Boninites are rare, high-Si, high-Mg, low-Ti lavas that have considerable tectonic significance, especially for recognizing and interpreting episodes of subduction initiation in the geologic record. Formal identification and classification of boninites may be carried out using MgO-SiO2 and MgO-TiO2 diagrams to find compositions that satisfy modified International Union of Geological Sciences (IUGS) criteria of Si8 > 52 and Ti8 < 0.5, where Si8 and Ti8 refer to concentrations of the oxides at 8 wt% MgO. However, screening of highly metasomatized rocks and accurate classification require precautions, including normalization to a 100% volatile-free basis. The MgO-SiO2 diagram can also be used for subdivision into low-Si boninites (Si8 < 57) and high-Si boninites (Si8 > 57). Satisfying one but not both of the boninite criteria are rocks with Si8 > 52 but Ti8 ≥ 0.5 (siliceous high-magnesium basalts) and rocks with Si8 ≤ 52 but Ti8 < 0.5 (low-Ti basalts). We tested the classification methodologies using ∼100 low-Ti lava suites dating from the present-day back to the Eoarchean. We conclude that, of those classifying as “boninite series,” Izu-Bonin-Mariana arc–type subduction initiation terranes provide the dominant setting only back as far as ca. 2 Ga, which marks the maximum age of extensive clinopyroxene-undersaturated melting and eruption of high-Si boninites. From 2 to 3 Ga, most boninites formed in intraplate settings by melting of refertilized, depleted cratonic roots. Prior to 3 Ga, hot, depleted mantle plumes provided the main boninite sources. Nonetheless, arc-basin boninites, though rare, do extend back to 3.8 Ga, and, together with the inherited subduction component in intracratonic boninites, they provide evidence for some form of subduction during the Archean.

Boninite is one of a small number of rock types (adakite is another obvious example) that are commonly linked to a particular present-day tectonic environment but (1) have more than one potential mode of origin and can therefore be generated in more than one tectonic setting, and (2) have likely, but unclear, significance for early Earth. Boninite is most commonly linked to embryonic arc volcanism following intraoceanic subduction initiation, i.e., the inferred setting of its type area, Chichijima, once known as Bonin Island, in the western Pacific (e.g., Kuroda et al., 1978; Taylor et al., 1994; Dobson et al., 2006). For this reason, the presence of boninites in Archean terranes has been cited as key evidence that subduction, and hence plate tectonics, operated during the Archean (e.g., Kerrich et al., 1998; Polat et al., 2002; Turner et al., 2014). There are, however, a number of caveats to this interpretation. First, these Archean boninites have typically undergone extensive alteration, leading to mobility of elements essential in boninite classification. Second, there is no single, robust classification scheme for boninites, therefore allowing “classification creep,” in which rocks that are not strictly boninites are described as “boninite-like” and treated as having settings similar to those of true boninites. Finally, even when correctly classified, boninites have now been identified in a range of settings, including oceanic plateaux (e.g., Ingle et al., 2007) and intracontinental rifts (e.g., Smithies, 2002; Srivastava, 2008), thus breaking the commonly assumed link between boninites and subduction initiation. The principal objectives of this paper were thus to detail the precautions needed in classifying boninites and to test the hypothesis that boninite is a useful indicator of past tectonic environments extending back to earliest Earth. In doing this, we will critically assess whether boninites do indeed provide the evidence claimed for pan-Archean plate-tectonic processes.

To achieve these objectives, the obvious starting point is the geochemical classification of boninite and an assessment of its sensitivity to the chemical alteration that has affected a high proportion of the basic and ultrabasic lavas in the geologic record. Historically, the name “boninite” can be traced back to Petersen’s (1891) work on the volcanic rocks of Bonin Island (now Chichijima). The petrogenetic and tectonic significance of boninites was recognized in the 1970s (e.g., Kuroda et al., 1978; Cameron et al., 1979; Meijer, 1980), culminating in the key publication of the book by Crawford (1989) on “Boninites and Related Rocks.” Crawford et al. (1989, p. 2), in their opening chapter to the book, define boninitic volcanic suites as “those in which the volumetrically dominant lavas either have >53wt% SiO2 and Mg# >0.6 [and TiO2 < 0.5%] or are demonstrably derived from parental magmas meeting these compositional requirements.”

The International Union of Geological Sciences (IUGS) subsequently included boninites in their reclassification of high-Mg and picritic volcanic rocks (Le Bas, 2000). They defined boninites as having SiO2 > 52 wt%, MgO > 8 wt%, and TiO2 < 0.5 wt%. Pearce and Robinson (2010) noted that a rectangular classification “box” does not satisfy the implicit Crawford et al. (1989) requirement that the definition should not change with differentiation. Thus, while accepting that the SiO2 and MgO criteria set out by the IUGS should be used in order to link the boninite definition with the existing total alkali-silica (TAS) classification scheme, Pearce and Robinson (2010) recommended that the IUGS classification should be modified, with boninites classified as volcanic rocks that have Si8 > 52 wt% and Ti8 < 0.5 wt%, where Si8 and Ti8 refer to values of SiO2 and TiO2 on fractionation trends at MgO = 8 wt%. In this paper, we developed this concept further by improving field boundaries using petrogenetic principles and emphasizing the importance of using data sets that are based on volatile-free totals that sum to 100 wt% with iron partitioned into ferrous and ferric iron oxides.

The problem of identifying altered lavas as boninites has largely been ignored to date, but it is nontrivial. Although Ti usually behaves as an immobile element, both Mg and Si are significantly susceptible to weathering and metamorphism. An effective way to classify altered rocks has been to define immobile element proxies for the mobile elements, for example, Zr/Ti or Co as a proxy for silica and Nb/Y as a proxy for total alkalis, to produce an alteration-insensitive alternative to the TAS classification diagram (e.g., Winchester and Floyd, 1977; Hastie et al., 2007). There are also a number of immobile element pairs that can act as a proxy for the MgO-TiO2 diagram in boninite classification, such as Cr-Ti (e.g., Todd et al., 2012) and Ti-V (Shervais, 1982). However, a major contributor to the Si variation in boninites is the incongruent melting of orthopyroxene (opx), for which no known proxies exist. We therefore propose, in this paper, to use data screening as well as immobile element proxies to counter the effects of SiO2 mobility.

With proper data reduction and classification procedures in place, we were then able to test the hypothesis that boninites have a useful paleotectonic significance by comparing examples from different, known tectonic settings and reexamining proposed boninites from the geologic record. We placed particular emphasis on the Archean, where boninites have been cited as key evidence for plate-tectonic processes in early Earth. Thus, we took the opportunity to test the subdivision into the major types of Archean boninite identified by Smithies (2002) and Smithies et al. (2004a): Mallina (intracontinental rift), Whundo (subduction-related), and Whitney (plume-related). We also tested the classification and significance of other proposed Archean boninites, including the oldest “boninite-like” lavas from Isua, Greenland (Polat et al., 2002), and Nuvvuagittuq, Canada (O’Neil et al., 2011; Turner et al., 2014).

Data Manipulation before Classification

One aspect of the IUGS classification schemes for volcanic rocks is that they apply to cation oxides only (i.e., not H2O or CO2). They also require that these oxides are recalculated to a major-element sum of 100% (Le Maitre, 2002). This procedure is needed to reduce variations resulting from degree of alteration and degassing. Thus, the diagrams presented here are also based on cation oxide totals that sum to 100%. Though this rule is not followed in many publications, it is particularly important to do so in classifying boninites because of the sensitivity of silica to the method of data reduction (e.g., a 2 wt% absolute difference in total propagates to an ∼1 wt% absolute difference in silica). In the case of full element analyses with iron analyzed separately as FeO and Fe2O3, and H2O+, H2O, and CO2 also determined quantitatively, the calculation simply involves the multiplication of concentrations by 100/x, where x is the sum of metal oxides.

It is more common, however, to fuse the sample under oxidizing conditions and report the mass loss as loss on ignition (LOI; Lechler and Desilets, 1987). The true volatile loss is then LOI plus the oxygen added during oxidation of FeO to Fe2O3 (where iron as Fe2O3 = iron as FeO*1.11). Thus, if T is the sum of oxides and LOI, L is the LOI, F is the measured Fe2O3, and r is the estimate of the original proportion of iron present as Fe2O3 (typically ∼0.2 in boninites; Brounce et al., 2015), then SiO2, MgO, and TiO2 must be recalculated as:

Analyses using other methods of quoting volatiles, totals, and iron oxides need to be recalculated using the same principles and the appropriate equation. All data used in this publication are volatile-free values with separate FeO and Fe2O3 (either measured directly or calculated with a 4:1 ratio) and a total of 100 wt%.

SiO2-MgO Covariations

The IUGS classification of boninites in Le Bas (2000) is simply the standard TAS diagram, in which the basaltic andesite, andesite, and dacite fields are annotated with the note that compositions are, instead, boninite if MgO > 8 wt% and TiO2 < 0.5 wt% (Fig. 1A). We believe that the definition needs to be more sophisticated than this. Our starting point is the SiO2-MgO diagram of Pearce and Robinson (2010). In this diagram, the lower boundary for boninites is MgO = 8 wt%, as proposed by both Crawford et al. (1989) and Le Bas (2000).

The left boundary begins at SiO2 = 52 wt% (Si8 = 52), to comply with the IUGS definition, but then it follows a negative slope on the premise that it should follow a fractionation trend rather than be a vertical line. Pearce and Robinson (2010) drew this boundary empirically as the high-Si boundary of present-day mid-ocean-ridge basalt (MORB), ocean-island basalt (OIB), and oceanic arc fields. In the revised version (Fig. 1B), the boundary is drawn as a tie line to an olivine composition, because MORB, arc basalts, and boninites with MgO > 8 wt% are typically dominated by olivine crystallization. The olivine composition chosen is MgO = 50 wt%, SiO2 = 41 wt%, corresponding closely to olivines in equilibrium with inferred primary magmas from the Troodos Massif (e.g., Cameron, 1985) and Chichijima (e.g., Dobson et al., 2006).

The right-hand (high-SiO2) boundary begins at SiO2 = 63 wt%, which matches the IUGS andesite-dacite boundary and is close to the high-Si limit of boninite compositions. We drew this boundary from MgO = 8 wt% as a tie line to orthopyroxene at MgO = 32 wt% and SiO2 = 57 wt% (the average composition of orthopyroxene crystals in equilibrium with primitive magma from Chichijima; Dobson et al., 2006) to reflect the fact that boninites with Si8 = 63 typically are dominated by orthopyroxene crystallization. The upper boundary depends on the primary magma composition, itself particularly a function of temperature, as well as the extent of crystal accumulation. We arbitrarily placed this boundary at MgO = 32 wt% to match the orthopyroxene MgO value, and because only a few boninite lavas with an extremely high proportion of accumulated crystals exceed this value.

There have been a number of efforts to separate the boninite field into subfields. Crawford et al. (1989) divided boninites into “high-CaO” and “low-CaO” types. This distinction is important for petrogenesis, but it is difficult to apply to past volcanic rocks because of the high mobility of Ca. We thus prefer the subdivision proposed by Kanayama et al. (2013) into low-Si boninites and high-Si boninites. However, we drew our boundary between low-Si boninites and high-Si boninites to pass through Si8 = 57 (rather than Si8 = 59 in Kanayama et al., 2013). At this Si8 value, it corresponds with the boundary between basaltic andesite and andesite compositions on the IUGS TAS diagram.

As in Pearce and Robinson (2010), we defined a high-Mg andesite field to occupy the region between the lower bound of the boninite field (8 wt% MgO) and the upper bound of the fields of basaltic andesites, andesites, and dacites (BADR series) from normal modern arc and intraplate settings. To improve ease of application of the diagram, there is no longer any overlap between the high-Mg andesite and basalt-andesite-dacite-rhyolite fields, and the revised field boundaries are linear and designed to match the TAS classification system. This link to the TAS system allows the high-Mg andesite field to be subdivided into high-Mg basaltic andesites, high-Mg andesites, and high-Mg dacites if required, although we treat it as a single field in this publication. Finally, we took the boundaries and names for the low-Si (<52 wt% SiO2) rocks to the left of the boninite field (komatiite, picrite, picrobasalt) directly from the IUGS classification scheme for high-Mg volcanic rocks (Le Bas, 2000).

