Genesis of high-Ni olivine phenocrysts of the Dali picrites in the Central Emeishan large igneous province

Abstract The Emeishan large igneous province (ELIP) in SW China is considered to be a typical mantle-plume-derived LIP. The picrites formed at relatively high temperatures in the ELIP, providing one of the important lines of argument for the role of mantle plume. Here we report trace-element data on olivine phenocrysts in the Dali picrites from the ELIP. The olivines are Ni-rich, and characterized by high (>1.4) 100×Mn/Fe value and low (<13) 10 000×Zn/Fe value, indicating a peridotite-dominated source. Since the olivine–melt Ni partition coefficient (KDNiol/melt) will decrease at high temperatures and pressures, the picrites derived from peridotite melting at high pressure, and that crystallized olivines at lower pressure, can generate high concentrations of Ni in olivine phenocrysts, excluding the necessity of a metasomatic pyroxenite contribution. Based on the Al-in-olivine thermometer, olivine crystallization temperature and mantle potential temperature (T P) were calculated at c. 1491°C and c. 1559°C, respectively. Our results are c. 200°C higher than that of the normal asthenospheric mantle, and are consistent with the role of a mantle thermal plume for the ELIP.

As one of the earliest crystallized minerals from typical mantle-derived magmas, the composition of olivine phenocrysts with high Fo values (atomic 100×Mg/(Mg þ Fe)) can provide important information on the primary magma. Recent studies suggest that some trace elements (such as Mn and Zn) in olivine from pyroxenite-derived and peridotite-derived melts have two distinct trends (Le Roux et al. 2010;Howarth & Harris, 2017) and can therefore be used to distinguish pyroxenite components in the mantle source region. Some studies recorded picritic rocks in the inner zone of the ELIP (Hanski et al. 2010;Kamenetsky et al. 2012;Yu et al. 2017;Yao et al. 2019), which were thought to have formed at relatively high temperatures (Xu et al. 2001;Zhang et al. 2004Zhang et al. , 2006Shellnutt & Pham, 2018). However, some other studies suggested that the picrites were derived from a metasomatized pyroxenite-bearing mantle source (Kamenetsky et al. 2012;Yu et al. 2017). The role of high-temperature primary magmas in the ELIP therefore remains contentious. In this paper, we investigate this aspect based on petrology and mineral chemistry of olivine phenocrysts from Dali picrite in the inner zone of the ELIP, with a view to evaluate the nature of the mantle source and the conditions of melting.

Geological setting
The ELIP is mainly located in the western part of the Yangtze craton in SW China (Fig. 1). Its southwestern margin is bounded by the Ailaoshan-Red River Fault, and the northwestern boundary is traditionally thought to be the Longmenshan-Xiaojinhe Fault. The volcanic successions in the ELIP trend N-S with a rhombic shape and cover an area of more than 2.5 × 10 5 km 2 , with a total erupted volume of more than 3 × 10 5 km 3 (Xu et al. 2001Ali et al. 2005Ali et al. , 2010. The thickness of the volcanic sequence ranges from c. 5000 m to several hundred metres, with the thickest zone located near the Lijiang and Binchuan sections in the western part of the ELIP and the thinnest area located in the eastern part of the ELIP Song et al. 2001Song et al. , 2004Xiao et al. 2003Xiao et al. , 2004Zhang et al. 2006). The ELIP comprises a succession of tholeiites, with minor picritic and rhyolitic/trachytic lava flows. Many maficultramafic intrusions are exposed in a belt extending along the Panzhihua-Xichang region in the SW part of the ELIP, and some of these intrusions host giant Fe-Ti-V oxide and Ni-Cu-PGE sulphide deposits (Shellnutt, 2014). Previous studies have indicated that the main phase of magmatism took place at c. 260 Ma over a short duration of 1 Ma (Zhou et al. 2002;He et al. 2007;Shellnutt et al. 2008Shellnutt et al. , 2012Zhong et al. 2009Zhong et al. , 2011Zhong et al. , 2014Sun et al. 2010;Xu et al. 2010;Jerram et al. 2016).
