Discovery of Ultra‐Depleted Melt Inclusion in Late Cretaceous Intracontinental Basaltic Andesites in South China: Implications for Recycling of Lower Oceanic Crust

Incompatible element ultra‐depleted melt (UDM) has occasionally been found as olivine‐hosted inclusion in plume‐related oceanic island basalts and mid‐oceanic ridge basalts and is genetically linked with the involvement of lower oceanic gabbroic crust. However, it has rarely been reported in intracontinental magmatism, and the contribution of such a recycled component to the origin of intracontinental magmas remains unclear. Here, we first report UDM inclusions hosted in olivine from Late Cretaceous intracontinental basaltic andesites in South China. These UDM inclusions are characterized by low SiO2 (46.4–47.9 wt%) and extremely high Al2O3 (18.47–21.79 wt%) contents, positive Sr‐Eu anomalies, and low 207Pb/206Pb and 208Pb/206Pb ratios. Such geochemical features resemble those of high‐pressure melts derived from young lower oceanic gabbro. These results, in combination with geochemical modeling, plate reconstruction, and seismic topography, suggest that tearing or fragmentation of the subducted paleo‐Pacific slab during Late Cretaceous may trigger melting of the refractory lower oceanic cumulates. Our results demonstrate that lower oceanic gabbroic crust could also play an important role in the generation of intracontinental magmatism, which may previously have been underestimated.

. The origin of UDMs has been ascribed to (a) the final stages of melting of a homogeneous upwelling mantle (Gurenko & Chaussidon, 1995;Neave & Namur, 2022;Portnyagin et al., 2009;Sobolev & Shimizu, 1993); (b) melting of an unrecognized refractory mantle (Jackson et al., 2015); (c) diffusive interaction between percolating melts and plagioclase-bearing cumulates in the lower oceanic crust (Danyushevsky et al., 2003;Kamenetsky et al., 1998;Peterson et al., 2014;Saal et al., 2007); (d) melting of recycled oceanic gabbros/mafic cumulates (Sobolev et al., 2000(Sobolev et al., , 2011. Although the genesis of UDMs remains highly debated, systematical geochemical analysis shows that most UDMs have relatively high Al 2 O 3 contents and Sr/Sr* (Sr/Sr* = Sr N /(Pr N × Nd N ) 0.5 , N denotes primitive mantle [PM] normalization) and Eu/Eu* (Eu/Eu* = EuCN/(Sm CN × Gd CN ) 0.5 , CN denotes chondrite normalization) ratios compared with MORBs and Hawaiian OIBs (Figures 1b and 1c), indicating a genetic link with the involvement of oceanic gabbroic crust either in the magma source or during magmatic differentiation. Because lower oceanic gabbro represents the ultramafic-mafic cumulate that is much more refractory and has much higher solidus than the upper oceanic basalt (Leuthold et al., 2018;Poli & Schmidt, 2002;Yaxley & Sobolev, 2007), its involvement either in the source of OIB or during the magmatic differentiation of MORB requires specific conditions such as high temperature and abundant plagioclase in source lithology. Nevertheless, UDM inclusion has rarely been reported in intracontinental basalts, and it remains highly debated about the contribution of lower oceanic gabbro to the evolution of intracontinental magmatism.
In the South China interior, Late Mesozoic intracontinental basalts are sporadically distributed (Zhou & Li, 2000). The genesis of these basalts has been ascribed to an enriched subcontinental lithospheric mantle (e.g., Chen et al., 2008; Y. J. Wang et al., 2008). However, experimental studies have suggested that melting of  (UDMs). Panel (a) shows the location of UDM found in previous studies and its geological setting (Danyushevsky et al., 2003;Gurenko & Chaussidon, 1995;Jackson et al., 2015;Kamenetsky et al., 1998;Kent et al., 2002;Laubier et al., 2012;Sobolev et al., 2000Sobolev et al., , 2011Sobolev & Shimizu, 1993). Panels (b and c) show the Al 2 O 3 content and Sr/Sr* and Eu/Eu* ratios of UDMs in previous studies. The data of MORBs and Hawaiian OIBs (collected from GEOROC database) and basaltic rocks from South China are also plotted for comparison (Data Set S1). The statistical analysis is calculated using the Bootstrap resampling method (see Supporting Information S1). Error bars denote two standard errors (SEs) of the mean. Panel (d) shows the distribution of Mesozoic magmatic rocks in South China. Inset figure shows the sampling location. an enriched lithospheric mantle generally produces high-K alkali basalts (e.g., Pilet et al., 2008), lithologically distinctive from sub-alkaline basaltic rocks in South China (Guo et al., 2021). Recent geochemical studies indicate a pyroxenite-dominated lithology in the melting source of Late Mesozoic basalts in South China, implying that the crustal components from the contemporaneously subducted paleo-Pacific slab may have been involved in the mantle source (Jia et al., 2020;Zeng et al., 2016). Besides, the geochemical signatures of both bulk-rocks and melt inclusions (MIs) of Late Cretaceous basalts in Ji'an area reveal the involvement of altered oceanic gabbros (Y. M. Wu et al., 2020). Yet, more robust evidence that confirms the contribution of such lower mafic crust to the source of Late Mesozoic intracontinental magmatism is lacking.
