The inception of plate tectonics: a record of failure

Abstract

The development of plate tectonics from a pre-plate tectonics regime requires both the initiation of subduction and the development of nascent subduction zones into long-lived contiguous features. Subduction itself has been shown to be sensitive to system parameters such as thermal state and the specific rheology. While generally it has been shown that cold-interior high-Rayleigh-number convection (such as on the Earth today) favours plates and subduction, due to the ability of the interior stresses to couple with the lid, a given system may or may not have plate tectonics depending on its initial conditions. This has led to the idea that there is a strong history dependence to tectonic evolution—and the details of tectonic transitions, including whether they even occur, may depend on the early history of a planet. However, intrinsic convective stresses are not the only dynamic drivers of early planetary evolution. Early planetary geological evolution is dominated by volcanic processes and impacting. These have rarely been considered in thermal evolution models. Recent models exploring the details of plate tectonic initiation have explored the effect of strong thermal plumes or large impacts on surface tectonism, and found that these ‘primary drivers’ can initiate subduction, and, in some cases, over-ride the initial state of the planet. The corollary of this, of course, is that, in the absence of such ongoing drivers, existing or incipient subduction systems under early Earth conditions might fail. The only detailed planetary record we have of this development comes from Earth, and is restricted by the limited geological record of its earliest history. Many recent estimates have suggested an origin of plate tectonics at approximately 3.0 Ga, inferring a monotonically increasing transition from pre-plates, through subduction initiation, to continuous subduction and a modern plate tectonic regime around that time. However, both numerical modelling and the geological record itself suggest a strong nonlinearity in the dynamics of the transition, and it has been noted that the early history of Archaean greenstone belts and trondhjemite–tonalite–granodiorite record many instances of failed subduction. Here, we explore the history of subduction failure on the early Earth, and couple these with insights from numerical models of the geodynamic regime at the time.

This article is part of a discussion meeting issue ‘Earth dynamics and the development of plate tectonics'.

1. Introduction

Plate tectonics underpins most interpretations of the Phanerozoic record, yet it is not clear whether it existed for all of Earth's history [1], or whether it emerged from an archaic geodynamic regime during Earth's history [2–4]. The geological and geochemical record on the matter is equivocal (e.g. [5,6]). Plate tectonics was conclusively demonstrated on the modern Earth in the oceanic basins. However, with little to no preserved Hadean/Archaean oceanic crust, one must explore the preserved continental cratonic record for plate tectonic indicators.

The hallmarks of plate tectonics have evolved throughout the literature (e.g. [7]). For example, large horizontal motions and shear zones (e.g. [8]) do not necessarily exclude modes of intra-plate tectonism [6], and may represent a regime different from plate tectonics. Traditional geochemical indicators of subduction [9] have been shown to be ambiguous, and subduction signatures sensu stricto, replete with lithologies such as boninites, are rather rare, occurring at only certain times, which may suggest intermittent plate tectonics only during certain periods. Similarly, palaeomagnetic evidence does suggest some continental drift (e.g. [10–12])—at some extreme velocities—only during discrete periods of time (e.g. 2.7 Ga) interspersed by periods of apparent polar wander (APW) quiescence.

Lastly, there has been the suggestion that modern plate tectonics, consisting of a globally connected subduction/plate boundary network, operating continually, replete with cold, steeply dipping subducting slabs, has been characteristic of the Earth only since the Neoproterozoic [13–16]. Prior to modern cold subduction, subducting slabs may have been hotter and prone to dripping and break-off (e.g. [17]). Using compilations of metamorphic gradients from the Neoarchaean to the Phanerozoic, Brown [14] suggested that a duality of subduction thermal gradients is a characteristic of modern plate tectonics, with ultra-high-pressure (UHP) and eclogite high-pressure granulites occurring coevally through the Phanerozoic as a result of tectonic temperature–pressure environments of modern slab and arc settings. Prior to the Neoproterozoic, this duality is absent, as are (generally) UHP rocks, and Brown & Johnson [15] note that this may indicate a transition from an earlier hot ‘proto’-plate tectonics regime, characterized by failed convergent zones [18], into modern subduction.

Together, the geological record indicates periods of tectonic activity extending into the Mesoarchaean, and arguably subduction [19]. But it is not clear whether (i) subduction resembled modern (cold) subduction zones, (ii) developed into a globally connected plate boundary network, (iii) continued, once begun, unabated to the present, or sporadically ceased [20], and (iv) if indeed it regularly proceeded past the subduction initiation hurdle. These points are, with some debate, hallmarks of the modern subduction regime, and inability to satisfy the last three would indicate a failure of both subduction, and the benchmark for continuous plate tectonic behaviour.

The interpretation of the Precambrian geological record has been reshaped in recent years by the addition of information from geodynamic simulations [3,10,17,18,21–23]. Typically, such simulations solve the basic constitutive equations for dynamic mantle and lithospheric systems, though with differing boundary conditions, rheological complexity and material properties, and internal physics. As a result, the models show a range of behaviour, but some generalizations based on more than one study can be identified:

• (i) The dynamic behaviour of plates is nonlinear [10,21], and under hotter mantle conditions it may become episodic [17,24,25]. The implication is that plate tectonics may be seen sporadically in the geological record, and that subduction, if initiated, may repeatedly fail.

• (ii) Under much hotter mantle conditions (ΔT > 200°C), all subduction and plate tectonics may shut down [18]. This is due to both the low viscosity of the hotter mantle reducing the coupling of convection with the overlying plates [10], and the propensity of subducting slabs to neck and break off [17], hindering their ability to transmit stress to the surface, and maintain continual tectonics. Under these hotter conditions, volcanism may dominate plate recycling as a heat loss mechanism, and such regimes may be more adequately described as volcano–tectonic variants [26,27].

• (iii) A tectonic system may exhibit hysteresis, and demonstrate history-dependence (e.g. [23,24,28,29]. Whether or not a system exhibits plate tectonics, or indeed when, may be a strong function of its history and initial conditions.

Few of these studies have specifically addressed the inception of plate tectonics. Many geological studies implicitly assume that the system behaves in a somewhat binary fashion—there is a tectonic ‘switch’, at which point the system transitions from an archaic geodynamic regime, into modern plate tectonics [30–32]. This conceptual transition is perhaps a little simplified, given both the evolution to modern tectonics observed in the geological record [15] and the range of dynamics observed in geodynamic models. Indeed, the latter suggest that the concept of the origin of plate tectonics needs to be better defined. Do we mean the first subduction events in the record, which, perhaps, preserve discrete events? Or, do we mean the inception of modern, or at least continual plate tectonics, without significant periods of quiescence?

Both definitions may be constrained by evidence. As an example, Turner et al. [33] suggest that the volcano–sedimentary succession observed in the Nuvvuagittuq greenstone belts in Quebec—which have model ages of 4.4 Ga [34], and are cross-cut by approximately 3.8 Ga tonalitic dykes—have similarities to subduction initiation sequences observed in the Izu–Bonin forearc [35]. If true, this may be the oldest indication of subduction initiation in the geological record. However, it is singular that it is the initiation that is preserved, rather than the signature of a mature subduction zone—particularly so given the scarcity of subduction initiation sequences today. This may suggest that the subduction zone did not mature, and subduction failed, or also that perhaps subduction initiation was a common event—which would have varied geodynamic implications.