The picritic lavas mainly occur in the central ELIP such as those in the Dali and Binchuan districts, and these areas are considered to represent the centre of the mantle plume (Zhang et al. 2006;Hanski et al. 2010;Kamenetsky et al. 2012;Li et al. 2015;Ren et al. 2017;Wu et al. 2018). The samples of Dali picrite for this study were collected from a road-cut section c. 20 km NE of Dali City (25°40 0 48″ N; 100°21 0 14″ E), the occurrence of which was reported by Hanski et al. (2010) and Ren et al. (2017). The picrites occur in the bottom part of the Emeishan basaltic sequence (Fig. 2a). A newly opened quarry from where the basalts and picrites are mined exposes fresh outcrops, where the picrites are easily distinguished from their darker colour compared with the associated basalt flows.
Most of the picrites are porphyritic (with 20-40 vol% phenocrysts), and the dominant phenocryst is olivine (c. 80 vol% of the phenocrysts) with minor clinopyroxene. The olivine phenocrysts are subhedral to euhedral, generally ranging from 0.2 to 2 mm across with the largest grains up to 4 mm in diameter. Some olivine crystals are altered to serpentine along the rims and cracks, but their cores remain unaltered. Minor euhedral to subhedral Cr-spinel grains occur within the olivine phenocrysts (Fig. 2c). The groundmass consists predominantly of microcrystallinecryptocrystalline olivine, anhedral clinopyroxene and small

986
W-C Cai et al.
plagioclase crystals. Some Cr-spinels and Fe-Ti oxide minerals occur in the groundmass. The basalts are porphyritic (with 15-40 vol% phenocrysts), and plagioclase is the dominant phenocryst (c. 80-90 vol% of the phenocrysts; Fig. 2d) with minor clinopyroxene (c. 10-20 vol% of the phenocrysts). The groundmass consists predominantly of microcrystalline plagioclase, anhedral clinopyroxene, some basaltic glass and a small amount of magnetite and Fe-Ti oxide.

Analytical methods
Polished thin-sections of picrite samples were prepared for the analysis. The olivine phenocrysts selected for analysis show little alteration with no zoning. The unaltered Cr-spinel inclusions without any fractures were selected for analysis.
Major and minor elements in olivines and Cr-spinels were analysed using an EPMA-1720 electron microprobe at the EPMA Laboratory, Institute of Earth Sciences, China University of Geosciences, Beijing (CUGB). For most elements, the accelerating voltage was 15 kV, beam current 20 nA and beam diameter 1 μm, with the on-peak counting time set for 10 s per element and 5 s for background. The Al element was analysed using a 60 s peak counting time and the background was set for 30 s. The standard samples are natural minerals and synthetic oxides produced by SPI Supplies of the United States of America. The precision is better than 1% for most major and minor elements.
Trace-element concentrations of olivine were determined by laser ablationinductively coupled plasmamass spectrometry (LA-ICP-MS) using the Agilent 7900 Quadrupole ICP-MS coupled with a Photon Machines Analyte HE 193 nm ArF Excimer laser ablation system. The analyses were carried out at the Mineral Geochemistry Lab, Ore Deposit and Exploration Centre (ODEC), Hefei University of Technology, China. In this study, the ablation protocol employed a spot of 30 μm diameter at 8 Hz with an energy of c. 4 J cm -2 fluence. Each spot was measured for 40 s, and each spot analysis was followed by a gas blank for 20 s. Helium was used as the carrier gas and argon was used as the makeup gas, and they were mixed via a T-connector before entering the ICP Wang et al. 2017). Reference materials GSE-1G, GSD-1G, BCR-2G and SRM-612 were used as external standards, and they were measured every 10-12 spot analyses to check reproducibility through the analytical session. Off-line selection and the integration of background and analytical signals, and time-drift correction and quantitative calibration, were performed with ICP-MS Data Cal (Liu et al. 2008). The precision for most trace elements in each spot was better than 10%.

