assimilation or to fractional crystallization, it has been proposed that the Coucal basalts originated in a source that was already enriched, or not yet depleted, in crustal components4.

Figure 1: Trace-element geochemistry.
figure 1

a, Concentration of selected trace elements in uncontaminated Coucal basalts (green) and Palaeoarchaean East Pilbara TTGs (pale orange field, with average composition in darker orange; n = 15) relative to primitive mantle28. For reference, the compositions of N-MORB and E-MORB28 are also shown. b, c, log(La/Yb) versus log(Sr/Y) (b), and absolute Nb concentrations versus log(Sr/Y) (c) for Palaeoarchaean East Pilbara TTGs (grey dots), superimposed on smoothed kernel density estimates of Archaean TTGs and potassic granitoids worldwide10.

PowerPoint slide

Full size image

Palaeoarchaean TTGs (3.47–3.30 Gyr old; average silica content is 69 wt%; potassium oxide/sodium oxide ratio = 0.40; Sr/Y ratio = 35) from the East Pilbara Terrane (Fig. 1a and Extended Data Table 2) are the likely melt products derived from Coucal-like basalts. These TTGs have strongly fractionated trace-element compositions (that is, higher Th/Yb ratios and LREE/HREE ratios), but show depletions in niobium and titanium (but not strontium), rather like the Coucal basalts. Notably, almost all Palaeoarchaean East Pilbara TTGs have trace-element compositions that correspond to MP- and LP-TTGs (Fig. 1b, c)10.

Figure 2 shows a pressure (P)–temperature (T) phase diagram calculated for the average composition of uncontaminated Coucal basalts (n = 10), assuming an Fe3+/ΣFe of 0.1 (ref. 24) and a water content (4.7 mol%) that allows minimal water-saturated melting (see Extended Data Fig. 2 for the full phase diagram). Within the P–T window of interest (0.35–1.35 GPa, 630–1,000 °C), substantial melting begins with fluid-absent reactions that consume biotite, hornblende and quartz, producing around 15% melt as biotite is exhausted, and a total of 25–30% melt as hornblende (±quartz) is exhausted at 800–870 °C (Fig. 2). Above these temperatures, melting continues by consumption of anhydrous minerals, decreasing the water content of the melt. At relatively low pressures (yellow fields in Fig. 2), neither garnet nor rutile is predicted to be stable; at intermediate pressures (green fields), garnet is stable without rutile; at high pressures (blue fields), garnet and rutile co-exist. These fields correspond to the LP-TTG, MP-TTG and HP-TTG subdivisions of ref. 10, although plagioclase is predicted to be stable throughout the P–T window of interest, consistent with experiments on quartz-rich amphibolite starting compositions17. Notably, we calculate that garnet—which defines the lower pressure limit for MP-TTGs—is stable in the average Coucal basalt at pressures that are much lower (by up to 0.8 GPa) than suggested from experimental studies of melting in amphibolite10,17.

Figure 2: Phase equilibria modelling.
figure 2

Simplified P–T phase diagram for an average uncontaminated Coucal basalt (n = 10), using an Fe3+/ΣFe ratio of 0.1 (ref. 24) and a water content just sufficient to saturate the solidus at 1.0 GPa. The red dashed lines show the proportion of melt (mol% on a one-oxide basis, approximating vol%); at low melt fractions (<10%), the melts are granitic. For higher-melt fractions, the yellow, green and blue fields show the stability of LP-, MP- and HP-TTGs10.The white dashed lines represent linear geotherms (°C GPa−1), with the 900 °C GPa−1 geotherm emboldened. The white dots show the P–T conditions of markers that represent partially molten, hydrated basalt formed in non-uniformitarian tectonic settings in a numerical model of Archaean geodynamics9.

PowerPoint slide

Full size image

The apparent geothermal gradient recorded by post-Mesoarchaean metamorphic rocks (those that formed after 2.8 billion years ago) is a function of the geodynamic setting in which they formed (for example, through subduction or accretionary/collisional orogenesis)25. Superimposed on Fig. 2 are linear apparent geotherms ranging upwards from 500 °C GPa−1—values that encompass more than 95% of the data from Archaean metamorphic rocks worldwide25 (Extended Data Fig. 3). Also shown are the P–T conditions of markers that represent partially molten, hydrated basalt formed in non-uniformitarian tectonic settings in models of Archaean geodynamics9. These settings include delamination and dripping of the lower crust into the mantle2,15, local thickening of crust either side of upwelling mantle, and small-scale crustal overturns, all of which are predicted to produce substantial volumes of felsic (TTG) magmas9.

