Is plate tectonics a post-Archean phenomenon? A petrological perspective

At the present day, magmatism linked to plate tectonics takes place in a range of settings: at divergent plate boundaries, at convergent plate boundaries, at ‘pull-apart’ sites at transform plate boundaries and in collision zones (both syn- and post-collision). The principal way to determine whether plate tectonics analogous to today operated in the Archean has been to recognize divergent and convergent plate boundary magmatism in the crustal archive. However, this is not a simple task as the criteria used to fingerprint igneous rocks formed in present-day plate-tectonic settings may not apply to an evolving Earth given the expected higher ambient mantle T_P in the Archean (Herzberg et al. 2010; Aulbach and Arndt 2019; Brown et al. 2022; Herzberg 2022). Moreover, it is not just a question of recognizing the products of potential plate margin magmatism, but also of showing that they can be distinguished from the range of igneous rocks that might be generated in non-plate-tectonic settings, where magmatism was probably dominated by mantle plumes (Fischer and Gerya 2016). Here, we test claims for Archean plate tectonics based mainly on the petrology and geochemistry of volcanic rocks, examining in turn evidence related to divergent margins (the ‘search for Archean oceanic crust’) and convergent margins (the ‘search for Archean volcanic arcs').

The method for identifying volcanic indicators of plate tectonics is not straightforward and is represented here in the form of a flow chart (Fig. 1). The chart highlights three routes to plate tectonics: the first (the ‘green route’) leads to oceanic divergent plate margins or oceanic off-axis settings; the second (the ‘red route’) leads to convergent plate margins; and the third (the ‘purple route’) leads to intraplate basalts derived from lithosphere carrying inherited subduction signatures. The ‘blue routes’ lead to basalts from intraplate settings that do not require a role for plate tectonics.

Each of the divisions on the flow chart have been termed a decision point (DP) based on geochemistry and/or geology (Table 1). Geochemistry focuses primarily on the crustal proxy, Th/Nb, a well-tested plate-tectonic setting discriminant, either in that form (Pearce 2008; Pearce et al. 2021), its reciprocal (Nb/Th; Condie 2003, 2005) or its petrogenetic equivalent (Th/Ta; Wood 1980). In this study, Th/Nb forms the diagonal on the Th/Yb v. Nb/Yb ‘crustal proxy diagram’ (Figs 2–5). The first level of subdivision (DP1) separates suites of mafic (basic and basic–intermediate) volcanic rocks with low to high Th/Nb from those with consistently high Th/Nb (DP1a) and those with consistently low Th/Nb (DP1b). In this context, ‘high’ means 95% of compositions fall within the volcanic arc basalt (VAB) array in Figures 2–5, ‘low’ means they fall within the oceanic basalt (OcB) array on these plots and ‘low to high’ means they cross the boundary between the OcB and VAB arrays. DP1a has already been used by Smithies et al. (2018) and Mole et al. (2021) in their two-fold subdivisions of Archean greenstones, and the terminology here is equivalent in concept to the ‘variable v. constant Th/Nb’ discriminator of Smithies et al. (2018), but with the added incorporation of the precise 95% OcB–VAB discriminant boundary (Th/Nb = 0.15; Pearce et al. 2021). DP1b was the focus of Pearce (2008) in his ‘search for Archean oceanic crust’. For both DP1a and DP1b, the high and low Th/Nb (red, purple and green) routes carry most of the potential plate-tectonic indicators.

The second level of subdivision incorporates indicators other than Th/Nb geochemistry, to relocate any exceptions to the first-order interpretations that (1) high Th/Nb is the route to subduction, (2) low to high Th/Nb is the route to a continental intraplate setting, commonly a plume-related large igneous province (LIP), and (3) low Th/Nb is the route to oceanic crust. For this, there are four decision points (see Table 1 for details). DP2 transfers from the subduction route to the intraplate route any high Th/Nb suites that may be inferred to be unrepresentative (crust-rich) samples of what is actually a low to high Th/Nb suite; DP3a transfers from the intraplate route to the subduction route any low to high Th/Nb suites that may be inferred to have a back-arc or forearc origin with variable subduction input and hence variable Th/Nb; (3) DP3b transfers from the intraplate to the oceanic route any low to high Th/Nb suites that can be shown to be oceanic crust modified by metasomatism during high-temperature metamorphism; and (4) DP4 transfers from the oceanic crust route to the intraplate route any low Th/Nb suites that can be shown to have belonged to a continental intraplate setting but to have appeared oceanic owing to limited crustal interaction.

The third level of subdivision deals with distinctions within the resulting ‘cleaned up’ three-fold (subduction, intraplate and oceanic crust) classification. Decision point DP5 on the high Th/Nb route in Figure 1 separates the red route to active subduction from the purple route to inherited subduction. This could involve, for example, geologically distinguishing the arc setting of the former from the intraplate setting of the latter, geochemically distinguishing the mantle wedge source of the former from the lithospheric source of the latter, or isotopically identifying the significant time difference between subduction input and melting in the latter. Along the low to high Th/Nb (intraplate) route, the key third-level decision point (DP6) is between plume–crust and plume–SZLM interaction, where ‘plume’ in this context means intraplate magma derived from upwelling mantle asthenosphere and ‘SZLM’ refers to subcontinental mantle lithosphere modified by subduction zone derived melts and fluids. Both plume–crust and plume–SZLM interactions have similar intraplate settings and so geological criteria are unlikely to be effective. However, on the crustal proxy diagram, the plume–crust trend typically lies between the OcB array and continental crust, whereas the plume–SZLM trend lies typically between the OcB array and volcanic arc compositions. However, this is not always diagnostic, requiring, for example, isotopic characterization to better distinguish continental crust and subduction endmembers.

It should be noted that this flow chart (Fig. 1), including the decision points, has been chosen to be as robust as possible in its applicability to the Archean. Thus, for the oceanic (green) route, higher ambient mantle T_p will increase the degree of melting, which will decrease dispersion along the OcB array. However, Th/Nb is unlikely to be affected owing to the similarity in mineral–melt partitioning between Th and Nb. Moreover, there is no indication that Th/Nb of the mantle has changed significantly with time, except perhaps in the earliest Earth as discussed below. As a consequence, the Phanerozoic OcB array should apply to all except perhaps some Eoarchean magmas. An important geological consideration is that higher ambient mantle T_P may have led Archean ridge-generated oceanic crust to resemble the crust of ridge-centric Phanerozoic oceanic plateaux in its structure, thickness (up to c. 25 km; Tetreault and Buiter 2014) and geological characteristics (Moores 2002). Higher ambient mantle T_P also probably limited the thickness of Archean oceanic lithosphere and this, combined with higher Archean mantle plume T_P, may have allowed Archean off-axis mantle plumes to start to melt well below the base of the lithosphere with potentially high degrees of melting leading to fewer ocean island hasalt (OIB)-like compositions and further thickening of pre-existing oceanic crust.

For the intraplate route, higher magma temperatures during the Archean might have increased the extent of plume–crust interaction, although higher magma fluxes may have created channelways with reduced access to unreacted crust. Moreover, the small difference between Phanerozoic felsic continental crust (FCC) in Figures 2 and 4 and Archean felsic continental crust (A-FCC) in Figures 3 and 5 (Rudnick and Fountain 1995) should produce similarly diagnostic trends from low to high Th/Nb. For the active subduction (red) route, it has been argued that warmer Archean slabs might have a reduced Th/Nb owing to a greater proportion of basaltic to sediment-derived melt in the subduction component (Van Hunen and Moyen 2012). However, while most subducted sediment has significantly higher Th/Nb than subducted altered oceanic crust, experimental melting of mid-ocean ridge basalt (MORB)-like basalts under hot (>1000°C) subduction conditions (Rapp et al. 1999) still generate melts that will react with the mantle wedge to produce primitive arc magmas that plot in the VAB array. Such experiments are supported by the Phanerozoic VAB array in Figure 4, which contains numerous arc magmas from ridge-subduction, subduction initiation and plume-proximal settings, which are believed to record zones of hot subduction. Thus, although the 95% probability tramlines for an Archean VAB array cannot be known precisely, it is likely that the Phanerozoic discriminant boundaries are a reasonable approximation.

Finally, for the plume–SZLM interaction (purple) route, there is no reason why any Archean subduction should not metasomatize mantle lithosphere in a similar way to Phanerozoic subduction, although the depth of the metasomatized ‘layer’ may decrease because of the shallower intersection of the lithosphere geotherm and the solidus of the invading melt. However, an important question is: when did mantle lithosphere become sufficiently stable to allow the resulting SZLM to be preserved as a mantle source reservoir? The answer does not affect our approach, but is does mean that an absence of SZLM characteristics does not necessarily equate to an absence of subduction, with profound consequences for the oldest (c. 3 Ga) SZLM signature detected, as discussed below

In practice, the principal questions of applicability relate less to the continuation of existing (albeit hotter) processes into the Archean, and more to the significance of non-uniformitarian processes (i.e. those restricted to the Archean) that might yield false subduction signals. We have added this possibility to the flow chart as dashed blue lines with a decision point DP7 that needs to be addressed before potential active subduction (at DP7a) or potential inherited subduction (at DP7b) can be treated as a definitive plate-tectonic indicator. In reality, as non-uniformitarian processes are poorly defined, so is the decision process at DP7 (Fig. 1). Notwithstanding, we consider two main non-uniformitarian processes that might give false subduction signals: crustal delamination as an alternative to plate subduction; and relict primitive (Hadean) mantle sources as an alternative to subsequent (at least Paleoarchean to Recent) mantle sources.

Crustal delamination as a subduction alternative is primarily based on geochemical and geophysical modelling. For example, Bédard (2006) argued that delamination of lower crust following plume activity can refertilize otherwise hot, residual mantle from preceding plume events and so lower the mantle solidus and promote further melting. Johnson et al. (2014) used geodynamic models to argue that delamination of overthickened primary crust by Rayleigh–Taylor instabilities provides a alternative to subduction in the Archean. Identifying the potential subduction setting in question as plume-related and using a full range of isotopes and trace elements to fully define the enrichment source (for example, identification of lower crustal sources that fit delamination better than subduction components) are among the most promising approaches to demonstrate non-uniformitarian processes.

The second process involves primitive mantle and magma–ocean cumulates in the mantle acting as subduction-like sources in Eoarchean settings. These materials, if predating crustal growth, can have higher Th/Nb than Phanerozoic equivalents, so giving their melting products apparent (i.e. ‘false’) subduction signatures. The availability of these sources and involvement of minerals that can fractionate Th from Nb may, it is argued, be assisted by ‘ultra-deep melting in a heat pipe Earth’ (Webb et al. 2020; Rollinson 2022). This possibility is examined subsequently in the text.

Most studies have equated the search for oceanic crust with the identification of Archean ophiolites (Moores 2002; de Wit 2004; Kusky 2004; Kusky et al. 2004; Dilek and Furnes 2011; Furnes et al. 2014; Furnes and Dilek 2022). This has usefully furnished the research community with a list of potential exposures of oceanic crust in the geological record and for some, though by no means all (Condie and Stern 2023), has provided definitive evidence for plate tectonics through Earth history. Some of the opposing arguments are detailed and involve specific interpretations of outcrops and their structural settings. However, two more general criticisms, articulated well by Hamilton (2007), seem to be particularly important. The first is the somewhat arbitrary expansion of the original definition of ophiolite, which, although having the benefit of allowing many more ocean crust outcrops to be identified, also allows at least some potential continental crustal outcrops unrelated to ocean basins to be classed as ophiolites. The second is that many interpretations of outcrops as representing oceanic crust seemingly have not considered whether non-plate-tectonic interpretations also provide viable explanations. Below, we address these criticisms for the subset of ophiolites interpreted as having formed in major ocean basins unrelated to subduction.

Ophiolite complexes have long been informally described as ‘oceanic lithosphere emplaced on land’, a description that forms the basis for its role as a plate-tectonic indicator. The formal definition remains that provided by the Penrose Field Conference participants as ‘layered sequences, some 6-km thick on average, comprising deep-sea sediment overlying submarine (commonly pillowed) lavas, sheeted dykes, cumulate rocks (isotropic gabbros, cumulate gabbros and cumulate ultramafic rocks) and tectonized peridotites’ (Anonymous 1972). Most diagnostic are the sheeted dykes (>95% dykes by volume, some with one-way chilled margins), and tectonized peridotites. This formal definition also allows for the fact that few on-land exposures retain an original stratigraphy and/or have all these characteristics, in which case they may be described as ‘fragmentary’ and/or ‘incomplete’ ophiolites, respectively. However, even if the ophiolite stratigraphy and/or these features are absent, this need not negate an ophiolite interpretation, as oceanic lithosphere produced in (very) slow-spreading systems can lack sheeted dykes and have more complex internal structures (Cannat et al. 2006).

‘Post-Penrose’ ophiolite studies have also considered the fact that oceanic crust can form in a range of settings, many of which also have non-Penrose structures. In these cases, the principal subdivision is into mid-ocean ridge (MORB or MOR) type and suprasubduction-zone (SSZ) type (Pearce et al. 1984), or non-subduction and subduction type (Dilek and Furnes 2011). Both types essentially subdivide ophiolites according to whether or not the oceanic lithosphere in question relates to divergent plate boundaries (mainly mid-ocean ridges, ridge-centred oceanic plateaux and the oceanic parts of ocean–continent transition zones) or convergent plate boundaries (mainly forearc and back-arc spreading centres). In this section, we focus on MOR-type ophiolites.

As emphasized above, the search for Archean oceanic crust also has to consider the alternatives. In an Earth without plate tectonics, most igneous rocks might be expected to be part of continental LIPs (Ernst 2014), many of which are related to continental rifts. Like oceanic crust, continental LIPs typically comprise volcanic sequences, dyke and sill swarms and underlying mafic–felsic cumulate sequences. If erupted in regions of lithospheric thinning, pillow lavas can be common. Therefore, methods of distinguishing oceanic from continental magmatism are critical in the search for Archean oceanic crust. Geologically, such methods include the search for autochthonous continental basement and relict continent-derived xenoliths and xenocrysts.

In summary, past oceanic crust can potentially be identified using the following non-geochemical criteria (Kerr et al. 2000; Kusky 2004). (1) Formation in an oceanic setting, as identified by the absence of intercalated terrigenous sediments, autochthonous continental basement and older, inherited continental crust-derived zircons and, if underlain by continental crust, a basal thrust fault. (2) If formed at a ridge axis, the presence of the full Penrose ophiolite sequence including a fully sheeted dyke swarm with one-way chilled dykes and the presence of mantle tectonites, or, if these features are not present, then a significant number of components of the ophiolite stratigraphy. (3) If formed in deep water (at a ridge axis or the lower part of an oceanic plateau), the presence of pillow lavas and other indicators such as low vesicularity, an absence of extensive pyroclastic rocks, and the presence of chert and deep-water sediments, sea-floor metamorphism and black smoker deposits. (4) If formed as part of an oceanic plateau on ocean crust basement, the absence of continental indicators as in (1), together with the absence of sheeted dykes, few interbedded sediments and pyroclastic units, predominantly submarine magmatism and predominantly basaltic composition, although komatiites may also be present.

