New Olivine Reference Material for In Situ Microanalysis

A new olivine reference material – MongOL Sh11‐2 – for in situ analysis has been prepared from the central portion of a large (20 × 20 × 10 cm) mantle peridotite xenolith from a ~ 0.5 My old basaltic breccia at Shavaryn‐Tsaram, Tariat region, central Mongolia. The xenolith is a fertile mantle lherzolite with minimal signs of alteration. Approximately 10 g of 0.5–2 mm gem quality olivine fragments were separated under binocular microscope and analysed by EPMA, LA‐ICP‐MS, SIMS and bulk analytical methods (ID‐ICP‐MS for Mg and Fe, XRF, ICP‐MS) for major, minor and trace elements at six institutions world‐wide. The results show that the olivine fragments are sufficiently homogeneous with respect to major (Mg, Fe, Si), minor and trace elements. Significant inhomogeneity was revealed only for phosphorus (homogeneity index of 12.4), whereas Li, Na, Al, Sc, Ti and Cr show minor inhomogeneity (homogeneity index of 1–2). The presence of some mineral and fluid‐melt micro‐inclusions may be responsible for the inconsistency in mass fractions obtained by in situ and bulk analytical methods for Al, Cu, Sr, Zr, Ga, Dy and Ho. Here we report reference and information values for twenty‐seven major, minor and trace elements.

The inner part of the xenolith (700 g) was crushed and sieved. Around 10 g of clean olivine fragments with grain sizes of 0.5-1 and 1-2 mm were hand-picked under a binocular microscope (Figure 1b). The fragments with a grain size of 1-2 mm were leached in 2 mol l -1 HCl for 2 h to remove surface contamination. Some olivine fragments were mounted in epoxy and polished for in situ microanalysis. A preliminary electron microprobe investigation of 240 polished fragments of olivine showed a fairly homogeneous composition for major (Fe, Mg, Si) and ten minor and trace elements (Na, Al, P, Ca, Ti, Cr, Mn, Co, Ni, Zn). Some micrometre-size mineral inclusions (one FeNi sulfide 60 lm in diameter, one Al spinel and several orthopyroxenes) were identified within analysed olivine fragments. Mounts with 120 fragments of polished olivine grains and four batches (0.5-1 g each) of clean olivine separates were distributed to six analytical laboratories for in situ and bulk analysis (Table 1)  Isotope dilution by ICP-MS: Analyses were performed at CODES Analytical Laboratories, University of Tasmania. Eight aliquots (combination of single grain and multiple grain aliquots) of whole olivine were picked and weighed on a high precision Satorius balance (d = 0.001 mg). Aliquots ranged in mass from~4 to~14 mg. Several international reference materials were also weighed for isotope dilution including BIR-1, BHVO-2 and DTS-2 (Jochum et al. 2005).
A 25 Mg and a 57 Fe isotope spike was added to each sample in a 7 ml beaker and precisely weighed on another balance (d = 0.01 mg). The 25 Mg spike was from Inorganic Ventures and is certified against NIST SRM 3131a for the concentration: 9.952 ± 0.058 lg ml -1 25 Mg. The certified abundances of Mg isotopes (certified by Oak Ridge National Laboratories) are as follows: 24 Mg: 0.00963, 25 Mg: 0.98814, 26 Mg: 0.00223. The total certified Mg concentration is 10.068 ± 0.059 lg ml -1 .
Between 4-6 g of 25 Mg spike and 0.5-2 g of 57 Fe spike were added to each sample depending on mass of sample and estimated mass fractions of Mg and Fe. This was done to ensure similar sample to spike ratios in the analyses. Olivine was digested using a HF-HNO 3 mixture and was heated at 110°C, and treated ultrasonically multiple times until the solution was visibly free of any olivine grain(s). The HF-HNO 3 mixture was then evaporated to dryness and refluxed several times in concentrated HNO 3 to ensure sample-spike equilibration and total digestion of the sample. Each sample was diluted into 2% HNO 3 to a final dilution of~35000, giving roughly 8.5 lg g -1 Mg and 2.2 lg g -1 Fe in solution for the olivine samples.
To better constrain the concentrations of Mg and Fe in the spike solutions, reverse isotope dilution was performed using high purity Fe 2 O 3 and MgO powders from Alfa Aesar.