MgO-TiO2 Covariations

We drew the MgO-TiO2 diagram (Fig. 1C) using criteria similar to those used to construct Figure 1B. The key point is (0.5,8.0), which marks the TiO2 maximum and MgO minimum for a sample to be classified as a boninite. Boninite-series magmas will typically be crystallizing olivine ± pyroxene at this point, so we back-extrapolated the boundary to the olivine composition at (41, 50), because this maximizes the size of the boninite field. Points lying to the right of this boundary by virtue of higher TiO2 will then classify as komatiites, picrites, or basalts, depending on their MgO concentrations. Those to the left will classify as boninite, provided Si8 is also >52. For compositions with ≤ 8 wt% MgO, this diagram does not distinguish between members of the basalt-andesite-dacite-rhyolite and high-Mg andesite series.

Ti8-Si8 Covariations

The IUGS criteria require that a rock must plot in the boninite fields of both Figure 1B and Figure 1C to be termed a “boninite.” We labeled rocks plotting only in the boninite field of Figure 1B as “siliceous high-magnesium basalts” after the terminology of Sun et al. (1989), and those plotting only in the boninite field of Figure 1C as “low-Ti basalts,” after the terminology of Polat and Kerrich (2006), among others. The Ti8-Si8 projection of Pearce and Robinson (2010) allows this distinction to be made with just one diagram (Fig. 1D). On this projection, a rock with Si8 > 52 and Ti8 < 0.5 classifies as “boninite,” and one with Si8 ≤ 52 and Ti ≥ 0.5 classifies as “basalt” (or picrite or komatiite depending on MgO). Of the rocks that satisfy only one criterion, those with Si8 > 52 and Ti8 ≥ 0.5 classify as “siliceous high-magnesium basalts,” and those with Si8 ≤ 52 and Ti8 < 0.5 classify as “low-Ti basalts.”

On the MgO-SiO2 diagram, the Si8 value may be obtained by downward extrapolation to the line MgO = 8 using the slope of the bounding lines:
On the MgO-TiO2 diagram, the Ti8 value may be obtained by projecting from olivine to the line MgO = 8 using the equation:

We can then represent a single lava suite on the Ti8-Si8 diagram (Fig. 1) either by plotting all the individual data points or by plotting its mean Si8 and Ti8 (with 1σ error bars, if space allows). A rock with MgO < 8 wt% would classify as basalt, basaltic andesite, andesite, dacite, or high-Mg andesite according its position on Figure 1B. The two parts of the classification (MgO > 8 and MgO ≤ 8) can be combined to modify the name of the rock series: For example, a cogenetic set of lavas that extends from the boninite to the high-Mg andesite field would constitute a “boninite–high-Mg andesite series.”

The principles of classification laid out for the IUGS rock classification system (Le Bas and Streckeisen, 1991) emphasize that a proper classification scheme gives the rock a name objectively and reproducibly. Thus the scheme must be independent of any other geochemical or geological information and, in particular, independent of petrogenetic interpretation. Although the boninite boundaries in Figure 1 have a petrogenetic rationale in that they fulfill the Crawford et al. (1989) requirement of fractionation independence, they are invariant lines and so independent of interpretation; i.e., a data point plotting in the boninite fields on the diagrams in Figures 1B–1D classifies as a boninite regardless of petrogenesis and tectonic setting. However, as will be seen, the projections used here (MgO-SiO2, MgO-TiO2, and Ti8-Si8) do also carry useful information on the genesis and tectonic settings of the rocks that classify as boninites. Therefore, we investigated the relationship between classification and petrogenesis.

Boninite Genesis by Melting with Residual Clinopyroxene

Exhaustion of clinopyroxene (cpx-out) is an important barrier during mantle melting. Experiments have shown that melting before cpx-out (typically at ∼20%–25% melting) involves only a small increase in temperature above the solidus (<2.5 °C per degree of melting; Kushiro, 1996). Most present-day primary magmas, including virtually all MORB and OIB, are clinopyroxene-saturated, having formed by melting of mantle before the cpx-out barrier (e.g., Kushiro, 2001). Melting beyond cpx-out requires either a significantly greater increase in temperature per unit melt fraction (Kushiro, 2001) or solidus depression by fluid addition (e.g., Parman and Grove, 2004). Here, we tested the commonly held view that boninite genesis requires mantle melting beyond cpx-out, typically involving a two-stage process (e.g., Duncan and Green, 1987).

We began by plotting P-F grids (where P = pressure in GPa and F = degree of melting as mass percent) for the MgO-SiO2 and MgO-TiO2 classification projections with F up to 25% at P ≤ 3 GPa, the likely maximum F for cpx-out in spinel lherzolites. We drew the MgO-SiO2 grid in Figure 2A for P < 3 GPa using the anhydrous batch melting trends in figure 1 of Kushiro (1998) and for P = 3–6 GPa using the data of Walter (1998). However, because the grid was constructed from experiments on a variety of mantle sources, each with a different TiO2 concentration, we drew the MgO-TiO2 grid in Figure 2B using MgO values directly from the experiments, but scaling TiO2 concentrations to a depleted MORB mantle (DMM) value of 0.175 wt%.

The melting grids in Figures 2A–2C demonstrate that anhydrous, isobaric batch melts extend into the boninite fields only at the lowest pressures in MgO-SiO2 space and not at all in MgO-TiO2 space. The grids also apply to pooled polybaric fractional melting, provided the grid pressure is taken as the average of the melting column, and the grid degree of melting is treated as slightly lower than the actual value (Hirose and Kawamoto, 1995). Overall, Figure 2C shows that, although low-pressure melting of mantle of DMM composition can generate high-Si magmas with >8 wt% MgO, the high TiO2 value means that these classify as siliceous high-Mg basalts, not boninites. The same applies to hydrous melting. At the concentrations measured in boninite primary magmas (<0.5 wt%), water has a negligible effect on the geometry of the grid in the MgO-TiO2 diagram and only shifts the grid to slightly higher SiO2 values in the MgO-SiO2 diagram (e.g., Parman and Grove, 2004). Under water-saturated conditions, however, Hirose (1997) demonstrated that primary high-Mg andesites could be generated even at low temperatures and pressures (1000–1050 °C and 1 GPa), as shown in Figure 2A.

One outcome of the batch melting experiments is that, to generate boninites by partial melting with residual clinopyroxene, the prime requirement is to reduce TiO2 in the mantle. This is particularly evident in the Ti8-Si8 projection (Fig. 2C), where the grid lies well above the boninite field. Low Ti8 requires fractional or stepwise incremental melting in which one or more early melt fractions are lost from the melting zone. The most relevant experiments are the incremental melting experiments of Hirose and Kushiro (1998). They simulated a melting column at a potential temperature of ∼1340 °C, which releases ∼5 wt% melt increments at 0.5 GPa intervals from 2 GPa to 0.5 GPa (the blue trends in Fig. 2). They obtained a low-Si boninitic fraction (SiO2 = 52.63 wt%; MgO = 9.89 wt%; DMM-scaled TiO2 = 0.38 wt%) from the final, shallowest, increment. The total degree of melting (at ∼21%) required to produce this increment is just insufficient to extend beyond cpx-out, so demonstrating that low-Si boninites can form by tapping the low-pressure aliquots from a fractional melting column even when clinopyroxene is a residual phase.

Boninite Genesis by Melting beyond Clinopyroxene Exhaustion

Falloon and Danyushevsky’s (2000) assertion that a component of melting beyond clinopyroxene exhaustion is critically important for boninite genesis was tested and supported by the experiments of Wood and Turner (2009). The latter described melting beyond cpx-out as progressive “clinopyroxene (cpx) undersaturation.” In Figure 3, we plotted cpx-undersaturation trends from the above publications and others (Parman and Grove, 2004; Hirose and Kawamoto, 1995) onto the (A) MgO-SiO2, (B) MgO-TiO2, and (C) Ti8-Si8 classification diagrams to understand the effects of these trends on boninite classification and genesis.

In Figure 3A, cpx-undersaturation trends extend from melt compositions at cpx-out on the batch-melting grid to higher SiO2 and MgO. The trends enter the boninite (or siliceous high-Mg basalts) field following cpx-out, immediately for the lowest pressures and after a small and variable interval for higher pressures, and then transect the boninite field. Typically, they reach opx-out (at ∼40% melting) while still in the low-Si boninite field, though some can reach the high-Si boninite field for reasons explained below. These trends are subparallel, close to the mean 2:1 (MgO:SiO2) gradient found by Wood and Turner (2009) to apply to progressive cpx-undersaturated melting.

Orthopyroxene makes the predominant contribution to the melt following cpx-out. Thus, all the melting trends plot toward the average composition of the orthopyroxene entering the melt, displaced toward olivine (+ol) if the reaction coefficient for olivine (pol) is positive (i.e., melting is along an olivine-orthopyroxene cotectic) and away from olivine (-ol) if the reaction coefficient for olivine is negative (i.e., melting is along an olivine-orthopyroxene peritectic). In Figure 3A, the experimental cpx-undersaturated trends run directly toward orthopyroxene at ∼2 GPa, indicating an overall olivine reaction coefficient (pol) close to zero for that pressure. We also found that shallower melts form trends pointing to the right of the orthopyroxene composition (pol < 0) and can predict that deeper melts should form trends pointing to the left (pol > 0).

The cpx-undersaturation trends can be viewed as mixing lines (Parman and Grove, 2004), in this case, between the low-SiO2 and low-MgO pooled melt present at cpx-out (the red band in Fig. 3A) and the new high-SiO2 and high-MgO melt generated by further melting of the residual olivine, orthopyroxene, and spinel assemblage (the yellow band). This concept provides a clear way to interpret the cpx-undersaturation trends and hence understand the variations in primary boninite compositions on the SiO2-MgO classification diagram.

For example, for batch melting, or pooled fractional melting, of fertile mantle, 20%–25% melt already exists at cpx-out, and so the melt composition does not reach the high-Si boninite field even after adding the ∼20% of melt produced in reaching opx-out. This is the case for anhydrous melting of a DMM composition (Parman and Grove, 2004, trend M), hydrous melting of a slightly depleted Kilbourne Hole lherzolite composition (Hirose and Kawamoto, 1995, trend K), anhydrous melting of the somewhat more depleted Tinaquillo lherzolite composition (Falloon and Danyushevsky, 2000, trend T), and the Wood and Turner (2009) anhydrous, low-pressure, cpx-undersaturated melting experiments (trend W). To reach the high-Si boninite field, mass balance requires the melt contribution prior to cpx-out to be very small relative to the second stage of melt generation. This is exemplified in Figure 3 by anhydrous melting of the highly depleted Troodos harzburgite (Falloon and Danyushevsky, 2000, trend H), for which only ∼5% melting is needed to generate magmas in the high-Si boninite field.

A series of studies, notably those generating the primary data for Figure 3 and others such as those by Gaetani and Grove (1998) and Lee et al. (2009), have shown that the gradients of the trends are minimally influenced by water, and this is evident in the similarities between the hydrous (e.g., K) and anhydrous (e.g., M and T) trends. Thus, the principal role of water (at the level present in boninite sources) is not to influence the silica activity of the melt and hence the gradient of the trends. Rather, it is in lowering the mantle solidus and so promoting a greater degree of melting beyond cpx-out and hence a greater probability of producing boninite magmas.