The aim of this contribution is to explore insights on the inception of plate tectonics garnered from numerical models, and contrast this with observations of the geological record on both the initiation of subduction and its history of failure. We will argue that, while the inception of local tectonics is well documented in some terranes, there is a repeated failure of subduction throughout the Eo-Mesoarchaean, and that this is a hallmark of a transitory regime. Furthermore, we will posit that the evolution from an archaic, perhaps ‘stagnant-lid’/volcanic regime, through this transitory regime, into modern plate tectonics, arguably took much of the Proterozoic.

2. Geodynamic modelling

(a). The subduction initiation problem

Subduction initiation represents something of a bottleneck for self-perpetuating plate motion. It involves an evolving force balance between the converging plates [36] (figure 1). However, the plate forces required for subduction initiation are very large, and it has been suggested that the forces required may only be generated by subduction itself [37,38], presenting somewhat of a paradox.

Figure 1.

(a) Static force balance on an incipient subduction zone. RP, ridge push; CR, collision resistance; DF, mantle drag force; SP, slab pull; SR, slab resistance; SU, slab suction. (b) Development of incipient subduction in a numerical simulation, in response to a mantle upwelling. Colours represent log of mantle viscosity (in Pa s). Here, mantle stresses (grouped DF, SP, SU and SR forces) generate thickening in the adjacent lithosphere and eventually downwelling. (Online version in colour.)

The development of numerical approaches to the problem led to a more nuanced approach to the development of the system's stresses [36,39] (figure 1). Toth & Gurnis [36] pre-imposed an existing weak fault and found that subduction initiation is possible for reasonable plate forces (3–8 × 10¹² N m⁻¹—in line with ridge push estimates), provided the shear stress along the subduction thrust fault is low (less than 10–15 MPa). Such low strengths are in line with estimates of stress drops from earthquakes [40] and heat flow anomalies across plate boundaries such as the San Andreas [41]. Gurnis et al. [39] found that the total cumulative convergence was important, and that models of subduction initiation along fracture zones that exceeded 100–150 km of convergence transitioned from forced to self-sustaining subduction.

The implications for subduction initiation in the past are these: (i) Resistive forces are likely to have been less. The rheological controls that result in low fault strength are not well understood, but likely related to liquid water [42,43], which existed on the surface for much of Earth's history [44]. In addition, the resistance to bending is likely to have been less. This was the dominant resistive force in Gurnis et al. [39]; on the early Earth, hot thin plates result in lower elastic lithospheric thicknesses, and less resistance to bending. (ii) Plate driving forces could be less. Static force balances struggle to deal with the evolving stresses during incipient subduction, and numerical models of incipient subduction demonstrate that the driving ‘induced’ convective stresses in the lid, in a system without subduction, are tightly coupled to mantle temperatures [10,17,24]. Higher mantle temperatures equate to lower effective mantle viscosity, and lower levels of induced convective stress—hence lower plate driving forces. This is the fundamental challenge for early Earth subduction initiation—generating the high stresses needed for subduction, in a low-stress mantle environment. As a result, processes which enhance the levels of lithosphere stress (even temporarily), and facilitate convergence, will enhance the system's capacity for subduction initiation.

(b). The inception of plate tectonics in early Earth models

A number of workers have explored the development of ‘natural’ subduction zones in models of the early Earth [10,21,24,45,46]. These have been differentiated in this section from models where subduction has been imposed (e.g. [17,18,47]), to allow us to examine the details of the emergence of subduction from natural convecting systems. The ultimate balance in all these models is that the driving forces of convection exceed the resisting forces of lithosphere to allow subduction to proceed [10].

The models of Bercovici & Ricard [21] explored the evolution of simple convecting systems incorporating a ‘damage’ rheology, whereby a Zenner-pinning mechanism in an olivine–clinopyroxene mix permits a grain-size feedback, and the accumulation of damage within the lithospheric mantle—leading to failure in these damaged zones. The time scale for the development of a mature subduction system in these simulations was of the order of approximately 1 Gyr. Similar systems were explored by Foley et al. [45] and Foley & Rizo [48], who suggested that, under hotter mantle conditions, the behaviour of the grain growth and damage would lead to slower plates. Foley et al. [45] developed a scaling theory to suggest that subduction in the Hadean may have proceeded slowly—of the order of 1 cm yr⁻¹.

By contrast, many workers have explored the development of tectonic systems using yield-stress rheologies, where, once a defined yield stress is exceeded, the plate deforms plastically, via a modified viscosity (e.g. [49]). O'Neill et al. [10,24] found that the viscous accumulation of strain within a stagnant lid eventually led to over-thickening in convergent zones (figure 2), and thinning in extensional regions, until the point at which the lithospheric strength fell below the yield stress. This resulted in widespread overturn and rapid subduction—a transitionary regime which had been dubbed episodic overturn [49]. O'Neill et al. [10] showed that convecting systems with a yield stress naturally fall into this type of regime for hotter mantle conditions. In O'Neill et al. [24], which incorporated evolving mantle heat production and core temperatures, and similar rheologies, it was shown that systems evolving from a hot initial state naturally pass through a transition from stagnant lid, to episodic overturn, and then to plate tectonics, and eventually to a cold stagnant lid. It was, however, noted that initial conditions were important for the tectonic evolution—similar systems with cooler starting temperatures evolved in a plate tectonic regime over most of their history, waning to a cold stagnant lid as the planet cooled. Nakagawa & Tackley [46] explored similar rheologies for systems with an evolving core and found that the system initially adopted a volcanically active stagnant-lid or ‘heat-pipe’ regime, before evolving into an episodic or plate tectonic regime. They noted that the presence of laterally variable crust was an important factor in subduction initiation in their models, and emphasized the effect of eclogitization of dense crustal material in facilitating subduction.

Figure 2.

Development of a transient subduction zone in an early Earth model (after [50]). Here, downwellings promote over-thickening of the lithosphere, which eventually yields (b) and develops into a short-lived subduction zone (c,d), characterized by frequent necking and slab break-off. (Online version in colour.)

The importance of crustal thickness for subduction viability was first noted by Nisbet & Fowler [51] and Arndt [52], and later quantified by Davies [53]. These authors noted that, under hotter mantle conditions, the oceanic crust would be significantly thicker (greater than 20 km)—which would impart a strong buoyancy to the oceanic lithosphere, perhaps even impeding subduction completely [53]. Resolutions to this problem included the high density of high-temperature mantle melts [52], which, for mantle temperatures of approximately 1600°C, would effectively be komatiite with MgO contents of greater than 20 wt%. Crust composed of these Mg numbers would have densities greater than 3200 kg m⁻³ [25], and thus would have little trouble subducting. Davies [54] made the point that an extensively depleted early mantle (due to widespread continent formation) may have produced a reasonably thin oceanic crust, and facilitated subduction, though the reality of an extensive early depleted mantle has been questioned [55]. Alternatively, van Thienen et al. [56] showed that if the basalt–eclogite transition is incorporated into dynamic models, the thickening of volcanic piles beyond the eclogite transition (greater than approximately 30–40 km, depending on Mg number) results in a density increase in basaltic material of over 600 kg m⁻³; 200 kg m⁻³ denser than peridotite. Their modelling suggested that this effect would facilitate effective recycling of early thick oceanic crust, alleviating a significant criticism against early Earth tectonics.

