Introduction

Subduction initiation (SI) remains a poorly understood phenomenon within the framework of plate tectonic theory1. The Izu–Bonin–Mariana (IBM) arc (Fig. 1), located in the western Pacific, provides a unique natural setting for investigating SI2. Westward subduction of the Pacific plate beneath the proto-Philippine Sea Plate started around 52 Ma3,4, and was accompanied by several major tectonic events occurring in close proximity to the nascent IBM arc. Seafloor spreading in the Western Philippine Basin (WPB) is estimated to have occurred between 55 and 35 Ma5 along the now-extinct spreading center known as the Central Basin Fault Rift (CBFR). This process may have resulted in the splitting of the proto-Philippine Sea Plate, a Mesozoic arc terrane, into northern and southern parts6. At approximately the same time, the Oki-Daito plume crossed beneath the CBFR, forming three rises consisting of Ocean Island Basalts (OIB) with EM2 isotopic signatures between 50 and 36 Ma: the Oki-Daito Rise, Urdaneta Rise, and Benham Rise7,8. The reconstructed Middle Eocene location of the WPB lies close to the present-day Manus Basin where petrological and geochemical evidence indicate the involvement of a mantle plume9. From 45 to 41 Ma, the Mesozoic remnant arc in the northern WPB was split into three terranes: the Amami Plateau, Daito Ridge, and Oki-Daito Ridge6,10. Around 25 Ma, back-arc spreading formed the Shikoku and Parece Vela basins, which separated the present-day WPB and the Palau–Kyushu Ridge from the IBM arc11. Past studies have proposed different models for SI in the IBM region, including spontaneous subduction of the older Pacific plate beneath the younger Proto-Philippine Sea oceanic plate12,13, Izanagi-Pacific ridge subduction beneath Eurasia altering Pacific plate motion to induce SI along the transform fault14,15,16,17, Pacific plate subduction beneath a Mesozoic terrane6,18, and a mantle plume beneath a Mesozoic ridge and basin terrane as a likely trigger for the Pacific plate to begin to subduct beneath that terrane2,5.

Fig. 1: Regional map of the Izu–Bonin-Mariana subduction system with sampling locations.
Fig. 1: Regional map of the Izu–Bonin-Mariana subduction system with sampling locations.
Full size image

A, B Locations of the International Ocean Discovery Program Expedition 352 drill sites in the Izu–Bonin forearc. The positions of the IODP Expedition 352 drill sites (U1440-1441) in the Izu–Bonin forearc, along with other ocean drilling sites in the West Philippine Basin (U1438, 446, 1201, 447, 291, and 292), are indicated by white and yellow cycles and white triangles. The locations of the West Philippine Basin (WPB), Kita-Daito Basin (KDB), Amami Sankaku Basin (ASB), Minami-Daito Basin (MDB), and Central Basin Fault Rift (CBFR) are labeled. C Simplified lithologic columns for the forearc basalt holes (U1440A & U1440B and U1441A). The depths of the samples examined in this study, as well as those from Li et al.20, are plotted alongside the lithologic columns. The depths of the most enriched and most depleted samples are highlighted. mbsf: meters below the seafloor.