Some prior mantle depletion is also essential for mantle-derived melts to plot in the boninite (or low-Ti basalt) field in diagrams involving TiO2 (Figs. 3B–3C). Batch melting of the DMM source (trend M) always produces melts with TiO2 concentrations that are too high for boninites, even at opx-out. At lower pressure (1 GPa), melts from the slightly depleted Kilbourne Hole (trend K) and Tinaquillo lherzolites (trend T) do just reach the boninite field, but only after extensive melting beyond cpx-out. For even more depleted mantle sources, such as the Troodos harzburgite (trend H), all melts lie in the boninite field on the MgO-TiO2 diagram. In consequence, cpx-undersaturated melt occupies a wide field in Figures 3A–3C, reflecting variability in pressure, prior mantle depletion, and the relative contributions of cpx-saturated and cpx-undersaturated melts.

Boninite Genesis by Melting of a Hybrid Mantle Wedge

It has further been established by a number of studies (e.g., Pearce et al., 1992; Falloon et al., 2008, Li et al., 2013; Kanayama et al., 2013) that the genesis of many boninites likely involved interaction of a siliceous slab melt (or silica-rich supercritical fluid) with depleted mantle. There are three main options: (1) Siliceous slab melt causes flux melting of the mantle to produce boninite; (2) siliceous melt first undergoes subsolidus reaction with the mantle to produce an orthopyroxene-rich pyroxenite, which then melts to form boninite; and (3) mantle upwelling and slab melting produce separate boninite and silicic magma reservoirs, which mix in a mantle wedge storage and homogenization zone.

  1. Flux melting: An outcome of the adakite-peridotite reaction experiments at 3.8 GPa (Rapp et al., 1999) is that flux melts are generated at high adakite:peridotite ratios. The resulting melts plot predominantly within the high-Mg andesite field in Figure 4A, with higher MgO and lower SiO2 than the pure adakites used in the experiments. In the MgO-TiO2 diagram (Fig. 4B), they plot outside the bounds of the plot, toward higher TiO2 concentrations, reflecting the high concentrations of TiO2 in high-temperature eclogite-facies MORB melts. Given that the Si and Mg concentrations of amphibolite melts are similar to those of eclogite melts while Ti concentrations at comparable temperatures are lower (Rapp et al., 1991), it is likely that experiments run at the (lower) pressures of boninite genesis would produce low-Ti, high-Mg andesite magmas.

  2. Pyroxenite melting: The experiments of Rapp et al. (1999) also demonstrate that interactions between slab melts and mantle at low adakite:peridotite ratios produce orthopyroxene-rich pyroxenites by subsolidus reactions. Such pyroxenites have been recovered as xenoliths in a few arc volcanoes (e.g., Bénard and Ionov, 2012; Bénard et al., 2017). Although there have been many pyroxenite melting experiments (e.g., Lambart et al., 2016), only a few of these simulated enhanced silica and hence high opx:cpx. Presently, the most relevant are the experimental melt compositions of Sobolev et al. (2007) for pyroxenite melting (3.5 GPa, 1400–1570 °C). These lie in the high-Mg andesite field of Figures 4A–4B at low to moderate degrees of melting. They then converge on the boninite fields as degree of melting increases, eventually reaching the edge of the low-Si boninite field at very high degrees of melting (Fig. 4C). We can hypothesize that the lower pressures and the more residual mantle sources characteristic of boninite genesis would displace this trend to lower Ti8 and higher Si8.

  3. Mixing of boninite and slab melts: Yogodzinski et al. (2015) developed a model whereby the upper mantle wedge is the site of a magma storage and homogenization (MASH) zone between a deeper, cooler, more siliceous slab-derived melt and a shallower, hotter, peridotite-derived melt. As Figures 4A–4C show, this mixing process will shift the peridotite-derived melt toward the high-Si boninite field. The precise trend is difficult to define precisely, however, as it depends on the compositions of the two end members, both of which can be highly variable.

Boninite Genesis by Magma-Crust Interaction

Although most boninite localities are intraoceanic, and therefore only minimally affected by crystal assimilation, a further potential mechanism for boninite genesis in intracontinental settings is by interaction of hot picrite or komatiite magmas with continental crust, a process that may have been prevalent in the Archean (e.g., Barley, 1986). Figures 5A–5B investigate this possibility using three end-member parent magma compositions: a komatiite from Belingwe, Zimbabwe, denoted by B (Nisbet et al., 1987), a picrite from Gorgona Island, coastal Colombia, denoted by G (Révillon et al., 2000), and a primary low-Ti basalt picrite, denoted M, from the Suvorov Trough locality on the Manihiki Plateau, southwest Pacific Ocean (Golowin et al., 2017a, 2017b). The crustal end-member composition has arbitrary values of SiO2 = 70 wt%, MgO = 1 wt%, TiO2 = 0.3 wt%, and H2O = 0.2 wt%. The models reported here were run at 2 kbar for initially dry magmas that increased in water content with progressive assimilation.

Bulk mixing of komatiite or picrite magmas with crust or crustal melt would drive the composition directly toward and across the boninite field (as for mixing of boninite magma and small mass fractions of slab melts in Figs. 4A–4B). However, for magma-crust interactions, heating of the crust, and inmixing of the resultant materials will cool the basic magma and cause crystallization, leading to a steeper vector. In Figures 5A–5B, we established a more likely upper limit of interaction by modeling energy-constrained assimilation and fractional crystallization (EC-AFC; DePaolo, 1981; Spera and Bohrson, 2001) using the “magma chamber simulator” (Bohrson et al., 2014), which focuses on energy balance to determine the relative rates of crystallization and assimilation in an evolving magma chamber. The crystallizing phases were constrained using rhyolite-MELTS (Gualda et al., 2012). The assimilant was added as a melt of crust once the melt proportion reached 5%, and its composition made to evolve as the magma chamber wall rock heated up and became depleted by earlier melt extraction.

The EC-AFC models demonstrate that komatiites and some picrites can potentially cross into the boninite field on the MgO-SiO2 plot (Fig. 5A) by assimilating crust. Key provisos, however, are that the parent magma has sufficiently high MgO and that the assimilant temperature is near its solidus when fractionation begins. Thus, the trends in Figure 5A are for a high (700 °C) starting temperature for the assimilant. Decreasing this temperature means that the AFC process is initially dominated by olivine crystallization, resulting in a steep vector and reducing the capacity for the evolving magma to enter the boninite field. For example, komatiite magma B interacting with crust at 500 °C (not shown) does not enter the boninite field at all. Note, however, that interaction can be increased by allowing the starting magma to lie above its solidus, by increasing the proximity of the starting magma to the boninite boundary, and/or reducing the solidus temperature of the continental crust.

Even if the magma composition crosses the boundary in MgO-SiO2 space, however, the product will belong to the siliceous high-Mg basalt, not boninite, series, unless the starting composition also has sufficiently low TiO2. The models in Figure 5B demonstrate that slight TiO2 increase is the norm, because TiO2 decrease through assimilation of low-Ti crust is masked by increases due to earlier crystallization of pyroxenes. In consequence, only the low-Ti basalt Manihiki compositions have EC-AFC trajectories that enter the boninite field in the Ti8-Si8 projection (Fig. 5C).

Boninite Evolution by Fractional Crystallization

The primary boninite melts generated by the processes outlined in Figures 24 typically evolve by fractional crystallization to less MgO-rich (and more SiO2- and TiO2-rich) boninite magmas and then to high-Mg andesite magmas once MgO decreases below 8 wt%. For the Ti8-Si8 diagram to be most useful, Ti8 and Si8 should be insensitive to fractional crystallization. Thus, an understanding of theoretical fractionation trends will aid in the interpretation of observed trends on the boninite classification diagrams.

To assess fractional crystallization trends, we used rhyolite-MELTS to construct fractional crystallization trajectories at crustal pressures for some selected primary boninite melt compositions of both low-Si boninite and high-Si boninite subtypes (Figs. 5C–5D). For each, we ran the models at QFM+1 (one logarithmic unit above the quartz-fayalite-magnetite redox buffer) for four conditions: 1 kbar and 3 kbar wet (2 wt% water); and 1 kbar and 3 kbar dry.

The results show that, at crustal pressures with and without water, high-Si boninites magmas, H and BH, crystallize olivine first, followed by orthopyroxene. The temperature range of olivine crystallization before orthopyroxene starts to crystallize increases with increasing water concentrations and decreasing pressure. Eventually, crystallization of clinopyroxene and plagioclase (and perhaps magnetite, depending on fO2) generates shallower trends that cross the high-Mg andesite field on the MgO-SiO2 classification diagram (Fig. 5A), although plagioclase crystallization is suppressed in high-Si boninites because of their low CaO and Al2O3 concentrations and significant water concentrations (e.g., Dobson and O’Neil, 1987).

Comparison of Figures 5A and 5B reveals that fractional crystallization has a much greater effect on Si8 than Ti8. Thus, on the Ti8-Si8 diagram (Fig. 5C), boninite fractionation trends are typically subparallel to the Si8 axis. Some high-Si boninite magmas could cross into the low-Si boninite field during crystallization of drier magmas at higher pressures due to orthopyroxene crystallization, while some low-Si boninite magmas could cross into the high-Si boninite field during fractional crystallization of hydrous magmas at low pressure due to extensive olivine crystallization. At the margins of the boninite field, only deep, dry crystallization will take compositions out of the boninite field. However, most suites of boninitic magmas have differentiation trends that do run subparallel to the classification boundaries, as seen in the examples plotted in Supplemental File 11. This is likely due to prolonged olivine fractionation promoted by high water concentrations coupled with mixing of magmas and crystals before eruption.

In Supplemental File 1 (see footnote 1), we plotted a large number of alteration-screened data sets on the oxide classification diagrams of Figures 1B and 1C. Here, we demonstrate the methodologies used, first for the alteration screening, and second for identification and classification.

Screening Highly Metasomatized Rocks

Mg and Si are both susceptible to the metasomatic effects of weathering and metamorphism. In consequence, an evaluation of their mobility is a necessary prerequisite for the identification of boninites. The recalculation of the rock compositions as “anhydrous” (based only on cation oxides recalculated to 100%) corrects for alteration to a significant extent, given that many high LOI values in altered rocks reflect dilution due to fluid addition. The further question is how to identify Mg and/or Si metasomatism and whether to correct for it. There are many options, none of which is perfect. Our preferred method, which we outline here, is to use Cr as an immobile proxy for Mg and then use Cr-MgO, Cr-SiO2.and Cr-TiO2 plots for classification (Fig. 4).

Cr is not the only element that may be used (Ni and Co are also compatible and usually immobile), but its bulk partition coefficient of typically ∼4 during melting and fractional crystallization of mafic assemblages is most similar to that of Mg (Pearce and Parkinson, 1993). Cr in its most common oxidation state of CrIII is typically immobile, although Cr can theoretically be mobilized under extreme reducing and oxidizing conditions as CrII and CrVI, respectively. Empirically, however, almost all studies of chemical alteration of basic volcanic rocks have demonstrated the immobility of Cr (e.g., Lahaye et al., 1995). This may be attributed in part to the relative stability of its main host primary minerals (chromite and clinopyroxene) and affinity of Cr for many alteration products (e.g., chlorite and amphibole).

To test for Mg mobility, a plot of Cr against MgO can be effective, even though care is needed when Cr and Mg do not exhibit identical behavior, as in komatiite series in which olivine crystallizes before chromite (Barnes, 1998). For fresh rocks, the MORB glass database (Jenner and O’Neill, 2012), which includes glass from plume- and subduction-influenced ridges, gives a linear trend (Cr = 81.9MgO – 355.6) with ∼300 ppm Cr at MgO = 8 wt% and a 2σ close to 100 ppm. Thus, we used Cr = 300 ppm as a proxy for MgO = 8 in calculating Si8 and Ti8. Cr = 750 ppm and Cr = 1500 ppm can also be used as approximate proxies for MgO = 12 and 18 wt%, respectively, in defining the bounds of the picrite compositional field. For any altered volcanic suite, unsystematic deviations from a crystallization trend can generally be explained by Mg mobility.