We focus here on putative Precambrian ophiolite localities. Moores (2002) compiled the original list of pre-Rodinian ‘ophiolites and possible ophiolites’ based on stratigraphy, rock type and, to a lesser extent, composition, noting that they cluster in age at 1.0–1.5, 1.8–2.3, 2.5–2.7 and c. 3.4 Ga. Nine of those listed by Moores (2002) are Archean in age, at Dongwanzi, China (c. 2.5 Ga); Wind River, Wyoming (c. 2.6 Ga); Kalgoorlie and Norseman, Australia (c. 2.7 Ga); Cameron–Beaulieu, Yellowknife and the Point Lake in the Slave Province, Canada (c. 2.7 Ga); and the Jamestown and Pietersberg complexes, South Africa (c. 3.4–3.5 Ga).

Detailed descriptions of many of the proposed Precambrian ophiolites (and the related synopsis) in the volume on Precambrian Ophiolites and Related Rocks (Kusky 2004) laid out the evidence for and against each being a fragment of ancient oceanic crust. In the same volume, de Wit (2004) made a list of Archean ophiolites, largely matching that of Moores (2002), but also making the case for preservation of oceanic crust in mélanges such as those reported from the East Pilbara Terrane and Isua Supracrustal Belt (Komiya et al. 1999, 2004). In examining ‘four billion years of ophiolites’, Furnes et al. (2014) expanded the list of potential Archean ophiolites to include, for example, the c. 2.7 Ga Wawa sub-province in the southern Superior Province, concluding that ‘half the Archean greenstones originated as oceanic crust’. All potential outcrops of non-subduction type oceanic crust, which form a significant thread of the argument in favour of plate tectonics on the early Earth, are evaluated in this paper by attempting to follow the ‘green’ routes to oceanic crust in Figure 1, either via decision points DP1b and DP4 or, if metasomatized at high temperature, via DP1b and DP3b.

Before carrying out the search for potential Archean oceanic crust, we examine the dispersion of Phanerozoic analogues on the Th/Yb–Nb/Yb diagram (the crustal proxy projection) to facilitate interpretations. For the purpose of finding plate-tectonic indicators, we define oceanic crust in its broad sense to include both the crust formed at ocean ridges and any off-axis edifices and intrusions subsequently built on and within it. In this context, ocean crust is made up of three tectonically defined magma types: mid-ocean ridge basalts (MORB), oceanic plateau basalts (OPB) and ocean island basalts (OIB). Each can be subdivided according to source depletion into D (depleted), N (normal) and E (enriched) and into tholeiitic (th) and alkalic (alk). The Th/Yb–Nb/Yb crustal proxy projection based on the combination of all these magma types from the known plate-tectonic (Phanerozoic) Earth is shown in Figure 2a, in which the 95% probability density limits of the combined MORB–OIB–OPB data give the boundaries of the resulting oceanic basalt (here labelled ‘OcB’) array (see Pearce et al. (2021) for the data points). Averages of some of the main oceanic magma types within the OcB array are plotted to highlight the along-array variability and the key fact that their Th/Nb ratios (the diagonal on this projection) are all broadly similar. An exception, not relevant here, is a subset of alkali basalts with EMII enriched sources that define a separate EM–OIB field at the upper edge of the OcB array.

The boundaries of the volcanic arc basalt (VAB) array are also based on 95% probability density limits, in this case based on the combination of oceanic and continental arc rocks of mafic (basic and basic–intermediate) compositions. Coincidentally, its lower boundary is almost the same as the upper boundary of the OcB array and so this also defines the discriminant boundary. However, the VAB array is not restricted to volcanic arcs, as Th/Nb is a crustal proxy rather than just a subduction proxy. Thus, it also contains the vast majority of continental crust compositions (felsic continental crust (FCC) has been plotted in Fig. 2a) and therefore also mafic volcanic rocks from other settings with a clearly detectable crustal input or with a subduction-metasomatized lithospheric mantle (SZLM) source. On this basis, we expect any oceanic crust formed at a major divergent plate boundary, or off-axis within the resulting oceanic basin, to plot between the 95% probability OcB tramlines on the crustal proxy projection, and this is the main potential oceanic crust-based plate-tectonic indicator. The projection in its older form (Pearce 2008) has already been used for this purpose (Dilek and Furnes 2011; Furnes et al. 2014).

However, as already emphasized, the crustal proxy projection provides only a geochemical indicator that volcanic suites within the OcB array are consistent with an oceanic crust origin. To be definitive, their composition must be distinguished from their possible alternatives, primarily compositions from continental intraplate LIPs. As Figure 2b shows (see Pearce et al. 2021 for data points), a significant part (almost 50%) of the continental LIP basalt field lies within the OcB array, with the remainder occupying the VAB array by virtue of having been derived from, or contaminated by, SZLM and/or continental crust. Many of these continental LIPs in the OcB array can be distinguished from oceanic LIPs and other oceanic basalts based on data dispersion, the continental LIPs typically having a component of variation from the OcB array to continental crust. Nonetheless, a small number of continental LIPs do form trends entirely within the OcB array. It is these suites that can be most easily mistaken for oceanic basalts.

Figure 2c gives some examples of Phanerozoic oceanic plateau basalts that might be considered the most useful comparators with proposed Archean oceanic crust. The key feature of oceanic basalts is that they lie almost entirely within the OcB array tramlines. Variations along the OcB array are mainly a function of mantle enrichment and depletion, coupled with degree of melting if in the garnet facies. Oceanic plateaux such as those from the Colombia–Caribbean LIP (CLIP) (Kerr et al. 1997) and the various components of the Greater Ontong–Java plateau (OJP) (Tejada et al. 1996, 2013; Golowin et al. 2017, 2018) experience the greatest variations, which can be summarized as follows: (1) most mafic lavas from oceanic plateaux cluster close to the average OPB composition and reflect broadly similar plume compositions that underwent a high degree of melting; (2) depleted sources are not uncommon and lead to a separate cluster around the D-OPB composition, which here includes the well-known Gorgona komatiites from the CLIP; (3) a subset of data typically plots towards enriched compositions as a result of local mantle enrichments and/or lower degrees of melting; (4) mantle heterogeneity can cause compositions to plot near the top (and base) of the array, as in the case of the main Mahihiki EM-rich sourced basalts in Figure 2c.

The difficulty in distinguishing between the oceanic crust plate-tectonic indicator and continental plume-derived magmas with no plate-tectonic connotations is highlighted in Figure 2d . The contrast between the two options is best seen in the typical example of a transition from continental to oceanic plume magmatism in the North Atlantic Igneous Province (NAIP) (Fitton et al. 2000). Here, the pre-rifting stage, where plume-derived magmas penetrate and erupt over thick continental lithosphere, gives rise to trends extending from the OcB array towards continental crust compositions and hence well into the VAB array. With continuing lithospheric extension, which here includes the main East Greenland dyke swarm (Hanghøj et al. 2003) and their offshore equivalents (Fitton et al. 2000), a higher proportion of compositions plot in the OcB array and hence fewer reach the VAB array. Compared with the oceanic plateau basalts, represented here by Iceland (Kokfelt et al. 2006), the continental basalts plotting in the OcB array do have a discernible crustal input represented by increased Th/Nb. However, that input is minor and only seen because both types of composition are available for comparison. It is thus likely that the large volumes of magma that rapidly traverse highly attenuated continental lithosphere experience limited interaction and hence a limited increase in Th/Nb. Moreover, as already noted, such localities may have some ophiolite characteristics such as pillow lavas, abundant dykes and basic intrusions. Thus, continental rifts that experienced considerable extension and magma flux, but that never extended sufficiently to form true oceanic crust, can be misidentified as oceanic plate-tectonic indicators without careful integration of geochemical and geological criteria. Making this distinction is represented in the flow chart by decision point DP4.

How the cluster of NAIP (and other continental LIP) data can be explained by a combination of two components is summarized in Figure 2e. The first component, the plume-derived endmember, lies along the OcB array at a point defined in particular by the degree of melting and degree of source depletion or enrichment. The second component, the continental crust contaminant (here represented by average felsic continental crust; FCC), typically lies in the upper right part of the VAB array. The interaction between the two endmembers is defined here by assimilation and fractional crystallization (AFC) trends that simulate the observed ‘plume–crust’ trends. The compositions of the endmembers and the relative rates, r, of assimilation to fractional crystallization (based on the equations of DePaolo (1981)) are most influential in defining the AFC trend. The degree of fractional crystallization, F, determines the degree of displacement from the plume endmember composition. The trends shown in Figure 2e are based on r = 0.3. Other forms of interaction, such as direct mixing of mafic and felsic magmas and migration of evolving magmas through crustal hot zones, give alternative (but not greatly dissimilar) pathways between intraplate magma and crust.

An important caveat in the search for oceanic crust, especially for Archean rocks, is the effect of high-temperature (upper amphibolite- or granulite-facies) metasomatism, which can give misleading interpretations of basalts (Fig. 2f). This process is easy to recognize as it is characterized by significant variations in Th/Nb on length-scales that are much shorter than those expected; for example, within single lava units. In the absence of external fluids, the amphibolite- to granulite-facies transformation typically leads to significant loss of Th owing to loss of supercritical fluids and melt. The liberated fluids and melts can infiltrate and react with rocks along their fluid pathways, commonly leading to Th-rich metasomatism and an increase in Th/Nb. Figure 2f focuses on the Catalina schist on Catalina Island in Southern California and the Shuksan metamorphic suite in the North Cascades (Sorensen and Grossman 1989, 1993). Both represent Jurassic MORB metamorphosed to amphibolite facies during accretion under hot (subduction initiation) conditions. In both cases, the compositions are displaced to variably higher Th/Nb that, without other information, might misleadingly imply a continental intraplate setting and the misidentification of rocks that actually are oceanic plate-tectonic indicators. However, the similar trend followed in Figure 2f by the non-oceanic, eclogite–amphibolite-facies continental margin metabasalt sills from the Chinese CCSD borehole in the Sulu ultrahigh-pressure (UHP) belt (Xiao et al. 2011; Yang and Pearce, unpublished data) demonstrates that low to high Th/Nb amphibolites that have undergone metasomatism cannot be assigned an oceanic crust setting without more detailed investigations.

In Figure 3a, we plot three well-documented Paleoproterozoic ophiolites, which are the oldest non-subduction ophiolite types described as having clear geological and geochemical indicators of an oceanic provenance. At 2 Ga, the Purtuniq ophiolite, from the Cape Smith fold belt in northern Quebec, is arguably the oldest ophiolite to have all the crustal members of the Penrose ophiolite sequence, including a sheeted dyke complex, although no mantle tectonite has been found (Scott et al. 1991, 1992, 1999). In Figure 3a, the lavas and dykes plot in the OcB array and are closer to E-MORB than N-MORB, which might also be due to a plume-proximal oceanic setting.

Jormua was chosen as a type-example of a ‘continental margin ophiolite’ by both Dilek and Furnes (2014) and Pearce (2014) based on geological reconstructions and its OIB to E-MORB geochemistry (Peltonen et al. 1996). In detail, the Jormua ophiolite is typical of ocean–continent transition zones such as the present Iberian margin, with some of the terrane having continental characteristics (inherited Archean zircons and exhumed continental lithosphere) and some having the full Penrose ophiolite sequence, including both the sheeted dykes plotted here, and mantle tectonites (Peltonen et al. 2003; Finlayson et al. 2023). In Figure 3a, the ophiolite basalt data plot in the OcB array despite the proximity to a continent and are more similar to E-MORB than N-MORB, which (as noted above) is characteristic of incipient ocean formation.

The third example is a set of lavas from the Amisk Collage (an accretionary complex) in the Flin Flon Belt, Canada, described as having N-MORB, E-MORB and OIB characteristics and interpreted as oceanic crust (Stern et al. 1995; Babechuk and Kamber 2011). After filtering out the recognized effects of Th- and Nb-metasomatism, these characteristics are confirmed by the dispersion of data within the OcB array from E-MORB and tholeiitic OIB to D-OPB in Figure 3a. As seen in Figure 2c and d, data dispersion of this type is typical of Phanerozoic oceanic plateaux such as the Ontong–Java Plateau (Golowin et al. 2017) and Iceland (Kokfelt et al. 2006), where mantle plume melting within the garnet facies coupled with complex mantle flow give mantle depletions and enrichments that can lead to a wide range of Nb/Yb ratios.

These studies are important from an igneous perspective and in the context of the theme of this paper as they show that, by 2 Ga, oceanic crust compositionally similar to that formed at the present day was being formed during the plate-tectonic ‘Wilson Cycle’ period of ocean opening and closing that led to the first major supercontinent (Nuna/Columbia). Between this time and the Archean (2.5 Ga), evidence for oceanic crust is more ambiguous and controversial, such that the search for subduction might be more definitive.

We present the results of our search for Archean oceanic crust in four parts: (1) proposed cases of Archean greenstones that are described as non-subduction type ophiolites but classify as continental intraplate basalts based on decision point DP1b in Figure 1; (2) proposed examples of Archean ‘MORB–OIB’ and ‘ophiolitic’ complexes that could be interpreted in several other ways based on Figure 1; (3) examples of mafic volcanic rocks affected by high-temperature metamorphism used to support Archean plate tectonics, but where metasomatism may have masked the true interpretation; (4) proposed examples of Archean oceanic crust that satisfy the low Th/Nb criterion (at decision point DP1b) but then need further assessment (at DP4) to determine whether they are truly oceanic.

The Yellowknife greenstone belt (southern Slave Province, Canada) is potentially the most significant of the proposed ophiolites as the lowermost (Kam) formation contains the only example of sheeted dykes reported from Archean greenstones (Baragar 1966; Helmstaedt et al. 1986). In Figure 3b, the lower Kam lavas from this belt plot on OcB–FCC trends, placing them clearly on a low to high Th/Nb, non-oceanic, trajectory at DP1b. An oceanic crust interpretation was similarly rejected by Bickle et al. (1994), who pointed out that the ‘sheeted dyke swarm’ was described as ‘closely spaced dykes with interspersed pillowed and massive basalts’, which is different from the 95% cross-cutting dykes with one-way chilled margins characteristic of ‘Penrose ophiolites’. Moreover, the dykes are not rooted in underlying gabbros as in ‘Penrose ophiolites’, but instead appear to intrude underlying continental basement. Bickle et al. also emphasized evidence for xenocrystic zircons in felsites from the same formation. The counter argument, advanced by Corcoran et al. (2004), is that the lower boundary is tectonic and that the greenstones were part of an oceanic plateau that was thrust over the basement. However, Northrup et al. (1999) questioned this hypothesis by showing that the dykes cutting the gneissic basement were contemporaneous with the ‘sheeted dykes’, an observation consistent with their continental intraplate interpretation based on Nd isotopes. Therefore, the Yellowknife ‘ophiolite’ has a very high probability of having formed as an LIP focused on attenuated continental crust, as deduced by Cousens (2000).

Several other Neoarchean greenstones, albeit without sheeted dykes, have been listed as examples of possible or probable fragments of accreted oceanic crust, but similarly plot on low to high Th/Nb trends towards continental crust in Figure 3b, which, when combined with geological criteria such as those of Bickle et al. (1994), lead to a continental intraplate setting based on Figure 1. This group includes other parts of the Slave craton (Corcoran 2001) as well as greenstones elsewhere that have also appeared on ophiolite lists, such as the c. 2.7 Ga Belingwe greenstones from the Zimbabwe craton (Shimizu et al. 2005), the c. 2.7 Ga Norseman and Kambalda greenstones from the Yilgarn craton (Said et al. 2010, 2012) and the c. 3.47 Ga lower Onverwacht volcanic series in the Kaapvaal craton (Robin-Popieul et al. 2012). Notably, like Yellowknife, the Belingwe greenstones have a well-defined autochthonous granitic basement (Bickle et al. 1994) with lavas containing continental xenoliths (Shimizu et al. 2004), whereas Norseman and Kambalda greenstones have intercalated terrigenous sediments (Said et al. 2012).