Forearc basalts (FAB) represent the earliest magmatism generated during IBM SI3,19. The source characteristics and melting mechanisms of the FAB should retain key information about SI. While FABs are broadly similar to mid-ocean ridge basalts (MORB) in their major and trace element compositions, indicating that their mantle source was largely unaffected by slab-derived fluids or melts20, they also exhibit several distinctive geochemical characteristics. These include lower TiO2 and incompatible element concentrations, as well as reduced Ce/Yb and Ti/V ratios compared to MORB21. Trace element modeling indicates that high-degree melting of a MORB mantle source cannot produce these low Ce/Yb signatures. Instead, an unknown prior melt extraction event must have occurred in the FAB source, followed by high-degree melting (up to 23%), to account for their depleted signatures21. This unknown melt extraction event may hold key information related to the mechanisms of SI. Mesozoic (~159.4 Ma) basaltic pillow lavas have been found beneath Izu FAB-related gabbro layers3, revealing the existence of Mesozoic melting events in the nearby mantle. Proterozoic (0.82–1.23 Ga) melting events have also been identified, based on peridotites recovered during drilling of the IBM forearc, which show unradiogenic 187Os/188Os ratios ranging from 0.1193 to 0.127322. Whether these recent or ancient melting events have formed a refractory mantle source, laying the foundation for the formation of FABs, remains uncertain. A series of drilling, dredging, and deep-sea sampling efforts has provided the basis for detailed investigations of the WPB mantle and its volcanic products. FAB lavas and dikes, located near the Izu–Bonin trench and dating back to 51.9–51.3 Ma4, were recovered from Sites U1440 and U1441 during International Ocean Discovery Program (IODP) Expedition 352 (Fig. 1). Similar FAB samples were recovered by the Japanese submersible Shinkai 6500 diving along the inner trench slope of the Izu–Bonin arc23. Site U1438 recovered basalts that formed shortly after SI (49.3–46.8 Ma) but were distal from the Proto-IBM, in the Amami Sankaku basin in the northern WPB near the Palau–Kyushu Ridge24,25. Drilling at Sites 1201, 447, and 291 recovered MORB erupted at 47 Ma, 44 Ma, and 43 Ma, respectively. Before the spreading of the WPB and Shikoku and Parece Vela back-arc basins, these sites were in close proximity to each other. Therefore, there may be some commonalities in their mantle composition and evolution. In addition, drilling at Sites 446 and 292 both recovered OIBs from the Minami-Daito Basin and Benham Rise7,25,26.

This study investigates the Nd-Hf isotopes and trace elements of IODP Expedition 352 FAB, with a particular focus on the poorly studied latest eruptions (Supplementary Data 1 and Supplementary Methods). While previous studies have primarily focused on tracing the earliest slab–mantle interaction processes during SI, such as whether the mantle source of FAB incorporated contributions from the Pacific slab19,20,21,27, this study shifts focus to the characteristics of the FAB mantle, its formation mechanisms, and the processes behind its melting. Site U1441 of Expedition 352 primarily recovered lavas, while U1440 also penetrated sheeted dikes21. The lava sequences at Site U1440 (Holes U1440A and U1440B) were categorized into lower and upper units with the break occurring ~166 meters below the seafloor (mbsf; Fig. 1C). The Lower FAB (L-FAB) from Site U1440 exhibit compositional similarities to the underlying dikes, while the Upper FAB (U-FAB) are compositionally different20,21. Site U1441 (Hole U1441A) samples of this study are also classified as L-FAB, and are lumped with U1440 L-FAB in our subsequent discussion. These data are compared to basement basalts from the WPB, Izu trench wall FAB and Mesozoic basalts, and OIBs (Oki-Daito plume). The results provide compelling evidence for the melting of young and ancient refractory mantle during FAB generation, triggered by rollback of the sinking Pacific slab.

Results

Unique trace elements and Hf–Nd isotopes for the FAB

The primitive-mantle–normalized trace element patterns of the FABs show stronger depletions in incompatible trace elements than those of average depleted MORB28 (Fig. 2A). Among the L-FAB samples, the most enriched and most depleted compositions constrain the range of incompatible trace element depletion. The depletion level of U-FAB falls between that of the most enriched and depleted L-FABs. The Nd and Yb concentrations in FAB are lower compared to those in most MORB (Fig. 2B, C). Though there is somewhat overlapping, the REE concentrations in U-FAB are generally lower than those in L-FAB, and the relative depletion in Nd is more pronounced. Both the FABs and other basalts from the WPB drill sites exhibit MORB-like Th/Yb versus Nb/Yb patterns, indicating minimal influence from the melting of the subducted slab, which has a high Th/Nb ratio (Fig. 3A). They also show low La/Yb and Dy/Yb ratios pointing to a more depleted mantle source than seen in MORBs (Fig. 3B).

Fig. 2: Trace elements for the forearc basalts.
Fig. 2: Trace elements for the forearc basalts.
Full size image

A Primitive mantle41 normalized trace element patterns for representative samples. B Nd versus Cr. C Yb versus Cr. The data for the U1438 basalts25, Sites 1201 and 447 MORB7,25,26, as well as average D-MORB28 and Pacific MORB42 are included for comparison.

