The garnet signature in the rare earth element (REE) abundances in adakites has been considered a key genetic indicator of these controversial rocks, whose proposed origins include direct melting of subducted oceanic crust (“slab melts”). We show that the garnet signature may be quantified using the shape coefficients of chondrite-normalized REE patterns. We applied this method to a global data set of Cenozoic and Quaternary volcanic samples described as “adakites.” The results indicate that many, but not all, suites of rocks labeled as adakites have undergone fractional crystallization of garnet, starting from parental melts attributable to partial melts of garnet-bearing sources. The extreme garnet signatures seen in many examples require hybrid sources, consisting of subducted sediment as well as igneous oceanic crust; however, extensive deep-crustal differentiation obscures the major and trace-element characteristics of these sources, casting doubt on their identification as primitive slab melts.

Adakites were originally proposed as a type of intermediate- to high-silica (≥56 wt%) volcanic/plutonic rock with high Sr, low Y (≤18 ppm) and Yb (≤1.9 ppm), and high La/Yb, derived from melting of subducted oceanic crust (OC) (Defant and Drummond, 1990). These are arguably the only igneous rocks whose definition rests entirely on trace-element characteristics. The trace-element characteristics that distinguish them from normal arc andesites, dacites, or rhyolites (ADRs; Defant and Drummond, 1990; Hastie, 2021) have been explained by garnet in the source (Rapp et al., 1991; Wolf and Wyllie, 1994), thus establishing pressure-temperature-composition constraints consistent with slab melting. Subsequent debate has focused mostly on whether their defining geochemical characteristics result from partial melting of subducted slab (Kay, 1978; Defant and Drummond, 1990; Yogodzinski and Kelemen, 1998; Castillo, 2012) or lower continental crust (LCC; Kay et al., 1991; Rudnick, 1995; Chung et al., 2003;), both with garnet residue, although a dissenting view is that fractional crystallization of minerals such as garnet or amphibole from basaltic magma causes such signatures (Müntener et al., 2001; Richards and Kerrich, 2007; Rodríguez, et al. 2007; Chiaradia et al., 2012). Here, we recommend that the term adakite should be reserved for suites of magmas that show a double garnet signature (i.e., in the source and during subsequent evolution). Such characteristics are likely only achievable by melting of a hybrid source consisting of both the igneous and sedimentary components of subducted OC.

The lack of consensus on the petrogenesis of adakites may be ascribed to two problems. First, the trace-element characteristics of an igneous rock are a product of the trace-element concentrations of its source as much as of subsequent petrogenetic processes. Conventional graphical methods of displaying trace-element data are ill-suited for addressing this complication, because they implicitly assume a particular source composition consistent with the model promoted. The second problem is that the defining characteristics of high Sr/Y and low La/Yb are not unique to the melting of garnet-bearing sources. Not only does Sr/Y vary considerably among common lithologies in both oceanic and continental crust, but amphibole fractionation also elevates Sr/Y (Rodríguez, et al., 2007; Müntener et al., 2001). Partial melting of sources containing pyroxene or amphibole result in elevated La/Yb irrespective of the presence or absence of garnet, since light rare earth elements (LREEs) are more incompatible than heavy REEs (HREEs) in these and most other rock-forming minerals other than plagioclase. The most distinctive petrogenetic signature of garnet is fractionation among HREEs. This signature stems from the compatibility of the HREEs in garnet (garnet/melt partition coefficients, graphic, >1) coupled with the large change in graphic among the HREEs. Although this signature is readily observable, there remains the problem of interpreting how much of a REE pattern in an igneous rock is due to the mechanisms of their petrogenesis, and how much is inherited from the REE patterns in the source.

An easily implemented method for comparing large numbers of REEs while also relating them to possible sources is to fit the chondrite-normalized patterns as a function of REE ionic radii to orthogonal polynomials:


with the terms of the polynomials, called “shape coefficients” (SCs) capturing the information in the patterns quantitatively (O’Neill, 2016). Convenient programs to calculate and plot SCs are available (Anenburg, 2020; Anenburg and Williams, 2021). The SCs suggest petrogenetic models and discriminate between hypotheses, including the ability to distinguish garnet signatures due to partial melting from garnet crystallization (Miller et al., 2022). This approach lends itself to describing large data sets, illuminating the relationships between different localities, and revealing aspects of petrogenesis not evident from studies of single localities. We applied this approach to a global adakite database (see our methods in the Supplemental Material1 and the adakite database in Table S2 therein) in order to clarify the interplay between garnet crystallization and source compositions in generating the REE patterns observed in these rocks.