Because of MgO mobility, we used Cr-SiO2 and Cr-TiO2 as proxies for MgO-SiO2 and MgO-TiO2, respectively. Cr-TiO2 is an immobile element diagram and can reliably replace MgO-TiO2 for most altered and unaltered samples. A minor difficulty is the lower bound of the boninite field, which can vary from 200 to 400 ppm. In contrast, Cr-SiO2 cannot generally be used to categorize samples into magma series because SiO2 lacks a direct immobile proxy. Instead, employing the same principles as those for Cr-MgO, unsystematic deviations can be used to screen samples when it is suspected that Si mobility may be significant.

Several of our test lava suites prove the importance of chemical screening of altered samples. For example, Figures 6A and 6B are plots of MgO-SiO2 and MgO-TiO2 for the Khizovaara “boninite” sequence in the North Karelian greenstone belt. The authors (Shchipansky et al., 2004) identified two suites, a “low-Ti basalt” suite and a “boninite” suite, which are plotted separately in Figure 6. Superficially, Figures 6A and 6B support this terminology. However, the Cr-MgO plot in Figure 6C reveals that, whereas these two elements in samples assigned to the low-Ti basalt group correlate significantly, those assigned to the boninite group plot off the low-Ti basalt trend to comparable Cr, but lower MgO, values. The most probable explanation is Mg loss.

The Cr-SiO2 plot (Fig. 6D) differs from the MgO-SiO2 plot (Fig. 6A) in that data no longer form a continuous trend from low-Ti basalts to boninite. Instead, the “boninites” are simply displaced from the low-Ti basalt trend, having similar Cr but variable SiO2. This does not follow any crystallization or melting trend and so is indicative of extensive and variable silica metasomatism.

Mg mobility is further supported by the “immobile” Cr-TiO2 plot in Figure 6E, where both low-Ti basalts and “boninite” samples form a common trend. This is quite different from the MgO-TiO2 trend in Figure 6B, where the low-Ti basalts and “boninite” groups are distinct. Taken together, Figures 6B and 6E demonstrate that all the samples belong to the same low-Ti basalt series; i.e., there are no boninites. This in turn means that there is a significantly lower probability that these lavas formed in a supra-subduction zone setting as originally proposed.

Another factor to be considered is the effect of summing major elements to 100% (the problem of closure). Thus, if silica metasomatism has added significant SiO2 to the rock, the apparent concentrations of MgO and TiO2 must decrease even if those elements were immobile. This is seen in the negative correlation between MgO and SiO2 in Figure 6A, as well as in the small displacement in TiO2 in Figure 6E.

In this paper, we have checked mobility as far as possible in all the plotted suites, and any obvious highly metasomatized or leached samples have been excluded from classification.

Classification of Boninite Suites

For this study, we classified low-Ti volcanic rock units from the geologic record by: (1) classifying individual samples using Si8 and Ti8 by applying Equations 2 and 3; and (2) recording the mean and standard deviation. If the mean and mean ± 1σ all gave the same classification, we assigned that classification uniquely to the units; otherwise, we reported the classification as transitional. Supplemental File 2 (footnote 1) provides templates of the diagrams and calculations, and Supplementary File 1 provides full details of the geological settings and MgO-SiO2 and MgO-TiO2 diagrams used for the classification, as well as data sources.

To demonstrate classification methodologies and issues that might arise, we focused on two examples, both from the type boninite province in the Izu-Bonin-Mariana forearc (Fig. 7; Figs. S1C and S1D [footnote 1]): a straightforward high-Si boninite suite from the type boninite locality of Chichijima (Taylor et al., 1994), and a more complex example from the inner wall of the Mariana Trench (Bloomer and Hawkins, 1987).

For the Chichijima boninites, data points plot fully in the high-Si boninite fields of the two oxide diagrams (Figs. 7A and 7B). Fractionation trends run subparallel to the boninite field boundaries, and so all data points plot in the high-Si boninite field of the Ti8-Si8 diagram with low dispersion (Fig. 7C). The water content of Chichijima primary magmas has been determined as ∼2 wt% (Dobson and O’Neil, 1987). From the rhyolite-MELTS trends in Figure 5A, one would thus predict crystallization of olivine followed by orthopyroxene. Instead, however, they follow a combined olivine + orthopyroxene trend. As these two minerals should not crystallize together at crustal pressures, the likely inference is that mixing between more- and less-differentiated boninites (or between boninites crystallizing at different pressures and with different water contents) was responsible. This inference is supported by mineral parageneses (Umino, 1986). Mixing here has, therefore, had a positive effect on the classification by smoothing out the more complex theoretical crystallization trends and producing a low-dispersion cluster on the Si8-Ti8 diagram. Overall, they classify unambiguously as a high-Si boninite–high-Mg andesite series. The estimated parent magma requires cpx-undersaturated melting at an approximate pressure of 1 GPa (Fig. 3), well within error of the 0.96 GPa calculated by Umino et al. (2017).

For the boninites from the Mariana forearc, the silica dispersion is much greater than for Chichijima (Fig. 7A). The mean composition is also clearly of high-Si boninite type, but the high standard deviation reflects the fact that the fractionation trend does not run parallel to the high-Si boninite boundaries. Instead, it begins in the low-Si boninite field before crossing the high-Si boninite field from top left to bottom right. This trend cannot be alteration, as silica correlates significantly with immobile elements such as Zr and Cr (Bloomer and Hawkins, 1987). Moreover, it does not match the cpx-undersaturated melting trends in Figure 2A. In addition, it does not match fractionation trends involving olivine and/or orthopyroxene (Fig. 5A), and it cannot be explained by AFC (Fig. 5C) because there is no continental crust to assimilate. Most likely, therefore, it is a mixing trend with a siliceous magma, either a cogenetic magma residing in the crust (e.g., a magma with the composition of proximal and coeval Saipan rhyolites; Reagan et al., 2008) or a slab melt within a hybrid mantle wedge (Fig. 4). These options may be distinguished isotopically, though that is outside the scope of this study.

In Table 1, therefore, we classify the Mariana Trench wall lavas as high-Si boninite transitional to low-Si boninite to indicate a dominant high-Si boninite composition derived from a low-Si boninite primary magma. The latter has slightly lower silica than the Chichijima primary magma, requiring a slightly smaller contribution from melting along an ol-opx peritectic, perhaps at slightly higher pressure (Fig. 7A). On the Ti8-Si8 classification diagram, the Chichijima boninites form a tight cluster, while the Mariana boninites form a trend from the low-Si through high-Si boninite quadrants toward low Ti8 and high Si8.

Here, we report the results of our classifications of boninites and related rocks of post-Archean age from known, or well-substantiated, tectonic settings. As will be seen, boninite, low-Ti basalt, and siliceous high-Mg basalt volcanic suites may be found in the wide range of settings, as described below. In the summary diagram (Fig. 8), we grouped these into two subduction-related groups (subduction initiation, and arc-basin systems; Figs. 8A and 8B), two intraplate groups (mantle plume–sourced and subcontinental lithospheric mantle–sourced systems; Fig. 8C), and one transitional group (plume-subduction interaction; right-hand edge of Fig. 8C).

In the text that follows, the full classification results are summarized in tabulated form for boninites from known geodynamic settings (Table 1) and for boninites from inferred or unknown settings (Table 2). They are also summarized graphically as Ti8-Si8 plots for post-Archean (Fig. 9) and Archean (Fig. 10) boninites.

Boninites from Oceanic Subduction Initiation Terranes (Izu-Bonin-Mariana Type)

All data for in situ boninites from this type of setting relate to subduction initiation events in the western and southwestern Pacific Ocean. The best covered location is its type locality, the Eocene Izu-Bonin-Mariana forearc. The youngest, and not fully evolved, is now believed to be the much smaller-scale, 2 Ma M&H (Matthews and Hunter) subduction initiation terrane at the southern edge (Figs. S2E–S2F [footnote 1]) of the Vanuatu arc (Sigurdsson et al., 1993; Patriat et al., 2019). As demonstrated and explained more fully in Figures S1A–S1F, subduction initiation was followed by creation of oceanic crust by a seafloor spreading event in which compositions evolved from tholeiitic basalt (forearc basalt) to low-Si boninite (Reagan et al., 2010, 2017). This was followed by construction of a protoarc, which was predominantly boninitic, commonly high-Si boninite, in composition (e.g., Taylor et al., 1994; Arculus et al., 1992; Kanayama et al., 2012). Finally, compositions reverted via transitional compositions to the basalt-andesite-dacite-rhyolite (BADR) series as subduction stabilized to give rise to a first arc, the beginning of normal arc magmatism. Overall, the period of the boninite-dominant magmatism extended for a period of some 5 m.y. (Ishizuka et al., 2011; Reagan et al., 2019), creating what Patriat et al. (2019) termed a subduction initiation terrane (SITER).

On the Ti8-Si8 diagram (Fig. 9A), the boninites from the protoarc form a diagonal field running across the boninite segment from upper left to lower right, i.e., with Ti8 decreasing as Si8 increases. The principal axis of this field may be explained by varying additions of siliceous melt to cpx-saturated magma. This siliceous melt could be: (1) mantle-derived melt as low-pressure, cpx-undersaturated melt from a second stage of mantle melting (Fig. 3); (2) slab-derived melt as amphibolite melt from fusion of MORB (Fig. 4); and/or (3) evolved, cogenetic melt in a crustal magma chamber (Fig. 7).

Despite the paucity of in situ examples of this boninite type, there are many well-documented analogues in the post-Archean geologic record, typically within or above suprasubduction zone ophiolites. This boninite-ophiolite association has long been used as an indicator, not just of a suprasubduction zone origin, but also of a subduction initiation setting (e.g., Stern and Bloomer, 1992; Whattam and Stern, 2011; Sklyarov et al., 2016). We plotted a subset of the many Phanerozoic examples on the oxide classification diagrams in Figures S5A–S5F (see footnote 1). The Ti8-Si8 diagram (Fig. 9E) shows that many of these are of high-Si boninite type, including the well-known examples from Papua New Guinea (Jenner, 1981), New Caledonia (e.g., Cameron, 1989), Newfoundland (e.g., Coish, 1989; Bédard, 1999), and Tasmania (Brown and Jenner, 1989). Low-Si boninites are also common, including well-known examples from the Troodos (e.g., Cameron, 1985) and Semail (Ishikawa et al., 2002; Kusano et al., 2017) ophiolites.

These Izu-Bonin-Mariana–type subduction initiation boninites extend, though with progressively fewer examples, through most of the Proterozoic. Examples in Table 2 and Figures S5G–S5J are from the 0.57 Ga Khantaishir ophiolite in western Mongolia (Gianola et al., 2017), the 0.78 Ga Adola mafic/ultramafic complex in southern Ethiopia (Wolde et al., 1996), and the 1.02 Ga East Sayan ophiolite from the Central Asian orogenic belt (Khain et al., 2002) The oldest of this class of boninites found so far is a ca. 1.9 Ga 200-m-thick unit from the Flin-Flon domain within the Trans-Hudson belt, Canada (Stern et al., 1995; Wyman, 1999).

Figure 9E demonstrates that the group of ophiolitic and protoarc boninites from post-Archean orogenic belts follows a very similar Ti8-Si8 trend to the Izu-Bonin-Mariana–type subduction initiation boninites in Figure 9A. It is thus evident that this subduction initiation setting provides almost all examples of the world’s high-Si, as well as some low-Si, boninites.

In terms of petrogenesis, it is well established that the Izu-Bonin-Mariana–type subduction initiation boninites are generated because residual mantle from this spreading event undergoes second-stage melting when fluxed by fluids and melts from the newly subducting plate (Fig. 8A). This melting event completes the exhaustion of clinopyroxene and promotes cpx-undersaturated melting to produce the low-Si and high-Si boninite compositions. The resulting melts create some oceanic crust and a significant proportion of the protoarc before upwelling of fertile mantle causes compositions to transition into those of the first (normal) arc.