These were dried in an oven at 80°C for several hours, then weighed into a Teflon digestion vessel and digested in Seastar HNO 3 (for the MgO) and Seastar HCl (for the Fe 2 O 3 ), and gravimetrically diluted to a final volume of 250 ml. Next, an aliquot from each bottle was diluted into a 100 ml vial to give similar Fe and Mg concentrations to those expected from the olivine solutions and this was spiked with the 25 Mg and 57 Fe spike.
Samples were analysed using an Agilent 7700 ICP-MS with a collision cell and helium gas to remove polyatomic species. To avoid any potential complications with different detector modes (pulse vs. analogue counting), the isotopes of Fe and Mg were collected only in the analogue mode of ion detection, since count rates were > 1 Mcps, and to avoid any issues in changing ion detection modes on the isotope ratios. Data were collected in fifteen replicates with 200 sweeps of the quadrupole per replicate and took about 4.5 min per analysis. Solutions of pure Fe and Mg were measured throughout the analysis to correct for instrument mass bias (exponential law used) and monitor drift in the isotope ratios. No instrument drift was observed for either the Mg or Fe isotope ratios.
Isotope dilution results were calculated based on the 24 Mg/ 25 Mg and 56 Fe/ 57 Fe ratios, and concentrations calculated using the reverse isotope dilution results from the Mg and Fe spike solutions. Errors were propagated from the counting statistic errors on the analyses, error on the concentrations of Mg and Fe in the spikes (from the reverse isotope dilution) and error on the fractionation factor.
Minor and trace element measurement by ICP-MS: Measurements were performed at four laboratories: ISTerre, CODES Analytical Laboratories, JAMSTEC and CAU (Table 1). The details of the instruments, analytical conditions and reference materials' reproducibility are summarised in Table 2 and Table S1.
-ISTerre, University Grenoble Alpes. Olivine fragments with grain size 0.5-1 mm were leached in 2 mol l -1 HCl for~2 h and powdered in an agate mill. Three separate 20-30 mg aliquots were dissolved in Parr bombs. Five measurements of each dissolution were made. The details of the analytical method are given in Chauvel et al. (2011). The ICP-MS signal was calibrated relative to the BHVO-2 contents compiled in Chauvel et al. (2011); individual element mass fractions were calculated using a BHVO-2 doped in Ni and a dilution of 5000 except for Al, which was calculated using a BHVO-2 not doped in Ni and with a dilution of 20000. Rock reference materials (BR24,BEN, 4 5 5 were run as unknowns during the same measurement session. Results are provided in online supporting information Table S1. Values for BEN are from Chauvel et al. (2011) andJochum et al. (2016); values for UB-N are from Chauvel et al. (2011).
-CODES Analytical Laboratories, University of Tasmania. Three aliquots (two of 0.5-1 mm size and one of 1-2 mm size) weighing~40 mg each of olivine were digested and analysed five times each. To minimise contamination, all grains were inspected under an optical microscope and only the cleanest grains without identifiable inclusions were selected for digestion. The grains were then leached in 1 mol l -1 HCl for~5 min to remove any surface contamination and then rinsed in DI H 2 O several times. The samples were digested in HF-HNO 3 (2 and 1 ml, respectively) mixture on a hot plate (pre-cleaned Savillex Teflon) for 24 h at 110°C.  Makishima et al. (1999Makishima et al. ( , 2002, Makishima and Nakamura (2006), others from Jochum et al. Samples were then dried, and concentrated HNO 3 was added and evaporated to dryness several times. Samples were then reconstituted in 4 mol l -1 HNO 3 and diluted to give a 2% HNO 3 solution and a 10009 dilution. The solutions were analysed using an Agilent 7900x instrument, with He as the collision gas, in timeresolved data acquisition mode. The primary calibration was done using a mixture of pure-element solutions and was forced through the origin with 25 Mg and 115 In used for internal calibration. The following secondary reference materials were analysed in the measurement session: BIR-1, W-2, JP-1 and DTS-1 (Table S1). DTS-1 and JP-1 were digested using Parr bombs at 210°C for 24 h to fully dissolve chromite, while other reference materials were digested using same procedure as for the olivine. Olivine data were corrected to rock reference materials BIR-1, W-2 and DTS-1 (the latter only for high abundance elements), while other reference materials were treated as unknowns. Since the analysis was done using a collision cell, no correction was made for 30 SiH interference on 31 P or for SiO interference on 45 Sc (Robinson et al. 1999, Yu et al. 2000, Norman et