To illustrate the changing composition of melts and co-existing minerals, we model the prograde evolution of the average Coucal basalt along the 900 °C GPa−1 geotherm (thick dashed line in Fig. 2), for which all melts in excess of 5% are predicted to have formed in equilibrium with garnet but not with rutile (that is, MP-TTGs; Figs 2, 3a and Extended Data Table 3). Corresponding data for apparent geotherms of 700 °C GPa−1 and 1,100 °C GPa−1 are shown in Extended Data Figs 4, 5 and Extended Data Table 3. The calculated compositions of melts (normalized on an anhydrous basis) formed by 5%, 10%, 15%, 20%, 25%, 30% and 40% (mol%, which approximates vol%) equilibrium melting in a closed system are shown in Fig. 3b, along with the average normalized compositions of Archaean LP-, MP- and HP-TTGs and potassic rocks worldwide10 (see also Extended Data Table 4).

Figure 3: For the average Coucal basalt, a plot of calculated modal proportions for melting along an apparent geotherm of 900 °C GPa−1, and a comparison of calculated melt compositions with natural TTGs.
figure 3

a, Changing abundance (mode) of phases along an apparent geotherm of 900 °C GPa−1 for the average Coucal basalt. b, Calculated composition of melt produced from the average Coucal basalt at various melt fractions (5–40%), normalized to the average Palaeoarchaean East Pilbara TTG, along a geotherm of 900 °C GPa−1. The grey shaded region shows the 2σ envelope on the average Palaeoarchaean East Pilbara TTG data. Also shown are the global average compositions of TTGs and potassic granitoids10 (dashed lines). FeOT is the total Fe oxide content considering all Fe as Fe2+.

PowerPoint slide

Full size image

At low melt fractions (5–10%), the calculated melts are water-rich and granitic, with K2O/Na2O ratios greater than 0.8 (Fig. 2). At 20% melting, the calculated K2O/Na2O ratio (0.54) of the melt is similar to the upper range observed in TTGs, and at 30% melting this ratio (0.38) is very close to the average measured value in TTGs10. The major oxide composition of the average East Pilbara TTG shows remarkable correspondence with average LP- and MP-TTGs worldwide (Fig. 3b). The calculated composition of melt produced from the average Coucal basalt fits best with East Pilbara TTG compositions at a modelled melt fraction of about 25% (Fig. 3b), consistent with existing mass balance constraints5. When using published partition coefficients15,26,27 (Extended Data Table 5) and the predicted melt residua (Fig. 3a and Extended Data Table 3), trace-element modelling of the calculated melt compositions along the 900 °C GPa−1 geotherm generally shows good correspondence with the composition of the average East Pilbara TTG (Fig. 4). The fit is least good for zirconium, for which concentrations in the modelled melts are up to twice those of the average East Pilbara TTG. The reasons for this are not clear, but may reflect heterogeneities in the inferred source rocks—the Coucal basalts—in which zirconium contents vary by a factor of three (Extended Data Table 1).

Figure 4: Trace-element modelling.
figure 4

Normalized (against primitive mantle) concentration of selected trace elements in the average Palaeoarchaean East Pilbara TTG (orange), compared with the composition of modelled melts of the average Coucal basalt at various melt fractions (5–40%). The grey shaded region shows the 2σ uncertainties on the average East Pilbara TTG.

PowerPoint slide

Full size image

Despite the major-element composition of the average Coucal basalt being broadly similar to that of N-MORB, it contains only about half of the magnesium oxide (4.0 wt% versus 8.3 wt% on an anhydrous basis) and has an Mg# of 35—lower than that of any of the amphibolite starting materials used in the experiments that underpin the threefold baric subdivision of TTGs (Mg# = 71–38)10,17. Calculated phase equilibria show that the low Mg# results in the stabilization of garnet and rutile at pressures of 0.7–0.8 GPa and 1.1–1.2 GPa, respectively—much lower than in the experiments. These pressures, equivalent to depths of around 20–35 km in mafic crust (Fig. 2), correspond to the lower levels of the thick plateau-like crust that would be predicted if mantle temperatures were considerably warmer in the Archaean eon than at present2,6,7,15.