Continental Subduction Initiation Terranes (Setouchi Type)

To our knowledge, the only intracontinental subduction initiation terrane containing boninites is the Miocene Setouchi belt in SW Japan, the type location of the sanukite rock type, defined as a form of high-Mg andesite (e.g., Tatsumi and Ishizaka, 1982; Tatsumi et al., 2002; Tatsumi, 2006). The precise setting of the Setouchi belt was the start of subduction of the young lithosphere of the Shikoku Basin (Tatsumi, 2006) and opening of the Japan Sea. Thus, it is unclear whether the sanukite compositions resulted from the embryonic nature of the subduction or the youth of the subducting plate, or a combination of the two. Both options are supported by the high slab-top temperatures implied by the isotopic evidence that melt from subducted sediment made a significant contribution to the genesis of these rocks (Shimoda et al., 2003). Tatsumi (2006) did not class these rocks as “boninites” because of their abundant groundmass plagioclase, but the “opx–high-Mg andesite” subgroup (in which orthopyroxene follows olivine as a crystallizing phase) does classify as boninite (high-Si boninite) using our classification scheme, while the “cpx–high-Mg andesite” subgroup (in which clinopyroxene follows olivine) classifies as siliceous high-Mg basalt (Figs. S1G–S1H [footnote 1]; Table 1).

In the Ti8-Si8 diagram (Fig. 9A), the sanukite lavas follow a trend from the uppermost part of the boninite high-Si boninite field across the siliceous high-Mg basalt field. They are therefore quite distinct from their oceanic equivalents. This may be because they lack a comparable mechanism for Ti depletion (mantle depletion at a preexisting spreading axis) while retaining mechanisms for silica enrichment (subducted sediment melt, high degrees of melting and crustal contamination).

Boninites from Oceanic Back-Arc Basins

In theory, back-arc basins should be ideal settings for boninite formation because they are commonly the sites of shallow, hydrous (and hence high-degree) mantle melting. As Pearce and Robinson (2010) have already shown, however, this is not the case. On the Ti8-Si8 projection (Fig. 9B), back-arc basin basalts follow a trend in which Si8 increases and Ti8 decreases with increasing trench proximity, likely as a consequence of increasing mantle depletion and increasing fluid flux. Of back-arc spreading segments with proposed or potential boninitic affinities, only two sets of arc-proximal segments fit the boninite classification: the Manus Basin East Rift (Sinton et al., 2003) and the Fonualai Rift in the NE Lau Basin (Figs. S2A–S2B [footnote 1]; Keller et al., 2008).

Boninites from Oceanic Forearc Rifts

The only recent example of a forearc basin, where rifting takes place on the trench side of an arc, is the Southeast Mariana forearc rift, dated at 3.7–2.7 Ma (Figs. S2C–S2D). It overlies the shallow subducting slab at the southern end of the Mariana arc (Ribeiro et al., 2013a, 2013b). Boninite compositions (actually transitional to low-Ti basalts; Fig. 9B) are found in olivine-hosted melt inclusions trapped during an early stage of extension (Ribeiro et al., 2015) when subduction fluids interacted with a depleted mantle wedge trapped between the trench and the arc. During the subsequent spreading event, upwelling of fertile mantle caused the magmas to revert to basalt compositions.

Another type of rifted forearc basin is the Eocene Izu-Bonin forearc basin, presently located trenchward of the active Izu-Bonin arc. It likely originated ∼10 m.y. after subduction initiation by splitting of the boninitic protoarc, attaining its forearc setting once the “first arc” became established in a position further from the trench. The Izu-Bonin forearc basin was sampled at two sites by drilling during Ocean Drilling Program (ODP) Leg 126 (Taylor et al., 1992): Site 793 lies on the eastern (trench) side of the basin and contains rocks classifying as low-Si boninites, while Site 792 lies on the western (arc) side and contains rocks classifying as transitional between boninite and siliceous high-Mg basalts (see Figs. S2C–S2D [footnote 1]; Fig. 9B). In terms of magma genesis, the Southeast Mariana forearc rift model of rift-related reactivation of preexisting depleted mantle wedge could also apply. Alternatively, mantle flowing into the shallow, relatively near-trench setting could undergo depletion along its trenchward flow path (Taylor et al., 1992).

Boninites from Oceanic Slab Edges

Slab-edge boninites are distinct from normal forearc or back-arc settings because they erupt close to ridge-trench-transform (RTF) triple junctions. In this setting, subduction takes place beneath actively spreading ridges or zones of diffuse, volcanically active rift basins. Moreover, this is a location where hot mantle from outside the subduction system can flow sideways into the shallow mantle wedge, driven by suction from slab roll-back. The type example is the northern edge of the Tonga subduction system, where boninites have been reported from the near-trench zone of diffuse spreading and rifting (e.g., Falloon and Crawford, 1991; Falloon et al., 2008), and from discrete submarine volcanic edifices such as Mata Seamount (Resing et al., 2011; Glancy, 2014). North Tonga is distinctive because of its proximity to the Samoan plume, so raising the possibility, supported by geothermometry, that the influx of depleted plume residue was instrumental in the genesis of the boninites (Sobolev and Danyushevsky, 1994; Falloon et al., 2007). It is thus a rare, possibly unique, recent example of boninite genesis by plume–subduction zone interaction.

Figures S2E–S2F (footnote 1) and Figure 9B show that plume-sourced North Tonga boninites have low-Si boninite character, in one location transitional to high-Si boninites. In Figure 9B, they form a shallow, incremental melting trend characterized by variable Ti8 but constant Si8 (cf. Fig. 2C) with the boninites plotting at the low-Ti8 end of the trend.

Boninites from Oceanic Arcs

The vast majority of oceanic volcanic arc basalts follow basalt-andesite-dacite-rhyolite (BADR) trends on the MgO-SiO2 diagram and have Ti8 > 0.5 on the MgO-TiO2 diagram (Pearce and Robinson, 2010). The three exceptions recognized to date are Volcano A (Cooper et al., 2010) and Tafahi (Beier et al., 2017) from the Tofua arc and the Sulu Ridge from the New Britain arc (Woodhead et al., 1998). All of these are described and plotted in Figures S3A–S3B (footnote 1). On the Ti8-Si8 projection (Fig. 9C), they plot as low-Si boninites just within the boninite quadrant at the low-Ti, high-Si end of an oceanic arc field.

The principal cause of low-Ti8 in oceanic arc boninites is preconditioning (McCulloch and Gamble, 1991), whereby mantle flowing toward the arc front loses melt fractions along its flow path, either in adjacent back-arc basins (Volcano A and Tafahi) or in the rear-arc part of the arc itself (Sulu Ridge). Preconditioning in back-arc basins may be episodic, as proposed for Volcano A (Cooper et al., 2010), or continuous, as may be the case in Tafahi. Within the arc melting column itself, Tamura (1994) argued that the final, shallowest melt fractions typically have boninite compositions, but that they normally mix with deeper and more voluminous melt fractions and so do not form separate eruptions. If so, this may help to explain why boninites make up a tiny proportion of arc magmas and are typically interbedded with normal calc-alkaline or tholeiitic arc lavas (Fig. 8B).

Small numbers of oceanic arc volcanoes from anomalous settings (not plotted) classify as members of the siliceous high-Mg basalt–high-Mg andesite series. These are from: the Ryukyu arc, associated with the opening of the Okinawa Trough in the late Miocene following resumption of subduction (Shinjo, 1999); the western Aleutian (Yogodzinski et al., 1995) and southern Vanuatu (Monzier et al., 1993) arcs in areas of oblique subduction; and Kavachi in the Solomon arc (Johnson et al., 1984), in an area close to the site of ridge subduction. In these cases, large slab melt fluxes may have led to an increase in Si8, but there was no mechanism for the mantle depletion that was also needed to form boninites rather than siliceous high-Mg basalts.

Boninites from Continental Arcs

For continental arcs, we found that no lava series classifies as boninite. The vast majority belongs to the basalt-andesite-dacite-rhyolite series, though a few classify as siliceous high-Mg basalts (Fig. 9C), including: Mount Shasta in the Cascades (Grove et al., 2005); Ruapehu (Price et al., 1999) and White Island (Cole et al., 2000) in New Zealand; and Sheveluch, Shisheisky, and Klyuschevsky in Kamchatka (Bryant et al., 2011). See also Figures S3C–S3D (footnote 1). The one common feature of the volcanoes that erupt siliceous high-Mg basalts is the link to regional extension, raising the possibility that either (1) extension-driven melting of metasomatized lithosphere or (2) subarc front melting of asthenosphere preconditioned by extensional, rear-arc magmatism (Grove et al., 2002) is responsible. In no case, however, is Ti depletion sufficient to produce boninites.

Boninites from Collision Zones

We are aware of only one example of collision boninites: in the segments of the West Bismarck island arc that are in collision with the Papua New Guinea margin (Johnson et al., 1985; Woodhead et al., 2010; Cunningham et al., 2012). The boninites from this setting classify as low-Si boninites, overlapping the low-Ti basalt field (Figs. S3E–S3F [footnote 1]; Fig. 9C). An increase in subduction flux during collision, and the presence of depleted melt residues in the mantle wedge, may have provided the conditions for boninite genesis. In continental collision zones, the closest rocks to boninites is a suite of late Miocene, postcollision, high-MgO lavas from the Aegean–western Anatolia system (Agostini et al., 2005), which classify as siliceous high-Mg basalts (Figs. S3E–S3F; Fig. 9C).

Boninites from Mid-Ocean Ridges

Boninites in oceanic, nonsubduction settings have been described as “extremely rare” (Natland, 2014). Our search of MORB databases (e.g., Jenner and O’Neill, 2012; see also Figs. S4A–S4B [footnote 1]) revealed that, with the exception of a tiny proportion of outliers, all samples from the world’s ridge system plot in the basalt quadrant of Figure 9D. This observation is consistent with the experimental data in Figure 2A, which demonstrate that pooled melts from fractional melting columns do not have boninite compositions. The one cited example of a mid-ocean-ridge “boninite” from the Somali Basin (e.g., Frey et al., 1980) classifies as siliceous high-Mg basalts here, but its original setting may have been an attenuated continental margin rather than an ocean ridge.

Boninites from Oceanic Intraplate Settings

Oceanic islands are typically derived from fertile mantle asthenosphere that melts beneath thick lithosphere, resulting in low to moderate degrees of melting at depth (e.g., Watson and McKenzie, 1991). In consequence, Ti8 is typically much higher, and Si8 is lower, than boninites. Some oceanic plateau lavas (e.g., those on Iceland) do have relatively low TiO2 values, but these still classify as basalts (Figs. S4C–S4D [footnote 1]). The only lavas so far described as “boninite” are from the Manihiki Plateau (Ingle et al., 2007; Timm et al., 2011; Golowin et al., 2017a, 2017b). Our classification scheme supports this description (Figs. S4C–S4D). Specifically, lavas from the Suvorov Trough classify as low-Ti basalts, but those from the Central Danger Island Troughs (DIT) classify as transitional between boninite and low-Ti basalts (Fig. 8D).

Golowin et al. (2017a, 2017b) proposed that the upwelling mantle plume beneath the (at the time) adjacent Ontong-Java Plateau underwent some 30% of melting to form the Ontong-Java Plateau, but a depleted fraction of the mantle plume was then refertilized and remelted beneath the Manihiki Plateau. This second stage of melting could be attributed to further decompression, e.g., if residual mantle flowed into a zone of thinner, possibly near-ridge, oceanic lithosphere (Fig. 8C). Alternatively, as Golowin et al. (2017a) proposed, the second stage of melting could have been due to input of heat from hotter upwelling mantle.