In this study, we conduct in-situ chemical and Pb isotopic analysis on olivine-hosted MI together with bulk-rock geochemical analysis on Late Cretaceous basaltic rocks in the South China interior. We for the first time identify olivine-hosted UDM inclusions that confirm the contribution of lower oceanic gabbroic component to their mantle source. Under the tectonic framework of the contemporaneous subduction of the paleo-Pacific slab, we then discuss the possible origin of this recycled oceanic component. Finally, we collect the geophysical data and conduct seismic tomography to search the torn paleo-Pacific oceanic slab in deep mantle, which may provide critical evidence for Late Cretaceous slab fragmentation.

Geological Setting and Samples
The South China Block comprises the Cathaysia Block to the southeast and the Yangtze Block to the northwest (Figure 1d), which were amalgamated along the Shaoxing-Jiangshan suture in Neoproterozoic (G. W. Zhang et al., 2013). During Late Mesozoic, subduction of the paleo-Pacific plate formed an Andean-type active continental margin along the eastern margin of the South China Block (e.g., Lapierre et al., 1997;Li & Li, 2007;B. Zhang et al., 2019). The subduction affected the tectonic evolution of South China, including the pervasive crustal deformation and widespread magmatism (e.g., Zhou & Li, 2000). It is generally considered that the paleo-Pacific slab subduction caused the transition of tectonic evolution from Middle-Late Jurassic compression to Cretaceous extension (e.g., J. H. Li et al., 2014;G. W. Zhang et al., 2013), which triggered the widespread Cretaceous extensional basins and magmatic activities (>250,000 km 2 , Zhou et al., 2006).
Most of the Cretaceous mafic igneous rocks are distributed along the coast of South China, and a few are distributed in the South China interior (Figure 1d). Systematic studies on petrology and geochemistry have shown that the Cretaceous mafic igneous rocks can be divided into two categories: (a) The Arc-type mafic rocks distributed along the coast of South China were formed in response to subduction of the paleo-Pacific slab (e.g., Xu et al., 1999); (b) The Cretaceous OIB-like basaltic rocks (except for the Wuyi and Jiangshan samples) are mostly distributed in the South China interior and have been regarded as intracontinental basalts (Guo et al., 2021). The temporal-spatial relationship between the OIB-like basalts and subduction of the paleo-Pacific slab makes it possible to track the involvement of recently subducted oceanic slab in the origin of intracontinental magmatism (Jia et al., 2020;Zeng et al., 2016).
In this study, fresh basaltic rocks were sampled from Guanshijie (N 26°43.5′, E112°53.8′) in Hengyang Basin (Figure 1d). High-precision 40 Ar/ 39 Ar dating shows that the Hengyang basaltic rocks erupted at about 71 Ma (Meng et al., 2012). These samples are gray-black in color and consist of euhedral to subhedral phenocrysts, mainly including olivine, clinopyroxene, and plagioclase ( Figure 2). Olivine phenocrysts are fresh and euhedral to subhedral in shape. Only partial serpentinization occurs along cracks within and on the margins of olivine. Abundant MIs are hosted in olivine phenocrysts and occur as rounded or elliptical shapes with a grain size of 30-100 μm (Figure 2c). The original MIs are mainly composed of glass with minor filamentous microcrystalline ( Figure 2b). Most heated (homogenized) MIs consist of a single glassy phase, expect for some heated MIs that contain bubbles.

Bulk-Rock Chemical and Sr-Nd-Pb-Hf Isotopic Analyses
We cut the samples into millimeter-size chips and, then, only selected fresh parts for further analyses. All analyses were conducted at the State Key Laboratory of Isotope Geochemistry, Guangzhou Institute of Geochemistry (GIG), Chinese Academy of Science (CAS). We determined bulk-rock major elements on fused glass disks 10.1029/2022JB025324 4 of 20 with a Rigaku RIX 2000 X-ray fluorescence spectrometer (XRF). We analyzed bulk-rock trace elements using a Perkin-Elmer ELAN 6000 inductively coupled plasma mass spectrometer (ICP-MS). The analytical error for the international standards GSR-3 and BHVO-2 are generally less than 5% for bulk-rock major and trace elements.