Mélanges interpreted as chaotic fragments of oceanic crust (and mantle) accreted during subduction have also been used as evidence for plate tectonics (Kusky et al. 2020). If the fragments are restricted to volcanic upper crust with MORB–OIB characteristics, they are commonly termed ‘MORB–OIB’ mélanges. If they also contain sufficient members of the Penrose ophiolite stratigraphy, they are commonly termed ‘ophiolitic mélanges’. In Fig. 3c, we evaluate the three examples that make up the most detailed claims for demonstrating Archean plate tectonics from the Mesoarchean to Eoarchean: Cleaverville (c. 3.1 Ga) and North Pole (c. 3.5 Ga) from the East Pilbara Terrane; and Isua, SW Greenland (c. 3.8–3.7 Ga).

The Cleaverville greenstones in the West Pilbara Terrane (part of the Regal Formation (Sun and Hickman 1999)) have been interpreted by Ohta et al. (1996) as an accretionary mélange comprising intercalated sediment and pillow lavas, with the lava compositions described as ‘A-MORB’ (Archean MORB). However, the crustal proxy diagram (Fig. 3c) shows the compositions to lie just above the OcB array owing to their relatively high Th/Nb ratios, a point made by Sun and Hickman (1999). On the flow chart in Figure 1, these follow a poorly defined low to high Th/Nb route. A back-arc basin origin was an option but, at decision point DP3a, inter-lava sedimentation did not support an oceanic crust origin (Kiyokawa et al. 2019), making plume–crust interaction the more probable explanation. In any case, there is little to support the ‘MORB–OIB’ interpretation.

The ‘MORB–OIB mélange’ from the North Pole region of the East Pilbara craton has been identified and mapped in detail by Komiya et al. (2002), who provided significant data in support of their interpretation, later augmented by additional trace element analyses (Nakamura et al. 2020). However, the GSWA (Geological Survey of Western Australia) mapped and sampled the whole northern Pilbara craton and concluded that the various greenstones (including the c. 3.5 Ga terrane mapped by Komiya et al. (2002)) were not part of an accreted mélange but were volcanic formations successively erupted onto a continental basement (Hickman 2016) to create a long-lived volcanic plateau (Van Kranendonk et al. 2015), an interpretation supported by the common presence of inherited zircons and crustal enclaves. Unfortunately, the ICP-MS data of Komiya et al. (2002) were affected by incomplete dissolution in sample preparation, so we have matched the units they sampled with equivalent units (North Star Basalt and Mt Ada Basalt) in the GSWA dataset (Smithies et al. 2007) for the Th/Yb–Nb/Yb projection (Fig. 3c). This low to high Th trend from the OPB endmember in the OcB array does not support an oceanic crust origin, instead favouring plume–crust interaction in a continental setting (see Fig. 2b), the preferred interpretation of Smithies et al. (2005a).

The ‘MORB–OIB mélange’ from the Isua Supracrustal Belt is similarly controversial, unsurprisingly given its tectonic complexity. Komiya (1995) and Komiya et al. (2004) claimed that their age and geochemistry defined them as ‘the oldest MORB and OIB in the world’. However, this description is confusing because the MORB and OIB assignments do not match present-day classifications. Specifically, the ‘OIB’ lacks the high Nb/Yb (residual garnet) characteristic of present-day OIB and (if oceanic) best resembles present-day OPB (see Fig. 2a). In addition, the ‘MORB’ has very low Nb/Yb indicative of a mantle source depleted in residual garnet, and hence not typical of a mid-ocean ridge setting. It also exhibits a small but significant enrichment in Th that causes the data to plot in the lower left-hand part of the VAB field. Following the flow chart, the sample set classifies as ‘high’ Th/Nb, following the ‘red route’ to subduction and so will be re-examined in the next section. Thus, although it is true that Isua compositions are potentially significant in terms of plate tectonics, there is no clear evidence for the proposed MORB–OIB bimodality.

Using the simple but effective criterion of significant, small length-scale Th/Nb variations (Table 1), it is evident that many amphibolite-facies volcanic rocks of Archean age have experienced Th enrichment or depletion relative to Nb during metamorphism, which can probably be ascribed to interaction with Th-rich supercritical fluids or melts. Figure 3d and e give examples of outcrops cited as representing Archean oceanic crust from (1) the end of the Archean (North China craton, 2.6–2.5 Ga) and (2) the beginning of the Archean (SW Greenland and northern Canada, c. 3.8–3.7 Ga), all of which emphasize the need to test for mobility of so-called immobile elements before interpreting Archean amphibolites.

The Wutaishan and East Hebei amphibolites from the North China craton have been used by a number of researchers to infer the presence of oceanic crust and hence plate tectonics towards the end of the Archean (2.6–2.5 Ga), despite Th/Nb exhibiting short length-scale variations indicative of metasomatic Th enrichment. For Wutaishan, Wang et al. (2004) subdivided amphibolites from the Wutai complex into MORB-like, island arc tholeiite (IAT)-like and adakitic. As Figure 3d shows, the classification fits the geochemistry in the sense that data plotting in the OcB array classify as MORB, whereas those plotting above it (in the VAB array) could classify as island arc tholeiite (IAT). However, in the absence of stratigraphic control, this is only one interpretation (Option 1). If all the data are treated instead as a single group, two further interpretations are possible. In Option 2, the amphibolites are all MORB but experienced variable metasomatism and hence variable Th enrichment (as in the accreted Jurassic MORB in Fig. 2f). In Option 3, the amphibolites underwent variable metasomatism combined with pre-existing magma–crust interaction (as in the Sulu Belt continental rift amphibolites in Fig. 2f). Options 1 and 2 favour plate tectonics whereas Option 3 does not. Moreover, if the plate-tectonic interpretation is correct, Option 1 provides evidence for both MORB and arc (divergent and convergent margin) magmatism, whereas Option 2 favours MORB (divergent margin) only.

The same problems arise from the work of Ning et al. (2022) on amphibolites from two other complexes of similar age from East Hebei: the Zunhua ophiolitic mélange and the Shangyin MORB–OIB-type mélange. They also form OcB–VAB trends (some Shangyin gabbro samples also showing some scatter best explained by cumulation), which could be explained in these three different ways. In the flow chart (Fig. 1), these samples follow the low to high Th/Nb route to decision point DP3, whereby they can potentially subdivide into ‘oceanic crust’ (at DP3b) or ‘SSZ oceanic crust’ (at DP3a) or continue along the intraplate route to plume–lithosphere interaction. However, at present, there is insufficient information to make this decision. Thus, a more sophisticated interpretation and sample selection is needed to ascertain whether these really are plate-tectonic indicators.

Eoarchean (c. 3.8–3.7 Ga) amphibolites from Isua (SW Greenland) and Nuvvuagittuq (northern Quebec) provide other well-studied examples of metasomatized amphibolites, in this case representative of some of the Earth's oldest volcanic rocks. In particular, Frei et al. (2002) and Polat and Hofmann (2003) demonstrated Th mobility relative to Nb during metamorphism of amphibolite-facies submarine lavas from Isua. Their data (plotted in Fig. 3e) show a similar trend to the Phanerozoic amphibolites in Figure 2f and Wutaishan amphibolites in Figure 3d and face the same ambiguities. For Nuvvuagittuq, O'Neil et al. (2011) and Turner et al. (2014) identified three lava units, and interpreted the lower two (‘high-Ti’ and ‘unenriched low-Ti’) as potential subduction initiation MORB-like and boninite-like oceanic crust related to subduction initiation; these precede an upper unit interpreted as arc-derived as discussed subsequently. However, short length-scale variations make it clear that Th has been mobilized, thus complicating evidence for the true tectonic setting. In Figure 3, their trends resemble those of Isua (Fig. 3e), the North China craton (Fig. 3d) and Phanerozoic accreted amphibolites (Fig. 2f), with further work needed to confirm or reject a plate-tectonic origin.

Of the many proposed examples of Archean oceanic crust, only a small subset satisfies even the first criterion of the flowchart in Figure 1, that of low and constant Th/Nb at decision point DP1b. This subset is plotted in Figure 3f. However, to be considered true oceanic crust and therefore potential plate-tectonic indicators, they also need to ‘pass’ decision point DP4, which is based on criteria needed to distinguish oceanic crust basement from highly attenuated continental crust.

The c. 2.7 Ga Superior Province greenstone belts located in the southern Superior craton provide some of the most convincing claims for having formed in an oceanic Archean plate-tectonic setting. In particular, a significant body of work (Fan and Kerrich 1997; Kerrich et al. 1999; Polat and Kerrich 2000; Polat 2009) has been used to argue that the Wawa sub-province is fully oceanic and made up of two parts, an older part best explained as oceanic crust and a younger part best explained as an (oceanic) island arc and back-arc basin (discussed in the next section). In this model, the Tisdale group represents an oceanic plateau and ocean islands and the Schreiber–Hemlo group is an accretionary complex containing komatiites and tholeiites of oceanic crust provenance.

Unlike the greenstones described in the previous sections, there has been less controversy over these interpretations, and even Bédard and Harris (2014), who favoured a different overall tectonic model driven by mantle flow rather than plate tectonics, treated these Superior Province greenstones as oceanic-like tracts between separating cratonic fragments. However, although Kerrich et al. (1999) originally claimed that there is no evidence for continental crust associated with these greenstones, the geochemical test is much less clearcut (Fig. 3f). The data for the ‘plateau’ units plot within the OcB array but with a greater scatter to high Th/Nb values than is normal for oceanic crust. Thus, the Th/Nb geochemistry is a better match for, say, the East Greenland continental margin than Iceland in Figure 2d. This could also be explained by analytical variance, especially as Th concentrations are low, but the reported data quality makes this interpretation unlikely. Moreover, subsequent work by Ayer et al. (2002) and Thurston (2002) supported an autochthonous continental basement for rocks of this age based on the presence of zircon xenocrysts and other geochemical indicators. On this basis, the route to oceanic crust in Figure 1 would fail to pass decision point DP4 and eventually be assigned a continental intraplate setting.

More detailed investigation of the full range of sub-provinces in the SE Superior craton (including Th/Yb–Nb/Yb plots for each) has been made by Mole et al. (2019), whose interpretation of a continental rift system driven by plume magmatism prior to c. 2.7 Ga matches previous work better than an oceanic basin. Of course, this does not negate the possibility that oceanic crust had formed but was not preserved, and so it need not negate the tectonic model of craton break-up and reassembly as proposed, for example, for the Wawa subprovince (Polat 2009) and the Wabigoon superterrane (Bjorkman et al. 2024).

The c. 2.8 Ga Kostomuksa and c. 2.9 Ga Sumozero–Kenozero greenstone belts in the NW and SE Baltic Shield (Puchtel et al. 1998, 1999) have been interpreted as having lower units with oceanic plateau characteristics and have been cited by Kerr et al. (2000) and others as examples of Archean oceanic plateaux. Specifically, Kostomuksa has been interpreted by Puchtel and co-workers as a partially accreted oceanic plateau, and Sumozero–Kenozero as a partially accreted plateau subsequently overlain by arc-like upper units. They reported a lack of field evidence for continental basement and a lack of geochemical features such as low $ε$Nd that are typically associated with interaction with continental lithosphere. Moreover, the compositions plot entirely, and uncommonly, fully within the OcB array on the crustal proxy projection (Fig. 3f). Despite a 100 Ma age difference, the two Baltic Shield greenstones have a narrow data dispersion near present-day average OPB with no clear vector towards continental crust. Of all the proposed examples of Archean oceanic crust, they are the most convincing, as demonstrated by a comparison between Figures 2c, d and 3f. The principal caveat is that key supporting studies, such as a search for inherited zircons, have yet to be carried out.

The c. 3.3 Ga Kromberg mafic–ultramafic sequence in the Barberton Mountain Belt was claimed to represent accreted oceanic crust based on evidence from structures (an allochthonous interpretation), geology (an absence of inherited zircons) and geochemistry (juvenile $ε$Nd and oceanic Th/Nb ratios) (Grosch and Slama 2017). Grosch and Slama also used the Th/Nb–Nb/Yb projection to demonstrate the similarity between the Kromberg sequence and the Abitibi/Wawa and Somozero greenstones, a feature also evident in Figure 3f. In detail, however, Th/Nb is still somewhat higher than expected for oceanic crust, which was attributed to a primordial (higher Th/Nb) mantle source. However, this seems unlikely as both older and younger parts of the Barberton greenstone belt include lavas with normal mantle sources (Fig. 3b). Thus, although the decision at DP4 in Figure 1 is not definitive, a continental intraplate setting would seem more likely.

As with the ‘search for oceanic crust’, there have been many attempts to identify subduction-related igneous rocks in the Archean and to use these as evidence for plate tectonics. These studies have also proved controversial, with interpretations ranging from pan-Archean arc magmatism (Polat and Kerrich 2006; Sotiriou et al. 2022), through intermittent periods of arc magmatism (Van Hunen and Moyen 2012) to an absence of arc magmatism (e.g. the ‘snArc’ of Bédard et al. 2013). As in the previous section, the aim here is to make an independent assessment based on geochemical proxies, notably Th/Nb, which (as already discussed) is one of the most robust in circumventing the likely temporal global variations in heat flow and continental crustal inputs.

Volcanic arc rocks may be defined as those derived from magmas generated in volcanic chains above zones of active subduction of oceanic crust. One obvious problem in interpreting past subduction-related magmas is that present-day studies have revealed that they can be highly variable within and between arcs (Schmidt and Jagoutz 2017). Two principal variables are the nature of the lithospheric basement (oceanic or continental) and the geodynamic status of the subduction system. For the former, continental arcs are typically marked by less partial melting, less mantle depletion and greater interactions between magmas and lithosphere (SCLM and crust) compared with oceanic arcs. For the latter, geochemical compositions change during the evolution of single arcs, from subduction initiation through a period of steady-state subduction interspersed with transient events such as ridge subduction and flat subduction, and ending with continental subduction in their final (syn-collisional) stage.

Volcanic arc magmas that erupt at the present day principally follow the basalt–andesite–dacite–rhyolite (BADR) sequence with basalts dominating oceanic arcs and andesites dominating continental arcs (Gill 2012). They can also be subdivided according to potassium content (at a given silica value) into a (low-K) tholeiitic series, calc-alkaline series and high-K calc-alkaline series (Peccerillo and Taylor 1976), or, more rigorously, into separate classifications based on K content (high-, medium- and low-K) and iron enrichment (tholeiitic and calc-alkaline or low-, medium- and high-Fe) (Arculus 2003), where low-K and high-Fe characterize the majority of oceanic arcs, whereas medium- to high-K and medium- to low-Fe characterize the majority of continental arcs (Garcia 1978). Ultrahigh-K magmas (shoshonitic series and lamproites) are also significant as indicators of particularly high slab fluxes into the mantle wedge or of highly metasomatized SCLM, as, for example, in continent–arc collisions, rear-arc settings and areas of slab failure (Morrison 1980). Adakite-related volcanic rocks with their definitive high Sr/Y (Defant and Drummond 1990; Castillo 2012) have been found to characterize specific volcanic arc settings such as ridge subduction, flat subduction and slab edges.