Our phase equilibria modelling demonstrates that the Palaeoarchaean East Pilbara TTGs could have been produced from 20–30% melting of a Coucal basaltic source along apparent geotherms of 700 °C GPa−1 or warmer (Fig. 2). Those TTGs with compositions corresponding to LP-TTGs (Fig. 1b, c) are consistent with apparent geotherms of greater than 1,100 °C GPa−1 (Fig. 2). These apparent geotherms are much warmer than those recorded in Neoproterozoic and younger metamorphic rocks from subduction zones (being less than 375 °C GPa−1; ref. 25), but match well with apparent geotherms recorded in other Archaean metamorphic rocks, most of which exceed 700 °C GPa−1 (ref. 25; Extended Data Fig. 3).

Major- and trace-element compositions show that the Coucal basalts are not primary mantle melts4. Negative anomalies in niobium and titanium indicate source rocks with residual hornblende, rutile and/or ilmenite12,26, but the lack of depletion in HREEs argues against the presence of residual garnet (Fig. 1a). Such patterns are consistent with source rocks that contained much more magnesium and had a much higher Mg#, requiring at least a two-stage process for production of the Coucal basalts. These observations are important, in that they suggest that the ‘arc-like’ trace-element signature evident in the Palaeoarchaean TTGs (Fig. 1a) was inherited through melting of the Coucal basalts. Further, our results require a protracted multigenerational process, from partial melting of the mantle to production of the first TTG magmas. This multistep melting process is not consistent with subduction, as shown by the contrast with the formation of modern-day adakites, which are thought to represent direct partial melting (together with subduction) of primary-mantle-derived crust14. Approximate ages of mantle melting, estimated using the samarium–neodymium isotopic system for rocks of the Warrawoona Group, are around 3.8 Gyr, assuming a first-stage 147Sm/144Nd ratio of 0.18, consistent with a mafic source. As the earliest East Pilbara TTG rocks are around 3.5 billion years old, these data suggest a period of about 300 million years between the production of primary mafic/ultramafic crust and the formation of stable continental crust. This is sufficient time for several cycles of intracrustal melting of evolving source materials, including recycling of the dense melt residua through delamination into the mantle2,15, and implies that TTGs had not only parents but also grandparents, and possibly great-grandparents before them.

The results of our phase equilibria modelling show that the residual mineral assemblages required by the composition of Palaeoarchaean TTGs could have stabilized in mafic source regions at much lower pressures than previously thought. An apparent geothermal gradient of more than 700 °C GPa−1 and temperatures of 850–900 °C may have resulted in widespread melting of Coucal-like basalts, and could thus account for the production of most of the Palaeoarchaean TTGs in the East Pilbara Terrane. These geotherms are attained in the lower levels of plateau-like mafic crust that would be predicted if Archaean mantle temperatures were considerably warmer than present2,6,7, and are consistent with numerical simulations investigating Archaean geodynamics and the generation of TTGs9. Melt compositions are controlled by several factors—principally source-rock mineralogy and the conditions and extent of partial melting—and are unlikely to be unique to any particular geodynamic setting. The arc-like geochemical signature of many Archaean rocks does not require that they formed in arcs. Hence subduction was not required to produce TTGs in the early Archaean eon.

Methods

Coucal basalt geochemistry

The Coucal basalts (n = 15; Extended Data Table 1) form a tholeiitic series that is characterized by substantial enrichment in incompatible trace elements relative to E-MORB, N-MORB, and other Pilbara Supergroup basalts with similar MgO, Cr and Ni contents (Fig. 1a)4. Pronounced negative anomalies are evident for Nb (and Ta), Sr and Ti. Depletions in Sr are accompanied by similar depletions in Ba and Rb, but not Eu; and Sr/Eu decreases with increasing content of volatile species (on the basis of analytical loss on ignition). This suggests that the destructive alteration of feldspar, under relatively oxidizing conditions, has led to an underestimate in the concentrations of many fluid-mobile trace elements. Of the Coucal basalts, a group (n = 5) that defines a trend to higher La/Nb and Th/Nb ratios with increasing silica and decreasing Ti/Gd is thought to have assimilated felsic crust4 (Extended Data Fig. 1) and was excluded. The remaining ten samples have low Th/Ce (average = 0.05), Th/Nb (average = 0.18) and La/Nb (average = 1.5) ratios, within the range for basalts considered to be free of major crustal influence22,23.