Boninites from Continental Intraplate Settings

Only a few continental intraplate localities have lavas with Ti8 < 1.0. They are a subset of the so-called “low-Ti” lavas associated with continental breakup (e.g., Peate and Hawkesworth, 1996). These have been shown to have been sourced from subcontinental lithospheric mantle, which was depleted but then refertilized during subsequent subduction events (e.g., Merle et al., 2013). Of the provinces related to continental breakup that led to formation of present-day ocean basins, most lavas with Ti8 < 1 classify as basalt. However, a small subset classifies as siliceous high-Mg basalts, notably those from the Ferrar Province linked to Antarctica-Australia breakup (Hergt et al., 1991; Siders and Elliot, 1985) and parts of the Central Atlantic magmatic province (e.g., Grossman et al., 1991; see also Figs. S4E–S4F [footnote 1]; Fig. 8D). None of these low-Ti, subcontinental lithospheric mantle–sourced lavas classifies as “boninite.”

Going back in time, we found no continental, intraplate boninites until the Paleoproterozoic at ca. 2 Ga. This age marks the first appearance of the mainly plutonic boninite-norite (BN) complexes in intracontinental rift settings (Srivastava, 2006, 2008) and an increase in the frequency of appearance of siliceous high-Mg basalt lavas and dikes (Sun et al., 1989). The consensus interpretation is that these siliceous high-Mg basalts originated by plume- and/or extension-driven reactivation of highly depleted, but re-enriched, subcontinental lithospheric mantle harzburgites (Fig. 8C). Generally, the critical factor for boninite genesis is having a sufficiently depleted mantle source prior to enrichment by low-Ti, subduction-derived fluids and melts.

We plotted a series of these Paleoproterozoic intracratonic rift volcanic suites in Figure S6 (see footnote 1) and Figure 9F. As Table 2 summarizes, our classification of these suites supports that of Sun et al. (1989) in mainly giving siliceous high-Mg basalt compositions (e.g., Wilson, 1982). The Bastar craton samples (Figs. S6A–S6B), which Srivastava et al. (2004) described as “boninite,” are transitional between siliceous high-Mg basalt and boninite (low-Si boninites). However, in the ca. 2.05 Ga Bushveld large igneous province, the earliest magmas (B1 sills) from the Bushveld Complex (Barnes et al., 2010) classify clearly as boninite (low-Si boninites). Broadly contemporaneous magmas from same large igneous province, the Trabazimbi sills (Rajesh et al., 2013), classify as siliceous high-Mg basalts, and later Bushveld magmas (B2 and B3 sills) classify as basalt (Figs. S6C–S6D [footnote 1]). In addition, Stubbs et al. (1999) reported one boninite composition from satellite dikes to the Great Dyke. In Figure 9F, this group of intracratonic lavas and dikes forms a near-vertical trend through the siliceous high-Mg basalt field and just into the boninite (low-Si boninite) field (Bushveld B1). This trend could be explained by shallow, incremental melting (Fig. 2C).

Using the same methodology as that used for the post-Archean boninites, we also examined the Archean rocks that have been described as “boninites” or “boninite-like” rocks to discover (1) whether they are boninites according to our classification, and (2) the post-Archean settings to which they best correspond. We subdivided them on the basis of published groupings, and with no implied setting, into: Barberton Al-undepleted komatiites (AUK) type, Commondale type, basaltic komatiite (BK) type, Whitney type, Whundo type, Mallina type, Kambalda type, and Nuvvuagittuq type. Figures S7–S12 (footnote 1) contain the oxide plots, and Figure 10 and Table 2 summarize the classifications. Interpretation of the Archean boninites in terms of their most probable tectonic settings requires incorporation of geochemical fingerprinting and, for a small subset of boninites of ambiguous settings, a more detailed examination of the petrogenetic evidence from Si-Mg-Ti systematics (Fig. 12).

For reasons of space, we restricted the geochemical fingerprinting to the Th/Yb-Nb/Yb projection (Pearce, 2008). This method was chosen because it is has already been applied successfully to determining the settings of ophiolitic boninites (Pearce, 2014) and to distinguishing subduction-related from intraplate Archean greenstones (Smithies et al., 2018). The diagram itself (Fig. 11) has two parallel arrays, one for magmas unaffected by subduction or crustal contamination (the MORB-OIB array) and the other for volcanic arc lavas (the arc array), while contaminated magmas from the MORB-OIB array form diagonal trends toward the crustal compositions. Here, each part of Figure 11 represents a different tectonically defined boninite group, on which we have plotted the compositions of our post-Archean examples and their potential Archean equivalents.

Barberton AUK-Type “Boninites”

In the oxide plots for representative Archean komatiite suites (Figs. S7A–S7B [footnote 1]), the Barberton Al-undepleted komatiites (AUK) is the only suite to classify as “boninite.” To investigate this result further, and to circumvent the problem of within-suite metasomatic variations, we took the estimates of primary komatiite magma compositions made by Sossi et al. (2016) and extrapolated these to obtain Ti8 and Si8 values (Fig. 10A). The AUK then formed a diagonal trend that starts in the komatiite field, passes through the edge of the siliceous high-Mg basalt field, and extends as far as the Barberton AUK series in the boninite quadrant. In contrast, the Al-depleted komatiites (ADK) plot entirely as siliceous high-Mg basalts, but they are displaced to high Ti8 compared to AUK compositions.

Sossi et al. (2016) interpreted these Barberton AUK-type komatiites as the product of 40% batch, or 50% fractional, melting of slightly depleted mantle at 9 GPa and a TP of 1950 °C, the highest estimated degree of melting for any komatiite suite. This is consistent with Figure 2, where the batch melting grid at high pressures is closest to the boninite field at the highest degrees (>40%) of melting. The Barberton AUK analyses plot in the center of the MORB-OIB array in the fingerprinting diagram (Fig. 11C), which is consistent with an undepleted to slightly depleted, plume-sourced magma with no discernible crustal contamination (Pearce, 2008).

Commondale-Type “Boninites”

Other komatiites with long-recognized boninitic affinities are the depleted, Commondale type found within the ca. 3.3 Ga Commondale greenstone belt (e.g., Wilson, 2002, 2003) and the Weltevreden Formation of the Barberton greenstone belt (e.g., Stiegler et al., 2012; see also Figs. S7C–S7D [footnote 1]; Table 2). “Depleted” refers to the extreme depletions in the most highly incompatible elements (Wilson et al., 2003). The Ti8-Si8 diagram (Fig. 10A) confirms that both suites classify as boninite (low-Si boninites), with the Weltevreden data plotting on an extension of the Sossi trend, and the Commondale data displaced to a lower average Ti8 value.

The sources and settings of these depleted komatiitic boninites are controversial, with advocates of both a shallow, subduction-driven origin (Wilson et al., 2003; Barr et al., 2009) and an intraplate plume origin (Stiegler et al., 2012; Puchtel et al., 2013). The fingerprinting diagram (Fig. 11D) shows that the Weltevreden and Commondale rocks have low-Nb/Yb compositions within, or just above, the MORB-OIB array. This is indicative of magmas derived from depleted mantle that has variably interacted with continental crust (Pearce, 2008). Stiegler et al. (2012) and Puchtel et al. (2013) proposed that these unusually depleted compositions were sourced from relict minerals from a magma ocean. While possible, our diagrams do not require the mantle source to be anything other than a hot, deep mantle plume similar to that producing the Barberton AUK boninites, except that it had undergone a prior, deep depletion event.

BK-Type “Boninites”

Cameron et al. (1979) first noted that the Archean lavas defined by Viljoen and Viljoen (1969) as “basaltic komatiites (BK)” exhibited many of the characteristics of Phanerozoic boninites. Parman et al. (2001, 2004) took these similarities further by proposing that “basaltic komatiites” from the ca. 3.5 Ga Komati Formation of the Barberton greenstone belt were Archean equivalents of Phanerozoic subduction-related boninites. Here, we examine that hypothesis from the perspective of boninite classification.

Apart from a few outliers, BK lavas from the Komati Formation (Viljoen and Viljoen, 1969) plot as “siliceous high-magnesium basalts” on the oxide diagrams in Figures S8A–S8B (footnote 1) and the Ti8-Si8 diagram in Figure 10A. However, one set of samples from the Hooggenoeg Formation higher in the Barberton sequence (section HV9 of Furnes et al., 2012) does classify as boninites on these diagrams. The Koolyanobbing greenstone belt in the 3.02 Ga Southern Cross domain of the Yilgarn craton (Angerer et al., 2013) is another possible example; its lavas classify as transitional between siliceous high-Mg basalts and low-Si boninites.

The tectonic setting of this type of boninite–siliceous high-Mg basalt association is debated. Parman and Grove (2004) and Furnes et al. (2012) both argued for a subduction origin, though most others (e.g., Arndt et al., 1998) have seen them as simply part of a normal plume-related volcanic sequence. Here, the vertical siliceous high-Mg basalt–boninite trend on the Ti8-Si8 diagram (Fig. 10A) is quite different from the diagonal trend followed by their proposed subduction initiation boninite analogues (Figs. 9A and 9E). On the fingerprinting diagram (Fig. 11D), the boninites from the Hooggenoeg Formation plot on the upper edge of the MORB-OIB array and distinctly below the arc array. These compositions are typical of continental basic lavas that are plume-sourced, with an intraplate setting, and with variable crustal contamination (Pearce, 2008).

Whitney-Type “Boninites”

Kerrich et al. (1998) described as “boninites” two lava units from the Whitney Township outcrops of the ca. 2.7 Ga Abitibi greenstone belt. Smithies et al. (2004a) classed these as “Whitney-type” boninites. They are intercalated within a volcanic sequence, which also includes Al-depleted and Al-undepleted komatiites and tholeiitic basalts. This apparent alternation of “boninites” and komatiites provided Fan and Kerrich (1997), among others, with supporting evidence for their much-discussed model of plume–subduction zone interaction (e.g., Wyman, 2012; Bédard et al., 2013). Figures S9A–S9B (footnote 1) provides oxide plots for the proposed members of the Whitney boninite group. Figure 10C and Table 2 summarize their Ti8-Si8 characteristics.

In MgO-SiO2 space (Fig. S9A), most of the Whitney Township lavas classify as “picrites” in IUGS terminology, with the few evolved samples classifying as high-Mg andesites. Thus, none of the published data actually classifies as “boninite,” although an evolutionary trend between the picrites and high-Mg andesite could theoretically intercept the corner of the boninite field and be explained by assimilation and fractional crystallization, as in the EC-AFC trends in Figure 5A. Nonetheless, it is evident that the parental magmas were not boninites, as the most basic liquid compositions plot well to the left of the boninite field on that diagram. Ti8 values are low, however, resulting in their classification as low-Ti basalts (Fig. S9B [footnote 1]; Fig. 8C).

Figures S9C–S9D and Figure 10B demonstrate that other lavas assigned by Smithies et al. (2004a) to the Whitney-type boninite group similarly plot as low-Ti basalts. These include Kidd Creek Lower lavas (Fan and Kerrich, 1997; Kerrich et al., 1998), the ca. 2.8 Ga North Karelian greenstones (Shchipansky et al., 2004), and the ca. 3.8 Ga Isua “boninite-like” amphibolites of Polat et al. (2002).

The fingerprinting diagram in Figure 11D reveals that the Whitney and Karelia low-Ti basalt lavas plot close to the Manihiki Plateau Suvorov Trough low-Ti basalt lavas, but displaced toward continental crust (Isua amphibolites have not been plotted because of Th metasomatism). These Archean low-Ti basalt series also have similar compositions to the Archean Commondale-type komatiitic boninites on this projection. This is consistent with the derivation of both from depleted mantle plumes. Our classification therefore supports that of Smithies et al. (2004a), who inferred a plume origin for the Whitney-type “boninites” on the basis of geology (e.g., the association with komatiites) and geochemistry (e.g., higher Al and heavy rare earth elements [HREEs]).