Compositional Analyses of Olivine and Melt Inclusion
We performed the heating experiment following Ren et al. (2017) to achieve the homogeneous composition of olivine-hosted MIs. Then, we chose the well-preserved MIs with no visible fractures for further analysis. Major element analyses and backscattered electron (BSE) imaging of minerals and olivine-hosted MIs were conducted using a JEOL JXA-8100 electron probe microanalyzer (EPMA) at GIG-CAS. Conditions of the EPMA for melt inclusion analysis include a beam diameter of 3 μm, a beam current of 20 nA, and an accelerating voltage of 15 kV. Conditions of the EPMA for olivine analysis include a beam diameter of 3 μm, a beam current of 300 nA, and an accelerating voltage of 20 kV. Calibration of elemental analyses was carried out using mineral standards from Structure Probe, Inc. (SPI) Company. During each batch analysis, we determined an internal olivine standard from Hannuoba mantle peridotite and an internal glass standard (JB-2) to monitor the instrumental drift. Analytical uncertainties of the olivine and glass standard are less than 4.9% and 5.7% for all oxides, respectively (Data Set S4).
We analyzed the trace element compositions of MIs using a CAMECA high-resolution secondary ion mass spectrometer (HR-SIMS-1280) following the procedure of Y. M. Wu et al. (2020). A current of 10 nA primary beam of − 2 ions was used to bombard the sample surface. The size of the ion spot on the sample surface was approximately 20 × 30 μm 2 . The secondary accelerating voltage was 10 kV, and the molecular interferences were filtered by an offset of 50 ± 10 eV. Each analysis took ca. 20 min and consisted of a 10-cycle running. We determined the elements including 28 Si, 30 175 Lu, 208 Pb, 232 Th, and 238 U. Then, we normalized the trace element compositions of MIs to 28 Si, which EPMA had already measured. We conducted the calibration through the values acquired by repeated analyses of standard BCR-2G, BHVO-2G, GSC-1G, GSD-1G, and BIR-1G. Repeated analyses show that the analytical errors of the monitor standards are generally less than 15% (Data Set S5).

Pb Isotope Compositions of Melt Inclusion
We performed in-situ Pb isotope analysis on MI through a Neptune plus MC-ICP-MS and RESOlution M-50 laser ablation system. The operating conditions include an energy of 80 MJ, a spot size of 45 μm, a repetition rate of 3 Hz, an attenuation value of 25%, and an integration time of 0.262 s (L. Zhang et al., 2014). We first cleaned the sample surface in an ultrasonic bath with dilute HNO 3 and Milli-Q water for 10-15 min. Then, we dried it with a nitrogen gas gun. We employed a large dry interface pump (100 m 3 h −1 pumping speed), a Jet sample cone and an X skimmer cone, and nitrogen gas with a flow rate of 2 ml/min to enhance the signal intensity of the Pb isotope (L. Zhang et al., 2014). We used standard-sample bracketing to correct mass bias and instrumental drift. Before analyzing, we measured the glass standards NKT-1G and BHVO-2G in turn to monitor the accuracy of the instrument. During each batch analysis, we detected the instrumental drift by measuring an internal glass standard (BHVO-2G). Because of the low relative abundance of 204 Pb and the isobaric interference from 204

Olivine Composition
We analyze the compositions of 234 olivine grains with euhedral texture (Data Set S2), of which 135 grains contain melt inclusion. The olivine phenocrysts have a narrow range of forsterite (Fo = 100 × Mg/(Mg + Fe) in atomic ratio) content from 70.3 to 75.0 (formulated as Mg 1.4-1.5 Fe 0.5-0.6 SiO 4 ). They have much higher CaO contents (0.170-0.303 wt%) than mantle olivine (CaO < 0.1 wt%; Thompson & Gibson, 2000), consistent with a magmatic origin (X. C. Wang et al., 2012). For given Ca and Ni contents, the olivine phenocrysts have lower Fo contents compared to olivines from primitive Hawaiian OIBs and MORBs, implying the effect of olivine and clinopyroxene fractionation (Figure 3).

Geochemistry of Olivine-Hosted Melt Inclusions
The compositions of olivine-hosted MI are corrected through the MIMiC program (Rasmussen et al., 2020), which employs the model of Toplis (2005) to calculate the Mg-Fe distribution coefficient and the olivine-melt thermometer of Putirka et al. (2007) to estimate the magma temperature. The detailed correction processes are described in Supporting Information S1. On the total alkali versus silica diagram (TAS) diagram, the studied melt inclusions are classified into basalt to andesite, plotting within the sub-alkaline series field (Figure 4), compositionally overlapping the range of bulk-rocks. The MIs with higher SiO 2 (49.2-60.1 wt%) are similar to the bulk-rock samples in terms of trace element patterns (Figures 5b and 5c), we hence term them as normaltype MI. The MIs with lower SiO 2 (46.4-47.9 wt%) have extremely higher Al 2 O 3 (18.47-21.79 wt%) contents and much lower contents of incompatible elements (Figures 5b and 5c), compositionally resembling UDM inclusions reported in previous studies (e.g., Jackson et al., 2015;Laubier et al., 2012;Sobolev et al., 2000Sobolev et al., , 2011, which show obviously positive Sr-Eu anomalies and LREE depletion relative to heavy rare earth elements (HREEs). The  Thompson and Gibson (2000). The black and gray fields represent the olivine compositions that crystallize from primary magmas of peridotite source and olivine-poor source (through 20%-40% addition of dacite melts), respectively (Herzberg et al., 2014). The trend of olivine compositions (black lines) that crystallize from the assumed peridotite and olivine-poor source-derived magmas is also shown. The olivine crystals from peridotite-sourced MORBs and pyroxenite-sourced Hawaiian OIBs are also plotted for comparison (Sobolev et al., 2005), respectively. UDM inclusions have much higher Nb/La ratios (1.6-3.7) than bulk-rocks (Nb/La = 1.2-1.3) and normal MIs (Nb/La = 1.2-1.6). The normaltype MIs show higher 207 Pb/ 206 Pb(i) (0.8529-0.8874) and 208 Pb/ 206 Pb(i) (2. 0839-2.1922) ratios similar to bulk-rock samples, whereas the UDM inclusions have distinctly lower 207 Pb/ 206 Pb(i) (0.8264-0.8489) and 208 Pb/ 206 Pb(i) (2.0011-2.1044) ratios, plotting within the field of Pacific MORBs and lower oceanic gabbros ( Figure 6).