Despite this wide variation in volcanic arc magma types, some geochemical features are characteristic of virtually all subduction zones. These include the element inputs and fractionations that take place to create high ratios of subduction-mobile large ion lithophile elements (LILE) relative to the less subduction-mobile high field strength elements (HFSE), which are evident in such features as high Th/Nb (the main subject of this study), negative HFSE anomalies on MORB-normalized plots and the decoupling of isotope values such as $ε$Nd and $ε$Hf. Also ubiquitously important are the consequences of added subduction fluids that include the high oxygen fugacities that influence the behaviour of redox-sensitive elements such as V during melting, so leading to distinctive Ti/V (Shervais 1982), as well as the sequence of phase appearance related to factors such as melt depolymerization and the enhanced role of hornblende crystallization. Geological criteria are also an integral part of the search for subduction. Criteria used for volcanic arcs (Garcia 1978) include: compositional peaks of basic to basic-intermediate rocks if oceanic but andesitic if continental; abundant andesitic pyroclastic rocks; highly vesicular lavas for their composition and (if oceanic) assumed depth of eruption; abundant volcanogenic clastic sediments; associated underlying plutons including batholiths if sufficiently long-lived; and porphyry-type ore deposits.

However, as in the search for oceanic crust, any significant deficiency in the correct recognition of plate tectonics via subduction-related magmatism in the Archean may lie less with the identification of potential subduction signatures in the igneous rocks and more with whether the same signatures can also be created in a plume-dominated Earth without plate tectonics. The method here with its focus on the Th/Nb crustal proxy is the same as that used to search for oceanic crust, but is now applied to the search for volcanic arcs and related subduction indicators, which requires finding volcanic suites that follow the ‘red and purple’ routes in Figure 1. As before, we first examine comparator examples from the consensus plate-tectonic Earth (the Phaneozoic), then find, as another comparator, the oldest consensus volcanic arc system, before finally carrying out the search for Archean volcanic arcs based on existing lists of potential Archean volcanic arcs, notably that of Polat and Kerrich (2006).

The Th/Yb v. Nb/Yb plot for oceanic basalts in Figure 2a is duplicated in Figure 4a but with more information on the compositions of mafic volcanic arc lavas (basalts and basaltic andesites) in the VAB array. The dashed fields distinguish the broad areas occupied by oceanic (Oc-VAB) and continental (Cont-VAB) compositions (see Pearce et al. 2021 for data points) and the red squares give the average compositions. The distinctive difference is that, overall, Cont-VAB have higher Nb/Yb than Oc-VAB at similar Th/Nb, which can be attributed in part to less depleted (higher Nb) mantle sources in continental settings, and in part to a suite of other factors that include more terrigenous sediments undergoing subduction, more continental crust to assimilate and lower degrees of deeper melting. It should be noted, however, that these fields apply only to subduction of oceanic lithosphere, as even Oc-VAB can have high Nb/Yb once continental lithosphere is subducted in a syn-collision environment.

Figure 4b repeats the LIP basalt field from Figure 2b but with the aim of highlighting the overlap between the VAB and mafic LIP fields. This overlap results because Th/Nb is a crustal proxy rather than a subduction-specific proxy, and so high-grade metamorphism and plume–lithosphere interactions can also lead to subduction-like signals. In the flow chart (Fig. 1), decision point DP1a separates most subduction-related mafic volcanic rocks, which follow the red and purple route based on consistently high Th/Nb (i.e. >95% plot within the VAB field), from most mafic, intraplate volcanic rocks, which follow the ‘blue route’ based on their low to high Th/Nb. As already covered, decision point DP2 then transfers out of the subduction route and into the intraplate route any high Th/Nb suites that can be shown to extend into the OcB field by extrapolation back to primitive magmas. Decision point DP3a transfers into the subduction route low to high Th/Nb suites that might have transitional subduction (marginal basin) compositions. In Figure 4c–f, we provide some type examples of Phanerozoic volcanic rocks to highlight the similarities and differences of those indicative of subduction compared with those having no subduction connotation. Where possible, we combine mafic compositions (basalts and basaltic andesites) with felsic compositions (andesites, dacites and rhyolites) to provide complete BADR trends.

The increase in Th/Nb owing to interaction between continental intraplate magmas and crust is highlighted in Figure 4c, showing the full BADR trends. The Yellowstone LIP (Hildreth et al. 1991) is a classic example of a bimodal (basalt and rhyolite) suite, made up of a basaltic plume-derived magma undergoing extensive interaction with the crust. Hildreth and coworkers recognized the extensive basalt–rhyolite mixing and basalt contamination and explained the rhyolites as the products of crustal melting in the lower crust aided by basalt injection followed by accumulation in upper crustal reservoirs. The products of these plume–crust interactions are the diagonal trends in Figure 4c.

The British Tertiary Province, part of the North Atlantic Igneous Province (NAIP), contains several examples of BADR series formed by plume–crust interaction that have been confirmed by isotope studies. A simple example is Ardnamurchan (Geldmacher et al. 1998), where the evolving compositions have been interpreted in terms of fractional crystallization coupled with shallow assimilation of ambient Precambrian metamorphic rocks. It differs from the Yellowstone trend in having a lower Nb/Yb plume endmember (closer to the average OPB) but a similar crustal endmember (close to FCC), which results in a shallower plume–crust interaction trend.

The Etendeke LIP, represented by the c. 132 Ma picritic Spitzkoppe dyke swarm in Namibia, provides a further example of well-constrained plume–crust interaction (Thompson et al. 2007). Thompson et al. found evidence for ‘limited interaction with metasomatized lithosphere’ before interaction with Damaran crust, although Figure 4c does not allow the two to be distinguished. The plume and crustal endmembers are similar to those of the NAIP example, but more interaction was required to reach evolved compositions, a possible consequence of the more picritic (magnesian) plume-derived inputs.

Although plume–crust interaction is unrelated to plate tectonics, plume–mantle lithosphere interaction may carry a ‘hidden’ record of past subduction if the lithospheric component in question is subduction-modified lithospheric mantle (SZLM). However, identifying plume–SZLM interaction can be difficult as the plume-derived magma is interacting with continental crust as well, resulting in a complex spread of data.

The examples shown in Figure 4d are distinct from plume–crust interaction in that they typically extend from the OcB array towards the centre of the VAB array rather than towards continental crust. If the OcB endmember has an E-MORB or tholeiitic OPB composition, they can form trends with negative slopes. This is the case for the upper units of the Siberian Traps (Lightfoot et al. 1990), the Bunbury basalts (Olierook et al. 2016) and the Durham Basin from the SW part of the CAMP LIP (Grossman et al. 1991). The latter forms a complicated cluster rather than a simple trend, which emphasizes the potential difficulty of identifying plume–SZLM trends. However, if they can be identified in the Archean, they may provide useful additional indirect evidence for subduction.

Some of the different types of volcanic arc formation that make up the VAB array, extended to include the more evolved rocks, are highlighted in Figure 4e. In each case, as already noted, the most basic composition depends on the degree of depletion of the mantle source and the composition and relative input of the subduction component. The length and orientation of the mafic to felsic trend depends on the crystallizing assemblage (particularly the incoming of hornblende with low Th/Nb), as well as on the composition of the continental crust and extent of crustal contamination. The set of examples in Figure 4e has been chosen to facilitate the interpretation of proposed Archean arc BADR series in the next section.

For normal oceanic arcs, Candlemas (Scotia Arc) and Soufrière (Montserrat, Lesser Antilles arc) represent oceanic arc BADR series. For Candlemas, a low-K, tholeiitic series, parent magma derived from a depleted mantle wedge crystallized an olivine–clinopyroxene–plagioclase assemblage with no significant crustal input (Pearce et al. 1995). In Figure 4e, this trend exemplifies a depleted arc endmember characterized by intermediate Th/Nb, very low Nb/Yb ratios and little mafic–felsic dispersion. For Soufrière, a medium-K calc-alkaline volcanic suite with a significant sediment-dominated subduction flux and an amphibole–plagioclase crystallization assemblage (Zellmer et al. 2003), amphibole fractionation together with interaction with older arc crust explain the greater spread of compositions from mafic to felsic. For ‘hot oceanic subduction’, Simbo (New Georgia Group, Solomon Islands) provides an example of picrites and high-Mg andesites (HMA) from a site of ridge subduction (König et al. 2007). Like the other oceanic arcs, they are characterized by high Th/Nb BADR trends.

The majority of continental arcs have medium-K, calc-alkaline character, and so are broadly similar to Soufrière, which (although oceanic) has developed on an oceanic plateau, lacks an actively spreading back-arc basin and is proximal to sources of continent-derived sediment. Thus, in Figure 4e we have plotted Vulcano from the south–central Aeolian arc (Francalanci et al. 2007; Peccerillo et al. 2013) to illustrate features of the high-K (shoshonitic) continental arc series. The setting of Vulcano is complex, on a continental rise on the edge of the Tyrrhenian Sea above a retreating oceanic plate, and the high Th contents cause it to occupy the upper right corner of the VAB array. As an example of ‘hot subduction’ in a continental arc setting, we have chosen Shasta (southern Cascade arc), a continental, medium-K calc-alkaline arc volcano incorporating high-Mg andesites and associated with subduction of very young crust (Grove et al. 2002, 2005). The reduced subduction flux (and hence low, although still arc-like, Th/Nb) combined with extensive interaction with continental crust gives a steeper BADR trend than most arc volcanoes. Some of this trend may also relate to the presence of two separate parental magma sources; namely, from the rear-arc and mantle wedge (Streck et al. 2007). As will be seen, this steeper trend has analogues in the Archean crustal record.

Arc-like magmas can result not only from active subduction but also from reactivation of subcontinental lithosphere enriched by past subduction events. Figure 4f highlights some type examples of inherited subduction-modified magmas: a continental LIP (the Ferrar LIP), an oceanic LIP (the Great Benham Rise), a post-collision volcano (Mt Ararat) and a zone of continental extension (the Basin and Range).

The continental Ferrar LIP (FLIP) is an elongate belt of Jurassic age some 3000 km long mainly made up of hypabyssal intrusive rocks of siliceous high-Mg basalts (SHMB) and HMA compositions, which were emplaced over a very short time interval at c. 183 Ma. It is one of the best-known, consensus examples of magma derived from a ‘fossil subduction zone’ (Choi et al. 2019). Its precise context is debated, although most publications ascribe it to a subduction-modified subcontinental mantle lithosphere source (Hergt et al. 1991). Interestingly, there is no plume-derived magma in the region, although the far-field effect of the near-contemporaneous Karoo plume may be important from a tectonic and thermal perspective. In Figure 4f, basic compositions plot near the centre of the VAB array.

The Great Benham Rise refers to a series of related, mainly Eocene, oceanic plateaux in the Philippine Sea linked to the upwelling of the Ancestral Manus mantle plume (Yeh et al. 2021). The initial impact was with Cretaceous oceanic arc lithosphere, resulting in extension and eruption of arc-like lavas in a rift-basin setting (Hickey-Vargas et al. 2013). For this study, the key point is that these lavas plot in the field of oceanic VAB magmas in Figure 4f (see Fig. 4a), demonstrating that inheritance can also be chemically dependent on the setting of the original subduction system.

The Basin and Range volcanic province in SW USA and NW Mexico provides a good example of post-subduction magmatism, in this case probably resulting from extension and mantle upwelling related to the change from a subduction to a transform (proto-San Andreas) plate boundary rather than a mantle plume (Putirka and Platt 2012). The high-K, calc-alkaline, lithosphere-sourced volcanic rocks from New Mexico plotted in Figure 4f characterize the early stages of magmatism (c. 40–30 Ma), after which time asthenosphere-sourced magmas took over (Davis and Hawkesworth 1994). In Figure 4f they plot in the centre of the arc array parallel to the discriminant boundary, within the range of active continental volcanic arc magmas.

Mt Ararat (eastern Turkey) is another example of post-subduction melting of lithospheric mantle, but here the setting is post-collisional and the lithospheric enrichment in the province can be spatially related to subduction prior to the collision (Pearce et al. 1990). The time interval from the end of arc activity to the start of post-collision magmatism is about 35 Myr. However, the compositions of the resulting magmas are similar to those of the preceding continental medium- to high-K, calc-alkaline arc magmas, which plot in the centre of the VAB field in Figure 4f.

Comparison between Figure 4e and f highlights the fact that distinguishing inherited from active arc signatures (decision pont DP5 in Fig. 1) can be difficult unless there are clear geological criteria that can be applied. As noted above, one obvious difference is that mantle enrichment for active arcs is usually synchronous with eruption, whereas for inherited arc signatures there may be a significant time difference between the two. Mantle xenoliths from the regions of proposed lithosphere-sourced magma, especially if characterized using the Re–Os isotope system (Walker et al. 1989), may also provide useful information.

The oldest, well-documented and largely undisputed post-Archean subduction events took place during the formation of Laurentia by accretion of Archean terranes during the Paleoproterozoic period (Hoffman 1988). Of these events, the best preserved in terms of volcanic stratigraphies and accompanying geochemical data are in the Trans-Hudson Belt of northern Canada, which may be considered part of a larger terrane extending from Wyoming through Canada, Baffin Island and Greenland to the Baltic Shield. It is believed to result from opening, and subsequent closure though subduction, of the so-called Manekewan ocean. Corrigan et al. (2009) concluded that ocean opening initiated at c. 2.17 Ga, that it began to close at c. 1.92 Ga and that closure was fully complete by c. 1.80 Ga. It is the volcanic arc magmas erupted during the 60 Myr period between 1.91 and 1.85 Ga that provide us with an opportunity to test methods and to act as a comparator for proposed Archean arcs.

Figure 5a focuses on the Flin Flon Belt in northern Manitoba–Saskatchewan (Canada) (Stern et al. 1995), which can be divided into an oldest ‘boninite–tholeiite’ suite (c. 1904–1900 Ma), followed by a ‘calc-alkaline’ suite (c. 1900–1890 Ma) and finally a ‘more potassic calc-alkaline’ suite (<1890 Ma). In Figure 5a, each volcanic unit plots within and along the VAB field accompanied by an increase from relatively low Th/Nb in the oldest suite through arc-average Th/Nb in the middle suite to still higher Th/Nb but also higher Nb/Yb in the youngest suite. Stern et al. (1995) interpreted this sequence as representing an oceanic arc that evolves from subduction initiation through to a mature arc, a sequence observed elsewhere in the Trans-Hudson Belt (Leybourne et al. 1997) and in many present-day arc systems. Continental arc systems are also present, as exemplified in Figure 5a by the volcanic rocks of the LaRonge Belt, which lies north of the Flin Flon Belt and includes the Lynn Lake and Rusty Lake greenstone belts (Chauvel et al. 1987; Watters and Pearce 1987; Thom et al. 1990). Like present-day continental arcs, these plot within the VAB array, but with higher Nb/Yb compared with their oceanic equivalents, and also follow a trend indicative of crustal interaction as well as fractional crystallization.