The buoyancy of volcanic piles under Archaean conditions is insufficient to preclude their subduction. The strength of Archaean lithosphere would be largely dominated by its weakest mechanism, which, while incompletely understood for modern Earth [43], can be constrained geodynamically [57] or thermally [41] to within effective bounds [10]. As plate driving forces might be anticipated to be lower, given lower mantle viscosities, subduction would be near a critical point, and convection models suggest that repeated subduction failures characterize this transitory regime.

(c). Primary drivers of subduction initiation

While modern subduction initiation is facilitated by the engine of plate tectonics today, the associated plate forces were not available to provide the impetus for initiation on a pre-plate tectonic planet. Lower lithospheric stresses, and weaker plates, of the early Earth have led some workers to invoke a ‘primary driver’, i.e. an extreme mechanism, to begin the process [22].

One example is the arrival of large plumes under the lithosphere (figure 3), which was explored by Gerya et al. [22]. They noted that, importantly, the process could operate in the absence of an existing plate network, and involves significant thermal thinning, and magmatic weakening, above the plume head. Lateral flow around the plume in this model causes thickening at the edges, and the development of a circum-plume subduction zone. They note a number of precursors necessary to the development of subduction by this mechanism: (i) An old, thick, negatively buoyant oceanic lithosphere. (ii) Magmatic weakening, which facilitates the penetration of the plume head through the lithosphere, and the generation of spreading ridges. For low levels of magmatic weakening, the plume simply spreads under the lithosphere. Alternatively, a similar effect can be enacted with a more vigorous hotter, larger plume. (iii) Weak oceanic upper crustal rheologies, associated with alteration of the oceanic crust. As noted by Gerya et al. [22], the style, dynamics, size and number of plumes on the early Earth are still largely speculative, but for reasonable assumptions, the mechanism would be operable.

Figure 3.

Development of incipient subduction in response to the arrival of a thermal plume at the base of the lithosphere. The thermal buoyancy of the plume drives lateral spreading at the surface (26.25 Myr), and the development of localized subduction at the edges (29.5 Myr). Also shown is a compositional field equating to ‘MORB’ concentration (purple is depleted, harzburgitic mantle; yellow equates to more MORB-like compositions). The eclogitization of dense basaltic drips facilitates subduction in over-thickened regions.

In addition to magmatic weakening, volcanic loading has been demonstrated to initiate plate fracture in places like Hawaii today (e.g. [58]), and the extreme volcanic piles observed in places like the East Pilbara (e.g. van Kranendonk et al. [6,59]) might facilitate lithospheric fracture and thus the development of subduction. A number of recent studies have explored the effect of voluminous magmatism on the cooling efficiency and tectonic regime of the early Earth [26,27,46,60]. Fischer & Gerya [61] extended the models of Gerya et al. [22] to explore the dynamics of continental crust during plume impingement, and found it induced a regime where thickened crustal piles generated eclogite at depth, which was episodically recycled into the mantle. They noted that full-blown subduction failed to initiate in their models, largely due to the weakness of the subducting material, which tended to neck and drip off, in a similar fashion to van Hunen & Moyen [17]. Lourenço et al. [60] showed that development of melting-induced crustal production in their models affected the yielding behaviour of the system, and hence the transition between tectonic regimes. Crustal production was shown to facilitate the development of active tectonics under present-day mantle conditions, with mobile or episodic lid behaviour more common in models with crustal production than those without. Volcanism was purely extrusive in these models, and the produced basalt was inserted at the top boundary, creating a load that affects the lid's stress state.

A model of the effect of extreme volcanism on mantle convection was presented by Moore & Webb [26], who explored a ‘heat-pipe’ model for the Hadean mantle. Here, mantle melt was efficiently transported to the surface, contributing to a thickening crustal pile. At the same time, melt transport efficiently advected mantle heat, which was lost at the surface during volcanic emplacement, effectively refrigerating the mantle, and leading to the development of a thick thermal lithosphere, over a hot convecting interior. As this melt advection (and cooling) mechanism is more effective at hotter mantle temperatures, this had the effect of preventing subduction initiation, and plate tectonics, in their models. Subduction could not progress until the mantle had cooled sufficiently that the mantle-melting, heat-pipe transport was less dominant.

Heat-pipe models implicitly assume that all mantle melt is erupted, and none emplaced within the crust or mantle. A ratio of volcanic to magmatic emplacement (the eruption efficiency) of 100% greatly exceeds estimates of intra-plate eruption efficiency today (approx. 20% [62]), and so a number of workers have extended this basic model to incorporate the emplacement of plutonic melt. Rozel et al. [27] used the eruption efficiency as a free parameter, to explore the effect of extrusive/intrusive mantle-derived magmatism on crustal temperatures, and thus trondhjemite–tonalite–granodiorite (TTG) formation. They assumed that intruded melts were emplaced at the base of the crust, and thus contributed to weakening the crust and lithosphere. Their results indicated that the thermal conditions likely to produce the observed ratios of low-, medium- and high-pressure TTGs were most likely to be met for intrusion efficiencies of 65–80%, depending on the modelled coefficient of friction. They suggested that the early Earth may have been in a plutonic ‘squishy-lid’ tectonic mode, with weak, warm lithosphere prone to dripping, and also global overturn events. The transition through a volcano–tectonic regime in some ways circumvents the subduction initiation issue of a strong stagnant lid, through the weakening inherent in these systems.

An alternative to endogenous processes for developing incipient subduction are bolide impacts [50,63–65]. Hansen [63] suggested that large meteorite impacts might drive local tectonic flow, and that impacts might drive tectonism on planets above a critical size. Price [64] and also Yin [65] have speculated that subduction on the Earth (and other bodies) may have been driven by bombardment effects. O'Neill et al. [50] (figure 4) developed a quantitative model for the thermal effects of bolide impacts on the mantle, based on the approach of Roberts et al. [66], and refined it using hydrocode simulation results [67]. They found that large, single impacts could generate a substantial thermal anomaly in the mantle, and the upwelling induced by this anomaly was sufficient to generate extensive spreading over the point of impact, and convergence in a zone circumferential to that. For large enough impacts (greater than approx. 300 km diameter), this could develop into a major subduction zone, often capable of driving a global lithospheric overturn. They also explored the effect in systems with both an evolving internal heat production and core temperatures (a prerequisite for Hadean conditions), and a series of evolving impact fluxes, based on early Solar System impacting rate estimates. They found that large early impacts could both drive temporary subduction events, and, due to their heat addition, prime the mantle system for subsequent tectonics. As a result, small upticks in meteoritic flux, as recorded at approximately 4.1 Ga in many Solar System cratering analyses, can act as triggers for plate tectonic activity. In their models, O'Neill et al. [50] showed that small impacts (less than 100 km in diameter) may trigger local subduction initiation in systems primed for an overturn, which may then evolve into regional, albeit rapid, subduction episodes.

Figure 4.

Development of subduction in response to the thermal effect of a giant (1000 km diameter) impactor (after [50]). (Online version in colour.)

Geodynamic systems without subduction may be pushed into a subduction regime by external drivers, such as giant impacts, the imposition of large plumes or volcanic effects. Interestingly, in many cases, the effects of these events may be ephemeral, with systems reverting to a stagnant-lid state as subduction wanes.