Mineral chemistry of olivine and Cr-spinel
The major elements and some minor elements (Ni, Mn and Al) were analysed by EPMA, and all the analytical spots were on fresh core regions of the grains (online Supplementary Table S1a,  Generally, the content of NiO (0.25-0.51 wt%) shows a positive correlation with the Fo values, whereas the MnO content displays a negative correlation with the Fo values (0.11-0.32 wt%). The CaO concentration is relatively high, ranging from 0.25 to 0.46 wt%, with no correlation with the olivine Fo values. The Al 2 O 3 content in olivine range from 0.03 to 0.11 wt%. All the olivine phenocrysts of the Dali picrites contain > 0.25 wt% CaO content, and exhibit no kinked band. We infer that they were crystallized from magma and are not mantle xenocrysts, and can therefore be used as an indicator of the primary melt.
With regard to the minor-and trace-element concentrations in the core of the olivine phenocrysts (analysed by LA-ICP-MS), the Ni concentration ranges from 2563 to 3897 ppm (0.32-0.49 wt% NiO) and shows a positive correlation with the Fo values (Fig. 3). The CaO concentration ranges from 0.25 to 0.50 wt%. MnO content ranges from 0.11 to 0.22 wt% (Fig. 4), whereas the Zn concentration range is 49.8-120.9 ppm, showing a positive Zn-Mn correlation ( Fig. 5a; online Supplementary Table S2, available at http://journals.cambridge.org/geo). In addition, the olivine contains 115.6-171.1 ppm Co, 5.8-21.1 ppm V and 4.4-8.9 ppm Sc. However, most rare earth elements (REEs), especially light REEs (LREEs), are present in quantities lower than the detection limits. Although some minor elements such as Ni and Mn were analysed by both EPMA and LA-ICP-MS, we discuss the LA-ICP-MS data as it is of greater precision. The Cr no. (atomic Cr/(Cr þ Al)) of Cr-spinel ranges from 0.570 to 0.685 (online Supplementary Table S1b). The content of TiO 2 is in the range of 0.4-1.8 wt%, and total iron as FeO (FeO*) is 19.5-29.2 wt%. The MgO contents vary from 9.7 to 14.3 wt%, and the Al 2 O 3 values range from 16.4 to 20.8 wt%.

5.a. Origin of high-Ni olivine
The Ni content of primitive mantle is estimated as 1960 ppm, which is similar to that of the depleted peridotite (McDonough & Sun, 1995;Herzberg et al. 2013), whereas the Ni concentration of olivine in mantle peridotite ranges from 2800 to 3100 ppm (Ionov, 2007;Herzberg et al. 2013Herzberg et al. , 2016. However, the olivine phenocrysts in Dali picrites, especially those grains with high Fo values, have much higher Ni concentration (with Ni concentration up to c. 3900 ppm, Hanski et al. 2010;Yu et al. 2017) than those from mid-ocean ridge basalts (MORBs) or mantle peridotite (Fig. 3). The excess Ni in olivine can result from the following processes. (1) The core-mantle interaction can produce a less-degassed Ni-rich source, and this source continues adding into the mantle plume, which finally formed picrites with high Ni content (Herzberg et al. 2013(Herzberg et al. , 2016. (2) The silica-rich melts formed from recycled crust might interact with mantle peridotite to form a secondary olivine-poor pyroxenite source. Magmas derived from such metasomatic pyroxenite source would therefore contain higher Ni content than those derived from peridotite source (Sobolev et al. 2005(Sobolev et al. , 2007Herzberg, 2006;Foley et al. 2011Foley et al. , 2013. (3) Higher melting temperature and pressure would decrease the partition coefficient of Ni between olivine and melts (K D Ni ol/melt ); this will

988
W-C Cai et al.