Effects of Crustal Assimilation and Fractionation
The Hengyang basaltic andesites show positive Pb and K anomalies in the PM-normalized trace element spidergrams (Figure 5a), lower Nb/U (29-32) and Ce/Pb (5.1-8.4) ratios than that of oceanic basalts (47 ± 7 and 25 ± 5, respectively) (Hofmann, 1997), and enriched Sr-Nd-Hf isotopic compositions relative to MORBs (Data Set S3). Such geochemical characteristics could be ascribed to the input of crustal components into the melting source or crustal assimilation and fractional crystallization during magmatic evolution (e.g., L. Zhao et al., 2019). Thus, it is critical to determine the effects of crustal contamination and fractionation for these rocks before the source characteristics can be discussed.
While ascending, mantle-derived magma may experience contamination by continental crust. The upper continental crust (UCC) generally contains a lower Sr concentration but a higher 87 Sr/ 86 Sr(i) ratio than the Late Mesozoic basaltic rocks in South China (Rudnick & Gao, 2003). Thus, UCC contamination and/or assimilation will produce a negative correlation between Sr concentration and 87 Sr/ 86 Sr(i) ratios, which is inconsistent with the positive correlation of the Hengyang basaltic andesites (Figure 7a). Because continental crust has high SiO 2 and 87 Sr/ 86 Sr, and low MgO and ε Nd (Rudnick & Gao, 2003), a positive correlation of MgO versus ε Nd (t), and a negative correlation of MgO versus 87 Sr/ 86 Sr(i) would be predicted if crustal contamination played an essential role in their petrogenesis. However, the Hengyang basaltic andesites show nearly constant 87 Sr/ 86 Sr(i) and ε Nd (t) for a range of MgO (Figures 7b and 7c), inconsistent with the crustal contamination trends. Besides, the Hengyang basaltic andesites have much higher Nb/ La ratios (1.2-1.3) than that of continental crust (Nb/La < 0.5; Rudnick & Gao, 2003), suggesting an insignificant role of crustal assimilation. Finally, the 207 Pb/ 206 Pb(i) and 208 Pb/ 206 Pb(i) ratios of the bulk-rocks and MIs are much lower than that of lower continental crust (LCC; Figure 6), further suggesting that the LCC contamination is yet negligible.
The Hengyang basaltic andesites show negative correlations of MgO and CaO versus SiO 2 , suggesting the role of olivine and clinopyroxene fractionation (Figure 8). When the SiO 2 content is less than 54 wt%, the Al 2 O 3 positively correlates with SiO 2 , precluding a significant role of plagioclase fractionation in this stage (Figure 8c), which is also supported by the positive Sr anomaly and absence of negative Eu anomaly in the Hengyang bulk-rocks (Figure 5a). With SiO 2 content of more than 54 wt%, the rapid decrease of Al 2 O 3 content in the normal-type MIs indicates the presence of plagioclase crystallization during the stage. Accordingly, the magmatic evolution of Hengyang basaltic andesites was dominated by olivine + clinopyroxene fractionation, and the compositional variations in the normal-type MI were affected by olivine + clinopyroxene + plagioclase crystallization.  (Sun & McDonough, 1989); lower oceanic gabbro (VanTongeren et al., 2021); and Cretaceous basaltic rocks in South China (Guo et al., 2021).

The Source of UDM Inclusion: Lower Oceanic Gabbroic Crust
The remarkable difference in chemical and Pb isotopic compositions of the UDM inclusions and normal-type MIs in the Hengyang basaltic rocks suggest they were likely derived from different melting sources (Figure 9). The UDM inclusions have low SiO 2 and extremely high Al 2 O 3 contents (Figure 9), remarkably positive Sr-Eu anomalies, and extensive depletion in most incompatible elements ( Figure 5). Such geochemical features may be the consequence of either grain-scale dissolution-reaction-mixing or assimilation in the magmatic system (Danyushevsky et al., 2003(Danyushevsky et al., , 2004, or diffusive interaction between melts and plagioclase-rich cumulates in the lower crust (Maclennan, 2008;Peterson et al., 2014;Saal et al., 2007), or melting of recycled oceanic gabbros/mafic cumulates in the mantle source (Kent et al., 2002;Sobolev et al., 2011).