(d). Importance of planetary history

The mantle's viscosity is extremely dependent on its temperature, and this was demonstrated by Tozer [68,69] to be an extremely efficient regulator of thermal evolution. If the mantle was hotter, due to inefficient heat extraction, or higher internal heating, the viscosity of the mantle would be less, leading to more vigorous convection, greater efficiency of heat loss and a lowering of mantle temperatures. If the mantle cooled too much, higher viscosities would lead to sluggish convection, and inefficient heat loss. Coupled with ongoing internal heating, this would raise mantle temperatures. Tozer [69] suggested that this would keep mantle temperatures at a level determined by the decay of internal heat production through time.

However, with the development of mantle convection models capable of simulating plates (e.g. [49,70,71]), it became apparent that the heat lost from a convecting system was a function not solely of viscosity, but also of tectonic state. Active or mobile-lid tectonic systems lose heat far more effectively than do stagnant-lid regimes, and thus are able to maintain systematically lower internal temperatures, and lower degrees of melting.

Stein et al. [72] and O'Neill et al. [10] both suggested that the tectonic state of a system may trend towards a hot stagnant-lid regime under high internal heat production conditions, such as anticipated on the early Earth. However, the exact nature of the transition depends strongly on a number of critical variables, including not only heat production and yield strength but also initial conditions [24,73]. Crowley & O'Connell [28] and Weller & Lenardic [29] both explored the effect of initial conditions on the tectonic state of a system, arguing that, for system parameters near tectonic transition boundaries, no one clear tectonic state existed, but that the system was ‘bi-stable’—capable of existing in one of two tectonic states, depending on its history. The transition boundaries themselves in these simulations depend on initial conditions. Weller et al. [23] argued that such systems display hysteresis—the system state depends on its prior history. For the early Earth, this suggests that the probability of subduction depends significantly on its initial state—something we have very little information about given the scarcity of samples from the Hadean. O'Neill et al. [24] explored a range of initial conditions in full evolutionary models, demonstrating that the early Earth (4.5–2.5 Ga) may demonstrate either a stagnant-lid to episodic mode, or plate tectonics, for reasonable ranges in starting parameters.

Clearly, the range of possible outcomes found in geodynamic simulations, and their potential range of dynamics, while illustrative for planetary models, is unsatisfactory for understanding the specific history of the early Earth. In the following section, we touch on the observational constraints on the particular planetary history of Earth.

3. The geological history of subduction initiation and failure

(a). Palaeomagnetics and plate motion

Palaeomagnetic data provide solid evidence for continental drift relative to a geocentric axial dipole, and thus provide key evidence for contemporary plate tectonics. Such data also have the potential to constrain its likelihood on the early Earth. The challenges, however, have been recovery of high-quality data from ancient cratons, comprehensive enough sampling to reconstruct valid APW curves, and high-quality age dating on the relevant, often mafic, units.

Extensive mapping of many cratons has led to a renaissance in their geochronology, which has benefited palaeomagnetic investigations. Although cratons have had a protracted tectonic history, many areas have avoided significant metamorphism above the Curie temperature which would have reset their magnetization (e.g. [74]), and while factors such as lightning exposure are still important [75], these processes are identifiable during demagnetization.

One of the earliest detailed studies of a single terrane was presented by Strik et al. [12], who showed a detailed palaeomagnetic path for the Pilbara terrane over the interval 2.86–2.71 Ga, and its continuation until approximately 1.70 Ga. The preservation of natural remanence magnetism was demonstrated by fold, conglomerate and reversal tests over a series of predominantly basalt samples spanning the interval 2.860 Ga to approximately 2.715 Ga. Their results indicate very little difference in palaeolatitudes in the interval 2.772–2.721 Ga, followed by an abrupt shift in palaeolatitude of 27.2° over an unconformity at 2.721–2.718 Ga. This translation equates to over approximately 3000 km motion, which, depending on the true time span of the unconformity (3–10 Myr), equates to minimum plate velocities of the order of approximately 100 cm yr⁻¹ at a minimum—an order of magnitude larger than typical fast plate velocities today [12]. Following this rapid motion, extension of the APW path into the Hamersley Basin using the data of Li et al. [76] suggests fairly limited palaeolatitudinal motion from 2.715 Ga until almost 1.70 Ga. Strik et al. [12] interpreted this APW as evidence for plate tectonics, albeit at a rapid pace; although the characteristics of the motion are more akin to an episodic regime previously discussed.

An attempt to link the APW velocities of a number of terranes over these time intervals was made by O'Neill et al. [10], who compiled high-quality (Q > 3) palaeopoles from the Pilbara, Kaapvaal, Fennoscandian, Slave, Superior/Laurentia and Siberian cratons. Available high-quality data on these cratons spans different age periods, and it was noted that different sampling densities can lead to, for example, lower APW velocities during intervals of few volcanic ages. For similar sampling densities, however, it was noted that different terranes display similarities in their APW motion. For example, Kaapvaal and the Pilbara both display fast APW velocities around approximately 2.7 Ga, and the Kaapvaal, Slave and Superior/Laurentia display increases in APW velocities at approximately 2.1 Ga. The change in the dispersion of the poles during these intervals was argued by O'Neill et al. [10] to be due to relative motion, rather than true polar wander (TPW). Between these periods of rapid motion are lulls—in many cases, the uncertainty ellipses of poles up to 100 Myr apart overlap, suggesting very slow, or no, plate motion. O'Neill et al. [10] suggest that this may be due to an episodic overturn-type regime being preserved in the geological record.

Following this, Piper [11,77] recompiled and analysed the Precambrian palaeomagnetic database for evidence of ‘lid tectonics’—long periods of stagnant lid, interspersed with occasional resurfacing events, akin to episodic convection in the modelling community. Piper [77] makes the point that palaeomagnetic tests for plate tectonics are difficult, as they comprise very detailed APW paths for different continental blocks that diverge through time. He notes the converse—a test for an immobile lid requires only a long progression of coincident poles—and is fairly easy to test. Piper [77] noted long intervals in the Precambrian between 2.7 and 2.2 Ga, 1.5 and 1.3 Ga and 0.75 and 0.6 Ga where the requirements for lid tectonics are met. These intervals are interspersed by large focused loops of polar wander, suggested by Piper [11] to reflect TPW contributions.

Two of the problems with analysis of APW velocities are (i) the difficulty in constraining uncertainties in such velocity estimates, which depend both on the quality of the pole and on the accuracy of the magnetization age, and (ii) the difficulty in propagating such estimates forward during intervals of sparse coverage, and understanding uncertainties during these periods. In figure 5, we have reanalysed the palaeomagnetic record of both the Pilbara and Kaapvaal cratons—two of the only cratons with records for ages greater than 3.0 Ga—which thus cut across the commonly proposed interval for plate tectonic initiation. The raw poles were taken from the Paleomagia database [79], filtered for pole quality (greater than 2), and also restricted age uncertainties (less than ±75 Myr). To estimate the uncertainties in the angular velocities for these intervals, we have adopted a Monte Carlo approach, where we construct a cloud of 100 points around a pole from a Fisher distribution based on the pole's A95 value. The angular distance from these points, and the swarm around the subsequent pole, can be used to create a distribution of angular distances, from which the mean and standard deviation can be derived. We use a similar approach to estimate the time differences, adopting a cloud of age estimates from a truncated normal distribution, limited by the upper and lower age estimates of the units, and subtracting the points from the subsequent age date distribution. Together, these data distributions can be used to derive the mean angular velocity, and its standard deviation, based on a full treatment of the pole's positional uncertainties, and age uncertainties. These are shown as blue diamonds in figure 5, with uncertainty bars representing one standard deviation.