result in a higher Ni concentration in primary melt, and the olivine crystallizing from this melt will have higher Ni content (Niu et al. 2011;Putirka et al. 2011;Matzen et al. 2013Matzen et al. , 2017. Olivine with high Ni content has also been recognized in other LIPs. The picrites in Baffin Island and West Greenland also have olivine phenocrysts containing high Ni concentration (c. 3800 ppm) (Sobolev et al. 2007;Herzberg et al. 2013), even 20% higher than the modern MORB source (2800-3100 ppm) (Ionov, 2007). These picrites are characterized by elevated 3 He/ 4 He ratios (Starkey et al. 2009). Picrites from Baffin Island and West Greenland show 3 He/ 4 He values up to 50 times the atmospheric value. Herzberg et al. (2013) proposed that a lessdegassed mantle reservoir formed by core-mantle interaction near the core-mantle boundary contains high 3 He/ 4 He, and they correlated the elevated 3 He/ 4 He associated with high Ni concentration to the less-degassed Ni-rich material from core-mantle boundary that was added to the mantle source region. For the ELIP, however, picrites from Dali, Binchuan and Lijiang area generally have higher Os concentration and slightly lower Re concentration than many other ocean island picrites, although the 187 Os/ 188 Os ratios are obviously lower than the mantle value with negative γOs values (Chen et al. 2007;Xu et al. 2007;Zhang et al. 2008;Hao et al. 2011), which are distinctly different from the Siberia LIP picrites that display relatively positive γOs values. The Os isotopic compositions of the picrites in the ELIP therefore suggest no major input from the core-mantle boundary to the source region.
If the silica-rich melt derived from recycled crust reacts with peridotite, an olivine-free pyroxenite source can be generated that would produce magmas with high Ni concentration (Sobolev et al. 2005). Olivine crystallizing from such a magma will generally be characterized by high Ni concentration and low Mn/Fe. The Mn, Fe and Zn in olivine mostly occur as divalent elements and their ionic radii is close to that of Mg 2þ (De Hoog et al. 2010); they therefore behave similarly in the olivine-melt system. Since Mn, Fe and Zn have similar olivine-melt partition coefficients (0.89, 1.1 and 1.09, respectively; Foley et al. 2013;Howarth & Harris, 2017), their ratios are relatively constant during partial melting or fractional crystallization. The ratios of these melts could therefore well constrain the nature of their mantle source (Sobolev et al. 2007;Herzberg, 2011;Foley et al. 2013;Herzberg et al. 2016). Experimental studies show that the olivine crystallized from melts derived from pyroxenite have 100×Mn/Fe values of 1.05-1.35, whereas those in melts derived from peridotite source show higher 100×Mn/Fe values (>1.5) (Sobolev et al. 2007;Foley et al. 2013). The Dali picrites have olivine phenocrysts with higher Mn content compared with those of Karoo or Hawaii, and show a similar trend to the Mn content of MORB (Fig. 4). Additionally, the high 100×Mn/Fe ratios (1.43-1.73) of olivine in Dali picrites are consistent with those olivines crystallized from melts derived from peridotite (Fig. 5b). This evidence indicates that the primary magma of the Dali picrites was more likely derived from a peridotite-dominated source. Sobolev et al. (2007) suggested that the percentage of metasomatic pyroxenite (X px ) in the mantle source could be calculated based on the Mn/Fe values of the olivine phenocrysts. The equation was proposed as: X px = 3.48−[2.071×(100×Mn/Fe)]. Combined with the trace-element data from LA-ICP-MS analysis (online Supplementary Table S2), we calculate an average X px in the Dali picrite mantle source of 22.6% (Fig. 6). For comparison, Sobolev et al. (2007) suggested an average X px value of 17% for MORBs. However, the data for olivine in MORB reported by these authors spans a somewhat larger range in X px values, which show a similar range to the Dali picrites (Fig. 6); this suggests that the mantle source of Dali picrites is not pyroxenite dominated. It is notable that high pressure may increase the partitioning of Mn between olivine and melts (K D Mn ol/melt ) (Matzen et al. 2017). Our calculation of X px using the method of Sobolev et al. (2007) is therefore very likely to be flawed. The ratio of inter-mineral exchange coefficient of Zn/Mn between olivine and clinopyroxene, K D (Zn/Mn) cpx/ol , defined as K D (Zn/Mn) cpx/melt /K D (Zn/Mn) ol/melt , is c. 0.26 (Le Roux et al. 2010). It has also been established that K D (Zn/Fe) ol/melt is close to 1, but for clinopyroxene (K D (Zn/Fe) cpx/melt ), the value is much lower (Le Roux et al. 2010). This suggests that Zn/Fe is minimally fractionated during partial melting of peridotite mantle, but strongly fractionated during melting of pyroxene. As demonstrated previously, olivine grains in magmas derived from peridotite-dominated sources and pyroxenite-dominated sources have significantly different variation trends when plotted in the 100×Mn/Fe versus 10 000×Zn/Fe diagram (Howarth & Harris, 2017). Moreover, olivine equilibrated with melts derived from pyroxenite was demonstrated to have 10 000×Zn/Fe values of >14 (Le Roux et al. 2010;Howarth & Harris, 2017). Olivines in Dali picrites have 10 000×Zn/Fe values of 8.0-12.7, showing a similar trend to olivine in picrites from Edenteka and Baffin Island, which were derived from a peridotite source (Fig. 5b). In addition, olivine crystallized from pyroxenite-derived melts have a relatively low Mn/Zn ratio (<14), while those from peridotite-derived melts exhibit a higher Mn/Zn ratio (>14). The Mn/Zn values of the olivine in the Dali picrite range from 13.6 to 18.4, with most Mn/Zn values >15, also indicating a peridotite dominate source (Fig. 5c).