Phase equilibria modelling

Phase equilibria modelling was undertaken in the ten-component Na2O–CaO–K2O–FeO–MgO–Al2O3–SiO2–H2O–TiO2–O (NCKFMASHTO) chemical system for an average uncontaminated Coucal basalt (n = 10), with a composition, in terms of the mol% of the oxides Na2O–CaO–K2O–FeO–MgO–Al2O3–SiO2–H2O–TiO2–O, of 2.954, 8.560, 0.360, 11.486, 6.221, 8.910, 54.929, 4.734, 1.269, 0.578 (see also Extended Data Table 1). Calculations used THERMOCALC version 3.45i (ref. 29) and the internally consistent thermodynamic data set ds63 (ref. 30; updated 5 January 2015). Activity–composition solution models were as follows: tonalitic melt, augite and hornblende19 with a reduced Darken’s quadratic formulation (DQF) value for the glaucophane end-member of −3 kJ mol−1 from 0 kJ mol−1; garnet, orthopyroxene, biotite and chlorite31; olivine and epidote30; magnetite–spinel32; ilmenite–hematite33; C¯1 plagioclase and K-feldspar34; and muscovite–paragonite with a reduced DQF value for the margarite end-member of 5 kJ mol−1 from 6.5 kJ mol−1. Pure phases include quartz, rutile, sphene (titanite), and aqueous fluid (H2O). The results are shown in an isochemical P–T phase diagram, or pseudosection, between 0.35 GPa and 1.35 GPa, and between 630 °C and 1,000 °C (Extended Data Fig. 2). The H2O content in the modelled composition was fixed to be just sufficient to saturate the solidus at 1.0 GPa (producing less than 1 mol% H2O-saturated melt). The quantity of H2O-saturated melt is less than 5 mol% at all modelled pressures in excess of 0.4 GPa, but greater than 5 mol% at pressures below this (Fig. 2 and Extended Data Fig. 2). Assemblage fields are labelled with stable phases, in which abbreviations are as follows: melt (L), garnet (g), augite (aug), orthopyroxene (opx), hornblende (hb), biotite (bi), magnetite (mt), ilmenite (ilm), plagioclase (pl), K-feldspar (ksp), muscovite (mu), quartz (q), rutile (ru), sphene = titanite (sph), and aqueous fluid (H2O). All fields contain plagioclase. The depth of shading of assemblage fields reflects increasing variance. The field with the highest variance (aug–opx–ilm–L) at high temperatures and low pressures has a variance of 7 (Extended Data Fig. 2). The software and data files used to generate the phase diagrams can be downloaded from http://www.metamorph.geo.uni-mainz.de/thermocalc.

Trace-element modelling

For trace-element modelling (Fig. 4), we used the average uncontaminated Coucal basalt4 as a starting composition and the calculated abundance of phases along the 900 °C GPa−1 geotherm at various melt fractions (Figs 2 and 3a, Extended Data Table 3). We consider the residua to contain a fixed amount (1%) of apatite, as well as an amount of zircon that was calculated from the zircon concentration in the average Coucal basalt (200 p.p.m.), the composition of the melt at the appropriate temperature (Extended Data Table 4), the melt fraction and a refined zircon solubility model35. The calculations predict 0.03 mol% zircon in the residua at melt fractions of 5%, 0.02 mol% at melt fractions of 10%, 0.001 mol. at melt fractions of 15%, and no zircon thereafter. Mineral/melt partition coefficients (D) are taken from the literature and tabulated in Extended Data Table 5. Most D values follow ref. 15 (wherein the values were calculated for anatexis of metabasites), with some important modifications. For the modelled hornblende compositions (Mg# = 37–51), we use DNb = 0.8 and DTa = 0.38 (ref. 27), to give a DNb/DTa ratio of 2.1, which is more appropriate for residua containing low-Mg amphiboles12. For ilmenite we use values for DNb and DTa of 45 and 40, respectively26, and for DTi a value of 150 (ref. 36).

Data availability

The data supporting the findings of this study are available within the paper (including Methods and Extended Data); data in Extended Data Tables 1 and 2 are provided as Supplementary Information.