Figure 5.

APW velocities for the Pilbara (a) and Kaapvaal (b) cratons, as well as a combined path (c). APW estimates (blue diamonds) and uncertainties are calculated using a Monte Carlo approach (see text), incorporating both spatial uncertainties in the poles and age uncertainties. We have also used a propagating one-dimensional Kalman filter on the data (magenta line, uncertainties shaded magenta). The Kalman filter implements a motion model (here velocity depends on previous velocities—see text), which both reduces variance when it encounters noisy observations, and facilitates the propagation of uncertainties into regions of sparse data sampling (shown as increasing uncertainty regions between APWV measurements). The plots show estimates of APWV far above background levels at certain periods of time (e.g. 3.5, 3.2, 2.7–2.8 and 2.1 Ga). In between these peaks, it is difficult to ascertain any clear APW motion above uncertainty levels (particularly prior to 3.0 Ga). Compiled poles/data and scripts are available from online repositories ([78]: https://doi.org/10.5281/zenodo.1310984). (Online version in colour.)

Two limitations of this data treatment are apparent. First is that the Monte Carlo error estimates are generally worst-case, and we know little about the APW velocities in times of poor sampling. To address these, we have performed a one-dimensional Kalman filter on these data. Kalman filters (e.g. [80]) are commonly used in engineering and geophysics, and take a known system dynamic behaviour to propagate solutions forwards. When combined with (noisy) observations on the system, the total uncertainty can be reduced. For our one-dimensional system, we make the standard dynamic assumption that the forward positions are related to previous positions and velocity, and velocity itself is largely predicated by previous velocities, as is the case for plate motion today [81]. However, plate motions do change, and to permit a weak dependence of future velocities on rate of change, we have included an acceleration term into our motion matrix. The full code involved in both steps is available from community repositories ([78], see https://doi.org/10.5281/zenodo.1310984). The final Kalman solution, with standard deviation errors shaded, is shown in figure 5.

The Kalman-filtered APW velocities (APWV, magenta lines) show a tendency to follow previous velocities, changing when new data become available. Uncertainties are tightly bound when the solution encounters an observation, and they increase with time during periods of limited sampling. Both the Pilbara and the Kaapvaal are limited by available data—and uncertainties in the Pilbara path in particular are very large during data gaps. Both cratons exhibit periods of rather rapid APW though, at approximately 3.5 and 2.7–2.8 Ga in the Pilbara, and at 3.2, 2.7–2.8 and 2.1 Ga in the Kaapvaal. 3.2 Ga is particularly interesting, as it has been suggested to represent the onset of plate tectonics in the Pilbara [82], and also represents a time of intense meteorite bombardment [83]. For illustrative purposes, we have combined both Pilbara and Kaapvaal datasets in figure 5c, to present a Kalman-derived APWV plot from 3.5 to 2.0 Ga. The implications of this plot are as follows: (i) There exist substantial periods during the Archaean of little (to no) plate motion. Of course, the uncertainties during these periods cannot preclude plate motion, and indeed APW velocities during the Phanerozoic can be of the order of 1 degree Myr⁻¹ [81]—within the errors here. (ii) There exist large excursions from these background APW velocities, over an order of magnitude above background rates. These events occur at times of substantial tectonism in the geological record [10,82]. Furthermore, many of these are well sampled, and their error estimates (from either the Kalman filter or the raw Monte Carlo estimates) imply APW motions well above uncertainty levels. Post the 2.7 Ga event, APW motions exhibit nonlinearity, but could permissibly concur with consistent plate motions, albeit with speed-ups (e.g. 2.1 Ga) and slow-downs (e.g. 2.2–2.35 Ga [84,85]).

APW excursions may be attributed to either plate motion relative to a geocentric axial dipole or else TPW (e.g. [86]). Here, we argue that the association of these excursions with known tectonics events, and the change in the dispersion of the poles during these excursions [10], suggest a plate motion origin for the APW velocities shown in figure 5. The implication is that the palaeomagnetic record does indicate periods of often extreme continental drift, indicating active subduction, yet it also preserves long stretches of coincident poles—consistent with periods of immobile lid behaviour (figure 5). Together, these suggest that if subduction got going on the early Earth (e.g. at approx. 3.2 Ga), it regularly failed.

(b). Differentiating subduction in the geochemical record

Many recent estimates have suggested an origin of plate tectonics at approximately 3.0 Ga, inferring either a monotonically increasing tectonic transition from pre-plates to a modern plate tectonic regime around that time or that regional tectonics transitioned to a global system then [31]. In this section, we cast a critical appraisal on the newly emerging geochemical evidence for much older episodes of subduction in the absence of global plate tectonics, beginning with a quick summary of the basic principles behind such studies.

Geochemically, modern arc lavas and continental crust in general are characterized by negative anomalies in high-field-strength elements (HFSE) and positive anomalies in large-ion lithophile elements relative to the rare-earth elements (REE) and Th. There is, as yet, no diagnostic isotope signature. Thus, Pearce [9] has championed the use of Th/Yb versus Nb/Yb systematics (thereby using elements that are immobile and are not subject to changes in oxidation state) in identifying processes that can be related to tectonic environment for the origin of magmatic rocks. As discussed by Pearce [9], the rigour with which this methodology can be applied to identify a subduction origin for very ancient rocks requires some caution. In addition, the sediments subducting beneath arcs bear negative HFSE anomalies and so it is less clear to what extent this ‘arc signature’ is newly generated beneath today's arcs versus recycled [87]. In principle, if the continents were formed early in Earth history [88], and not necessarily by subduction, the negative HFSE anomalies may have been largely recycled ever since. In practice, some arc lavas show no evidence for sediment contributions (e.g. [89]) and for these, at least, the HFSE must have been retained by a residual Ti-rich phase during melting. One thing that is not currently in dispute is the evidence from ¹⁴²Nd anomalies that parts of the mantle underwent significant differentiation within the first 250 Ma of Earth's history [90].

In the light of the preceding observations, it is, of course, prudent to seek complementary and/or alternative geochemical approaches. Arndt [91] has noted that the continental crust is largely composed of granite and hence buoyant compared to surrounding oceanic crust. There is a vast experimental literature to show that granites are most commonly formed by hydrous melting—a process favoured in subduction settings. Of course, this can be a two-stage process in which the subduction process introduces hydrated basalts into the crust that can be melted at a later stage to produce granitic magmas. While hydrous basalts are not absent in intra-plate settings, it is likely that they themselves recycled subduction-modified lithospheric mantle [92]. In other words, there may be a simple petrological argument that the continental crust was dominantly formed through subduction at some time and place.

Distinctive rocks called boninites have widely been invoked as the hallmark of subduction initiation and many have looked for boninites in the ancient rock record. However, silica, which is important in the definition of boninites, is susceptible to alteration and metamorphism. Pearce & Reagan [93] have recently developed refined major element criteria by which to identify ‘true’ (i.e. subduction-related) boninites and then applied this to a wide range of Archaean rocks.