The samples identified as “adakite” in the global database (Fig. 1) have a larger range of silica (52–72 wt%), MgO (0.07–14 wt%), and Al2O3 (13–19 wt%) than the range originally proposed as adakite (SiO2 ≥ 56%, Al2O3 ≥ 15%, MgO ≤ 3%) by Defant and Drummond (1990), with only about half the data within this range (Fig. 2). Likewise, samples have a large range of Sr/Y and La/Yb, including values <20, in the ADR field (Figs. 2B and 2C). The ages of “adakites” in the global database vary from Archean to Cenozoic; however, we only discuss the Cenozoic samples. Localities are shown in Figure 1, with Figure 2 defining major- and trace-element characteristics. The REE pattern shapes of filtered data (Table S1) are plotted (Fig. 3) by locality, to better display the diversity in their REE patterns. For comparison, we’ve plotted the ocean floor basalt (OFB; Jenner and O’Neill, 2012) array. All adakite groups plot to higher λ1 (steeper slope) than most of the OFB, with the quadratic curvature (λ2) similar to or higher than enriched OFBs (higher λ1 and λ2). Given geologically feasible sources, such patterns are qualitatively consistent with garnet signatures, the origin of which will be examined herein.

Partial Melting

We tested the hypothesis that the REE patterns of adakites are due to partial melting of garnet-bearing sources. To first order, the concentration of a trace element in a partial melt ([M]) is related to that in its source ([M]o) approximately by the batch melting equation with constant crystal/melt partition coefficients (Shaw, 2006; O’Neill, 2016; Table S1). We compared the calculated shapes of REE patterns produced by these partial melting scenarios from the hypothesized sources to the global database (Fig. 3). Adakites of island arcs (Figs. 3A and 3B) were plotted with partial melting modeling starting from oceanic crust (OC) sources (upper OC [UOC]; Niu and O’Hara, 2009). Without garnet in the source, the petrogenetic process vectors (PPVs; Table S1) for partial melting shift patterns to higher λ2 with higher λ1, producing PPVs that parallel the OFB array and do not produce the adakite pattern shapes. Although partial melting with garnet in the source does produce a shift to steeper slopes without a large increase in curvature, its effect is insufficient to account for most adakite REE patterns, in both island arcs and continental arcs: the majority of adakites plot well to the right-hand side of the partial melting PPVs starting from plausible source compositions, especially for OC sources, and even for sources with continental crust garnet patterns (Fig. 3E and 3F). Note that small degrees of partial melting produce REE pattern shapes that are almost identical, whether garnet or amphibole holds REEs in the source (Figs. 3A and 3B). Contrary to previous conclusions, partial melting of garnet-bearing sources cannot achieve the strong garnet signatures seen in most adakites. An additional petrogenetic process is required.

Simple Fractional Crystallization

Following partial melting, magmas evolve by crystallization, the common adduced mechanism being simple fractional crystallization (SFX), as described by the Rayleigh equation (Shaw, 2006; Table S1). Calculated PPVs for SFX of single-phase amphibole and garnet are shown in Figure 3. Clinopyroxene PPVs plot in the same direction as amphibole, but are weaker (O’Neill, 2016). Even though SFX of amphibole, clinopyroxene, or garnet all increases λ1 (i.e., LREE enrichment), the effect of amphibole or clinopyroxene on the curvature (λ2) is almost orthogonal to that of garnet, making it easy to distinguish garnet crystallization from that of amphibole or clinopyroxene. The PPVs of garnet are an order of magnitude stronger than those of amphibole for a given mx (mass fraction of crystalline phase x in the source), and only modest amounts of garnet fractionation shift REE pattern shapes in the appropriate direction. Specifically, crystallization of ~1–5% garnet is sufficient to produce the observed pattern shapes (e.g., Figs. 3D3F), starting from 5%–20% partial melting of lower continental crust (LCC) (Rudnick and Gao, 2003), with higher amounts of garnet fractionation required for sources of UOC. Further evidence for garnet fractional crystallization comes from the trends seen in some individual suites with a series of differentiated samples, such as from the Honshu arc (Fig. 3B) and the Central American volcanic arc (Fig. 3D). Evolution along the garnet crystallization trend is not accompanied by increasing Mg# (defined as molar Mg/ΣFe) as shown in Figure 4A.

Mixed Sources

Assimilation of marine sediments has also been proposed to explain the geochemistry of adakites (Castillo, 2012). Possible effects on adakite REE patterns may be assessed using modern global subducting sediment (GLOSS-II; Plank, 2014) as an example, although modern sediments differ considerably from those subducted in the past, especially in their biogenic components (e.g., pelagic marine carbonates are post-Paleozoic). Because the orthogonal polynomials are based on logarithms of REE concentrations, mixing curves are not linear on a λ2 versus λ1 diagram (O’Neill, 2016). Adding GLOSS produces trends similar to amphibole control, providing an alternative mechanism to produce a high λ1, λ2 parental magma prior to garnet fractional crystallization, without obviating the need for the latter. Compared with garnet and amphibole PVVs, various sources (e.g., GLOSS) impact more on λ1, the slope of the REE pattern shapes.