Mallina-Type “Boninites”

The ca. 3.0 Ga lavas and minor intrusions of the Mallina Basin formed in an intracratonic rift near the margin of the East Pilbara craton. Sun et al. (1989) first studied in detail one of the units, the Negri volcanics, noting its siliceous high-Mg basalt character. Smithies (2002) and Smithies et al. (2004b) subsequently discovered that some units have boninite compositions that differ in a number of respects from their Whitney and Whundo boninite types. They also recognized similarities with the siliceous high-Mg basalt–boninite associations of the early Proterozoic intracratonic rift volcanics (Fig. 9F). Our classifications (Fig. S9C; Fig. 10C) match those of Sun et al. (1989), Smithies (2002), and Smithies et al. (2004b). In Figure 10C, the various lava suites form a vertical trend from the siliceous high-Mg basalts (Negri and South Mallina basalts) into the boninite field (Loudon and other boninite-like rocks). This trend is consistent with a shallow (<2 Gpa) mantle source variably depleted by fractional or incremental melting (Fig. 2A).

Srivastava (2006, 2008) identified the Stillwater complex (ca. 2.7 Ga) as another example of an Archean intracratonic boninite. Data are limited, however, to a small set of sills and dikes beneath the complex (Helz, 1985). Helz (1995) considered the range of compositions to reflect mixing two end members: a magnesian norite (MN) and a magnesian gabbronorite (MGN; Figs. S10C–S10D [footnote 1]). This interpretation is supported by the MN-MGN trends in Figures S10C–S10D, which follow mixing trajectories in which SiO2 decreases with decreasing MgO. On the Ti8-Si8 projection, the BN end member classifies as boninite (low-Si boninites), while the MGN end member is off scale above the basalt field. The logical explanation is that two mantle sources, an asthenospheric (plume) source and a subcontinental lithospheric mantle source, underwent melting simultaneously (Fig. 8C).

In the Th/Yb-Nb/Yb diagram (Fig. 11E), the Mallina Basin lavas plot in the arc array together with the Paleoproterozoic intracratonic rift lavas. This is a well-known feature of inheritance in which the subcontinental lithospheric mantle source retains a history of past subduction-related enrichment, perhaps linked to craton accretion. No whole-rock trace-element data are yet available for the Stillwater minor intrusions.

Kambalda-Type “Siliceous High-Mg Basalts”

Redman and Keays (1985) separated non-komatiitic lavas from parts of the Yilgarn and Superior cratons into low-Mg basalts, high-Mg basalts, and siliceous high-Mg basalts. Figures S8C–S8F and Figure 10D compare our classification scheme with that of Redman and Keys for the type locations of Kambalda (Redman and Keays, 1985) and Munro Township (Arndt and Nesbitt, 1982). Comparisons show that the two classification schemes give similar results.

Despite evidence for crustal contamination in Kambalda (e.g., Arndt and Jenner, 1986; Said et al., 2010), the absence of diagonal trends on the MgO-SiO2 plot (cf. Fig. S8C [footnote 1]; Fig. 4B) indicates that crustal contamination may not have been instrumental in explaining the distinction between basalts and siliceous high-Mg basalts. This conclusion was also reached by Said and Kerrich (2009), who explained the siliceous high-Mg basalt compositions at the top of the Kambalda sequence not by assimilation but by melting of depleted plume mantle made heterogeneous by crustal recycling. This is possible, though we note that Said and Kerrich (2009) interpreted the setting as plume magmatism at a craton margin, in which case, the siliceous high-Mg basalts may be of Mallina-type, sourced by metasomatized craton lithosphere. Support for this latter interpretation is provided by Figure 11E, where the siliceous high-Mg basalt lavas plot in the arc array close to other subcontinental lithospheric mantle–sourced lavas and hence are attributable to an inherited subduction component. The low-Mg basalt lavas (not plotted) lie in the MORB-OIB array and so can be assigned a probable asthenosphere source.

Whundo-Type “Boninites”

Smithies et al. (2004a, 2005) treated the 3.12 Ga Whundo volcanic sequence of West Pilbara as their type example of Archean boninites of subduction origin. They showed that the boninites are intercalated with calc-alkaline rocks within the lowermost of three volcanic sequences. In Figure S11A (footnote 1), the rocks described as “boninites” plot in the low-Si, low-Mg part of the boninite (low-Si boninite) field, forming a fractionation trend, which extends into the high-Mg andesite field. In contrast, the interbedded “calc-alkaline” lavas and the overlying “tholeiitic” lavas classify as members of a basalt–high-Mg andesite series. These are not typical of most present-day arc series in that respect, as the latter more commonly follow basalt-andesite-dacite-rhyolite series trends (cf. Fig. S3). In their papers, Smithies and coworkers argued that this high-Mg andesite character, together with a small set of high-Nb lavas and some acidic rocks with adakitic compositions, is indicative of hot subduction.

Smithies et al. (2004a, 2005) also categorized the “boninite-type” volcanic rocks reported by Boily and Dion (2002) from the ca. 2.8 Ga Frotet-Evans greenstone belt of the Opatica subprovince in Quebec as Whundo-type boninites (Figs. S11C–S11D [footnote 1]). Both the Whundo and Frotet-Evans “boninites” classify as low-Si boninite on the Ti8-Si8 plot (Fig. 10E), where they occupy the same part of the classification diagram as recent arc-basin boninites (Fig. 9B). The lavas from both localities similarly plot well within the arc array on the fingerprinting diagram (Fig. 11B).

The “boninites” of the Neoarchean Gadwal greenstone belt (ca. 2.7 Ga) from the eastern Dharwar craton in India have also been interpreted as subduction-related rocks (Manikyamba et al., 2005, 2007; Khanna, 2013). Superficially, they resemble the Whundo-type boninites, with similar Ti8-Si8 (Fig. 10E) and Th-Nb-Yb (Fig. 11B) systematics. However, their parent magma compositions had much higher MgO (Figs. S11E–S11F [footnote 1]). Figure 12, a plot of the parental magmas from all intraplate and subduction-related boninites, demonstrates that present-day arc-basin and Whundo-type boninites are characterized by low-MgO parental magmas, consistent with low-pressure melting and a small relative contribution from cpx-undersaturated melting. In contrast, the parental boninites from the Gadwal craton much more closely resemble parental intraplate plume-sourced boninites of the BK type. However, the apparent subduction signature better fits the intracratonic Mallina-type boninites, making the Gadwal suite difficult to categorize.

Nuvvuagittuq-Type “Boninites”

The Nuvvuagittuq area in northern Canada hosts a small (∼9 km2) sequence of amphibolite facies lavas dated as >3.75 Ga (O’Neil et al., 2011) and divided into Lower, Middle, and Upper units. Turner et al. (2014) proposed that they provide evidence for a subduction initiation event. In their Izu-Bonin-Mariana–based model, the Lower Unit is tholeiitic and represents ocean crust created by spreading following subduction initiation, the Middle Unit is boninitic representing the start of arc volcanism, and the Upper Unit is calc-alkaline representing further evolution into a volcanic arc. If true, not only is it the oldest boninite location, but it also provides evidence for plate-tectonic processes in the earliest Archean. Here, we used our methods of alteration screening and classification to test this model further. Full details of both are given in Figure S12 (footnote 1).

The Ti8-Si8 diagram in Figure 10F, based on alteration-screened data (Figs. S11B–S11D), summarizes the large geochemical contrast between the three volcanic units, in which the Lower Unit classifies as basalt (basalt-andesite-dacite-rhyolite series), the Middle Unit classifies as transitional between low-Si boninite and low-Ti basalt, and the Upper Unit classifies as siliceous high-Mg basalt. In terms of geochemical fingerprinting, Th/Yb-Nb/Yb plots by O’Neil et al. (2011) and Turner et al. (2014) revealed a wide dispersion due to Th mobility during amphibolite-facies metamorphism. However, lavas from the Upper Unit plot consistently in the arc array regardless of alteration intensity (Fig. 11B), supporting interpretations that the Upper Unit does have real subduction characteristics.

This evolution up-sequence from basalt to the low-Ti basalt–boninite boundary to siliceous high-Mg basalts (Fig. 10F) could, therefore, denote Izu-Bonin-Mariana–type subduction initiation as Turner et al. (2014) proposed. However, this is in marked contrast to the Izu-Bonin-Mariana evolutionary trend, which extends directly from fore-arc basalt to low-Si and high-Si boninite and then returns to basalt as normal arc volcanism begins (Fig. S1 [footnote 1]; Fig. 9A).

As with the Gadwal boninites, the MgO-SiO2 plot of intraplate and subduction-related parental boninites in Figure 12 is informative. On this diagram, Izu-Bonin-Mariana–type subduction initiation boninites plot along low-pressure cpx-undersaturated trends (cf. Fig. 3A). In contrast, the most primitive Middle and Upper Unit Nuvvuagittuq lavas both plot in the central part of the arc-basin field in Figure 12B and thus more closely resemble Whundo-type boninites. This does not rule out a subduction initiation interpretation, however, because subduction initiation on continental or ocean plateau crust is still an option as discussed further below.

Our aim in this publication has been to (1) establish a protocol for rigorous classification of boninites, (2) apply this protocol to samples throughout the geologic record that have been reported as having boninitic affinities, and (3) assign those rocks classifying as boninites, siliceous high-Mg basalts, or low-Ti basalts to their most probable tectonic settings. In Figure 13, we summarize our results in the form of a boninite time line made up of a set of individual time lines for boninites of different tectonic affinities. These time lines highlight our principal conclusions, namely, (1) boninites may originate in a range of both subduction and intraplate settings, and (2) the relative importance of these settings has varied through geological time. As discussed below (and with the obvious caveat that data are sparse), we found that, whereas arc-basin boninites likely extend as rare magma types throughout the geological record, others occupy specific time windows. We focus here on: the Izu-Bonin-Mariana–type subduction initiation boninite window from 0 to 2 Ga, the intracratonic boninite–siliceous high-Mg basalt association window from 2 to 3 Ga; and the plume-derived boninite, siliceous high-Mg basalt, and low-Ti basalt window from 3 to 4 Ga.

0–2 Ga: Principal Period of Izu-Bonin-Mariana–Type Subduction Initiation Boninites

We have shown that, at the present day, boninites are rare rock types in commonly preserved tectonic settings (island arcs, continent–island arc collisions, oceanic plateau, and continental subduction initiation) and common rock types in rarely preserved tectonic settings (forearc basins and northern Tonga–type slab edges). Only in Izu-Bonin-Mariana–type of subduction initiation terranes are boninites common rock types in a setting commonly preserved in the geologic record. Thus, this setting has been providing the great majority of boninites, including almost all high-Si boninites, extending back over a significant period of time. However, this period ends with the ca. 1.9 Ga boninites of the Flin Flon terrane (rounded up to 2 Ga on the time line). The distinctive petrogenetic feature of boninites of Izu-Bonin-Mariana subduction initiation type is their derivation from shallow, subduction-related, cpx-undersaturated melting of peridotites that have already undergone significant depletion. We found no high-Si boninites earlier than 2 Ga (Fig. 10), and, while there are a small number of good examples of subduction-related Whundo-type boninites stretching back through the Archean, these are dominated by cpx-saturated melting (Fig. 12) and so are more similar to boninites from established arc-basin settings.

The absence of any Archean analogues for this principal category of post-Archean boninites was the surprising result of this study. One explanation is that Izu-Bonin-Mariana–type subduction initiation boninites were as common in the Archean as in the Phanerozoic, but poor preservation means that they still await discovery. The opposite alternative is that Earth was a one-plate planet throughout the Archean with no subduction zones to initiate. In that case, the subduction signature would result from translation of crust into the mantle by other means such as delamination, underthrusting, and sagduction (e.g., Zegers and van Keken, 2001; Bédard et al., 2013; Johnson et al., 2014).