The grain-scale dissolution-reaction-mixing processes are generally observed in high-Fo (>85) olivine and may have extensive effects on the composition of MI trapped by olivine phenocryst that crystallizes from hot primitive mafic magmas in the early stage of magmatic evolution (Brahm et al., 2022;Danyushevsky et al., 2004). These processes were thus unlikely to have affected the composition of olivine-hosted MIs in the Hengyang samples (Fo < 80). Even so, the UDM inclusions may be generated by trapping plagioclase in a hot, strongly plagioclase-undersaturated magma or assimilation of gabbroic crust into the primitive magma (Laubier et al., 2012). Reactions between plagioclase (±clinopyroxene) and hot olivine result in the formation of aluminous spinel and low Na 2 O contents (<1.6 wt%) of the inclusions (Bédard & Hébert, 1998;Danyushevsky et al., 2003). Except for Eu, REEs are incompatible in plagioclase with much more incompatibility of HREEs than LREEs (Bédard, 2006;C. G. Sun et al., 2017), the assimilation of plagioclase can lead to a stronger decrease of HREE than LREE of the resultant melt (Laubier et al., 2012). This is inconsistent with the much higher concentrations of HREE than LREE in the Hengyang UDM inclusions (Figure 5c). Also, the BSE images show no remarkable difference in appearance between the Hengyang UDM inclusions and normal-type MIs, with no spinel found in the UDM inclusions ( Figure 2c). Finally, the Na 2 O contents (2.35-2.83 wt%) of Hengyang UDM inclusions are similar to the average values of bulk-rock samples and normal MIs, inconsistent with a significant contribution of calcic plagioclase assimilation that may lead to a Na 2 O decrease of the assimilated magma ( Figure 9d). All these observations indicate a minor effect of plagioclase assimilation.
The low contents of incompatible trace elements of the Hengyang UDMs make them susceptible to modification by melt-cumulate interaction. Thus, the "ghost plagioclase" signature of UDM inclusions might be produced by the diffusive interaction of MORB-like melts with a plagioclase-rich lower crust (Anderson et al., 2021;Peterson et al., 2014;Saal et al., 2007). A melt-cumulate diffusive interaction model that considers the approximate solution to the diffusion equations in a finite crystal-melt geometry was employed in this study to evaluate the effect of trace element diffusion (Hart, 1993;Saal et al., 2007). The detailed calculation procedure is described in Supporting Information S1. The input and output data of the model are listed in Table S3 in Supporting Information S1. The results show that diffusive interaction between a depleted melt derived from the depleted MORB mantle (DMM) source (Workman & Hart, 2005) and a plagioclase-rich cumulate can generate obvious positive Sr-Eu-Pb-K anomalies ( Figure S1 in Supporting Information S1). Due to the different diffusion coefficients of elements in plagioclase (e.g., Cherniak, 2003;Cherniak & Watson, 1994;Giletti & Casserly, 1994), there lacks a unified timescale to interpret the compositional variations among Sr, Eu, K, and Pb observed in the Hengyang UDM inclusions (Figure 5b). For example, the fast diffusion of Sr in plagioclase requires a short time (ca. 5 years) to create a positive Sr anomaly similar to the Hengyang UDM inclusions, whereas the timescale may exceed 20 years to create the positive Eu anomaly of the Hengyang UDM inclusions due to the slow diffusion of Eu ( Figure S1 in Supporting Information S1). Once the diffusion time is more than 20 years, the resultant melts have exceptionally positive Sr-Pb-K anomalies ( Figure S1 in Supporting Information S1). Such a prediction is inconsistent with the trace element features of Hengyang UDM inclusions. On the other hand, the DMM-derived melt usually has a Na/La ratio of ca. 0.8 (Workman & Hart, 2005), much lower than the Hengyang UDM inclusions. Since the diffusion coefficient of Nb is 2 orders of magnitude lower than La in plagioclase (Table S3 in  (Stracke et al., 2003), Mesozoic Pacific MORB (Hauff et al., 2003), lower oceanic gabbro from Pacific (Amórtegui et al., 2011), Atlantic (Barling et al., 1997) and SE India Ridge (Hart et al., 1999;Holm, 2002), SC LCC (Shen et al., 2014) and NC LCC (Qian et al., 2017 and references therein) are also plotted here. Note that the high 207 Pb/ 206 Pb and 208 Pb/ 206 Pb ratios of the SE India Ridge gabbro were likely attributed to recycling of lower continental crust (Escrig et al., 2004). Abbreviations: LCC, lower continental crust; NC, North China; SC, South China.