Overall, therefore, there is evidence here for a full subduction cycle from birth to death of a subduction zone, with geochemical characteristics not obviously distinguishable from Phanerozoic subduction cycles. In the flow diagram (Fig. 1), most of the magmas may be assigned to active subduction at decision point DP5 given the high Th/Nb characteristics and the abundant geological evidence for a subduction rather than an intraplate setting. However, some of the youngest magmas may be the product of post-collision subduction inheritance owing to melting of mantle lithosphere enriched during the preceding subduction episode, a common feature of Phanerorozoic subduction systems (Pearce et al. 1990). Thus, plate tectonics resembling that at the present day probably was possible back to at least 2 Ga. Our evaluation strongly supports the assertion of Thom et al. (1990) that processes and compositions have probably not changed much over the past 2 Gyr. However, we are not aware of any comparable subduction systems in the Archean in terms of longevity and compositional range.

We present the results of our search for Archean subduction-related rocks in five parts: (i) examples of Archean greenstones described as volcanic arcs but which classify as having continental intraplate settings; (ii) examples of Archean greenstone suites with likely inherited subduction signatures; (iii) an example of an Archean greenstone suite with possible evidence for plume–SZLM interaction; (iv) examples of Archean greenstone suites with likely Archean active subduction signatures; (v) examples of Archean amphibolite suites with potential subduction signatures.

There are many examples of Archean plume–crust interactions that follow the ‘blue (plume/intraplate) route’ through the flow diagram in Figure 1, and several were reported in the section on ophiolites (Fig. 2b). A wide range of further examples have been published by Barnes et al. (2012) and Smithies et al. (2018) from the Yilgarn and Pilbara cratons (Australia) and by Mole et al. (2021) from the Superior craton. Here we have highlighted three of their best examples of BADR sequences resulting from plume–crust interaction rather than plate subduction: the Agnew Greenstones (2.7 Ga) from the Kalgoorlie Terrane in the Yilgarn craton (Barnes et al. 2012), the Duffer and Panorama groups (3.46–3.43 Ga) from the East Pilbara Terrane (Smithies et al. 2005a) and the Frotet–Evans greenstone belt from the Opatica sub-province from the Superior craton (Boily and Dion 2002; Mole et al. 2021). We also added a fourth well-studied example, the Belingwe greenstones (2.7 Ga) from the Zimbabwe craton (Shimizu et al. 2005).

These examples are very similar, all forming a trend between two endmembers: a plume endmember close to average OPB composition in the OcB array and a crustal component close to Archean felsic continental crust (A-FCC) in the upper right segment of the VAB array. All four exhibit a bimodal distribution in Figure 5b: a mafic mode composed predominantly of komatiite, basalt and basaltic andesite compositions, and a felsic mode predominantly of andesite to rhyolite compositions. This bimodality differs from the unimodal distribution with an andesitic peak that is common in Phanerozoic continental volcanic arc terranes, a distinction further indicative of an intraplate origin.

Some well-documented examples of high Th/Nb volcanic rocks that follow the ‘purple route’ to inherited subduction (Fig. 1) are shown in Figure 5c. As the aim is to investigate inherited Archean subduction, the volcanic rocks themselves need not be of Archean age, although the metasomatism of the lithosphere, by definition, must be. Two of the chosen examples of this type are Paleoproterozoic and both plot in the centre of the VAB field in Figure 5c. The youngest example (2.06 Ga) comprises the Bushveld B1 ‘chills’ (Maier et al. 2016) and the penecontemporaneous Dullstrom Volcanics (Buchanan et al. 1999). A key feature in these, and many other Archean lithosphere-sourced magmas with subduction-like signatures, is their boninite–HMA, or SHMB–HMA signatures, the likely result of melting of highly depleted and refertilized lithospheric sources. The second example is the Matachewan LIP dyke swarm (2.48–2.44 Ga) in the Superior province, Canada (Ernst and Bleeker 2010). As one of the world's largest dyke swarms, it clearly has a plume-related LIP origin. However, unlike the Bushveld example, these are not boninitic, implying that the lithosphere was less depleted prior to refertilization.

Pearce and Reagan (2019) found that there are many examples of magma from depleted and refertilized sources within the period 3–2 Ga, all of which link to LIP or non-plume continental rift settings. The Mallina Basin (Smithies et al. 2004) contains what are possibly the oldest clear examples of inherited subduction signatures. Like many other Archean and Paleoproterozoic examples in this age range, they mainly classify as SHMB–HMA suites and form a trend in Figure 5c within the centre of the VAB array. The setting (within a sedimentary basin) and composition both support an inherited subduction component within depleted continental lithospheric mantle.

The final example is the (enriched) Paringa Basalts from the Kalgoorlie Terrane, Australia. In their compilations of volcanic rocks from the Yilgarn Block in Australia, Said and Kerrich (2009), Barnes et al. (2012) and Smithies et al. (2018) all found the Paringa Basalts, and notably the ‘enriched’ suite, particularly distinctive with ‘unusual’ features such as low Ba/Th. In Figure 5c, their data plot entirely within the VAB array. Thus, according to Figure 1, they follow the ‘red’ potential subduction route with active arc and inherited arc both options until decision point DP5 is reached. There, even apart from geological indicators, their ‘unusual’ geochemical features (negative anomalies of P, K and Ba, in particular) are readily explained by melting of metasomatized lithosphere with residual phlogopite and apatite.

The examples cited above are not the only occurrences of inherited subduction in the Archean record. Most Archean cratons appear to have refertilized lithospheric sources that continue to provide sources of magmas through to at least 2 Ga. However, whether they can be regarded as plate-tectonic indicators will depend on decision point DP7 on the purple route in Figure 1, where non-uniformitarian ways of enriching lithosphere by means other than plate subduction need to be considered. Here, as elsewhere in the Archean, there is no definitive answer as non-uniformitarian processes lack undisputed examples. However, given that the potential Archean inheritance examples in Figure 5c are so similar to their Phanerozoic equivalents in Figure 4f, and given that inherited subduction is extremely common in the Phanerozoic, the implication must be that subduction inheritance is the higher probability option for the examples just cited. That said, back-tracking through dating and isotope geochemistry to the sources of the proposed inherited magmas would provide important confirmatory information.

Plume–SZLM interactions are difficult to recognize. They are found either where plume-derived magma interacts with SZLM or when plume-derived and SZLM-derived magmas undergo mixing. However, these processes can be difficult to distinguish from crustal contamination. One potential example (Fig. 5d) is the low-Ti and boninitic compositions from the Frotet–Evans greenstone belt (Boily and Dion 2002; Mole et al. 2021). The Frotet–Evans BADR series, shown in Figure 5b, is a simple trend from the OcB array to Archean felsic crust and does not provide evidence for subduction. However, the associated low-Ti suite is different. It defines a near-vertical trend that could indicate interaction between an intraplate magma and metasomatized mantle lithosphere, either in the mantle or by magma mixing in the crust. If correct, it could indicate that the lithosphere beneath the Frotet–Evans greenstones experienced (otherwise hidden) subduction at some time in its history.

A number of c. 2.7 Ga greenstones within the the Superior craton have been listed by Polat and Kerrich (2006) as examples of Archean active arc volcanism. Two of these, from the Wawa Domain (Polat and Kerrich 2000) and the Birch–Uchi Confederation (Hollings and Kerrich 2000), are plotted in Figure 5e. They form BADR trends entirely within the VAB array, thereby following the high Th/Nb route in Figure 1 consistent with the volcanic arc intrerpretations.

The third example in Figure 5e is from the 3.12 Ga Whundo Group (West Pilbara Terrane) (Smithies et al. 2005b, 2018). The Whundo section is described as comprising boninites and a calc-alkaline BADR sequence interspersed in places with tholeiitic basalts, all of greenschist-facies metamorphic grade (Smithies et al. 2005b). In Figure 5e, the ‘calc-alkaline and boninite series’ lavas form a trend fully within the VAB array (the tholeiites are transitional and not discussed here).

In terms of the flow chart in Figure 1, DP5 is the key decision point between active and inherited arc magmatism. In all three examples in Figure 5e, limited outcrops and metamorphism and deformation make it difficult to apply the DP5 criteria listed in Table 1, although the terrane-scale volcanic stratigraphy may be particularly significant. The Wawa and Uchi Confederation examples are each part of a sequence of continental rifts and volcanic plateaux followed by an inferred volcanic arc and finally by collision (orogenic and post-orogenic) magmatism. This is a similar sequence to that in the Paleoproterozoic Trans-Hudson Belt, even though the proposed arc events in the former may have been shorter-lived. Given that the late-stage collision volcanism has abundant examples of subduction inheritance, the c. 2.67 Ga diamond-bearing lamprophyres (Wyman and Kerrich 1989) being the best of many examples, it is logical that a preceding active arc was the source of the inherited metasomatizing melts and fluids. The c. 3.1 Ga Whundo sequence similarly lies between a long, older period of regional LIP volcanism (including the c. 3.45 Ga Apex/Panorama volcanics in Fig. 5b) and a younger period of inherited subduction magmatism (including the c. 3 Ga Mallina Basin in Fig. 5c).

The decision to assign these volcanic rocks to the active arc group is not totally unambiguous. At decision point DP7a, non-uniformitarian translation of crustal materials into the mantle, for example of the type proposed for the Abitibi terrain by Bédard (2006), remains a topic in need of further investigation. A further point of note is that all three potential examples of active arc volcanism have been interpreted by the authors cited above as ‘oceanic arcs’. Although not conclusive, comparison of Figure 5e with the active arc comparator diagram in Figure 4e shows that their compositions resemble more closely continental arcs. The increase in Th/Nb and Nb/Yb from mafic to felsic and the resulting convergence on crustal compositions is particularly diagnostic, with all three Archean examples best matching the continental, hot-subduction Shasta comparator.

Two key localities for lavas with clear subduction-like signatures but insufficient data to confirm as active or inherited subduction are the ‘low-Ti basalts’ and ‘boninites’ of Isua and the Upper lavas of Nuvvuagittuq, both probably of 3.8–3.7 Ga age. In both cases, intense deformation, high metamorphic grade, limited outcrop size and unclear geological context make decision points DP5 and DP7 hard to apply and it is unsurprising that interpretations are controversial. Despite this, the crustal proxy diagram does provide some constraints (Fig. 5f).

The Isua supracrustal belt is one of the most controversial locations for tectonic fingerprinting. Polat et al. (2002) and Jenner et al. (2009) highlighted the presence of boninites and depleted tholeiites, which they interpreted in terms of subduction initiation and oceanic arc settings. As established in the previous section, Th mobility during metamorphism is a major issue in data interpretation (Fig. 2f) as is data quality at such low abundances of Th and Nb. However, by carefully selecting the least metasomatized samples from the study of Polat et al. (2002) and by carrying out high-precision analysis, Hoffmann et al. (2010) do appear to have circumvented these problems. Their ‘filtered’ dataset, plotted on their figure 6g crustal proxy projection (and reproduced in our Fig. 5f), plots at the lower edge of the island arc tholeiite segment of the VAB array (see Fig. 4a). They are the only Archean basalts we have so far found to do so. It may also be significant that analyses by Hoffmann et al. (2011) of samples from the ‘sheeted dyke’ outcrop identified by Furnes et al. (2007) have a similar composition. If the outcrop is made up of sheeted dykes, which is by no means certain (Nutman and Friend 2007), the protoliths may indeed have been boninites from SSZ ophiolites associated with subduction initiation, as they and others have inferred.

Webb et al. (2020) and Rollinson (2022) presented the main alternative view, that Isua's apparent plate-tectonic indicators based on Th/Nb are unrelated to subduction and, instead, are the product of ultra-deep melting involving residual perovskite residues in a heat pipe Earth. For that model to work, it also requires the magma source to comprise primitive mantle (silicate mantle prior to any crustal extraction), or magma ocean cumulates derived from primitive mantle, to provide sufficiently high Th/Nb mantle sources. In detail, the best estimate to date for the Th/Nb ratio of primitive mantle may be 0.136 from Lyubetskaya and Korenaga (2007), which is slightly higher than (though within error of) the commonly used value of 0.119 of Sun and McDonough (1989). However, both are significantly higher than the estimate of 0.053 for present-day depleted MORB mantle (DMM) (Workman and Hart 2005) and, as a result, they plot close to the upper edge of the present-day OcB array where Th/Nb = 0.15. Further, the mantle source must have been depleted much deeper than any mantle wedge to explain the HREE depletions, and Th/Nb must be increased still further than the primordial mantle ratio: the proponents argue that both features are potentially achievable in a heat pipe Earth. Moreover, the closer the age to the Hadean, the more likely it is that materials with primitive mantle compositions were available to provide a magma source.

The Nuvvuagittuq Upper Lavas have been assigned a low-Ti (enriched) classification, which distinguishes them from the underlying high-Ti and low-Ti (depleted) lavas investigated in Figure 3e (O'Neil et al. 2007; Turner et al. 2014). Unlike the underlying lavas, where the interpretation is masked by Th mobility (Fig. 3e), the Upper Lavas do not follow a low to high Th/Nb contamination trend. Instead, despite similarly pervasive metasomatism, the trend is of a more constant high Th/Nb indicative of potential subduction. From this, it is evident that the pre-metasomatic Th values must have been sufficiently high that Th addition during metamorphism had only a limited effect. Thus, despite metasomatism, the interpretation by Turner et al. (2014) that these rocks had arc-like geochemistry is supported by the crustal proxy projection where the data plot in the centre of the VAB array (Fig. 5f). As a result, the Nuvvuagittuq trend does resemble the other potential examples of Archean arc-like rocks in Figure 5e and f and is very different from Isua, looking, for example, much more like a continental, as opposed to an oceanic, arc.

On the flow chart (Fig. 1), lavas from both Isua and Nuvvuagittuq follow the high Th/Nb (red) route to potential subduction, but there are insufficient data to decide (at decision point DP5) whether that is active rather than inherited subduction and whether (at DP7) non-uniformitarian processes were involved. Consequently, there are still important decisions to be made if Eoarchean plate tectonics is to be confirmed. Clearly, these relatively small outcrops of amphibolite carry key information about early Earth tectonics that future work will hopefully help to unravel.

Metamorphic rocks record evidence of change in pressure (P) and temperature (T) with time (t), commonly expressed as a P–T–t path, which represents the record of heating and cooling during burial and exhumation derived from mineral assemblages and compositions combined with geochronology (Brown 1993, 2001). The peak pressure may be achieved before the maximum temperature (clockwise P–T–t path), the maximum temperature may be achieved before the peak pressure (counter-clockwise P–T–t path) or the path may reach a coincident peak in pressure and temperature. The P–T–t path records the changing spectrum of transient geotherms characteristic of a particular tectonic setting (England and Richardson 1977; Richardson and England 1979; Thompson and England 1984), and the thermobaric ratio (T/P) at the metamorphic peak is characteristic of different plate-tectonic settings (Brown and Johnson 2018, 2019a, b, c). Insofar as metamorphic pressure can be translated to depth, the thermobaric ratio (T/P ≈ T/z) can be viewed as a proxy for the transient geothermal gradient recorded by a metamorphic rock at the metamorphic peak.