Fig. 3: Trace elemental ratios for the forearc basalts.
Fig. 3: Trace elemental ratios for the forearc basalts.
Full size image

A Th/Yb vs. Nb/Yb. B Dy/Yb vs. La/Yb. The data from Izu trench wall FAB23,25, U1438 basalts25, Sites 447, 291 and 1201 MORB7,25,26, Oki Daito Plume, Izu forearc Mesozoic basalts3, and Pacific and Indian MORB42 are included for comparison.

The FAB samples show clear stratigraphic Nd-Hf isotope variations (Fig. 4). L-FAB samples have relatively uniform εNd (8.6–9.0) and εHf values (19.0–19.7), aside from our single most enriched L-FAB sample at ~320 mbsf, which shows lower values (εNd = 8.1, εHf = 17.8), and the most depleted L-FAB sample at ~180 mbsf, which displays higher values (εNd = 9.7, εHf = 20.4). The L-FAB samples generally exhibit higher εNd and lower εHf values compared to the U-FAB samples, with some limited overlap. The U-FAB samples display distinctive Nd-Hf isotopic variations across a comparatively narrow range of sampling depths. At 165–166 mbsf, εHf values increase from 19.3 to 22.1, with little change in εNd. Upsection, εNd values show a scattered increase, from 7.3 to 9.1, while the εHf values decrease from 21.6 to 19.2.

Fig. 4: Stratigraphic variations of Nd and Hf isotopes.
Fig. 4: Stratigraphic variations of Nd and Hf isotopes.
Full size image

A εNd. B εHf. mbsf: meters below the seafloor.

The FAB all plot above the εHf versus εNd terrestrial array29, along with most other WPB samples (Fig. 5), showing significantly higher εHf values at a given εNd as compared to Pacific Ocean MORB. The degree of deviation from the terrestrial array decreases from Expedition 352 FAB, to Site U1438 basalts, and then to WPB MORB. The εNd values of the U-FAB samples show a negative correlation with εHf. The FABs and Site U1438 basalts, excluding the most enriched L-FAB sample, deviate from the εNd versus Sm/Nd trend of MORBs, with relatively low εNd as compared to the most depleted MORBs (Figs. 6 and 7).

Fig. 5: Hf and Nd isotopic correlations for the forearc basalts.
Fig. 5: Hf and Nd isotopic correlations for the forearc basalts.
Full size image

The calculated Hf and Nd isotopic compositions of melting residues derived from the most enriched MORB mantle in the West Philippine Basin—preserved and evolved until ~50 Ma—are shown for reference. Melt extraction is assumed to have occurred in the spinel facies at 160 Ma (light blue solid line) and 1000 Ma (red solid line), respectively. The 160 Ma melting residue, subsequently re-enriched shortly after depletion by a small-degree melt (~0.4%) with trace-element characteristics similar to Site 291 MORB, is represented by the dark blue broken line. Numbers along the lines represent melting degrees (%). Mixing between Site 291 MORB and 5% re-melting products of the 1000 Ma and 160 Ma refractory mantle domains at 50 Ma is illustrated by the red (① and ②) and light blue (③ and ④) broken lines, respectively. The data from Izu trench wall FAB23,25, U1438 basalts25, Sites 447, 291 and 1201 MORB7,25,26, Oki Daito Plume (Sites 292 and 446 OIBs7,10,43), as well as Pacific and Indian MORB42, and Arctic Mid-Atlantic Ridge basalts35,44 are included for comparison.

Fig. 6: The effects of melt–rock reaction on the Lu–Hf and Sm–Nd elemental and isotopic systematics.
Fig. 6: The effects of melt–rock reaction on the Lu–Hf and Sm–Nd elemental and isotopic systematics.
Full size image

A, C εNd vs. Sm/Nd. B, D εHf vs. Lu/Hf. A and B illustrate the interaction between Site 291 MORB and the 160 Ma refractory mantle, while C and D depict the interaction with the 1000 Ma refractory mantle. The numbers represent the residual melt fractions. The data sources are identical to those used in Fig. 3.