Supporting Information S1), diffusive interaction between a depleted melt and plagioclase will further reduce the Na/La ratio of the resultant melt ( Figure 10). This differs from the high Na/La ratio (ca. 1.6-3.7) observed in the Hengyang UDM inclusions (Figure 10). Furthermore, the diffusive interaction can change the Pb isotopic compositions of the assumed depleted melt due to the fast diffusion of Pb in plagioclase ( Figure 10d). As discussed above, the lower 207 Pb/ 206 Pb and 208 Pb/ 206 Pb ratios of the Hengyang UDM inclusions relative to the Hengyang normal-type MIs and LCC of South China argue against a significant effect of plagioclase-driven kinetic diffusion occurring in the lower crustal level. Finally, the geochemical variations among the Hengyang bulk rock, normal-type MI, and UDM inclusion are different from the plagioclase-diffusion-affected example, as evidenced by Galapagos Archipelago (Peterson et al., 2014), in which the different types of MI and bulk-rock samples show overlapping Pb isotopic compositions among them. In summary, the "ghost plagioclase" signature of Hengyang UDM inclusions might have been influenced by the plagioclase-melt diffusive interaction, but this influence was quite limited.
Alternatively, melting of recycled lower oceanic gabbro could produce the typical geochemical features of UDM inclusions (Kent et al., 2002;Sobolev et al., 2000Sobolev et al., , 2011. Geochemical data of the Southern Samail (Oman) ophiolite show that the lower oceanic gabbroic crust is significantly depleted in incompatible trace elements and exhibits distinctly positive Sr-Eu anomalies relative to the upper oceanic crust consisting of both basalt and gabbro (VanTongeren et al., 2021), which are compositionally consistent with the Hengyang UDM inclusions. We further compare the Hengyang UDM inclusions with those experimental melts derived from the mafic protoliths under low-(within the plagioclase stability field) and high-(within the garnet stability field) pressure conditions (Figures 11a and 11b). The results show that the melts derived from garnet clinopyroxenite and gabbro constitute different trends of Al 2 O 3 and CaO versus SiO 2 , respectively. The Hengyang UDM inclusions are plotted in or near the green fields defined by experimental melts of garnet clinopyroxenite and/or websterite (Figures 11a  and 11b). The broadly positive correlations of Al 2 O 3 and CaO versus SiO 2 of the Hengyang UDM inclusions are consistent with experimental melts in equilibrium with residual garnet. A simple batch melting calculation on the PM-normalized trace element patterns also suggests that the obviously positive Sr-Eu anomalies of Hengyang UDM inclusions can be formed in the garnet stability field (Figure 11c). Furthermore, melts derived from coesite-absent garnet clinopyroxenite have low SiO 2 (basaltic) contents (Y. Wang & Xu, 2020;Yaxley & Sobolev, 2007). Compared to the upper oceanic gabbros that represent the solidified rocks of differentiated mafic magmas, the lower oceanic gabbros are the ultramafic-mafic cumulates and hence have much lower SiO 2 contents (VanTongeren et al., 2021). Therefore, the unique chemical compositions of Hengyang UDM inclusions suggest an origin from the lower oceanic gabbroic crust that probably underwent eclogitization. Finally, the Hengyang UDM inclusions have MORB-type 208 Pb/ 206 Pb and 207 Pb/ 206 Pb ratios, indicating a juvenile nature of the source possibly related to the contemporaneous subducted paleo-Pacific slab.

The Normal-Type MI Source: Subducted Sediment-Metasomatized Mantle
The bulk-rock geochemistry of the Hengyang basaltic andesites shows some "crust-like" signatures (Figure 5a), such as the positive K and Pb anomalies in the PM-normalized spidergrams, and highly radiogenic Sr and chondritic Nd compositions, possibly derived from an enriched mantle with significant involvement of recycled sediments (B. Zhang et al., 2019). The statistical results show that 97% MIs (137 of 141) belong to the normal type, indicating that the bulk-rock geochemistry is mainly governed by these normal-type MIs. In contrast, the proportion of the UDM inclusion is very low and its unique geochemistry is easily lost in bulk-rock samples by mixing/ mingling processes during the residence time of single olivine crystal, as observed in the Hawaiian basaltic lavas (Sobolev et al., 2000(Sobolev et al., , 2011. The Hengyang basaltic andesites show elevated ε Hf (t) at a given ε Nd (t) (Figure 12a) and an unusual negative correlation of ε Hf (t) versus ε Nd (t) (Figure 12b). Such isotopic features are likely attributed to the input of recent pelagic sediments rather than old terrigenous sediments into the mantle source (Chauvel et al., 2008). However, these rocks also show distinctly low Lu/Hf (0.07-0.08) and high Th/La (0.21-0.22) and Th/Yb (2.26-2.58) ratios, which are similar to continental arc magmas with the input of dominant terrigenous sediments into the mantle source (Guo et al., 2021;. This paradox likely reflects that the transport and deposition time of the terrigenous sediments (e.g., eroded materials from a new arc crust) is too short to produce the obvious decoupling of Nd-Hf isotopes (L. Zhao et al., 2019). Moreover, the global subducting sediment is dominated by terrigenous sediments (∼76 wt%) and subordinate pelagic sediments (Plank, 2005;Plank & Langmuir, 1998). Therefore, the input of recent pelagic and terrigenous sediments in the mantle source of Hengyang basaltic andesites was likely related to subduction of the contemporaneous paleo-Pacific slab, which transferred the marine sediments into the South China interior to form an enriched mantle source (Figures 12c  and 12d).
