During the subduction process, the basaltic oceanic crust may become garnet plagioamphibolite while the terrigenous sediments (mud) may transform into slate at 1.5–2.0 GPa. As the oceanic crust could then partially melt at high geothermal gradient (at approximately 1.5–2.0 GPa and ∼900°C), we chose 1.5 GPa and 850–1,000°C as run conditions in this study (Zheng et al., 2015; Wei and Zheng, 2020; Zhang Z. et al., 2020).

Two metamorphic rocks were chosen to prepare the starting materials. The 18TH-02 is a garnet plagioamphibolite and contains 50 vol% Amp, 35 vol% plagioclase (Pl), 12 vol% garnet (Grt), and trace amounts of V-Ti magnetite. Its major element composition is very similar to the average composition of oceanic crust, and several important trace elements (such as V, Cr, Rb, and Sr) of 18TH-02 are also close to the average composition of oceanic crust. The terrigenous sediment is represented by a plagioclase slate (18TH-07), this sample contains 85 vol% feldspar (orthoclase + albite + plagioclase), 10 vol% mica, and trace amounts of quartz. Their major element composition and selected trace elements are very close to the Upper Continental Crust (UCC) (the data of these two metamorphic rocks is listed in Supplementary Table S1).

Both of these rocks were collected from Susong, Anhui Province, China. The two rocks (18TH-02 and 18TH-07) were ultrasonically cleaned in distilled water and alcohol successively, and then crushed to powder to ∼120 meshes separately. In order to simulate the partial melting process of the oceanic crust with or without terrigenous sediments thereon during subduction process, two starting materials were used: PURE and MIXT. The PURE is representative of pure basaltic oceanic crust, which was made by 18TH-02 power only. The MIXT is representative of basaltic oceanic crust with terrigenous sediment, which was made by mixture 90 wt% 18TH-02 power and 10 wt% 18TH-07 power. The proportion was determine by typical geochemical studies (Gan et al., 2010; Niu, 2013), for more detail refer to Zang et al. (2020). Then these two materials were placed in an agate mortar and ground for >3 h separately, ensuring the particle size of their power was inner 10 μm. Before being used, these prepared powers were stored in a 110°C drying oven for more than 48 h to remove adsorbed water.

The sample assemblage consisted of NaCl sleeve, Pyrex Glass, graphite furnace (with a cap), magnesium oxide (after 1,000°C heating), and the sample capsules. The power of the starting material was filled in gold sample capsules (with 3 mm outer diameter, 2.6 mm inner diameter, and 2.5 ± 0.2 mm height) and sealed, and then the capsule was covered by a thin hexagonal boron nitride sleeve and placed into the graphite heater. The centre of the capsule is 1 mm below the geometric centre of the graphite heater (close to the stainless steel) to get a stable thermal field (Xia et al., 2014). In order to minimize adsorbed water, all the assemblies used in our experiment were heated at 110°C for 48 h before being used. Then the sample was fitted into a 10 mm diameter pressure plate one by one as shown in Figure 2.

All experiments were performed on a Rocktek-PC-1 type Piston-cylinder apparatus located in National Engineering Research Centre of Coal Mine Water Hazard Controlling, Suzhou University. The temperature was measured and controlled via NiCr-NiAl thermocouples with an accuracy of ±3°C. The temperature gradient of our sample was about 5°C, based on similar study results (Xia et al., 2014). The pressure was checked by NaCl melting (falling ball method), following the melting curve offered by Tingle et al. (1993). The error was within 0.1 GPa. According to our assessment of similar experiments, the oxygen fugacity within the experiment was close to FMQ (Tao et al., 2015), which is similar to the basaltic oceanic crust in the subducted margins.

In one run, the pressure rose to about 800 PSI by hand first. We then started heating with a 10°C/min increasing speed. When the temperature was up to 400°C, the pressure would continue to rise with a 10 PSI/min to the aim value. After the pressure rising to aim value, the temperature rose to the aim value with 10°C/min increasing speed. The pressure and temperature were maintained for 168 h, then quenched by turning the power off. The temperature fell to room temperature in 1 min. We then depressurized and removed the run product carefully from the capsule and mounted it in epoxy, which was polished for observation and analysis.