The occurrence of metamorphic belts characterized by the juxtaposition of different sequences of mineral assemblages in orogens led Miyashiro (1961) to the concept of paired metamorphism, although we now know that the individual belts need not be contemporaneous, and one may be far-travelled relative to the other (Brown 2010). Paired metamorphism has been related to convergent plate boundaries, where the asymmetric (one-sided) subduction of ocean lithosphere depresses isotherms, creating a cool (low T/P) metamorphic environment characterized by low heat flow. The breakdown of hydrous minerals in the descending slab generates fluids and/or melts at subcrustal depths that promote magma generation in the overlying mantle wedge, creating a warm (high T/P) metamorphic environment characterized by high heat flow in the orogenic back-arc or hinterland (Oxburgh and Turcotte 1971; Hyndman et al. 2005). Blueschists and eclogites, which are characterized by low T/P ratios, are generally related to subduction, as confirmed by the transport of Paleogene blueschist fragments to the surface in a serpentine-mud volcano located above the subducting Pacific plate at the southern end of the Mariana trench (Maekawa et al. 1993; Tamblyn et al. 2019; Miladinova et al. 2024). Ultrahigh-pressure metamorphic rocks of continental affinity are also related to subduction, as exemplified beneath the Pamir (Schneider et al. 2013). By contrast, lower crustal granulite xenoliths brought to the surface in the Kilbourne Hole maar in the Rio Grande Rift record high T/P ratios. They demonstrate that the contemporary lower crust is at granulite-facies conditions, in this case because crustal thickening was succeeded by lithospheric thinning and heating of the lower crust to granulite-facies conditions during orogenic collapse (Cipar et al. 2020).

Data collected during the last 50 years have been used to show that regional metamorphism is characterized globally by an increasingly divergent bimodal distribution of T/P ratios since the early Proterozoic (Holder et al. 2019). This bimodality reflects the different thermal gradients generated by tectonic coupling between the cold lithosphere returning to the mantle via subduction and the enhanced heat flow beneath the thinned lithosphere of the overlying orogenic hinterland (Brown and Johnson 2018; Holder et al. 2019; Holder and Viete 2024). To assess overall changes in the trend of crustal metamorphism through time, Brown et al. (2020b) applied locally weighted scatterplot smoothing (LOWESS) to the time series data of Brown and Johnson (2019c) to create a polynomial regression through the T/P ratios and applied sequential analysis (by cumulative sum) to quantify changes. T/P was generally high during the Neoarchean–Mesoproterozoic with a drop in T/P in the Orosirian, related to the formation of Nuna/Columbia, and a second, more dramatic two-stage drop in T/P in the Ediacaran–Cambrian, associated with the widespread appearance of blueschists and ultrahigh-pressure metamorphic rocks. These interpretations from the metamorphic record have been combined with other geological data to conclude that plate tectonics has been operating on Earth since at least the early Paleoproterozoic (Brown et al. 2020b).

The metamorphic dataset used in this study (Table S1) is based on the 28 February 2018 dataset of Brown and Johnson (2019c), which has been updated after an extensive review of the literature up to 3 November 2023. The dataset comprises localities with robust quantitative estimates of pressure (P), temperature (T) and age, and assumes equilibration at the P and T of interest. This is justified because prograde dehydration and melt loss generally produce mineral assemblages that are difficult to retrogress or overprint without fluid influx, and when evidence of post-peak reaction is present its effects generally can be accounted for, as detailed by Brown and Johnson (2018). The dataset has not been filtered based on presumed inferior quality or by removing data deemed to be outliers. As in previous versions, the dataset is restricted to crustal protoliths, so orogenic peridotites and occurrences of ultrahigh-pressure minerals in chromitites associated with ophiolitic complexes, and mantle xenoliths, are excluded.

One change from the previous iteration is the deletion of two low T/P (UHP) localities from the Nordre Strømfjord shear zone in the Paleoproterozoic Nagssugtoqidian Orogen of western Greenland (Glassley et al. 2014). Data from these localities have recently been called into question by Schönig et al. (2023), who found no evidence of low T/P (especially UHP) metamorphism in heavy minerals separated from modern sands representing eight catchments that drain the area. Another exclusion is the location at the northwestern end of the Isua supracrustal belt where rare lenses of mantle dunite occur in schistose ultramafic rocks (Nutman et al. 2020). Olivine in these rocks has intergrowths of titano-clinohumite replacing titano-chondrodite and is in equilibrium with antigorite, features that apparently record decompression through P–T conditions of 2.6–3.0 GPa and 500–700°C. However, how these rocks were emplaced into the crust is unknown, and without additional evidence of UHP metamorphism in the Isua supracrustal belt and a precise age the tectonic significance of these dunite lenses is ambiguous.

The current dataset includes 197 new localities, some of which replace a previous locality in the same geographical area, and incorporates updates to existing localities where new data have been published; overall the dataset is about 30% larger than its predecessor. However, in comparison with datasets used in other studies of secular change, the metamorphic dataset used here remains small with only 728 localities. These localities are globally distributed (Brown and Johnson 2019a) and include 75 localities that record Archean metamorphism, although most are Neoarchean in age, with only 10 of Mesoarchean, 12 of Paleoarchean and two of Eoarchean age.

The new metamorphic dataset has a similar P–T distribution to the previous dataset (Fig. 6a). Of 728 localities (Fig. 6b), 293 (c. 40%) record high T/P metamorphism (T/P > 775°C GPa⁻¹), 198 (c. 27%) record intermediate T/P metamorphism (T/P = 375–775°C GPa⁻¹) and 237 (c. 33%) record low T/P metamorphism (T/P < 375°C GPa⁻¹). The record of ocean basin evolution since the Triassic allows metamorphic belts younger than 195 Ma to be related directly to a plate-tectonic setting. For orogens older than 195 Ma in age, we interpret low T/P metamorphism to be related to subduction-related sutures, intermediate T/P metamorphism to record the thickened lithosphere of former mountain belts and high T/P metamorphism to be mostly related to late orogenic extension (Hyndman et al. 2005; Brown and Johnson 2018; Hyndman 2019; Agard et al. 2023). Blueschists, all <850 Ma in age, and orogenic eclogites, mostly <850 Ma in age, are commonly interpreted as related to subduction (Stern 2005; Brown 2006; Tamblyn 2021) and are dominantly of low T/P type (Fig. 6c). Notably, older eclogites tend to record higher T/P ratios on average than younger eclogites (Fig. 6c). Furthermore, for data <850 Ma where there is consensus that plate tectonics has been operating, the metamorphic record is characterized by a bimodal distribution of T/P values that has been related to asymmetric subduction (Fig. 6d; see Brown 2006; Holder et al. 2019).

A plot of number of localities v. age (Fig. 7) shows the data are not uniformly distributed but cluster coincident with the supercontinent cycle since the early Paleoproterozoic or the formation of supercratons during the Neoarchean (Brown and Johnson 2018; Liu et al. 2021). The peaks in the distribution of the metamorphic data correspond to peaks in the global distribution of igneous and detrital zircon U–Pb ages that probably reflect the cyclicity of crustal aggregation and break-up (Puetz et al. 2018). Furthermore, similarity in the age distribution of pre-Neoarchean metamorphic rocks and contemporaneous gneisses of broadly tonalite–trondhjemite–granodiorite (TTG) composition (Johnson et al. 2019) suggests similarly low rates of preservation before the crustal recycling rate decreased in the Neoarchean as the cratons stabilized (Dhuime et al. 2012; Aulbach and Smart 2023).

The temporal distribution of blueschists and eclogites, which are generally accepted as related to subduction (Brown and Johnson 2019b; Holder and Viete 2024), shows that the former are almost exclusively Phanerozoic in age whereas the latter are dominantly Phanerozoic in age (Brown 2023). The evolving style of plate tectonics through the Proterozoic into the Phanerozoic, which involved a change from warmer to colder subduction, may explain the sporadic distribution of eclogites in the Proterozoic and the appearance of blueschists since the Cryogenian (Sizova et al. 2014; Brown and Johnson 2019b; Holder et al. 2019; Chowdhury et al. 2021; Holder and Viete 2024). Alternatively, the widespread occurrence of eclogites and the first appearance of blueschists in the rock record could identify the emergence of modern plate tectonics on Earth (Stern 2008).

To evaluate secular change in T, P and T/P we consider the dataset as a single entity rather than by metamorphic type and use locally weighted scatterplot smoothing (LOWESS) to calculate continual smoothed curves and residuals (Fig. 8). Confidence intervals for the LOWESS curve were calculated using a t-distribution. We also calculate 95th and fifth percentiles of the data using rolling windows of the nearest 5% data. For the T, P and T/P data, next we performed bootstrap sampling 500 times on a random selection of 1000 data points using a CUSUM model. The changepoints were combined and fitted with a kernel density estimation (KDE) to identify the most significant changepoints (Supplementary material Fig. S1). Although the precise position of the KDE peaks differs owing to the inherent randomness of the bootstrapping approach, the T, P and T/P data all show prominent KDE peaks (changepoints) at around 2.5, 1.9–1.8, 1.1–1.0, 0.4 and 0.08 Ga.

The average T increases from the mid-Paleoarchean to the Proterozoic, and remains high through the Paleoproterozoic and Mesoproterozoic, before decreasing until the late Cretaceous (Fig. 8a). Average P increases slightly through the Archean, associated with the stabilization of the cratons, to the Orosirian, associated with the formation of the Nuna megacontinent (as emphasized by the 95th percentile curve), then drops to its lowest value during the Mesoproterozoic (Fig. 8b). From the Tonian to the Permian average P rises to a high of c. 2.2 GPa prior to a decrease during the Mesozoic and Cenozoic (Fig. 8b). The average T/P, which reflects these changes in T and P, increases slightly through the Archean before decreasing slightly in the early Paleoproterozoic during the formation of the Nuna megacontinent (Fig. 8c). Average T/P reaches its peak in the early Mesoproterozoic before a stepwise decrease to the end of the Paleozoic (Fig. 8c).

To characterize secular change in the distributions of metamorphic T/P, we binned the data using natural breaks in the time series (vertical orange bars in Fig. 7) as follows: >2.85, 2.85–2.25, 2.25–1.35, 1.35–0.85, 0.85–0.195 and <0.195 Ga (see Holder et al. 2019). Histograms and probability distribution functions (PDFs) for T/P (°C GPa⁻¹) plotted logarithmically to emphasize variations in the distribution are shown in Figure 9. Rather than fitting a Gaussian mixing model to each distribution (Holder et al. 2019), we used the FindDistribution function in Mathematica (https://reference.wolfram.com/mathematica) to identify a simple functional form to fit each distribution of log₁₀T/P. The data >2.85 Ga in age and for the period 2.85–2.25 Ga define single (unimodal) normal distributions (Fig. 9a and b), the data for the period 2.25–1.35 Ga exhibit a broad unimodal left-skewed distribution whereas the data for the period 1.35–0.85 Ga have a uniform distribution with the development of a distinct low T/P mode (Fig. 9c and d), and the data for the period 0.85–0.195 Ga define a broad trimodal distribution with the most prominent mode at lowest T/P whereas the data <0.195 Ga in age show a broad bimodal normal distribution, with distinct lower and higher T/P peaks (Fig. 9e and f). In essence, the data show a gradual transition from a narrower unimodal normal distribution of metamorphic T/P in the Archean eon to a broader bimodal normal distribution since the Triassic. The histogram distributions are qualitatively comparable with those of Holder et al. (2019) using the Brown and Johnson (2019a) dataset. In both cases, the low T/P peak becomes prominent only after the Mesoproterozoic, consistent with the first appearance of blueschists and UHP metamorphic rocks.

For metamorphic T/P data older than 2.25 Ga, when the prevailing tectonic mode is debated (Brown et al. 2020a), the log₁₀T/P data define single (unimodal) normal distributions that include both intermediate T/P and high T/P types of metamorphism (Fig. 10a). Here we compare the natural data with P–T information retrieved from markers (as analogues for rocks) in 2D thermomechanical models of geodynamics for which plate-tectonic behaviour is not predicted (Fig. 10b).

First, we consider P–T information for 545 markers taken from different settings in an experiment performed using a model of a stagnant–deformable lid tectonic mode with initial conditions appropriate to the Archean (T_P = 250°C higher than the present day), no lateral lithological variation in the initial set-up but with a small temperature perturbation, and a model run time equivalent to c. 200 Myr (Sizova et al. 2015). The plots show clearly that the natural data are comparable with T/P values for markers from small-scale crustal overturns and downwelling of inter-domal crust (sagduction) and local thickening of the crust in the experiment (Fig. 10). Downwelling or dripping of crust into the mantle generates a range of T/P values at low T that overlap the low T part of the range defined by small-scale crustal overturns. Whether any of the natural data correspond to this tectonic setting is unclear.

More recent experiments have used a model that assumes subduction and continent–continent collision as the starting point to investigate what happens during continuing convergence after initial collision for a range of ambient mantle T_P (Chowdhury et al. 2017). In these experiments, the prescribed velocity field leading to the initial collision is removed towards the end of subduction and the subsequent convergence evolves spontaneously. Experiments at ambient mantle T_P appropriate to the Neoarchean (T_P = 200°C higher than the present day) predict a ‘peel-back’ style of behaviour that is transitional between stagnant-lid and plate-tectonic modes. Markers in these experiments also reproduce the P–T conditions of natural data >2.25 Ga in age well, as shown by Chowdhury et al. (2020) in their figures 3 and 5.

Modelling has also been used to investigate the role of an initial thinning perturbation of an otherwise uniformly thick lithosphere in the style of mantle convection and crustal thermal regimes (Capitanio et al. 2019a). At an ambient mantle T_P argued to be appropriate to the Archean (T_P = 260°C higher than the present day), a mobile-lid mode that differs from plate tectonics is predicted. However, in this experiment markers record predominantly intermediate T/P metamorphism and the model did not generate high T/P metamorphism (Capitanio et al. 2019a, fig. 3B).

Importantly, the range of T/P values derived from these experiments using thermomechanical models is generally comparable with the range of T/P values retrieved from the early crustal archive. This similarity demonstrates that the single (unimodal) normal distributions of log₁₀T/P data for ages >2.25 Ga could have been generated in any of a variety of non-plate-tectonic modes.

If Earth has had plate tectonics essentially since crystallization of the last magma ocean, features that are considered diagnostic of plate tectonics since the Triassic should be recognizable (much) further back in time. We first assess whether the temporal distribution of diagnostic igneous and metamorphic rocks is broadly representative or, rather, is biased to the point of being unreliable. Then we discuss secular changes in the igneous and metamorphic records and consider the implications for the emergence of plate tectonics on Earth.

Given that geographical areas are not equally accessible and samples are generally selected non-randomly in both space and time, sometimes with preferences for some locations and/or rock types (and constituent minerals) over others, spatial and temporal bias in the petrological record is inevitable. In addition, the fidelity of the petrological record could be affected by differential recycling of crust back into the mantle, which could have biased the surviving crustal archive.

Hawkesworth et al. (2009) argued that the correlation between peaks in the distribution of zircon crystallization ages for continental crust and the supercontinent cycle or the formation of supercratons was a function of preservation, although this interpretation has been disputed, most recently by Puetz and Condie (2021). Similarly, the crustal record of metamorphism is related to the supercontinent cycle and the formation of supercratons (Brown 2007a), consistent with the expectation that most regional-scale metamorphic belts are related to convergent plate margins and collisional orogenesis (Liu et al. 2022). Prior to the formation of supercratons in the Neoarchean, preservation of crustal rocks was probably related to craton stabilization (Jaupart et al. 2014; Beall et al. 2018; Wang et al. 2018; Tang et al. 2020).

For oceanic crust, even Phanerozoic MORB are poorly preserved because they have the thinnest lithosphere and are typically subducted. Those that are preserved commonly occur in subduction–accretion complexes (e.g. the Franciscan; Wakabayashi 2015). Oceanic plateau basalts are better preserved, as oceanic plateaux are more difficult to subduct; remnants of accreted oceanic plateaux are found, for example, all around the Pacific Ocean (Safonova and Santosh 2014). In the Archean, subduction of oceanic crust could have been more difficult, but accretion more likely (Davies 1992).