Fig. 7: The effects of melt–melt mixing on the Sm–Nd and Lu–Hf elemental and isotopic systematics.
Fig. 7: The effects of melt–melt mixing on the Sm–Nd and Lu–Hf elemental and isotopic systematics.
Full size image

A, C εNd vs. Sm/Nd. B, D εHf vs. Lu/Hf. In panels (A) and (B), the blue pentagon represents the re-melting (F = 5%) of a refractory mantle formed by 10% prior melt depletion at 160 Ma. The dark blue dotted lines illustrate the mixing of Site 291 and Site 1201 MORB with melts (F = 5%) from the 160 Ma 10% melting restite. The light blue lines show the mixing of Site 291 and Site 1201 MORB with melts (F = 5%) from a refractory mantle that underwent 18% prior melt depletion at 160 Ma. In panels (C) and (D), the green lines depict the melting of the Site 291 MORB mantle. Dark red lines represent the mixing of Site 291 MORB with melts (F = 5%) from the refractory mantle produced by 18% prior melting at 1000 Ma. Light red broken lines represent the mixing of Site 291 mantle melts (F = 20%) with melts (F = 5%) from the refractory mantle produced by 18% prior melting at 1000 Ma. Numbers denote percentage. The data sources are identical to those used in Fig. 3.

Simple re-melting of pre-depleted mantle fails to account for the formation of FAB

The decoupling of Hf–Nd isotopes, characterized by extremely high εHf values, has been observed in abyssal peridotites and cratonic mantle peridotites. This pattern is typically argued to indicate the involvement of an ancient (>1.0 Ga) refractory mantle component30,31,32,33,34. This kind of Hf–Nd isotopic decoupling is rarely observed in basalts. The one exception is in MORBs from the Mohns and Knipovich Ridges along the Arctic Mid-Atlantic Ridge, where their Sr-Pb isotopes and trace element systematics support the re-melting of ancient refractory mantle34,35 (Fig. 5). To better understand the source signatures of the FABs, we modeled the effects of ancient and recent melting events on the residual mantle, focusing on Hf–Nd isotopes, as well as Sm/Nd and Lu/Hf ratios. Considering that prior melt extraction results in an increase in εNd in the residual mantle, and that other MORB generally have higher εNd values than FABs and ODP Site 291 basalts, our calculations assume an enriched MORB mantle in the WPB, with mineral compositions as estimated by Niu36, trace element compositions as outlined by Workman and Hart37, and Hf–Nd isotopes similar to those of the ODP Site 291 basalts. Mantle melting is considered to have occurred in the spinel facies, taking into account the variation in Yb concentrations between L-FAB and U-FAB (Fig. 2C). We followed the critical fractional melting model, assuming a mantle porosity of 2%. The detailed calculation parameters and methods are provided in the Supplementary Table 1. We selected 1.0 Ga as the age of an ancient mantle depletion event, inferred from the ¹⁸⁷Os/¹⁸⁸Os model ages (0.82–1.23 Ga) of forearc peridotites22, and 160 Ma as a more recent event, recorded by basalts (~159.4 Ma) beneath the Izu FAB-related gabbro layers3, to evaluate whether these melting episodes influenced the FAB mantle. The melting residues are assumed to have been transformed into lithospheric mantle, thereby allowing them to be preserved and to undergo radiogenic ingrowth until ~50 Ma. In general, earlier melting events or higher degrees of melting lead to elevated εNd and εHf values in the residual mantle (Supplementary Table 2). If melting had depleted this mantle at 160 Ma, then a ~10% melting event would produce a residual mantle with an Hf–Nd isotope signature similar to that of the most depleted L-FAB sample (Fig. 5). However, this more recent pre-depletion event would not generate a mantle source consistent with either the U-FAB or our most enriched L-FAB sample. Assuming this melt depletion occurred earlier, at 1.0 Ga, it is also impossible to produce a mantle source consistent with any of the FABs. Clearly, direct re-melting of a pure pre-depleted mantle source will not explain our FAB signatures.