The whole rock compositions (major and trace element) analyses of the two metamorphic rocks (18TH-02 and 18TH-07) were carried out in the State Key Laboratory of Ore Deposit Geochemistry, Institute of Geochemistry, Chinese Academy of Sciences, Guiyang, China. Their major elements were measured by an X-ray fluorescence spectrophotometer (XRF, PW4400, Axios), and the trace elements were measured by an inductively coupled plasma mass spectrometer (ICP-MS), in accordance with the analytical procedures given by Qi et al. (2000). Back-scattered electron (BSE) images and the major element content of minerals in these two metamorphic rocks and in the experimental products were obtained using a JOEL JXA-8100 electron microprobe at the Key Laboratory of Mineralization and Dynamics, Chang’an University, Xi’an, China. The beam used for the analyses of the melt was 5–10 μm, with an accelerating voltage of 15 kV and beam current of 2 nA. The possible losses of Na and K in the melt were addressed using the same method in Zang et al. (2020) Trace-element concentrations within the melts (glass) of run products were measured by laser ablation ICP–MS using a GeoLasPro laser ablation system and an Agilent 7700x ICP–MS instrument at the Sample Solution Analytical Technology Company, Wuhan, China. The analytical procedures used are outlined in detail by Liu et al. (2008).

The run conditions and mineral assemblages in their products are listed in Supplementary Table S2. Representative BSE photographs of the experimental products are shown in Figure 3. There is no difference between the mineral assemblages of these two starting materials at all run conditions. In runs at 850–950°C, there are Amp + Cpx + Pl + Grt + Melt in all run products. While in runs at 1,000°C, the Amp disappeared and the mineral assemblage is Cpx + Pl + Grt + Melt. However, melts (quenched to glass) in run products of PURE are very rare. The contents of them are about 2–3 vol% in runs at 850 and 900°C. This content increases to about 5 vol% in run product at 950°C and about 10 vol% in run product at 1,000°C (Figure 3A). The content of melts in run products of MIXT is different from those of PURE at over 900°C. In the run at 850°C, the melt in run products of MIXT is also very rare, which is consistent with the run from PURE at 850°C. In runs at 900°C, the content of melt in MIXT’s run products increases to about 10 vol%. This content increases to 20 vol% at 950°C and 35 vol% at 1,000°C (Figure 3B). The melting trend of MITX is consistent with the result reported by Zang et al. (2020). The mixture starting materials will increase in melting percentage at over ∼925°C.

The minerals in run products of PURE at 850°C are all very small; the size of all minerals (Amp, Cpx, Grt and Pl) is all about 15 μm. In run products at over 900°C, the size of mineral crystals increases with run temperature. The size of Amp increases most obviously with the run temperature increasing, it is about 40 μm in run product at 900°C and about 100 μm in run product at 950°C. The size of the other three minerals is also increased with run temperature. The size of Cpx is 20, 30, and 50 μm respectively in runs at 900°C, 950°C, and 1,000°C. The size of Pl is 20, 40, 50 μm respectively in runs at 900°C, 950°C, and 1,000°C. The size of Grt is 20, 25, and 50 μm respectively in runs at 900°C, 950°C, and 1,000°C. The minerals in run products of MIXT at 850°C is also very small. The size of all minerals is the same as those of PURE, at about 15 μm. The size of minerals in MIXT’s run products is also increased with run temperature. The size of Amp is 25 and 50 μm respectively in runs at 900 and 950°C. The size of Cpx is 20, 30, and 50 μm respectively in runs of 900°C, 950°C, and 1,000°C. The size of Pl is 20, 20, and 30 μm respectively in runs at 900°C, 950°C, and 1,000°C. The size of Grt is 20, 30, and 60 μm respectively in runs at 900°C, 950°C, and 1,000°C. In all run products, Grt is euhedral crystal and other kinds of minerals are subhedral crystals.

The major and trace element composition of partial melts from run products reported in this paper is listed in Supplementary Table S3. The variation trend with run temperature rising is shown in Figure 4. As the run temperature rose, the SiO₂ contents of melts in all run products decreased, whereas the FeO contents of these melts show a reverse trend. The CaO content of melts from PURE showed a similar variation trend as SiO₂ contents, while the CaO content of melts from MITX at over 900°C showed a reverse trend. The K₂O content of partial melts from MIXT shows a decreasing trend as run temperature ring, while the K₂O content of partial melts from PURE is almost unchanged due to their starting materials having so little K₂O. The Al₂O₃ content of melts in all runs is between 15 and 25wt% is affected by the amount of feldspar. The Na₂O content of partial melts is about 0.5–2.5wt%. Chondrite-normalized REE patterns for the partial melts are shown in Figure 5. The REE patterns of the partial melts are enriched in light REE, compared with their parents, the HREE of partial melts is more depleted (Figure 5A). In chondrite-normalized spidergram, the partial melt depleted in Ti, Y, Yb, and Lu compared with starting material, as shown in Figure 6.