All the features presented above suggest that Dali picrites are more likely to have formed from peridotite-dominated mantle source. However, how a peridotite source can produce a picrite with high-Ni olivine phenocrysts needs to be evaluated further. According to recent experimental studies, the partition coefficient of the Ni element between olivine and melts (K D Ni ol/melt ) is mainly controlled by temperature and pressure. Under conditions of high temperature and pressure, the Ni partition coefficient (K D Ni ol/melt ) will decrease and result in less Ni left in residual olivine. As the melts rise to a shallower level or even to the sub-surface, the partition coefficient of Ni between olivine and melts (K D Ni ol/melt ) will increase due to the decreasing pressure. This may result in higher Ni concentrations in olivine phenocrysts, which are even higher than those in residual olivine in mantle (Li & Ripley, 2010;Niu et al. 2011;Matzen et al. 2013Matzen et al. , 2017. Matzen et al. (2013) presented results from experimental work which allowed them to investigate the effects of temperature (T) and pressure (P) on K D Ni ol/melt , independent of substantial changes in liquid composition. The partition coefficient of Ni between olivine and melts is described by the equation: In ðK D Ni ol=liq Þ ¼ ½ðÀΔ r H Tref ; Pref Þ=ðRTÞ þ ðΔ r S o Tref ; Pref =RÞ À In½XðMgOÞ liq =XðMgSi 0:5 O 2 Þ ol where −Δ r H Tref, Pref /R and Δ r S°T ref, Pref /R are constants independent of pressure and temperature, and T is the olivine liquidus temperature in degrees Kelvin. Putirka et al. (2018) reproduced the calibration data (n = 17) with a standard error of estimate of ±1.3 with an R 2 of 0.82, using the high-precision work of Matzen et al. (2013). The above equation can be simplified to: where the olivine liquidus temperature T is in degrees Celsius. From a rough estimate of the source region pressure via the inversion of lanthanide REE data and olivine-melt equilibrium, the pressure at the start of melting is estimated as 4 GPa (Xu et al. 2001;Zhang et al. 2003Zhang et al. , 2006. Tao et al. (2015) used the olivine and clinopyroxene composition in ELIP picrites to determine the P-T condition when magma crystallization occurred, and reported that the olivine phenocrysts crystallized at a pressure of 1 GPa. As we calculate later in Section 5.b, the olivines in Dali picrites crystallized at a temperature of c. 1491°C, and the mantle potential temperature T p is c. 1560°C. When these temperature estimates are introduced into the simplified equation above, the resulting Ni partition coefficient K D Ni ol/melt is c. 3.8 and 4.3 for melting and crystallizing, respectively. The Ni content of melts derived from peridotite is defined as: where Ni is in ppm and MgO in wt%, provided by Niu et al. (2011).