Ancient refractory mantle that becomes re-enriched shortly after depletion by a small-degree melt will undergo a decrease in Sm/Nd with little effect on Lu/Hf, such that its 143Nd/144Nd ratio increases more slowly than 176Hf/177Hf. Consequently, after a period of radiogenic ingrowth, this re-enriched mantle will plot above the mantle array in εHf–εNd space31. This process can generate a mantle endmember with extremely low εNd and high εHf, similar to that inferred for the FAB. According to our calculations, a ~160 Ma ~18% melting residue that was re-enriched immediately after depletion by ~0.4% of a melt with trace-element characteristics similar to Site 291 MORB can evolve to match the Hf–Nd isotopic compositions of the FAB with the lowest εNd at 50 Ma. However, because this mantle source retains excessively high Lu/Hf ratios (~1.15), its subsequent re-melting cannot reproduce the Lu/Hf ratios observed in the FAB (0.27–0.42; Supplementary Fig. 1). If the melting and re-enrichment events occurred earlier, the resulting mantle would yield even higher εHf and εNd values. Therefore, simple re-melting of refractory mantle that was immediately re-enriched after depletion by a small-degree melt cannot account for the source of the FAB.

Melt–refractory mantle interaction cannot account for FAB formation

We further examined interactions between melts from the asthenospheric mantle and a refractory mantle at 50 Ma, as this can also lead to the decoupling of Hf–Nd isotopes. Since refractory mantle should have much higher Hf/Nd ratios than basaltic melt, the εHf of the reacted melt should quickly evolve toward that of peridotite, while εNd largely retains the characteristics of the melt itself30,31,32,33. We calculated the melts resulting from the reaction of Site 291 MORB with either the 160 Ma or 1.0 Ga refractory mantle. The calculations utilized the assimilation-fractional crystallization (AFC) model of DePaolo38, which has effectively explained the decoupling phenomenon of Hf–Nd isotopes in abyssal peridotites31,32,33. This model simulates a melt migrating through a peridotite column, modified by partial assimilation of peridotite and the crystallization of clinopyroxene, with a constant ratio of assimilated mass to crystallized mass of 0.99, similar to the parameters set in Sanfilippo et al.33. Our results indicate that while reactive exchange between the melt and 1.0 Ga refractory mantle can rapidly increase its εHf and εNd within the range of residual melt fractions between 95% and 85%, the Sm/Nd and Lu/Hf ratios become significantly lower than those of the original melt, rendering it incapable of explaining the elemental and isotopic composition of FAB (Fig. 6). Reactive exchange between melts and 160 Ma refractory mantle shows that the reacted melts undergo a rapid decrease in Sm/Nd and Lu/Hf ratios at high residual melt fractions (>90%), followed by a sharp increase in both ratios once the residual melt fraction drops below 90%. These reacted melts do not overlap into the FAB data region in the εNd versus Sm/Nd diagram (Fig. 6). Therefore, reactions between MORB melts and refractory lithospheric mantle will not explain the generation of FAB magmas.

Mixing of asthenosphere-derived melts with re-melting of old refractory lithospheric mantle contributed to FAB formation

We have also modeled an alternative scenario: the effects of mixing melts derived from different mantle sources. Since melts from an ancient refractory mantle will have much higher Hf/Nd ratios than MORB source melts, their incorporation should lead to a significant increase in the εHf of the resultant mixed melt, with comparatively little change in εNd35. In these calculations, we assume that this refractory mantle component underwent 5% melting during FAB generation, regardless of the extent of prior melting it may have experienced.

It appears that mixing MORB melts, ranging from more enriched (Site 291) to more depleted (Site 1201), with melts derived from a ~160 Ma refractory mantle that had previously undergone >10% melting can fully account for the observed Hf–Nd isotopic variability in the L-FAB. Specifically, mixing ~20% Site 1201 MORB with ~80% melts from the 160 Ma refractory mantle can reproduce the Lu–Hf and Sm–Nd elemental and isotopic systematics of the most depleted L-FAB (Fig. 7A, B). Notably, if the extent of asthenospheric melting was higher during SI, the resulting melts would exhibit elevated Lu/Hf and Sm/Nd ratios, thereby reducing the proportion of refractory mantle–derived melts required in the mass-balance calculations. This process can also explain the εHf–εNd correlation observed in the L-FAB (Fig. 5). However, this process cannot fully account for the U-FAB, as many of their εHf values extend beyond the predicted mixing range (Fig. 7A, B).