al. 2003 since it is expected that most of the Si was lost during the initial evaporation to dryness of the HF-HNO 3 for digestion. All reagents used were Seastar purity and Milli-Q DI water. -Japan Agency for Marine-Earth Science and Technology (JAMSTEC). Olivine grains were inspected under optical microscopy and only the cleanest grains lacking identifiable inclusions were picked for digestion. The grains were then leached in 1 mol l -1 HCl for~1 h to remove any surface contamination. One aliquot (0.5-1 mm size) of the olivine was digested and measured twice. Olivine grains were weighed in a 23 ml PFA Teflon vial. After adding concentrated HClO 4 /HF (v/v: 25/75), the vial was capped tightly and placed on a hot plate at 130-140°C for 3 days. HClO 4 instead of HNO 3 was used in this step, as it produces a more effective digestion of refractory minerals by improving the efficiency of the HF. The sample was then evaporated to incipient dryness to remove volatile SiF 4 . Concentrated HClO 4 was added again, and the vial was closed and placed on a hot plate at 160°C for 1 day, then opened to dry the sample at a gradually increasingly temperature of up to 190°C, to drive out excess HF and to convert fluorides into chlorides. The residue was refluxed with 2 ml 6 mol l -1 HNO 3, moderately heated for 2 h and then dried down at a temperature of 120°C to incipient dryness. The final sample residue was dissolved in 5 ml 2% HNO 3 and diluted to 2000 times for trace elements and 20000 times for major elements prior to analyses. An Agilent 7500ce in normal nebulisation mode was used for analysis. Element standard solutions (SPEX) were used to generate calibration curves. Isobaric overlap correction factors were determined using synthetic solutions. Reference materials JP-1, JB-2 and BIR-1 (Table S1) were analysed together, and the results exhibit reasonable fits with the reference values (Chang et al. 2003, Nakamura andChang 2007). Since 45 Sc was within 9.6% RD for JP-1, no isobaric overlap correction was made.
-CAU Institute of Geosciences, Kiel. Two aliquots (0.5-1 and 1-2 mm grain size) of olivine grains were weighed in triplicate within 50 mg into 15 ml PFA (perfluoralkoxy) vials. After addition of mixed concentrated acids (HF-HNO 3 -HCl) samples were digested on a hot plate overnight, and the resulting digest solutions were repeatedly evaporated to dryness and finally taken up in 20 ml of 3% v/v sub-boiled nitric acid. Prior to analysis, digest solutions were diluted twofold and spiked with 2.5 ng g -1 Be, In, Re for internal standardization. Subsequent analysis was done by ICP-MS using an Agilent 7500cs instrument in standard mode after calibration with freshly prepared multi-element standards (Garbe-Sch€ onberg 1993). Results were blanksubtracted means of three runs. Analytical quality was monitored with procedural blanks during both digestion and sample preparation for analysis, and three replicate measurements of one olivine digest solution were used for assessing measurement precision. Basalt reference materials BIR-1 and BHVO-2 were used as secondary reference materials (Table S1) for checking the accuracy of the calibration and applying correction factors where necessary.
X-ray fluorescence spectrometry: XRF analyses were performed at CODES Analytical Laboratories, University of Tasmania on a PANalytical Axios Advanced WDS spectrometer using standard operating conditions. This technique was used to determine major and some minor elements. Two lithium borate fusion discs were made from the MongOL Sh11-2 olivine, consisting of 0.2 g of olivine, 0.3 g of high purity SiO 2 , 4.5 g of 12/22 lithium borate flux (mixture of lithium metaborate and tetraborate) and 0.0606 g of LiNO 3 . Each disc was analysed fifteen times, and greater variability between discs was seen than between repeated measurements with the RSD < 1% for all elements that were > 0.1% oxide in mass fraction. The olivine sample was diluted with the high purity SiO 2 due to the limited amount of the olivine sample. This had the advantage of bringing Mg and Ni values within the range of the instrument calibrations; however, minor elements such 4 5 7 as Ca, Al and Co were diluted to the point where they were close to the detection level. To assess the accuracy of the method, multiple (at least two) discs of several ultramafic reference materials (JP-1, PCC-1, DTS-1, DTS-2) were prepared with the same dilution of 0.2 g sample and 0.3 g SiO 2 . A secondary correction was applied to the average of these ultramafic materials (from the primary calibration of the instrument) to account for: (a) the slight drift in the calibration for some elements with time and (b) any matrix effects of diluting the samples with the SiO 2 .