In summary, two types of mantle sources have been identified based on the heterogeneous chemical and isotopic geochemistry of olivine-hosted MIs from the Hengyang basaltic andesites: (a) a lower oceanic gabbroic crust; (b) a subducted sediment-modified mantle. Mixing of the minor lower oceanic gabbro-originated and predominant enriched mantle-derived melts formed the Hengyang basaltic andesites.

Geodynamic Implications for the Paleo-Pacific Slab Subduction
How can the oceanic crust be involved in the mantle source of the South China interior, especially with the absence of a hotspot or mantle plume that can drive the upwelling of recycled oceanic crust from deep mantle? Melting of mafic oceanic crust requires an extremely hot young subducting slab (Peacock et al., 1994), which is We assume that the depleted melt is formed through 15% batch melting of the depleted MORB mantle (DMM) source (Workman & Hart, 2005), and the plagioclase-rich cumulate is crystallized from an enriched melt, which is estimated as the average compositions of the Cretaceous OIB-type basalts in South China (Guo et al., 2021). The Pb isotopic compositions of the depleted melt represent the average of the Pacific DMM (Stracke et al., 2003) and those Pb isotope ratios of the plagioclase-rich cumulate denote the average lower continental crust of South China (Shen et al., 2014). Pb/Pb* = PbN/(CeN × PrN) 0.5 , where N denotes normalization by primitive mantle (Sun & McDonough, 1989). The detailed description of the calculation procedure can be seen in Supporting Information S1. paradoxical to the old age (>70 Ma) of the subducting paleo-Pacific slab (Seton et al., 2012). Previous studies suggest that melting of the lower mafic oceanic crust requires intense thinning and/or necking of the subducted slab possibly in association with lithospheric detachment, slab breakoff and/or tearing (Guo et al., 2021;. Plate reconstructions (Figure 13a) show that South China was located at the junction realm between the Pacific-Panthalassa Ocean and the eastern Tethys Ocean during Cretaceous (Seton & Müller, 2008;J. Wu et al., 2022). The reconstructed southern limit of paleo-Pacific (i.e., Izanagi) subduction was located in eastern Eurasia between mid-Cretaceous and early Cenozoic times along the sinistral transform fault (Figure 13a), which divides the paleo-Pacific domain from another subduction realm to the south at ca. 30° near present Hangzhou, China (J. Wu et al., 2022) or to the further south near Fuzhou, China (Seton & Müller, 2008). The presence of a transform fault provides a suitable situation for slab tearing, especially during the rollback of the subducted slab (Cui & Li, 2022;Hale et al., 2010;Rosenbaum et al., 2008). In addition, recent studies suggest that pervasive slab weakening and pronounced segmentation in the outer-rise region can occur in the self-evolution of slab subduction due to brittle-ductile damage, which may cause randomly distributed slab fragmentation in active continental margins (Gerya et al., 2021).
Oceanic slab tearing or fragmentation has been widely documented by a series of tomographic studies beneath different sites in the Japan Sea and adjacent areas, such as the Changbaishan volcano in northeastern China (Tang et al., 2014), the Izu-Bonin region (D. Zhao et al., 2017), and the central Honshu, Japan to the Korean Peninsula (M. C. Sun et al., 2022). To better understand the dynamics of the paleo-Pacific slab subduction, we further conducted a tomographic analysis (Figures 13b and 13c)   . Panel (c) shows the comparison of the trace elements (Sr and Eu anomalies) in different melting conditions. Partial melting calculations are based on simple batch melting models, with assumed modal mineral compositions of 80% cpx +20 grt and 13% ol + 37% cpx +50% pl, respectively. The initial composition represents the average lower oceanic gabbro (VanTongeren et al., 2021). The description of the calculation and partition coefficients used here are presented in Supporting Information S1. Data sources: UDM inclusions (literature) (Danyushevsky et al., 2003;Gurenko & Chaussidon, 1995;Laubier et al., 2012;Saal et al., 2007); melts of garnet clinopyroxenite at high pressure (Hirschmann et al., 2003;Keshav et al., 2004;Kogiso & Hirschmann, 2006); melts of garnet clinopyroxenite and websterite (Y. Wang & Xu, 2020); melts of gabbro with the pressure range from 1 atm to 4.5 GPa (Koepke et al., 2004;Kvassnes & Grove, 2008;Yaxley & Sobolev, 2007); and melts of peridotite (Davis et al., 2011;Laporte et al., 2004;Schwab & Johnston, 2001;Walter, 1998;Wasylenki et al., 2003). Mineral abbreviations: Co, Coesite; Grt, Garnet; Opx, Orthopyroxene; Sp, Spinel. tomographic slices at 28°N, 30°N, 32°N, and 34°N show that the subducting Pacific slab has stagnated in the mantle transition zone, whereas the paleo-Pacific slab has penetrated deeper into the lower mantle (Figure 13c). The geometry of high-velocity anomalies in the lower mantle at 32°N and 34°N is very different from that at 28°N (Figure 13c). This implies that the paleo-Pacific domain might have separated from another subduction realm to the south at ca. 28°-32°N. More importantly, we identified two isolated high-velocity anomalies at depths of 1,000-1,500 km at 30°N (Figure 13c), which may originate from the fragmented paleo-Pacific oceanic slab. Considering a whole-mantle slab sinking rate of 1.52 cm/yr in NW Pacific (Peng & Liu, 2022), the sinking duration for 1,000-1,500 km ranges from 66 to 99 Ma, indicating that the subducted paleo-Pacific slab might be torn or fragmented during Late Cretaceous beneath the South China interior.