We further considered the potential contribution from ancient (~1.0 Ga) refractory mantle. To generate the lower εNd of the U-FAB, melting a more enriched MORB type mantle appears necessary. Mixing Site 291 MORB melts with melts from ~1.0 Ga refractory mantle explains the Hf–Nd isotopic correlation among U-FAB (Fig. 5), however, it does not explain their εHf versus Lu/Hf or εNd versus Sm/Nd systematics, as these mixed melts exhibit low Sm/Nd and Lu/Hf ratios (Fig. 7C, D). We have modeled the Sm/Nd and Lu/Hf systematics for a mantle source consistent with Site 291 MORB Hf–Nd isotopes that has experienced different degrees of melting at ~50 Ma (Fig. 7C, D). In our calculations, the source is assumed to possess trace-element compositions, mineral assemblages, and melting reactions comparable to those of an E-MORB mantle (see Supplementary Table 1). At melting degree >20%, this mantle source produces melts with elevated Sm/Nd and Lu/Hf ratios approaching those seen in the U-FAB. Mixing melts derived from this more enriched mantle source with melts (~30%) from an ancient (~1.0 Ga), depleted refractory mantle (i.e., mantle that has undergone 18% prior melt extraction) can generate the lowest εNd value seen in U-FAB. However, this process cannot systematically account for the genesis of both U-FAB and L-FAB, because in the εHf–Lu/Hf correlation the U-FAB requires a greater contribution from ancient refractory mantle relative to the L-FAB, whereas in the εNd–Sm/Nd correlation the U-FAB requires a smaller contribution from the ancient refractory mantle (Fig. 7C, D).

The L-FAB and U-FAB cannot be produced by mixing MORB with melts derived from a single, homogeneous refractory mantle source; they need to be considered separately. We propose that the re-melting of deep, young (~160 Ma) refractory lithospheric mantle contributed to the generation of the L-FAB. This process systematically accounts for why their Hf–Nd isotopes deviate more strongly from the terrestrial array than local MORBs, as well as for their anomalously depleted trace-element characteristics. In contrast, the U-FAB require an additional low-εNd component compared to the L-FAB. This component most likely originated from high-degree melting (>20%) of the asthenospheric mantle, followed by mixing with low-degree melts derived from shallow, ancient (~1 Ga), highly depleted lithospheric mantle. This process also explains why U-FAB have lower Nd and Yb concentrations at a given Cr content compared to L-FAB, as the shallow ancient lithospheric mantle had previously experienced very high degrees of melting (>18%) and is therefore much more depleted in these elements than the deep younger refractory lithospheric mantle, which had only undergone ~10% prior melting (Fig. 2). Ishizuka et al.3 identified ~159.4 Ma volcanic rocks in the deep Bonin forearc. These rocks are geochemically similar to U-FAB (Fig. 7A, C), exhibiting depleted trace element patterns and Pb-isotopic signatures with an Indian Ocean mantle fingerprint, with no clear evidence of subduction influence. This observation supports the presence of ancient refractory mantle in the forearc mantle domain, formed no later than ~160 Ma.

Discussion

Based on the εHf versus εNd systematics of the FAB (Fig. 5) and their εNd versus Sm/Nd correlations (Fig. 7A, C), this study finds that both younger (~160 Ma) and more ancient (~1.0 Ga) refractory mantle sources likely contributed to the formation of the FABs as well as the Site U1438 basalts. The younger refractory mantle component may have been confined to the area east of U1438, which was close to the forearc during SI. Our findings support the idea that the Izu–Bonin forearc mantle represents a long-lived arc lithospheric mantle formed via Proterozoic and Mesozoic melting events3,22. The stratigraphic evolution revealed in the Hf–Nd isotope systematics (Fig. 4) indicates that melting of the ancient ~1.0 Ga refractory lithospheric component was dominant only during the final formation stage of U-FAB. This may indicate that this ancient lithospheric mantle is preserved at shallower depths relative to the Mesozoic lithospheric mantle.