In situ analytical techniques
Electron probe microanalysis: This technique was used at two laboratories to determine the mass fractions of major, minor and some trace elements, and to perform a major element homogeneity test of olivine fragments. Analyses were made on two different instruments (JEOL JXA-8230 and Cameca SX100) using different procedures for matrix correction, both laboratories used San Carlos olivine (USNM111312-44 (SCOL), Jarosewich et al. 1980) as a control reference sample (Table 3).
-ISTerre. Over 240 fragments of olivine with sizes from 0.5 to 2 mm were analysed in polished epoxy mounts using JEOL JXA-8230 electron probe using the trace element analytical method of Batanova et al. (2015) ( Table 3). Accelerating voltage and probe current were 25 kV and 900 nA. The beam diameter was 2 lm. The ZAF correction procedure was applied to correct for matrix compositional effects. San Carlos olivine (USNM111312-44 (SCOL), Jarosewich et al. 1980) and ISTerre internal XEN olivine (Batanova et al. 2015(Batanova et al. , 2018 were run as unknowns three times after every batch of 30-40 measurements, in order to monitor potential instrumental drift and to estimate accuracy and precision. Additionally, ten grains were analysed using a 5 9 5 grid with a step from 100 to 300 lm. -Central Science Laboratory, University of Tasmania. Analyses were performed on Cameca SX100. Operating conditions were as follows: accelerating voltage 20 kV; beam current 30 nA; beam diameter 5 lm. Calibration was performed using simple oxide reference materials (periclase for MgO, spectrosil for SiO 2 , Smithsonian magnetite for Fe) and the 'Probe for EPMA' software (Probe Software, Inc.) with the Armstrong-Love-cott matrix correction method.
Laser ablation-ICP-MS: Minor and trace element analyses were performed at four LA-ICP-MS laboratories (Table 1) CODES Analytical Laboratories, JAMSTEC, CAU and CIW. An overview of the instruments, measurement conditions, reference materials and approaches to  Ar, 14 l min -1 Ar, 13 l min -1 15 l min -1 Ar, 14 l min -1 Auxiliary gas flow Ar, 0.8 l min -1 Ar, 0.7 l min -1 0.85 l min -1 Ar, 0.6 l min -1 Carrier gas flow Ar, 1.05 l min -1 Ar, 1.0 l min -1 0.85 l min -1 Ar Ar, 1.0 l min -1 Sample cone Pt cone Normal (Ni) Ni (1 mm quantification is presented in Table 4. Three laboratories used 193-nm Excimer lasers and one used 266-nm femtosecond laser operated at 10 Hz. The same 120 olivine fragments were analysed in each laboratory. -CODES Analytical Laboratories. The laser microprobe was a RESOlution S-155 instrument equipped with a coherent 193 nm excimer laser of 20 ns pulse width. Ablation was performed at a fluence of 10 J cm -2 , with a 70 lm beam at 10 Hz. A 90 s ablation was preceded by a 30 s gas blank. A pre-ablation of five laser pulses was done prior to each analysis, and a 20 s washout between analyses was used. An Agilent 7900 ICP-MS was coupled to the laser and tuned for ThO/Th < 0.2 and U/Th of 1-1.05 using a line ablation of the NIST SRM 612 glass. The dwell times ranged from 5 to 20 ms depending on expected abundance of isotopes in olivine giving a total sweep time of 0.67 s. Gas flows were 0.35 l min -1 He through the ablation cell, which was mixed with Ar flowing at 1.05 l min -1 immediately after the ablation. The signal from the ablation cell was smoothed using the 'squid' signal-smoothing device (M€ uller et al. 2009). Calibration was performed on NIST SRM 612 for all elements except Fe and P, for which BCR-2G was used. 25 Mg was used as the internal standard element.
Calibration reference materials were analysed twice after every ten analyses of the unknowns, using the same conditions as the unknowns. Data reduction was done using an in-house macro-based Excel workbook. Matrix effects were assessed and corrected by analysing BCR-2G and GSD-1G as secondary reference materials under the same conditions as the unknowns, after every ten analyses of the unknowns. All mass fractions for NIST SRM 612, BCR-2G and GSD-1G were taken from the GeoReM preferred values. Quantification was performed using conventional approaches (Longerich et al. 1996), with normalisation to 100% total of oxide components. A correction was applied for the 30 Si 1 H and 29 Si 16 O interference on 31 P and 45 Sc, respectively, by analysis of high purity Spectrosil silica glass analysed throughout measurement sessions. The correction was~0.7% and2 .2% for 31 P and 45 Sc, respectively. Total uncertainty of the reported mass fractions includes uncertainties of correction for instrumental drift during the session, matrix correction and the published values for secondary reference materials.