Although metamorphic rocks with a wide range of T/P ratios have survived since the Mesoarchean (Fig. 8c), orogenic eclogites are conspicuously absent from the Archean record and occur only sporadically in the Proterozoic until the Cryogenian, after which their occurrence, together with the appearance of blueschists, becomes widespread (Fig. 7). One reason why the preservation of orogenic eclogites should not be a first-order problem is that buoyant felsic continental crust, within which they commonly occur as boudinaged relics, resists recycling into the mantle (Sizova et al. 2012, 2014; Wang et al. 2021). Subduction of this buoyant crust requires it to be strong enough to remain coupled to the downgoing slab until the metamorphic peak, when sufficient weakening must occur that it can decouple from the slab but not be lost to the mantle (Warren et al. 2008). The temporal distribution of eclogites and blueschists suggests that these conditions were met sporadically during the Paleoproterozoic and widely since the late Tonian, consistent with secular change in ambient mantle T_P (Brown 2014, 2023). The gap in orogenic eclogites from the Statherian to the Ectasian was probably due to a plate slowdown (O'Neill et al. 2022), leading to warmer subduction.

It is commonly argued that information is increasingly lost going back through time owing to erosion of the older rock record (Möller et al. 1995; Willigers et al. 2002), or more specifically as a result of emplacement of low T/P rocks (blueschists and eclogites) at high structural levels in orogens (Copley and Weller 2024). However, the depth of erosion in Proterozoic and Phanerozoic orogens is variable, and low T/P rocks are common in some deeply eroded Phanerozoic orogens whereas older orogens do not necessarily expose deeper crust (Weller et al. 2021). Thus, if formed, some relicts of Archean orogenic eclogite should have survived. This conclusion is supported by the results of experiments using a 2D thermomechanical model that simulate oceanic subduction then continental collision at ambient mantle T_P > 100°C warmer than the present day, which predict large-scale trench rollback and slab breakoff at depths that generally preclude formation of low T/P metamorphic rocks (Sizova et al. 2014; Perchuk et al. 2019). Thus, the absence of orogenic eclogites from the Archean rock record probably indicates that an appropriate tectonic setting was not available (Sizova et al. 2014; Chowdhury et al. 2020; Tamblyn et al. 2021).

By contrast, cratonic eclogites preserved in the lithospheric mantle record T/P of 180–220°C GPa⁻¹, comparable with Phanerozoic blueschists and orogenic eclogites (Aulbach and Smart 2023). However, although the mechanisms by which the cratonic lithospheric mantle formed and cratonic eclogites were preserved are uncertain, the peak P–T conditions locate them at depths above the lithosphere–asthenosphere boundary (Aulbach and Smart 2023). Thus, the low T/P values recorded by cratonic eclogites probably represent conditions during craton formation.

It is possible that warmer ambient mantle T_P (Herzberg et al. 2010; Herzberg 2022) combined with slower average metamorphic cooling rates (Willigers et al. 2002; Chowdhury and Chakraborty 2019; Brown et al. 2022; Zou et al. 2023) and a different tectonic style (Sizova et al. 2014; Chowdhury and Chakraborty 2019) earlier in Earth history could have led to more widespread overprinting of eclogites during exhumation, whereas cooler ambient mantle T_P and faster average cooling rates since the Mesoproterozoic (Zou et al. 2023) could have limited retrogression of such rocks. However, as was pointed out a long time ago by Coleman et al. (1965), there are ‘eclogites and eclogites’. Overprinting is common in eclogites characterized by higher T/P values, which are generally associated with high-pressure granulites and migmatites; these are the ‘Group B’ eclogites of Coleman et al. (1965) that occur throughout the Proterozoic and Phanerozoic rock record (Wang et al. 2021). The overprinting, which is generally partial rather than complete, may be related to a combination of slower cooling rates and longer exhumation times for these eclogites (Wang et al. 2021). By contrast, overprinting is both less common and less pervasive in eclogites characterized by lower T/P values (‘Group C’ eclogites of Coleman et al. 1965), which are characterized by faster cooling rates and shorter exhumation times (Wang et al. 2021). That eclogites with low T/P values are largely absent from the record during the mid-Proterozoic, a time when mountain belts tended to be hotter and thinner owing to a warmer mantle (Spencer et al. 2021; Zou et al. 2023), should not be a surprise. Notwithstanding, overprinting should not generate bias through time.

The appearance of blueschists in the rock record has been interpreted to register a change to colder subduction and the beginning of the modern plate-tectonic regime (Brown 2006, 2007a, b; Sizova et al. 2014). Alternatively, it has been proposed that blueschists have become more prevalent since the late Precambrian owing to secular change in the composition of oceanic crust (Palin and White 2016), which became less magnesian because of secular cooling and decreased mantle melting (Johnson et al. 2014). This hypothesis is based on the calculated P–T stability of glaucophane, which is modelled to be stable in rocks with MgO contents ≤11.2 wt%, but is not predicted to form in rocks with higher MgO contents at low T/P (Palin and White 2016). However, more than half of basalt compositions in ancient greenstone belts have MgO-poor (≤11.2 wt%) compositions (dataset of Condie et al. (2016); n = 3414). Furthermore, such MgO-poor basalts occur as blocks in ophiolitic mélanges along an inferred arc–passive continental margin suture formed at c. 2.5 Ga in the North China craton (Wang et al. 2019; Jiang et al. 2020). Thus, if tectonic settings that could have generated low thermal gradients were widespread during the Precambrian, blueschists would be expected to have formed, although whether they could have all been lost (recycled) is an open question. The absence of blueschists prior to the late Tonian was not due to an absence of suitable rock compositions but to the prevailing tectonic mode (Foley 2020).

Ultrahigh-pressure (UHP) metamorphic rocks are variably distributed but occur throughout the geological record since the Tonian with no protracted gaps (Brown and Johnson 2018, 2019a, b). By contrast, except for a single eclogite xenolith in a Paleoproterozoic carbonatite in the North China craton (Xu et al. 2018), UHP metamorphic rocks are absent from the rock record before the Cryogenian. Subduction of continental lithosphere to mantle depths requires that slab breakoff occurred after the leading edge of the continent had reached UHP conditions and below the subducting continent, otherwise the continental lithosphere could have been recycled back into the mantle (Sizova et al. 2014). This condition has been met only since the Cryogenian (Brown and Johnson 2019b). Consistent with this interpretation, there is no evidence from the mantle isotope record of recycled upper continental crust before the Phanerozoic (Doucet et al. 2020; Jackson and Macdonald 2022), which suggests that shallow slab breakoff or a change in subduction dynamics may have been the limiting factor in the secular distribution of UHP eclogites (Sizova et al. 2014; Foley 2020; Chowdhury et al. 2021; Gunawardana et al. 2024).

Our search for oceanic crust in the geological record found that the oldest examples of non-subduction ophiolites that satisfy both geochemical and geological criteria for plate-tectonic behaviour are the well-studied Paleoproterozoic (c. 2 Ga) examples from northern Scandinavia and Canada. The great majority of Archean greenstones identified as potential ophiolites do not fall within the low Th/Nb oceanic basalt (OcB) array. Instead, they form trends towards continental crust indicative of continental intraplate settings (Fig. 3b and c). Part of the problem may be the informal expansion of the definition of ophiolite to allow submarine volcanism and associated subvolcanic intrusive rocks in continental rift zones to be included as ophiolites.

Moreover, even the rare ‘ophiolites’ that do satisfy geochemical criteria for oceanic crust commonly have geological criteria (e.g. matching ‘ophiolitic dykes’ with dykes intruding continental basement; presence of crustal zircon xenocrysts; presence of clastic, detrital sediments) that favour highly attenuated continental rift settings ahead of ocean ridges or plateaux. From a geochemical perspective, the most promising examples of oceanic crust are from the Baltic Shield, although these await full evaluation of geological criteria. Because of insufficient geochemical data, we did not consider a small number of terranes where the geological criteria point to an oceanic setting; for example, the Aberdeen Lake Supracrustal Belt in the Central Rae terrane, Canada (Hunter et al. 2018). Nor did we consider several examples of proposed oceanic crust that have experienced upper amphibolite-facies metamorphism leading to Th mobility during high-temperature deformation, metamorphism and anatexis (Fig. 3d and e). These amphibolites include Neoarchean examples from the North China craton and Eoarchean examples from northern Canada and Greenland, any of which could be found to have oceanic affinities following more detailed assessment. However, at the time of writing, we are not aware of any Archean greenstones that pass all our geochemical and geological tests for an oceanic crust origin.

The lava- and dyke-based study presented here is also consistent with studies of the deeper parts of ophiolite complexes that, although beyond the scope of this study, are typically less diagnostic and can even give misleading results. For example, although mantle tectonites in ophiolites are good plate-tectonic indicators, they can be difficult to distinguish from metamorphosed and deformed ultramafic cumulates from plume-derived large igneous provinces. The fact that several of these proposed Archean oceanic mantle outcrops have since been shown to have a cumulate origin (Szilas et al. 2015; Waterton et al. 2022; Zhang and Szilas 2024) emphasizes the likely dearth of oceanic lithosphere in the Archean. In our view, the only clear example of an Archean mantle tectonite lies within the chromites and peridotites of the c. 2.5 Ga Dongwanzi ophiolite (Kusky et al. 2004), although that is not of MOR-type.

It is difficult to escape the conclusion that almost all proposed examples of Archean oceanic crust were the product of plumes, resulting in mantle melting beneath variably extended continental lithosphere. With increased extension, the magmas had less and less crust to assimilate and the volcanic rocks thus appear increasingly oceanic in their geology and geochemistry. Eventually, in a subset of cases, the beta value approached infinity and oceanic crust, in the likely form of a volcanic plateau, was the result. The key question is whether that plateau developed into a stable and long-lived spreading axis, the crust from which has since been lost during subduction, or whether extension and oceanic crust generation stopped before any sizeable ocean formed. In the latter case, the volcanic plateau might subsequently have become accreted and preserved when the local tectonic regime reverted to convergence, but this would not be a plate-tectonic indicator.

To understand better this Archean transition between rifting and potential spreading, an important consideration from an igneous perspective is the relative role of active (plume- or uplift-driven) v. passive (horizontal, tensional stress-driven) rifting (Merle 2011). The present understanding is that most of the Archean was dominated by active rifting, although the later stages of such rifting (i.e. at the continent–ocean transition) might still have developed a passive rifting component. Putirka (2016), using direct determinations of T_P from ultramafic and mafic Archean igneous rocks, obtained a bimodal T_P distribution made up of a larger (plume T_P) and a smaller (ambient T_P) peak, which could be interpreted, in part at least, as reflecting contributions from both active and passive rifting. Combining T_P measurements with geochemical proxies of the type used in this study might provide useful new information on the transition from rifting to possible sea-floor spreading in the Archean.

As with the search for oceanic crust, our search for arc magmatism in the Precambrian geological record finds the oldest clear example of a volcanic arc in the Paleoproterozoic (c. 2 Ga) Trans-Hudson Belt of northern Canada. Here, volcanic arc magmas erupted during a long period of ocean closure as part of a tectonic (Wilson) cycle that included earlier sea-floor spreading and subsequent terminal collision leading to the formation of the Nuna/Columbia supercontinent. However, in our geochemical evaluation of the Archean volcanic units purported to be subduction-related, we find most to be the products of interaction between intraplate magmas and crust. Of the remainder, Archean volcanic rocks related to active subduction are rare, but include the 3.12 Ga Whundo Terrane in the Pilbara craton and various units of c. 2.7 Ga age within the Superior Province. Although described as ‘island arcs’, these more closely resemble continental arcs on the projection used.

Further, there is good evidence for active arc magmatism in the Eoarchean, with parts of both the Isua and Nuvvuagittuq terranes having arc-like characteristics, although alternative, non-uniformitarian options that may explain the data still need to be fully addressed. An additional, but important, plate-tectonic indicator is the inherited subduction signature found in lavas sourced from metasomatized mantle lithosphere and erupted in intraplate, post-subduction settings. Some of these can be traced to an earlier period of active subduction, whereas others cannot, the latter suggesting that subduction was more widespread than the relatively few convincing examples of active arc magmatism might imply.

There is insufficient space to cover the role of plutonic rocks in the search for plate tectonics, although much of the relevant Archean granitoid literature has been summarized recently, including by Halla et al. (2017), Moyen and Laurent (2018), Moyen (2020) and Laurent et al. (2024). However, the links between these granitoids and the different types of mafic–felsic volcanic series (non-arc, active arc and inherited arc) identified in Figure 5 are less well understood, but considered briefly on the Th/Yb–Nb/Yb projection shown in Figure 11. The starting point is the base diagram in Figure 11a, which shows density contours based on the 566 granitoids with all three Th, Nb and Yb analyses from the Moyen and Laurent (2018) Archean granitoid database. Average values for the main granitoid types (from table 2 of Moyen and Martin 2012) all lie within the 80% contour and are closely spaced on this projection. Figure 11b–d then compares three examples of granitic rocks that are coeval with each of the three main types of Archean volcanic series, as follows.

1. The intraplate trend (Fig. 11b) focuses on the East Pilbara greenstones in Figure 5b, expanded to all suites in the c. 3.47–3.45 Ga age range, together with granodiorites from the coeval (c. 3.46 Ga) part of the Shaw batholith (Bickle et al. 1983). The BADR volcanic trend ends with overlap between rhyolitic volcanic rocks and granodiorites, which can be inferred to represent the crustal endmembers of a plume–crust interaction trend.

2. The inherited subduction trend (Fig. 11c) focuses on the Mallina Basin (Bookingarra group) volcanic rocks in Figure 5c (Smithies et al. 2004), together with data from the high-Mg diorite (sanukitoid) intrusive rocks of similar age (c. 3 Ga) from the same tectonic setting (Smithies and Champion 2000). This SHMB–HMA volcanic trend extends towards the sanukitoids, which Smithies and Champion (2000) showed to have a mantle origin, formed by fractionation of the mantle-derived SHMB series (coupled with variations in composition and melting of the SZLM source) with limited crustal input.

3. The active arc trend (Fig. 11d) focuses on the 2.74 Ga Uchi–Confederation volcanic sequence in the Birch–Uchi greenstone belt plotted in Figure 5e (Hollings and Kerrich 2000) together with the coeval hornblende tonalite–granodiorite component of the adjacent Berens River intrusive sub-province (Stone 1998; Stevenson et al. 2009). This volcanic BADR trend ends with overlap with the tonalites and granodiorites, which would be expected, by analogy with present-day arcs, to be the product of interactions between mantle wedge-derived magma and continental crust in deep crustal hot zones and discrete magma chambers.

On the projection shown in Figure 11b–d, the principal Archean plate-tectonic indicator is the distinction made here (and, in part, by Smithies et al. 2018) between the mafic lavas of the subduction-related series with constantly high Th/Nb and the mafic intraplate lavas with low-high Th/Nb. However, as compositions become more felsic, the trends merge and their effectiveness as plate-tectonic indicators becomes less. For these examples, the distinction between inherited subduction and active subduction is negligible, in which case the distinctive mineralogical and geochemical characteristics of sanukitoids described in detail by Smithies et al. (2019) provide an effective identifier of inherited subduction at decision point DP5 in Figure 1. In terms of the granite classification of Moyen (2020): (1) the granitoid endmembers of the intraplate series are C (crustal) type, mainly low- and medium-pressure TTGs, which have little obvious tectonic significance; (2) the granitoid endmembers of the inherited series are of the M (mantle) type, which have the tectonic significance noted above; and (3) the active arc granitoid endmembers of hybrid type, the significance of which will probably depend on whether or not the crustal input has masked the mantle-wedge-derived magma contributions.