Refractory mantle should be more difficult to melt, and previous studies have only observed such melting along slow-spreading mid-ocean ridges35 and in mantle plume environments39. Melting in the IBM forearc reflects extensive lithospheric thinning due to rollback of the subducted Pacific slab (Fig. 8). During subduction initiation, the surrounding mantle may experience elevated temperatures due to the influence of Oki-Daito or Manus mantle plumes9,21. As the early subducting Pacific slab rolled back, pronounced lithospheric extension occurred, leading to intense decompression melting of the local asthenospheric mantle. With melting this locally upwelling asthenosphere rapidly becomes more refractory, increasing in its solidus temperature to match that of the refractory Mesozoic lithospheric mantle (Fig. 8D). Melting of both asthenospheric and Mesozoic lithospheric mantle under high-temperature conditions and with an extremely thin lithosphere generates the L-FAB. In its latest stages, the local asthenospheric mantle becomes even more refractory, allowing for melting of local ancient lithospheric mantle sources at shallow depths, which leads to the generation of the U-FAB.

Fig. 8: Cartoon illustrating the generation of the forearc basalts.
Fig. 8: Cartoon illustrating the generation of the forearc basalts.
Full size image

A Tectonic events occurred in close proximity to the nascent IBM arc. Around 52 Ma, prior to the initiation of IBM subduction, the Oki-Daito plume interacted with the proto-Philippine Sea Plate, a Mesozoic arc terrane with ancient lithospheric mantle component, potentially causing it to split along the Central Basin Fault (CBF). B Map view and C profile view of the SI process and FAB generation. During the earliest stages of IBM SI (52–50 Ma4), early rollback of the newly subducting Pacific slab triggers considerable lithospheric thinning within the overriding proto-Philippine Sea Plate, leading to intense decompression melting of the asthenospheric mantle before convective overturn in the mantle could establish. D Thermal profiles at ~52 Ma and during early SI at 52–50 Ma. This upwelling asthenospheric mantle rapidly became highly refractory, increasing its solidus temperatures. When solidus temperature in the asthenospheric mantle aligned with those in adjacent lithospheric mantle, melting occurred in both asthenospheric and lithospheric mantle domains under conditions of an extremely thin lithosphere, forming the L-FAB. In the latest stage, continued extensive melting in the asthenospheric mantle became coupled with melting in ancient refractory lithospheric mantle in the IBM forearc, generating U-FAB.

Methods

Calculations for Hf–Nd isotopes of the refractory mantle

The calculations for Hf–Nd isotopes of the refractory mantle involve several steps. First, we determine the Hf–Nd isotopes of a given mantle, such as the enriched MORB mantle, for a specific geological period, for example, 1.0 Ga. Next, we calculate the trace elements of the residual mantle using the parameters provided in Supplementary Table 1. This calculation is based on the critical fractional melting model, assuming a mantle porosity (retained melt) of 2%, and is constructed using the pooled fractional melting equations40:

$${{{{\rm{C}}}}}_{{{{\rm{L}}}}}/{{{{\rm{C}}}}}_{0}=(1/{{{\rm{F}}}})\ast [1-{\left(1-{{{\rm{PF}}}}/{{{{\rm{D}}}}}_{0}\right)}^{1/{{{\rm{P}}}}}]$$
(1)
$${{{{\rm{C}}}}}_{{{{\rm{S}}}}}/{{{{\rm{C}}}}}_{0}=[1/(1-{{{\rm{F}}}})]\ast {(1-{{{\rm{PF}}}}/{{{{\rm{D}}}}}_{0})}^{1/{{{\rm{P}}}}}$$
(2)

Where C0 = concentration of the element in the source; CL = concentration of the trace element in the melt fraction; CS = concentration of the trace element in the residual; D0 = bulk partition coefficient of the starting mineral assemblage; P = the bulk reaction coefficient, which is the sum of individual partition coefficients weighted according to the reaction coefficients; F = the fraction of solid melted. The retained melt is treated as a mineral phase with a partition coefficient of 1 and a reaction coefficient of 0.

Afterwards, we calculate the 147Sm/144Nd and 176Lu/177Hf ratios of the residual mantle based on its 143Nd/144Nd and 176Hf/177Hf ratios at the specified geological period, along with its trace elemental composition. Finally, in the last step, we calculate the 143Nd/144Nd and 176Hf/177Hf ratios of the residual mantle when it has evolved to 50 Ma.