Recently, Dhuime et al. [30] used time-integrated Rb/Sr ratios of a large dataset of continental rocks to suggest that juvenile silicic crust only began to form in significant volumes around 3 Ga and did not reach a typical thickness of 30 km until around 2 Ga. Thus, crustal elevations for most of the Archaean were inferred to lie below sea level. Interestingly, a plot of SiO₂ versus time for rocks from the same dataset does not show significant, if any, secular variation in silica (figure 6). This implies a trade-off between initial crustal composition and its later-stage re-melting/differentiation. One explanation might be the upward redistribution of heat-producing elements in modern differentiated arc-like juvenile crust, resulting in lower deep crustal temperatures. In a similar vein, Greber et al. [95] have used Ti isotopes to argue that continental crust has been emergent and silica-rich since at least 3.5 Ga, from which they infer an early initiation of subduction.

Figure 6.

Plot of SiO₂ (wt%) versus age using the database (approx. 70 000 analyses) of Keller & Schoene [94]. Grey dots are individual analyses with a running mean shown by the black line.

An alternative approach is to look for geochemical stratigraphy instead of an individual HFSE signature [33]. Stern et al. [96] have shown that subduction initiation in the Izu–Bonin–Mariana arc is characterized by a rock sequence involving (mid-ocean ridge basalt, MORB-like) fore-arc basalts (FABs) followed by boninites and subsequently calc-alkaline arc rocks. In principle, identification of the same sequence in ancient rocks may, therefore, provide a more robust argument for subduction (initiation).

(c). Subduction initiation sequences in the Precambrian

Straightforward application of the geochemical criteria outlined above reveals an increasing number of observations that suggest that some form of subduction, or subduction initiation, occurred perhaps multiple times and very early in Earth history.

Much of the oldest silicic continental crust is represented as TTGs and there has been much debate as to whether any, some or all of these might have formed during subduction or melting of stacked oceanic slabs of hydrated basalts [19]. Many have negative HFSE anomalies but, as noted above, these signatures can also be inherited. For example, several recent studies of TTGs have invoked partial melting of basalts within the crust in their petrogenesis. Adam et al. [97] undertook an experimental study of tonalite from the Nuvvuagittuq belt in Canada and Johnson et al. [98] studied TTGs from the East Pilbara. A key observation of both studies is that the identified protolith basalts had arc-like signatures (e.g. negative HFSE anomalies) and, due to the large extent of partial melting required [97], this is efficiently transferred to the resultant TTGs. Thus, if the question is ‘Were these TTGs formed by subduction?’, the answer is no; but if the question is ‘When did subduction occur?’ then in both cases, the answer must be significantly before the age of the TTGs and in fact when their protoliths were formed. The Coucal basalts from the East Pilbara are 3.5 Ga in age, suggesting some form of subduction in this region at that time.

The age of the Nuvvuagittuq mafic rocks may be more debatable, but because these are cross-cut by tonalites and gabbros of approximately 3.8–4.1 Ga age, 4.1 Ga is arguably a minimum age for these mafic rocks [99]. Moreover, Turner et al. [33] showed that, not only do the Nuvvuagittuq mafic rocks overlie modern arc rocks on a plot of Th/Yb versus Nb/Yb, but they also record the same geochemical stratigraphy as the lowermost sequences in modern subduction zones. The recent analysis of Pearce & Reagan [93] further reinforces the subduction-like nature of the mafic Nuvvuagittuq mafic rock sequences (figure 7), and so there is an increasingly strong case that these reflect some form of subduction at, or prior to, 4.1 Ga.

Figure 7.

Classification of amphibolites from the oldest proposed boninites, from Nuvvuagittuq (reproduced from [93]). Panels (a)–(e) demonstrate the importance of chemically screening the altered rocks using the methodology detailed in Pearce & Reagan [93]. Panels (f) and (g) use carefully filtered samples from the Lower, Middle and Upper Nuvvuagittuq Units to confirm basaltic affinities for the Lower Unit and boninite affinities for the Middle Unit, and indicate SHMB affinities for the Upper Unit. The trend from basalts to LSB to SHMB is consistent with the concept of Archaean subduction initiation [33]. BA, basaltic andesite; A, andesite; D, dacite; LSB, low-Si boninite; HSB, high-Si boninite; HMA, high-Mg andesite; SHMB, siliceous high-Mg basalts; LOTI, low-Ti basalts. (Online version in colour.)

Before 4.1 Ga, we unfortunately have a dearth of rocks and rock relationships to study and so the main record of crust from these times is contained in detrital zircons. Belousova et al. [100] have argued that the excursion to low ε_Hf isotope values in greater than 3 Ga detrital zircons requires that some silicic crust had developed much earlier in order to be recycled. The isotopic compositions of the oldest detrital zircons known, from the Jack Hills in Australia, have been used to argue for the presence of continental crust and oceans as far back as 4.4 Ga [44]. The tectonic affinity of these zircons has proved harder to decipher. In order to resolve this shortcoming, we have used the available trace-element concentrations of the Jack Hills zircons to calculate the trace-element concentrations of the magmas from which they crystallized. While the Jack Hills zircons crystallized from a felsic magma, as shown in figure 8, they inherit negative HFSE anomalies and plot within or above (i.e. at larger negative HFSE anomalies) the field for modern arc rocks and are strongly displaced from the fields for intra-plate or mid-ocean ridge lavas. To our knowledge, this is the first time this has been attempted and the implication is consistent with the observations from Nuvvuagittuq that some form of subduction may have occurred at least as far back as 4.2 Ga.

Figure 8.

Th/Yb versus Nb/Yb diagram (after [9]) for calculated melts in equilibrium with 3.3–4.2 Ga Jack Hills zircons (data from [101]). Black circles used trace-element partition coefficients from Burnham & Berry [102]; grey circles used partition coefficients from Nardi et al. [103]. Thus, the observation is not dependent on whose partition coefficients are used. (Online version in colour.)

Finally, it should be emphasized that the rapidly growing evidence for Hadean to Archaean subduction need not be synonymous with the onset of plate tectonics, merely subduction initiation. If modern plate tectonics were already in operation, then plate movement would likely have been slower [53], and so increased recycling since 3 Ga [30] may simply reflect this rather than the actual onset of plate tectonics. In terms of ways forward, if subduction initiation events were caused by meteorite impacts, as recently suggested by O'Neill et al. [50], a fruitful avenue for future research might involve searching for iridium anomalies in any time-equivalent marine sedimentary rocks. Conversely, the recycling of intra-plate rocks may be revealed by elevated HFSE (a characteristic of ocean island basalts) in associated sediments. It would also seem to be a very worthwhile future endeavour to conduct further immobile trace-element analyses on the Earth's oldest zircons.

(d). Greenstones: a failure to launch

Smithies et al. [19] applied the concepts of Pearce [9] to greenstone belts within the Yilgarn and Pilbara cratons. They explored the variation in Th/Yb versus Nb/Yb to identify ‘constant’ Th/Nb trends—defined by an array of measurements that run parallel to but above the MORB array (figure 8), which reflects melting of a subduction-modified source and fractional crystallization, and thus may be used as a proxy for subduction processes. This contrasts with what they defined as ‘variable’ Th/Nb trends (steeper than the MORB array in figure 8), which indicate a mixing trend between MORB and a crustal component (here TTGs). Smithies et al. [19] suggest that constant Th/Nb trends are rare in Archaean rocks of the Pilbara and Yilgarn terranes, occurring mostly in suites between 3.13 and 2.95 Ga in the Pilbara, and between 2.82 and 2.71 Ga in the Yilgarn. Such suites are fairly uncommon worldwide, with the Whundo group in the Pilbara the only known example from the Palaeo- to Mesoarchaean. These lithologies are associated with mafic to dacitic volcanics, and are (from 2.95 Ga) associated with sanukitoids [19]. As such, these observations are most consistent with an association of these particular rocks with mature subduction that has evolved past the initiation stage.