The Effects of Accessory Minerals on REE Pattern Shapes

Accessory minerals involved in adakite petrogenesis with high REE concentrations include zircon, titanite, and apatite (Castillo, 2012). The PPVs for the crystallization of titanite and zircon are shown in Figure 4A, starting from the source composition of Hidalgo et al., (2011). The vectors for titanite and apatite fractionation are in the opposite direction of garnet, but zircon moves the melt REE patterns in a similar direction to garnet, calculated using values of graphic from Boehnke et al. (2013). By assuming that all the zirconium goes into zircon rather than other phases on the cotectic of a crystallizing adakitic melt, the maximum extent of zircon crystallization can be estimated as <0.016 wt% (Table S2), which would have an insignificant influence on REE pattern shapes.

The hypothesis that crystallization of garnet may explain the HREE depletions often encountered in more-evolved convergent margin magmas can be traced to Green and Ringwood (1968) and Green (1972), who showed experimentally that garnet was a cotectic phase on a variety of natural-analogue compositions at high pressure. Subsequent experiments (e.g., Green, 1992; Müntener et al., 2001; Alonso-Perez et al., 2009) have confirmed this finding. More recently, Lee and Tang (2020) pointed out that garnet crystallization would simultaneously account for both the Fe depletion of the calc-alkaline differentiation trend characteristic of convergent margin magmatism and the increase in redox state of magmas along this trend. The trend of garnet crystallization (Fig. 4A) does not couple with increasing Mg#, indicating that garnet crystallization may not dominate Fe depletion of convergent margin magmatism. Alternatively, Holycross and Cottrell (2023) showed that garnet crystallization does not drive oxidation at arcs. While strongly supporting the garnet crystallization hypothesis in many cases, our assessment of REE patterns also shows that a non-negligible portion of the magmas labeled as “adakites” do not show pronounced garnet signatures (Figs. 3 and 4). Generalized conclusions on “adakite” petrogenesis based on one or a few suites as examples are clearly unwarranted. The distinction between adakites and the non-adakitic ADR varieties of convergent margin magmatism may be due to their deep crustal evolution, differences in sources notwithstanding. Garnet on the liquidus of upwelling magma strongly implies garnet as a residual phase in the source, and the garnet signature in erupted samples is likely a mix of these two processes.

Regarding sources, the observed REE patterns referenced to UOC source compositions (Fig. 3) show that many adakites require more garnet crystallization and low degrees of partial melting of the source than if the source were LCC. The average composition of the igneous OC, being more depleted in LREEs, would make an even less likely source composition. The importance of subducted sediment in convergent margin magmatism has long been recognized from correlations in the geographical variation of radiogenic isotopes between subducted sediment and erupted magma (e.g., White and Dupré, 1986). Figure 4B shows that the combination of subducted sediments with UOC produces a source with REE characteristics identical to the LCC, showing that the source composition for adakite petrogenesis cannot be identified with any certainty from REE patterns alone. If the term “adakite” is to be of any value, it should be restricted to those rocks from convergent margin settings showing evidence of garnet fractional crystallization, without requiring slab melting (Castillo, 2012). It is evident in the global adakite database that the labeling of many samples as “adakites” has been misleading, as they do not show any garnet signature. Consequently, they do not share the same petrogenetic history, and lumping them together with garnet-signature adakites obscures any relationships that these special rocks may have with tectonic environment or geological time. One reason for this past confusion is the inadequacy of simple ratios such as Sr/Y and La/Yb as diagnostic tools, use of which is fraught with danger. For example, the Sr/Y ratio in rocks with cumulate or xenocrystic plagioclase can reach high levels not representative of the true melt composition. The use of REE shape coefficients will reduce ambiguity in defining “adakites.” It provides a simple method to distinguish between the normal ADR magmatism at convergent margins, which mostly show no garnet signature, and the special cases, worthy of note, that show not only a garnet signature but one due to garnet fractional crystallization. Generating these extreme garnet signatures requires partial melting of a source composed of both igneous and sedimentary components of oceanic crust recrystallized in the garnet field.

1Supplemental Material. Methods, Table S1 (partial melting models and REE partition coefficients used in the calculations), and Table S2 (adakite database). Please visit to access the supplemental material; contact with any questions.

This study was supported by a China Scholarship Council and an Australian National University joint scholarship award to Y. Gao. We thank R. Arculus, A. Burnham, I. Campbell, and R. Chandler for discussions during the creation of this paper. We thank three anonymous reviewers and R. Holdsworth for his editorial handling.

Gold Open Access: This paper is published under the terms of the CC-BY license.