-JAMSTEC. A 266-nm wavelength, < 170 fs pulse width, 12 J cm -2 fluence laser pulse at 10 Hz was applied using an OK-Fs2000K laser ablation system (OK Lab, Tokyo, Japan) equipped with a Solstice onebox Ti: Sapphire 800-nm fs regenerative amplifier with TP-1A THG frequency tripling harmonic generator (Spectra-Physics, Santa Clara, CA, USA). The beam diameter was set at 90 lm and a circular raster protocol (15 lm radius, 10 lm s -1 raster velocity) was performed using a high precision sample translation stage to obtain a flat-bottomed crater of~100 lm diameter and 50 lm depth after 60 s of ablation. An Element XR sector field ICP-MS (Thermo Fisher Scientific, Bremen, Germany) was modified by an additional high-efficiency interface vacuum pump and operated with N-sampler H-skimmer cones and guard electrode (GE) disconnected to obtain low oxide ThO + /Th + < 0.2% and U/Th =~1.05 values. The laser aerosol carrier gas was He (at 1.2 l min -1 ), which was mixed with Ar sample gas (at 1.0 l min -1 ) in a mixing chamber (70 cm 3 inner volume) prior to the ICP torch. Analyses were performed in time-resolved mode with 20 s for gas blank, 60 s of LA signal acquisition and 80 s washout using gas blanks 15 s before ablation and after washout. Mass scan speed was~2.6 s per cycle and acquisition was made in low-resolution mode (M/DM = 400) using a dualmode ion counter for both major and trace elements. The reference material used was the USGS basaltic glass BHVO-2G for all the elements and was analysed before and after each five unknowns for calibration and drift correction. Laser ablation efficiency was corrected using total-100% normalisation using major oxides. The details of the interference corrections are in Table 4. Reference material glass BCR-2G was monitored for repeatability and reproducibility tests (Kimura and Chang 2012).
-CAU. Analyses were performed with an Agilent 7500s quadrupole mass-spectrometer coupled to a 193 nm excimer laser ablation system (GeoLas Pro; Coherent, G€ ottingen, Deutschland) using a 90 lm laser spot, a pulse frequency of 10 Hz and laser fluence of 10 J cm -2 . All analyses were performed in a large volume 'Z€ urich' ablation cell. The carrier gas was He (1 .05 l min -1 ) with addition of H 2 (14 ml min -1 ), which were mixed with Ar (0.85 l min -1 ) before introduction into the mass spectrometer. Oxide production rate, estimated as [ThO] + /[Th] + , was < 0.3%. Analyses were performed in time-resolved mode and included 20 s background measurement followed by 20 s sample ablation and signal measurement. Dwell time was 20 ms for all elements. Scan speed was 0.68 s per cycle. All spectra were processed with GLITTER software. Mass fractions were quantified from the measured ion yields normalised to 25 Mg, 29 Si, 57 Fe and Mg, Si and Fe mass fractions from EPMA microprobe data. The data obtained using different reference elements were averaged. Analyses of MPI-DING reference glass KL2-G as well as one of glasses GOR-128G, GOR-132G and BM90/21G were performed every twenty olivine analyses and used for calibration and drift correction (Jochum et al. 2005(Jochum et al. , 2006. Isobaric interference of 29 Si 16 O on 45 Sc was monitored and corrected by using data from Sc-free synthetic optical-grade quartz, which was measured together with reference glasses every twenty analyses. The details for the interference corrections are in Table 4. Typical Si oxide production rate on mass 45 was 0.035-0.065% (n = 14). In the absence of Al-free reference samples, Zn mass fractions were quantified from the calibration using Al-bearing reference glasses corrected for the interference of 67 Zn with 27 Al 40 Ar. Matrix correction was applied for Al and Ca mass fractions based on analyses of an in-house (reference) pressed nanopowder of San Carlos olivine characterised previously by ICP-MS and EPMA.