The SiO₂, Al₂O₃, CaO, and Na₂O content of Pl in the starting material and run products are shown in Figure 7A–D. Compared with the Pl in the starting material, the Pl in run products from PURE enriched in CaO and depleted in Na₂O, which was the same as the Pl in the run product at 1,000°C from MIXT. The other Pl in run products from MIXT have the same variation range with their starting material. On the Or-Ab-An classification map, all Pl in run products are plagioclase series (Figure 8A). The SiO₂, Al₂O₃, FeO, MgO, CaO, and Na₂O content of Amp in starting material and run products are shown in Figure 7E–J. Compared with starting Amp in 18TH-02, Amp in all run products is enriched in SiO₂, FeO, MgO, and Na₂O, and depleted in Al₂O₃ and CaO. This difference is also very obvious in the classification map of amphibole (Leake et al., 1997) as shown in Figure 8B. The SiO₂, Al₂O₃, FeO, MgO, and CaO content of Cpx in run products are shown in Figure 7K–O. As the run temperature rises, the Al₂O₃ content of Cpx in all run products increases, while the MgO content shows a reverse trend. In the classification map of clinopyroxene (Morimoto, 1988), the Cpx from runs at 850 and 900°C are enriched in En and others enriched in Wo (Figure 8C). The SiO₂, Al₂O₃, FeO, MgO, and CaO content of Grt in starting material and run products are shown in Figure 7P–T. Compared with the starting Grt in 18TH-02, the Grt in all run products was enriched in FeO and MgO, and depleted in CaO, which is consistent with the Alm-Pyr-Gro classification map, as shown in Figure 8D.

We then compared melts from PURE and MITX. With the sediment addition, the MITX increased its melting percentage at over 900°C, which is the same as previous studies (Mccarthy and Patiño Douce, 1997; Zang et al., 2020), but the melting percentage in this study is very small compared with others. The reasons for this may be the following: 1) there is no free water added in our experiment: the water in our run products is from dehydration of Amp in staring materials (18 TH-02). The water content of our experiment is lower than similar experiments with water added before the Amp is destroyed. 2) These two staring materials are metamorphic rocks, unlike basalt, there is no glass in them. By comparing the melting percentage of PURE and MITX, the melt percentage of MIXT is much higher than the one of PURE in runs at 950and 1,000°C, and the increasing value of MITX is more than the total content of addition metamorphic subducted terrigenous sediment. This result indicates the effect of sediment addition on increasing melting percentage needs a “staring temperature” (950°C in this paper). The addition sediments significantly help the partial melting of subducted oceanic crust over this staring temperature.

We then compared the chemical composition of melt from PURE and MIXT over the staring temperature, to consider which one could be the real magma. As shown in Figure 4 the melt from MIXT are enriched in Na₂O and K₂O and depleted in Al₂O₃ in major elements than the melt from PURE in runs at 950and 1,000°C. For trace elements, the melts from the mixture enrich more in Rb than the PURE because the MIXT has a very high content of Rb compared to PURE.

Possible ways in which the silicic arc magma is generated include: 1) that it is generated by partial melting of the subducted altered oceanic crust; or 2) that it could have evolved from mantle-derived mafic magmas (see Clemens et al. (2021) and references therein). It seems unlikely that the majority of silicic melt could be evolved from basalt for the following reasons: 1) silicic melt has a high viscosity and crystallization differentiation is difficult (even this process could be achieved with a large number of chadacrysts that are not completely separated from silicic magma, but the chadacrysts are not so widely distributed); 2) the crystallization differentiation of magma could be detected via geochemical data, such as separation of feldspar can lead to Eu anomalies, however, there are no obvious Eu anomalies in current data relating to rocks (SiO2 ≥ 52wt%) from Tonga and Kuril (Figure 5B). The partial of subducted altered oceanic crust with or without sediments could be an important way of generating silicic magma at convergent plate margins.

In terms of the effect of sediment addition on silicic arc magma generation, we compared the major element content of rocks from field observation and melt from partial experiments on basic rocks and basalt-sediment mixtures via Harker diagrams (Figure 9). For major elements, melts from the experiment seem to all cover the variation range of arc silicic magma located on Tonga and the Kuril Islands. The melts from MIXT also match the field observation better on the SiO₂-K₂O diagram because the PURE (and the oceanic crust) is depleted on K₂O. The high K₂O silicic magmas may thus generate from the mixture of magmatic sources. Besides the major elements, the trace element of rocks from field observation and melt from partial experiments are also compared via plots of Sr/Rb—Rb (Figure 10). Unlike the major elements, the melts from the mixture match the field observation better. Note that the trace element data on melts from pure basic rocks (Xiong et al., 2006; Qian and Hermann, 2013) are out of the variation range shown in Figure 10. Furthermore, the REE abundance patterns of partial melts from magmatic sources with sediment addition have an obvious right-leaning than the one from the pure magmatic source. The partial melts from the mixture magmatic sources match the silicic magma from Kuril Island better than the melts from pure magmatic sources. This result indicates that the arc silicic magma is mainly generated from the partial melting of a mixture of basic oceanic crust and sediments. The sediments also offer great help in establishing the partial melting percentage. This may also contribute to the formation of silicic arc magma.