Slab fragmentation can create gaps for asthenospheric upwelling to heat the subducted oceanic lithosphere and trigger the melting of different portions of the subducted slab (Figure 14a). This process may partially contribute to the geochemically heterogeneous features of Cretaceous basaltic rocks in the South China interior (Figure 14b), for example, the OIB-like dolerite dykes (ca. 108 Ma) from the Wuyi basin suggests that the mantle source was enriched by both subducted sediment and altered oceanic basalt (B. Zhang et al., 2020); whereas the bulk-rock positive Sr and Eu anomalies and low-δ 18 O olivine phenocryst in the Late Cretaceous Ji'an basalts (ca. 90 Ma) Figure 12. Identifying the contribution of pelagic and terrigenous sediments. Panels (a) and (b) show the plot of ε Hf (t) versus ε Nd (t) for the Hengyang basaltic rocks. The present-day average compositions of oceanic basalt and sediments are shown with black and orange squares, respectively. The mantle array and time-integrated evolution paths of Nd-Hf isotopes (red line) refer to Chauvel et al. (2008). Panels (c) and (d) respectively show ε Nd (t) and ε Hf (t) versus Hf/Sm for the Hengyang basaltic rocks (yellow filled triangles, this study) and other late Cretaceous basaltic rocks (gray filled squares; Meng et al., 2012;Y. J. Wang et al., 2003), which indicate the subducted sediments (pelagic and terrigenous sediments) have been incorporated in the mantle source of Hengyang basaltic rocks.
indicate the involvement of an altered gabbroic oceanic component in the melting source (Y. M. Wu et al., 2020). The occurrence of UDM inclusions in the Late Cretaceous Hengyang basaltic rocks indicates that the melting of the subducted paleo-Pacific slab had extended to the lower oceanic gabbroic crust. Therefore, the contribution of different portions of subducted oceanic slabs, starting from the top surface sediment, then upper oceanic crust to  (Seton et al., 2012) and present. (c) Vertical cross-sections of MIT-P08 tomography model (C. Li et al., 2008) along the four profiles shown in (b). The positive velocity anomalies outline the shapes of the subducting Pacific slab and the subducted paleo-Pacific slab. lower oceanic gabbro during slab tearing or fragmentation, may be an important geodynamic process to produce geochemically heterogeneous mantle sources.

Conclusions
We conduct comprehensive geochemical analyses on bulk rocks and olivine-hosted MIs of the Late Cretaceous Hengyang basaltic andesites in South China. The Hengyang basaltic andesites experienced a predominant fractionation of olivine + clinopyroxene with negligible crustal contamination. Two types of MI have been identified, that is, the normal and UDM types. The identification of UDM inclusions with geochemical features resembling lower oceanic gabbroic crust provides robust evidence for the involvement of the contemporaneous subducted paleo-Pacific slab in the mantle source of the intracontinental magmatism in South China. By integrating geochemical modeling, plate reconstruction, and seismic tomography results, we propose that slab tearing or fragmentation in response to the rollback of the subducted paleo-Pacific slab along the transform fault is the most likely mechanism to trigger melting of the refractory lower oceanic cumulates. This geodynamic process might produce geochemically heterogeneous mantle sources and the resultant intracontinental basaltic magmatism in South China.

Data Availability Statement
The research data associated with the manuscript are listed in the Supporting Information and can be found on Zenodo (https://doi.org/10.5281/zenodo.7326093). Free software packages MATLAB, Generic Mapping Tools (Wessel et al., 2013), and GPlates (Boyden et al., 2011) are used for creating the figures.  (Leuthold et al., 2018;Poli & Schmidt, 2002;Yaxley & Sobolev, 2007). (b) A tectonic cartoon showing tearing or fragmentation of the subducting paleo-Pacific slab and its magmatic responses during Cretaceous. The triangles with different colors indicate the contribution of different portions of oceanic crust for the Cretaceous basaltic rocks.