According to the olivine-bulk-rock equilibrium and melt inclusions data, previous studies suggested that the primary melt of the Dali picrites had MgO content of 20-23 wt% (Li et al. 2012Ren et al. 2017). When we use the primary melt MgO content in the equation above, the Ni content is 776-1006 ppm. The Dali picrites melted at high pressure (c. 4 GPa), adiabatically ascended to the near-surface environment at low pressure (c. 1 GPa) and started to crystallize olivine. In this case, the Ni partition coefficient K D Ni ol/melt = 4.3. In order to crystallize olivine with c. 3900 ppm Ni concentration, there should be at least c. 907 ppm Ni in the melt. This is consistent with the calculated Ni concentration in the primary melt of Dali picrite, suggesting that it is possible for the Dali picrite melt to crystallize high-Ni olivines. In other words, the Dali picrites were produced by partial melting of a peridotite source. We therefore propose that the high concentration of Ni in olivine in the Dali picrites might be attributed to the relatively low K D Ni ol/melt during peridotite partial melting at conditions of high temperature

990
W-C Cai et al.
and high pressure, which resulted in enrichment of the primitive magmas with Ni, and the high K D Ni ol/melt when the deep mantlederived magma rose to a shallower environment of lower temperature and pressure, which resulted in crystallization of those Ni-rich olivines. There is therefore no need for eclogites or pyroxenite in the generation of Dali picrites.

5.b. Estimation of mantle potential temperature
Based on the temperature-dependent behaviour of Al between coexisting olivine and Cr-spinel, the newly established Al-inolivine thermometer proposed by Wan et al. (2008) and Coogan et al. (2014) can be used to estimate the olivine crystallization temperature. Compared with the traditional olivine-melt equilibrium thermometer, Al-in-olivine thermometer has the advantage of being independent of crystallization pressure, parental melt composition, oxygen fugacity and water content (Wan et al. 2008;Coogan et al. 2014). In addition, Al is expected to diffuse much more slowly through the olivine lattice than Mg and Fe, so the initial composition of the olivine-spinel pair is preserved (Spandler & O'Neill, 2010). It is therefore much more useful and convenient for us to determine the crystallization temperature of olivines in Dali picrites. The experimental results provided by Wan et al. (2008) and Coogan et al. (2014) suggested a restricted range for this thermometer; only Cr-spinel with Cr no. of 0-0.69 (Cr/(Cr þ Al), mol) can be used for a reliable result from the Al-in-ol thermometer. The spinels in this study exhibit Cr no. values in the range of 0.570-0.685, all within the calibration range of the thermometer. We can therefore use these olivine and Cr-spinel composition data to estimate the crystallization temperature of olivines (°C), and the thermometer calibration equation is expressed: T olÀspl ¼ 273:15 þ 10 000=½0:575 þ ð0:884 Â Cr no:Þ À 0:897 The calculated temperature for Dali picrites ranges from 1221 to 1491°C (by EPMA data, online Supplementary Table S1c). Accordingly, we also calculated the uncertainty of the thermometer by error propagation, and the calculated 2σ error ranged from ± 57 to ± 65°C (online Supplementary Table S1c). The maximum crystallization temperature is estimated as 1491 ± 65°C (Fig. 7), which constrains the liquidus temperature to T ol/liq ≥ 1491°C. When compared with olivine crystallization temperatures reported from other plume-related settings, such as continental flood basalts or other LIPs, the hottest Dali olivine-spinel crystallization temperatures are c. 50°C higher than the maximum olivine crystallization temperatures for ELIP samples from Yongsheng, Binchuan and Dali estimated by Xu & Liu (2016) (1440 ± 63°C). The samples of Dali picrite for this study were collected from a road-cut section located close to the Dali sections where Xu & Liu (2016) collected their samples, and the higher crystallization temperature estimated in Dali picrites might be attributed to their higher Fo value of the host olivine than the samples of Xu & Liu (2016). For comparison, the highest Dali picrite olivine crystallization temperature is 1491°C, which is similar to the highest olivine crystallization temperature estimated for Karoo, Madagascar and Etendeka, but about 100°C higher than those estimated for Iceland and SE Greenland (Coogan et al. 2014;Heinonen et al. 2015;Xu & Liu, 2016;Matthews et al. 2016;Jennings et al. 2019; Table 1). This temperature is about 200°C higher than the maximum crystallization temperature of MORB (1270°C) calculated by the same thermometer (Coogan et al. 2014), and provides robust evidence for a pronounced thermal anomaly in the mantle source of the Dali picrites.