By contrast, the variable or ‘high’ Th/Nb population was argued to represent a mixing relation between typical low Th/Nb asthenospheric mantle and a crustal component (e.g. TTG) with high Th/Nb. Smithies et al. [19] argue that these rocks are not subduction-related, and also demonstrated that they dominate greenstone assemblages worldwide, including in the intervals 2.71–2.66 Ga in the Yilgarn, and between 3.49 and 3.23 Ga, and 2.77 and 2.68 Ga, in the Pilbara.

Smithies et al. [19] also note that, in many cases of suggested subduction influence, lithologies are made up of typical subduction-initiation sequences. An example is the Polelle group (and over/underlying groups) in Murchinson terrane of the Yilgarn, which has a boninite-like affinity, and was suggested to have formed from a Th-enriched mantle source. The association of boninites and Ti-depleted tholeiites in many places, such as the Abitibi greenstone belt of the Superior Province [104], has been used to argue for a subduction origin [104]. Models for these depleted tholeiites require a rehydrated mantle source [104]. Smithies et al. [19] showed that the low-Ti tholeiites of this group do have a constant Th/Nb array, which would tend to support this. However, Smithies et al. [19], following Arculus et al. [105], suggest that such highly depleted volcanics are more typical of the pre-arc subduction-initiation stage, and conclude that the subduction system recorded by the low-Ti tholeiites in the Abitibi failed.

The Palaeo- to Mesoarchaean greenstone record is dominated by lithologies with Th/Nb trends that do not appear subduction-like, and instead indicate mixing between asthenospheric mantle and a TTG-like crustal component. A few mature subduction-like systems are observed in the Pilbara (3.13–2.95 Ga), and in the Yilgarn (2.82–2.71 Ga), and apparent failed subduction zones between 2.8 and 2.74 Ga (Polelle group, Yilgarn), similar to that observed in the Abitibi at approximately 2.7 Ga [19,104].

(e). Tonalite–trondhjemite–granodiorite suites: natural and experimental observations

TTG magmas are the dominant felsic rock suite found in the Archaean and form up to 60% of the continental crust [106]. TTGs are characterized by Al₂O₃ contents greater than 15% at 70% SiO₂, Sr > 300 ppm, Y < 20 ppm, Yb < 1.8 ppm, Nb ≤ 10 ppm and low-Mg amphibole, and are often thought to be partial melts of a basaltic precursor [107,108]. Indeed, one of the most popular theories of early continental crust development is that an older basaltic protocrust was partially re-melted to produce felsic magmas plus a residue of garnet + pyroxene + rutile, with the felsic melts now represented by the TTG granitoids that dominate most surviving Archaean terrains.

One of the oldest possible examples of this basaltic protocrust is the Nuvvuagittuq greenstone belt of Ungava, Québec [109,110]. As noted previously, ¹⁴²Nd measurements on the greenstones suggest that they may have formed at approximately 4.28 Ga and could therefore represent the oldest mafic crust preserved on Earth [110]. The dominant lithology of the belt are heterogeneous gneisses that are composed of variable proportions of cummingtonite + plagioclase + biotite +quartz ± garnet ± cordierite (Ujaraaluk unit). The Nuvvuagittuq greenstone belt is surrounded by Eoarchaean TTGs that have geochemical and isotopic compositions consistent with their formation through the melting of a mafic source similar in composition and age to the Ujaraaluk unit [111]. The majority of the greenstones have mafic igneous compositions, but they can be divided into two distinct geochemical groups that are stratigraphically superimposed [110]. The greenstones near the base of the sequence are relatively high in Ti and basaltic in composition, with tholeiitic affinities characterized by flat REE profiles. These are succeeded up-section by low-Ti greenstones that range in composition from basalt to andesite and have calc-alkaline affinities characterized by REE-enriched profiles. The most primitive low-Ti greenstones are depleted in many incompatible trace elements and display U-shaped REE profiles that resemble those of modern boninites [33,109].

Melting experiments carried out on one of the most primitive and least altered of the ‘boninitic’ low-Ti greenstones investigated its relationship to the enclosing 3.6 Ga tonalites [97]. The results have duplicated those previously published for tertiary boninites, with orthopyroxene and olivine being the only near-liquidus phases of the hydrous natural composition at pressures from 0.5 to 1.5 GPa. At higher pressures (2.0–3.0 GPa), moderate degree (approx. 35%) partial melts are tonalitic and compositionally similar to the Nuvvuagittuq tonalites, which are typical of other Archaean TTGs. Although the Nuvvuagittuq tonalites have higher overall concentrations of incompatible trace and minor elements than the experimentally produced melts, the relative concentrations of their trace elements are similar. Partial melting of the more evolved, low-Ti greenstones that are found in the Nuvvuagittuq belt would reproduce the absolute as well as relative concentrations of the tonalites. An important conclusion from this is that, if the tonalite formed by melting of the Nuvvuagittuq greenstones (or compositionally equivalent rocks), its distinctive incompatible element concentrations would be largely inherited from its source and could be only partially attributed to fractionation during partial melting of greenstones. By implication, many of the chemical characteristics of early felsic continental crust would have to be ascribed to fractionation that pre-dated the partial melting that produced them from older mafic crust.

Experimental evidence is consistent with a fractionation that pre-dated the partial melting as described above. When experimental partial melt data are summarized and compared to natural Archaean TTGs (figure 9), it is clear that the TTG compositions can be made by partial melting, but of only certain kinds of basaltic compositions. TTGs may be produced from partial melting of hydrated low-MgO basaltic compositions (i.e. amphibolite ± garnet [117]). However, low-MgO protolith compositions require that the thick primary crust must have differentiated, possibly by partial melting and loss of melt. This would then require a multi-stage model for TTGs generation [118]. The occurrence of primitive MgO-rich lavas (komatiites), the huge volumes of TTG magmatic suites, and the volume of continental crust produced during the Archaean are, minimally, consistent with a hotter Earth, operating under an alternative dynamic regime [30,32,119,120].

Figure 9.

Major element data for natural and experimental TTG compositions: (a) CaO versus Na₂O and (b) MgO versus SiO₂. Open symbols are Archaean TTGs. Closed circles are experimental partial melts of basaltic starting material, from 11 different studies, from 17 different starting materials and over 126 individual experiments (data from our compiled database). The experimental melts form a much wider distribution than the natural TTGs, showing that there are only specific compositions that produce TTG-like melts. Natural data are from Moyen & Martin [112], Nutman et al. [113], Huang et al. [114], Laurent et al. [115] and O'Neil & Carlson [116]. Natural compositional data are filtered after Moyen & Martin [112].