A further alternative is that plate subduction did take place in the Archean, but without significant trench retreat and hence without the spreading events that preceded the post–2 Ga boninitic protoarcs. It is possible, for example, that weaker Archean asthenosphere and lower plate density and strength favored deep penetration of slabs into the mantle but inhibited trench retreat (Agrusta et al., 2018). In that case, there would be no reservoir of hot, residual (harzburgitic) mantle trapped in the embryonic mantle wedge and hence no newly depleted source available to form a boninitic protoarc (Fig. 8A; Reagan et al., 2017, 2019). One would then expect the first erupted compositions to be normal arc magmas, as happens in modern-day non–Izu-Bonin-Mariana–type subduction initiation settings such as the Tertiary ancestral South Sandwich arc (Dalziel et al., 2013). While not conclusive, subduction initiation without spreading could also explain, for example, why Nuvvuagittuq carries good chemostratigraphic evidence for evolution from a nonsubduction to subduction setting (Turner et al., 2014) but lacks the types of boninite typical of Izu-Bonin-Mariana–type subduction initiation systems.

2–3 Ga: Principal Period of Intracratonic Boninite–Siliceous High-Mg Basalt Associations

Although both the arc-basin boninite and the plume-related low-Ti basalt time lines continue through the Archean, intracratonic (or craton edge) boninites are most characteristic of the period 2–3 Ga (Fig. 13). Sun et al. (1989) and Srivastava (2006, 2008), in particular, developed the concept of a period of Archean–Paleoproterozoic, global-scale, intracratonic magmatism dominated by boninite and siliceous high-Mg basalt magmas. We support their work, specifically finding that the rocks in question classify as a boninite–siliceous high-Mg basalt association with boninite less abundant than siliceous high-Mg basalts.

As Sun et al. and Srivastava recognized, the intracratonic boninite–siliceous high-Mg basalt association originated from source regions that were of limited depth (<100 km; Fig. 12), significantly depleted by a prior episode of partial melting (Figs. 9F and 10C), and subsequently refertilized by hydrous, siliceous melts of likely subduction origin (Fig. 11E). These characteristics match those of many intracratonic mantle xenoliths (e.g., Griffin et al., 1999; Pearson and Wittig, 2008), thus pointing to an intracratonic subcontinental lithospheric mantle magma source. However, the depleted, ca. 3.3 Ga Commondale-type komatiitic boninites have high-MgO parental magmas indicative of asthenospheric temperatures and depths and no discernible subduction components (Figs. 11D and 12A) and thus probably do not belong to the intracratonic boninite type. Thus, the oldest clear examples of intracratonic boninites, and hence the upper boundary of the time line, are the Mallina Basin boninites and siliceous high-Mg basalts at ca. 3 Ga (Smithies, 2002). The lower boundary of the time line is marked by the Bushveld B1 chilled margins at ca. 2 Ga (Barnes et al., 2010). There are a number of siliceous high-Mg basalts younger than this, but, as yet, no associated boninites have been found.

The explanation for this period of intensive boninite–siliceous high-Mg basalt magmatism is beyond the scope of this paper. Essentially, we concur with Srivastava (2008) in envisaging the 2–3 Ga time line as a period of craton accretion followed by the melting out of the more fusible (i.e., metasomatized) parts of the thickened cratonic lithosphere in response to plume activity. It is possible that a subset of the siliceous high-Mg basalt magmas originated by crustal contamination of komatiite magmas (Sun et al., 1989), but we did not find evidence to confirm that hypothesis.

One further question is why 2 Ga marks the young end of the boninite–siliceous high-Mg basalt time line. Factors in favor of boninite, rather than siliceous high-Mg basalt, genesis in an intracontinental setting include: (1) the extent of mantle depletion (more depletion reduces Ti8 as in Fig. 3); (2) the composition of the refertilization component (shallower components typically have lower Ti8, as in Fig. 4); (3) the mass fraction of the refertilization component (a greater mass fraction increases Ti in the source, though this is balanced by added fluids, which increase the degree of melting and so reduce Ti8); (4) the cratonic geothermal gradient at the time of refertilization (high gradients allow shallower, and hence more depleted, mantle to be refertilized); and (5) the temperature of impacting plumes and degree of extension or delamination of the lithosphere (higher values give more melting and hence lower Ti8). Thus, there may be more than one explanation for the end of the period of intracratonic boninite magmatism.

3–4 Ga: Principal Period of Plume-Sourced Boninite–Siliceous High-Mg Basalt and Low-Ti Basalt Magmatism

Although the time line for arc-basin boninites extends to the earliest Archean with the volcanic units in the Whundo and Nuvvuagittuq sequences, the 3–4 Ga period is best characterized by the abundance of plume-derived, low-Ti magmas. These include the plume-sourced boninite–siliceous high-Mg basalt associations (Barberton AUK type, BK type, and Commondale type), which are restricted to the period 3.0–3.5 Ga, and part of the low-Ti basalt time line that extends back to the Isua “boninite-like” lavas at ca. 3.7 Ga. Here, using the data from Sossi et al. (2016), we found that plume-sourced magmas form a compositional array in Ti8-Si8 space that extends from the komatiite through the siliceous high-Mg basalt to the boninite field. Significantly, the hottest mantle plumes (those most capable of melting sufficiently to produce magmas with high-Si and low-Ti characteristics) generated primary boninite and siliceous high-Mg basalt magmas. If the plume was depleted before partial melting (as is likely for the Commondale type of boninite), then the probability increases that boninites rather than siliceous high-Mg basalts would be the product. Tentative support for this link between plume temperature and boninite genesis is the fact that the 3.0–3.5 Ga time line for the plume-sourced boninite–siliceous high-Mg basalt association matches the period of peak mantle plume temperatures in some thermal models (e.g., Herzberg et al., 2010).

The corollary is that, when cooler plumes undergo partial melting, they produce primary magmas with lower Si concentrations than the hotter plumes, i.e., komatiites, picrites, or basalts. If these cooler plumes undergo depletion prior to partial melting in an intraplate setting, then low-Ti basalts will likely be the product (cf. Fig. 2C). Thus, a key inference from the Ti8-Si8 plots in Figures 10A and 10B is that a depleted, hot plume would more likely produce a siliceous high-Mg basalt–boninite association (e.g., BK type), whereas a depleted, cooler plume would more likely produce a basalt–low-Ti basalt association (e.g., Whitney type). Low-Ti basalt–boninite associations may also be produced if lithospheric thinning were to cause a depleted mantle plume to undergo additional cpx-undersaturated melting (Fig. 8C). The fact that the only recognized example of such an association is from the Phanerozoic, intraoceanic Manihiki Plateau raises the possibility that melting beneath thin oceanic lithosphere is a prerequisite for genesis of this boninite type.

  1. Proper recognition and classification of boninites using Si-Mg-Ti systematics are important precursors to tectonic interpretation. For a rock sequence to be boninitic, it must satisfy both the Si8 > 52 and Ti8 < 0.5 criteria. Rocks that satisfy only the first criterion may be termed “siliceous high-Mg basalts,” and those that satisfy only the second may be termed “low-Ti basalts.” The term “boninite-like” is misleading because many rocks from disparate tectonic settings fit this description.

  2. Our recommended plots for classification are MgO-SiO2 and MgO-TiO2, with the orientation of the former designed to match that of the IUGS total alkalis-silica (TAS) classification. As with related IUGS-linked classification methods, proper classification first requires that the composition be manipulated to be volatile free, with iron partitioned between FeII and FeIII oxides and then totaled to 100%. A subdivision of boninites into low-silica boninites and high-silica boninites with a boundary at Si8 = 57 also fits into the IUGS classification scheme. However, the classification scheme here deviates from the IUGS scheme in its boninite boundaries, which lie along mineral control lines rather than along lines in which Si and Ti are constants. It also includes a high-Mg andesite (HMA) field between the basaltic andesite-andesite-dacite-rhyolite (BADR) fields and the boninite field.

  3. In terms of petrogenesis, batch melting or pooled fractional melting in melting columns cannot normally generate compositions that satisfy both boninite criteria. However, incremental or fractional melting can create boninite melt fractions even without clinopyroxene exhaustion. As is well known, however, melting of an already depleted source (two-stage melting) is more effective for boninite genesis. The type of boninite produced is then a function of the relative contributions of cpx-saturated and cpx-undersaturated melt, with high-Si boninites requiring the latter to be predominant. When present, siliceous slab melts may also help to increase Si8. Crustal assimilation can generate boninites from picrites and komatiites, but only if the crust is near its solidus and the melt starts with sufficiently high MgO and SiO2 and sufficiently low TiO2.

  4. For highly altered rocks, Cr is a useful proxy for MgO, and Cr-Ti is arguably more effective than MgO-TiO2 for boninite interpretation. However, there is no universal immobile proxy for silica, so careful data filtering is needed using a Cr-SiO2 diagram and/or other approaches. Significant numbers of “boninites” in the geologic record have been misclassified as a result of alteration, especially silicification.

  5. At the present-day, and likely through most of the Phanerozoic and Proterozoic, only a very small proportion of rocks from ocean-ridge, back-arc, collision, intraplate, or volcanic arc settings classifies as “boninite.” Thus, the “boninite” rock type has been dominated during the last 2 b.y. of Earth history by boninites associated with Izu-Bonin-Mariana–type subduction initiation, where an initial spreading event created a reservoir of depleted mantle that was subsequently fluxed by fluids and melts from the newly subducting oceanic crust. Forearc rifts and RTF triple junctions at slab edges are less common but are also dominated by boninites. Thus, oceanic forearc terranes and their ancient (commonly ophiolite-related) equivalents continue to be the predominant boninite localities in the Phanerozoic and Proterozoic rock record.

  6. Although Archean boninites have commonly been used to locate Archean subduction zones, this study reveals that many of these are either misclassified, are the product of silicification, or were misinterpreted in terms of tectonic setting. Most likely, Archean boninites formed in a number of tectonic settings. Results indicate that the 2–3 Ga period was marked by boninite–siliceous high-Mg basalt associations formed by melting of refertilized, depleted cratonic mantle in intracratonic rift or rift-edge settings. The 3–4 Ga period, in contrast, was characterized by boninite–siliceous high-Mg basalt associations, and low-Ti basalt magmas, derived from melting of depleted mantle plumes.

  7. Despite our downgrading of the commonly perceived importance of subduction in Archean boninite genesis, two boninite types do support Archean subduction in some form: arc-basin boninites, which resulted from active subduction events, and intraplate subcontinental lithospheric mantle–sourced boninites, which carry subduction components likely inherited from craton accretion. Whether subduction was episodic or continuous (e.g., van Hunen and Moyen, 2012), and whether subduction was linked to a plate-tectonic planet (see discussions in Condie and Pease, 2008) cannot be resolved by this study, however. Significantly perhaps, the high-Si boninites and cpx-undersaturated melting trends that best characterize Izu-Bonin-Mariana–type subduction initiation boninites have yet to be found in the Archean rock record.

We are grateful to the Ocean Drilling Program Leg 125 and International Ocean Discovery Program (IODP) Expedition 352 science teams, in particular, Richard Arculus, Sieger van der Laan, John Shervais, and Julie Prytulak, for the many discussions about boninite genesis and classification. Sarah Lambart kindly provided advice on pyroxenite melting. Pearce and Reagan acknowledge funding through Natural Environment Research Council (NERC, UK) grant NE/M012034/11485 and U.S. National Science Foundation grant OCE0840862, respectively. We thank the referees (Steven Barnes, Jean Bédard, and Gene Yogodzinski) for their many helpful comments, and Shanaka de Silva and Bob Stern for their editorial contributions.

1Supplemental Files. Supplemental File 1 (.pdf document): Geological settings and SiO2-MgO-TiO2 classification plots for the “testing set” of Anthropocene to Eoarchean low-Ti lavas and dikes. Supplemental File 2 (Excel spreadsheet): Templates for plotting the boninite classification diagrams developed in this paper. Please visit or access the full-text article on to view the Supplemental Files.
Science Editor: Shanaka de Silva
Guest Associate Editor: Robert J. Stern
Gold Open Access: This paper is published under the terms of the CC-BY-NC license.