As for the mantle potential temperature (T P ), there is a simple method for evaluating T P of a peridotite source from the olivine crystallization temperature. It is generally considered that progressively ascending melts are nearly adiabatic before olivine begins to crystallize. Because of the enthalpy of melting and adiabatic cooling, T P should be higher than the maximum crystallization temperature. In order to calculate T P of the Dali picrites, a temperature correction for melt generation and adiabatic cooling is required (Putirka et al. 2007;Herzberg & Asimow, 2015;Matthews et al. 2016;Jennings et al. 2019). Putirka et al. (2007) and Herzberg & Asimow (2015) assumed a simple two-step approach to calculate the mantle potential temperature from the crystallization temperature: first, correct back to the liquidus temperature at 1 atm along an adiabat, then correct the temperature which dropped through melting and crystallizing at 1 atm.
The water content of magma also needs to be taken into consideration, due to its potential effect on the T p calculation. As reported by Liu et al. (2017), the H 2 O contents in Dali picrites were calculated to be 2-4 wt% based on the study on  clinopyroxene-melt equilibrium. However, considering that the crystallization of olivine and spinel usually occurs earlier than that of clinopyroxene, an increase of H 2 O in the residual melts can be expected, and the reported H 2 O contents might be an overestimate. As our samples lack other hydrous primary minerals (e.g. amphibole, mica) and the basalts coexisting with Dali picrites (Fig. 2a) contain large amounts of plagioclase phenocrysts (Fig. 2d), we propose that the water content of our Dali pictrite samples was probably relatively low. Based on experimental data, Herzberg & Asimow (2015) defined the effect of pressure on increasing olivine liquidus temperature as: where T 1 ol/liq (°C) is the olivine liquidus temperature at a pressure of 1 atm and T(P) ol/liq (°C) is the olivine liquidus temperature at pressure P in gigapascals. Tao et al. (2015) suggested that the olivine crystallization pressure is c. 1 GPa, and the liquidus temperature T(P) ol/liq is constrained to ≥1491°C, resulting in T 1 ol/liq ≥ 1439°C. The mantle potential temperature T P is related to the olivine liquid temperature at 1 atm (T 1 ol/liq ) along the solidus line by the equation (Herzberg & Asimow, 2015): T p ¼ 1:049 Â T 1 ol=liq À ½0:00019 Â ðT 1 ol=liq Þ 2 þ ½1:487 Â 10 À7 Â ðT 1 ol=liq Þ 3 : Using T 1 ol/liq in the above equation, we obtain a T P of 1559°C. We also consider the uncertainty of the crystallization temperature in the calculation, yielding a propagated uncertainty in T P of c. þ95/-90°C. We therefore infer that the mantle potential temperature of the Dali picrites is ≥ 1559 þ95/-90°C, which is consistent with previous studies (c. 1620°C; Zhang et al. 2006). For comparison, Matthews et al. (2016) used a similar method to calculate a T P of 1480 þ37/-30°C for the Iceland mantle plume and 1318 þ44/-32°C for the MORB from the olivine-spinel crystallization temperatures. The mantle potential temperature of the Dali picrites is c. 200°C higher than that of normal MORB, and is therefore consistent with a plume-head origin (even if our method represents a simplification and neglects the uncertainty in water content, oxygen fugacity and the melt fractions). The mantle potential temperature T p is definitely higher than the olivine crystallization temperature, and T ol-spl = 1491°C is still much higher than the mantle potential temperature of the normal MORB T P (MORB) = 1318 þ44/-32°C).

Conclusions
1. The trace-element data of primitive olivine in Dali picrites are consistent with a peridotite-dominated source for the primary magma. 2. The high Ni content in olivine phenocrysts can be explained by the relatively low K D Ni ol/melt during partial melting in the deep mantle and the high K D Ni ol/melt during crystallization in the relatively shallow level. This model suggests that a significant contribution from a metasomatic pyroxenite to the formation of high-Ni olivine is not a prerequisite.