4. Implications and conclusion

One of the challenges in interpreting the Precambrian geological record is the conflicting evidence of subduction at different times (figure 10). Geochemical evidence [9,93,104] presents a compelling case for subduction at periods in the Archaean, which, together with evidence of continental drift from palaeomagnetic data, large preserved shear zones and the lateral movement of tectonic units [59], create a strong argument for subduction processes. This conflicts, though, with the extensive volcanic record from greenstones and TTGs demonstrating periods which do not show subduction involvement [9,19,98], and further thermal and mixing constraints limiting the rate of subduction on the early Earth [3].

Figure 10.

Timeline of major observational constraints on subduction initiation and plate tectonic proxies discussed in the text. Included (upwards from bottom) are an example of the proposed onset of plate tectonics [30], evidence of meteoritic bombardment (see [50]), earliest subduction stratigraphy [33], Archaean Nb/Th from the Pilbara and Yilgarn ([19]; black indicates no subduction involvement, orange indicates subduction involvement, red indicates possible subduction initiation), palaeomagnetics (see text; red indicates episodic APW velocities, blue indicates possible plate tectonic speeds), the Magmatic gap identified by Condie et al. [84], the ‘Boring billion’ [121], and evidence for UHP metamorphism since the Neoproterozoic [14]. (Online version in colour.)

If one assumes that subduction progresses in a linear fashion, and that around approximately 3 Ga the system ‘switches’ from an archaic pre-subduction regime into a modern plate tectonics regime, this presents a paradox. Evidence for and against subduction at different times in the geological record is not consistent with a monotonic transition into plate tectonics, and 3 Ga might instead represent the time when plate tectonics and subduction started to exert a dominant effect on the crustal record [31]. Similarly, one might argue that the evidence for subduction is preserved only locally, and the apparent lack of subduction processes is simply a consequence of regional preservation—terranes without subduction signatures were simply distally removed from active subduction zones. However, the time duration of preserved subduction involvement is generally short in the Archaean (e.g. [9,17]), and would seem to argue against long-lived mature modern-type subduction zones.

The value of geodynamic modelling to this debate has been not to constrain one particular end-member of Earth evolution, but to provide insights into the dynamics of such systems under early Earth conditions, and by analogy into the possible behaviour of the Earth. A number of important insights can be summarized from these models, the most important being the transition into plate tectonics was likely episodic, with numerous periods of subduction initiation and failure [17]. Additionally, early plate tectonics was likely extremely time-dependent [10,84], with frequent speed-ups and slow-downs. The other important observation is that, in transitional systems where the conditions for subduction are generally not met, singular events capable of invoking severe mantle stresses—dubbed ‘primary drivers’ here—may drive regional or sporadic subduction. These may include large plumes [22] or impact events [50]. It has been argued that both of these would be more prevalent on the early Earth [50], and even into the Mesoarchaean [83], and so our earliest subduction record may represent these singular events, rather than a self-sustaining heat loss mechanism.

Plate tectonics does not occur with a constant velocity today—the average plate speed on the surface of the Earth has fluctuated through time [122]. At their extreme, these fluctuations would lead to intermittent plate tectonics [20] as previously argued. Extending this into the Proterozoic, however, suggests that large speed-ups or slow-downs are possible. Condie et al. [84] suggested, on the basis of the zircon record, volcanic history, erosional contacts and sediment recycling, that the period between 2.4 and 2.2 Ga might represent a plate tectonic slow-down. This period is associated with the widespread glaciation of the Earth and the rise of oxygen, and a diminished volcanic influx associated with this slow-down might have facilitated these transitions. It was later shown that the gap in zircon U–Pb ages is not complete [123], and that zircon populations exist within this range from a number of regions. They are, however, volumetrically small, and the association with widespread unconformities and diminished passive margin abundance [124,125] suggests a physical system change. Recently, Spencer et al. [85] presented evidence of a coincident gap in orogeny, mafic volcanism and passive margin sedimentation over this time interval. They also present U–Pb/Hf zircon data suggesting a lull in the distribution, observed in kernel-density estimates of the data, at approximately 2.31–2.21 Ga, and ¹⁴³Nd/¹⁴⁴Nd evidence that mafic volcanism forms two discrete populations roughly separated by a mantle-extraction lull associated with the LIP (large igneous province) gap. They suggest that this supports the reality of the lull, and that it was driven by a global geodynamic slow-down.

The transition from a pre-plate tectonics regime may arguably have extended until the Neoproterozoic [14,16]. Cawood & Hawkesworth [121] noted that the period dubbed ‘Earth's middle age’ or the ‘Boring billion’, approximately 1.7–0.7 Ga, is characterized by the following: (i) A lack of glacial deposits, iron deposits and phosphate deposits, indicating a period of unusual environmental stability. (ii) A lack of passive margin development [125]—including none during the putative formation of Rodinia. (iii) A dip in the Sr isotope ratios of seawater [126], indicating a decrease in continental input, consistent with the absence of orogenic gold, which is a product of continental orogenesis. ε_Hf in detrital zircons likewise indicates limited crustal reworking. (iv) Widespread massif-type anorthositic volcanism, related anorogenic granites (Rapakivi granites) and associated volcanism, which are largely constrained to this period, and indicate widespread deep (40–50 km) and hot (950–1000°C) melting of the lower crust [127]. Taken together, these proxies argue for a period of extended tectonic stability. Cawood & Hawkesworth [121] argue that the core components of Nuna were minimally reconfigured during the assembly of Rodinia. The culmination of this period is marked by the breakup of Rodinia at 0.75 Ga, and the appearance for the first time of UHP subduction assemblages in the metamorphic record [14]. While it is clear that plate tectonic processes proceeded during this period, it seems that the tempo of some aspects of this cycle (development of passive margins and continental breakup, continental input into the oceans from Sr isotopes) may have changed. As a result, some authors have suggested that the origin of modern deep, cold subduction sensu stricto has its origin in the Neoproterozoic, and that the transition from a pre-plate tectonics Earth, into quasi-continual modern plate tectonics, arguably has taken at least that long.

Here, we have outlined the evidence for subduction on the early Earth, and noted that many of the arguments for subduction concomitantly suggest that subduction was short-lived. Palaeomagnetic constraints, geochemical signatures in greenstone belts and identification of subduction initiation sequences in the Precambrian record all suggest that subduction, while probably occurring, was generally short-lived. TTGs, previously suggested to be subduction melts, have been argued in recent analysis to develop in a stagnant-lid setting. Both observations are in line with geodynamic models which suggest that long-lived subduction is difficult, and that mantle mixing constraints noting subduction episodes, rather than ongoing subduction, can best explain the evolution of mantle reservoirs. Such geodynamic models suggest that the transition from a pre-plate tectonic Earth to modern plate tectonics was nonlinear and protracted, and characterized by sporadic transient subduction events during the Archaean, eventually overlapping to form time-dependent plate tectonics (e.g. throughout the Proterozoic ‘slow-down’ and the Boring billion), finally transitioning to a more continual style of modern cold subduction some time in the Neoproterozoic. Ultimately, the question of the origin of plate tectonics, and the subtly different question of the start of subduction, are amenable to advancement through falsification—the testing of different geodynamic scenarios against observational constraints.

Data accessibility

This article has no additional data. Accompanying data and software are available from the online respository Zenodo [78]: http://doi.org/10.5281/zenodo.1310984.

Competing interests

We declare we have no competing interests.

Funding

We received no specific funding for this study. C.O'N. is grateful for ARC funding and NCI computing support.