Overall, the igneous data evaluated here indicate that crustal materials were almost certainly subducted into the Archean mantle, both into mantle wedges to produce active arc magmas and into subcontinental mantle lithosphere producing subduction signatures that were inherited by subsequent continental intraplate magmas. There is some periodicity to this process in areas studied to date, but a more complete picture awaits a broader geographical coverage and a more thorough tracing of inherited subduction signals back to their sources. Nevertheless, the rarity and ephemeral nature of subduction-related magmatic events in the Archean best supports the concept of short-lived episodes of convergence (Van Hunen and Moyen 2012), perhaps lasting just a few million years as modelled by Foley (2020), that became of global extent only in the Paleoproterozoic.

Metamorphism from the Neogene back to the mid-Tonian is characterized by two prominent features, a bimodal (or trimodal) normal distribution of T/P values with a prominent low-T/P peak (Fig. 9a and b), and all known blueschists and most orogenic eclogites (Fig. 7). This period corresponds to the modern plate-tectonic regime of Brown (2006). From the mid-Tonian back to the early Paleoproterozoic, the lower T peak is first diminished and then eliminated (Fig. 9c and d), corresponding to the Proterozoic plate-tectonic regime of Brown (2006), leaving an Archean era in which metamorphism is characterized by a single (unimodal) normal distribution of high T/P values (Fig. 9e and f) and uncertain tectonic mode. However, these data can also be explained by the variety of tectonic processes that operate in a stagnant–deformable (squishy) lid tectonic mode at higher ambient mantle T_P than today (Fig. 10; Sizova et al. 2015). Overall, the distribution of metamorphic T/P has gradually became broader and more distinctly bimodal since the early Paleoproterozoic; orogenic eclogites are not known from the Archean rock record. The change from a unimodal distribution of metamorphic T/P values in the Archean to a distinctly bimodal distribution since the Mesoproterozoic has been interpreted as recording the emergence and evolution of plate tectonics during the Paleoproterozoic and Mesoproterozoic into its modern style owing to secular cooling of the mantle (Holder et al. 2019; Brown et al. 2022).

We argue that the appearance of blueschists in the late Neoproterozoic and eclogites in the early Paleoproterozoic eras is not the result of bias introduced through recycling. However, the absence from the Archean rock record of these rock types, which are regarded by many as diagnostic of subduction, may be insufficient to falsify the hypothesis that plate tectonics has operated throughout geological history, given auxiliary assumptions (Cleland 2013) and secular changes in, for example, ambient mantle T_P and mantle degassing and regassing (Herzberg et al. 2010; Parai and Mukhopadhyay 2018). Notwithstanding, if we accept the metamorphic record at face value, an alternative hypothesis holds that the first appearance of eclogite then blueschist records fundamental shifts in global tectonics with secular cooling (Brown 2006; Stern 2008), consistent with other geological evidence in the Paleoproterozoic, including the occurrence of contemporary-style continent–continent collision in the 2 Ga Limpopo belt (Yin et al. 2020). Whether this was a change in the style of plate tectonics or a transition from sluggish lid to plate tectonics is uncertain (Höink et al. 2013; Fuentes et al. 2019).

An interesting outcome arising from these studies is the extent to which the igneous and metamorphic perspectives reach the same conclusions about the advent of widespread subduction and the emergence of global plate tectonics. Both perspectives recognize the break-up of several supercratons and their dispersion in the early Paleoproterozoic, followed by convergence and accretion leading to the creation of the Nuna/Columbia supercontinent in the late Paleoproterozoic, as providing the first clear evidence of global plate tectonics. Based on a near-absence of thick volcanic arc sequences and extensive calc-alkaline intrusive systems (the igneous perspective) and of low T/P metamorphic rocks (the metamorphic perspective), neither approach finds support for anything other than short-lived and local plate-like behaviour in the Archean.

The igneous perspective highlights the possibility that much of Archean tectonics is based on crustal extension (divergence) and subsequent shortening (convergence), where crustal terranes may sometimes be extended sufficiently to form oceanic crust. Where this may have been the case, as in the Baltic Shield (Puchtel et al. 1998, 1999), the oceanic crust in question most resembles present-day oceanic plateaux and, although these may be accreted, there is no evidence for bona fide volcanic arcs. Overall, Archean arc volcanic sequences are rare and, although the presence of subduction-modified lithosphere supports the concept of Archean subduction, these probably reflect short-lived (‘failed’) subduction incapable of producing much arc volcanism (Smithies et al. 2018). This concept of local divergence and convergence driven by mantle flow is similar in several respects to the model of Bédard et al. (2013) for the Abitibi Belt in the southern Superior craton. Although it is conceivable that there were larger oceans than envisaged here and that Archean subduction dynamics resulted in a series of short-lived subduction events, we find no evidence for this interpretation.

From the metamorphic perspective, the principal consequence of these short-lived subduction episodes in the Archean is that the processes were essentially ‘frozen’ at the subduction initiation stage. Furthermore, if subduction was sluggish and drip-like, as modelled by Foley (2020), then the slab had more time to heat at shallow depths, inhibiting formation of low T/P metamorphic rocks. Thus, Archean subduction was probably of the ‘hot subduction type’, as proposed by Polat and Kerrich (2006) and others. Notably, hot subduction results in amphibolite-facies metamorphism, not the blueschist- or eclogite-facies conditions that have been the focus of the search for subduction-related metamorphism prior to the Proterozoic. Amphibolite-facies oceanic basalts supporting this assertion may be found throughout the Pacific margins (Sorensen and Grossman 1989, 1993) (Fig. 2f) and in the metamorphic soles of many SSZ ophiolite complexes (Cowan et al. 2014).

In the Western Pacific forearcs, the type examples of modern subduction initiation, although blueschists record oceanic crust accretion during mature subduction (Maekawa et al. 1993; Tamblyn et al. 2019; Miladinova et al. 2024), amphibolites were accreted during the initial stages of subduction (Shiraki et al. 1978). Similarly, boninitic rocks with geochemical signatures of shallow slab melting in the amphibolite facies (e.g. high Hf/Sm) form the earliest Western Pacific protoarcs, whereas the more recent volcanic arcs carry signatures of deeper, garnet-facies slab melting and/or dehydration (low Hf/Sm and Hf/Yb) (Li et al. 2022). This may reflect the temporal change in the thermal gradient at the top of the subducting slab from higher T/P at subduction initiation to lower T/P as the mantle wedge cools. Thus, in the Archean, an absence of long-lived subduction could help explain the lack of low T/P metamorphic rocks.

The temporal evolution of tectonic style is presented schematically in Figure 12, which shows a plot of divergence v. convergence over time. Depending on the extent of divergence, subsequent convergence yields different outcomes. At one end of the spectrum (Path 1), extension does not continue beyond the continental rift stage, in which case convergence creates a moderate T/P metamorphic fold and thrust belt with no (or limited) magmatism. With further extension, to an oceanic crust transition zone (Path 2), some of the resulting terrane may be accreted and partly underplated before terminal collision. Further divergence creates an oceanic plateau with no continental basement, but which does not develop into an ocean of significant size (Path 3). This may allow sufficient subduction to form a short-lived volcanic arc (protoarc) before collision, but under intermediate T/P (high-P amphibolite- or granulite-facies) conditions and without sufficient mantle flow or slab pull to drag the colliding margin to eclogite-facies depths. Paths 4 and 5 then represent closure of major ocean basins via long-lived subduction of oceanic lithosphere that probably created volcanic arcs. These two paths are interpreted to represent a global plate-tectonic system, with the difference being the colder thermal gradients of Path 5 as recorded by the appearance of blueschists and UHP metamorphic rocks in the late Neoproterozoic. The longer subduction continues, the lower the T/P and the greater the likelihood of blueschist or UHP metamorphism. We argue that paths 1–3 are characteristic of most of the Archean, Path 4 is characteristic of at least parts of the Paleoproterozoic and Path 5 is characteristic of present-day long-lived subduction.

A plethora of geological, geochemical and geophysical data has been used to argue for a change in tectonic mode sometime during the Mesoarchean to Neoarchean (3.2–2.5 Ga), as reviewed recently by Cawood et al. (2022). Although a comprehensive review of the literature on this topic is beyond the scope of this article, we briefly consider the results and implications of a few recent studies.

The isotope systematics and trace-element geochemistry of Archean komatiites strongly indicate that the contemporary mantle comprised chemically diverse domains as relics of accretion and magma ocean crystallization. The disappearance of these mantle heterogeneities by the end of the Archean indicates that they had largely been eliminated by up to 2 Gyr of convective mixing (Puchtel et al. 2022). Secular change in the isotopic composition of the mantle from the Archean to the Proterozoic includes a decrease in $δ$⁴⁹Ti in mantle-derived rocks from +0.053‰ around 3.8–3.5 Ga to +0.001‰ by c. 2.7 Ga, which corresponds to the modern depleted MORB mantle composition (Deng et al. 2023). This secular decrease in $δ$⁴⁹Ti through the Archean requires progressive recycling of both TTG and their residues, with mixing in the upper mantle sufficient to ensure isotopic homogeneity in subsequent mantle melting events (Deng et al. 2023). Furthermore, resolvable ¹⁴²Nd and ¹⁸²W anomalies present through the Archean were sufficiently diluted by the Proterozoic that they are unrecognizable (Carlson et al. 2019; Nakanishi et al. 2023). In each case, among several possible explanations, a transition from a precursor tectonic mode to plate tectonics is plausible. Additional support for such a transition comes from noble gas studies (particularly Xe) that show that recycling was unlikely to have been effective before 2.8 Ga, and by 2.5 Ga the convecting mantle had shifted from net degassing to net regassing consistent with subduction of hydrated oceanic crust (Parai and Mukhopadhyay 2018; Péron and Moreira 2018; Zhang et al. 2024b).

Palaeomagnetic data have been used to argue for differential motion between supercratons and/or cratons for short periods between the mid-Neoarchean and mid-Paleoproterozoic (Mitchell et al. 2014; Liu et al. 2021; Salminen et al. 2021), and for cratonic drift rates that were comparable with or exceeded current plate velocities as far back as the Mesoarchean (Brenner et al. 2023; Kasbohm et al. 2023). Liu et al. (2021) proposed that the data provide evidence for a fundamentally different style of geodynamics in the Neoarchean compared with the Proterozoic–Phanerozoic, whereas Salminen et al. (2021) interpreted these data as evidence of plate tectonics. These contrasting interpretations may reflect datasets that are not global in extent, and both could be consistent with a transition to plate tectonics, which was probably completed as the Nuna megacontinent began to assemble (Wan et al. 2020). Consistent with this conclusion and the first widespread appearance of orogenic eclogites in the Paleoproterozoic crustal record, the first direct evidence of cold deep subduction into the lower mantle has recently been reported from mineral inclusions in sublithospheric diamonds retrieved from a kimberlite in the Slave craton (Zhang et al. 2024a).

Lastly, we consider how early subduction zones could have formed and propagated into a network of plate boundaries. On modern Earth, subduction initiation may be spontaneous and due to large lateral density contrasts across lithospheric weaknesses, occurring, for example, at transform plate boundaries or along a plume head margin, or induced by plate convergence, including polarity reversals (Stern and Gerya 2018). By contrast, initiation of the first subduction zones in the Hadean or Archean is poorly understood, with both endogenic (mantle plume) and exogenic (impact-driven) processes as plausible triggers (Gerya et al. 2015; Davaille et al. 2017; O'Neill et al. 2017, 2020).

In numerical models of contemporary Earth, subduction initiates where a density instability develops and where the local stress exceeds the yield stress (Rolf and Tackley 2011; Coltice 2023). Stronger lithosphere increases the probability of subduction initiation in the vicinity of continental margins (Ulvrova et al. 2019) or along microcontinental edges (Zhu et al. 2023). Although the collapse of passive margins is poorly understood, weakening by grain damage (reduction in grain size) is the likely mechanism of subduction initiation at such sites (Mulyukova and Bercovici 2018; Bercovici and Mulyukova 2021). The formation and stabilization of cratons may have been a prerequisite for propagating a plate boundary network. As Earth's mantle began to cool and cratonic lithosphere strengthened (Rey and Coltice 2008) and stabilized (Laurent et al. 2014; Aulbach and Smart 2023), new subduction could have initiated along their margins (Rey et al. 2024). Other factors including continental emergence (Flament et al. 2008), increased rates of sedimentation (Sobolev and Brown 2019) or the deposition of dense banded iron formation (Zhang et al. 2023) could have facilitated the transition to global plate tectonics. As subduction zones lengthened and spread from one or more centres, a connected plate boundary network was arguably completed by the early Paleoproterozoic, although whether plate tectonics has been continuous since is contentious (Brown et al. 2020a; Stern 2020, 2023; Spencer et al. 2021; O'Neill et al. 2022; Roberts et al. 2022).

The most likely interpretation of the changes discussed above is that they record different traces of a common cause (Cleland 2013), the emergence of plate tectonics by the Proterozoic. Whether they do should be tested quantitatively using numerical models (e.g. Seales et al. 2022).

The distinctive petrological features of recent ocean crust, subduction-related magmatism and regional metamorphism generally cannot be unambiguously identified in the Archean geological record. Based on the petrological record, a global plate-tectonic mode can plausibly be extended back to the Paleoproterozoic but not clearly into the Neoarchean. Whatever the mode, Archean mafic magmatism was dominantly related to plumes rather than divergent or convergent plate margins. Metamorphic rocks commonly considered diagnostic of subduction (blueschists and orogenic eclogites) are absent from the Archean. Overall, the Archean metamorphic record is consistent with a stagnant–deformable (or squishy) lid (Sizova et al. 2015; Lourenço et al. 2020). However, the formation of Neoarchean supercratons requires mobility, supporting a sluggish lid (Fuentes et al. 2019) or multi-mode regime, such as lid and plate (Capitanio et al. 2019a).

Other types of geochemical and geophysical data record secular changes, but these occur at different times, from before the Mesoarchean into the early Paleoproterozoic, suggesting a long transition to global plate tectonics. Thus, from a petrological perspective, global plate tectonics is probably a Paleoproterozic and younger phenomenon, contrary to results of recent thermal modelling (Seales and Lenardic 2020; Seales et al. 2022). That interpretations based on the Archean petrological record and the modelling are seemingly contradictory provides motivation for future research.

We thank R. Arculus, K. Putirka and H. Rollinson for review comments and Y. Dilek for his sympathetic editorial handling; any remaining errors or misconceptions are ours.

MB: formal analysis (equal); JAP: formal analysis (equal); TEJ: formal analysis (equal).

M.B. and T.E.J. acknowledge funding from Australian Research Council Discovery Project DP200101104 and support from the State Key Laboratory for Geological Processes and Mineral Resources, China University of Geosciences, Wuhan (Open Fund GPMR201903).

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

All data generated or analysed during this study are available either in accessible cited articles or included in this published article and its supplementary information files.