-CIW. A Photon Machines 193 nm ArF excimer laser was coupled to a Thermo iCapQ quadrupole mass spectrometer. Analytical conditions were as follows: 7 mJ laser energy; 10 J cm -2 fluence; 50 lm diameter laser beam; 10 Hz repetition rate. Each analysis involved five laser shots pre-ablation followed by a wash-out of 40 s, 25 s of data acquisition of gas background (laser off)  Table 4. Calibration reference materials included MPI-DING and USGS glasses (Jochum et al. 2005(Jochum et al. , 2006. The San Carlos olivine (USNM111312-44; Jarosewich et al. 1980) was used as a secondary reference material to correct for instrumental drift, and to perform secondary reference material corrections using the observed differences between the measured and accepted values.
Secondary ionisation mass spectrometry: Analyses were performed at CIW on a Cameca IMS 6F ion microprobe using energy filtering techniques (Shimizu and Hart 1982). The primary Oion beam had a current of 15 nA; the crater diameter was 30 lm. A field aperture was applied to mask surface contamination. A mass resolution power of 3500 was sufficient to resolve 29 SiO from 45 Sc, but not enough to resolve Ca dimers from 88 Sr, 89 Y and 90 Zr. The elements were determined using the ratio of their chosen isotope to 30 Si. Each measurement was preceded by 5 min of pre-sputtering. The masses measured were as follows: 7 Li, 9 Be, 11 B, 23 Na, 26 (Jochum et al. 2005(Jochum et al. , 2006. The San Carlos olivine (USNM111312-44; Jarosewich et al. 1980) was used as a secondary reference material to correct for instrumental drift.

Assessment of the homogeneity of olivine fragments
Chemical homogeneity can be defined as variation in element mass fraction, which does not exceed the measurement uncertainty of the analytical method (e.g., Boyd et al. 1967, Jarosewich et al. 1980, Potts et al. 1983, Jochum et al. 2000, Gilbert et al. 2013, Harries 2014. As suggested by the key international guide for the characterisation of reference materials (ISO Guide 35:2017), to examine homogeneity of olivine fragments in major elements (Si, Mg, Fe), the F-test for comparison of two population variances was applied. The three sets of EPMA measurements of MongOl Sh11-2 (each containing 36-95 separate fragments) were compared with repeated measurements (9-24) of single fragments of the San Carlos olivine USNM111312-44 (Jarosewich et al. 1980) that were run together with each set. The results of the F-test indicated that the standard deviations of all three populations of the different fragments of MongOl Sh11-2 are equal at the 95% confidence level to the standard deviations of the population of analyses of a single fragment of San Carlos olivine USNM111312-44 for all major elements (Table S2).
Additionally, in this study we used the homogeneity index (H) to assess homogeneity of minor and trace elements. H represents the ratio of the measurement uncertainty to the expected value of the total combined uncertainty (e.g., Boyd et al. 1967, Harries 2014, Pankhurst et al. 2017. A value of 1 for the index implies that the sample is homogeneous within the analytical uncertainty of individual measurements. A value > 3 for the index indicates significant chemical heterogeneity (e.g., Boyd et al. 1967, Jarosewich et al. 1980, Potts et al. 1983, Harries 2014, Pankhurst et al. 2017. The H value can be considered as a particular case of an F-test when the degree of freedom of each population approaches infinity (Harries 2014).
The assessment of homogeneity for minor and trace element was made using LA-ICP-MS data on 120 individual grains obtained at CODES Analytical Laboratories (Figure 2). The average within-run analytical uncertainty of individual measurements includes the uncertainties of the signal on the sample and reference materials, matrix correction and uncertainty related to the instrument drift during the session (e.g., Gilbert et al. 2013).
The observed variations in the uncertainty of individual analyses are due primarily to differences in element mass fractions (lower mass fractions result in higher signal noise) and isotopic abundance. Only phosphorus showed significant heterogeneity with a homogeneity index of 12.    Several elements showed minor inhomogeneity with homogeneity indices of < 2: Li (1.5), Na (1.8), Al (1.6), Sc (1.3), Ti (1.6), Cr (1.6) and Sr (1.3) (Figure 1b). All other elements were found to be homogeneous within the analytical uncertainty.

Measurement results and suggested reference values
A total of over 1020 in situ analyses were performed in this study on 120 olivine grains by EPMA, LA-ICP-MS and SIMS. Eight aliquots were analysed by ID-ICP-MS, nine aliquots were analysed by solution ICP-MS and two aliquots by XRF. Tables 5-7 list all analytical results provided by each laboratory, including analytical uncertainties expressed as two relative standard deviations in per cent (2 RSD), displaying the dispersion of the data (the measured reproducibility). Additionally, in Table 7 we show the overall analytical uncertainty (U) that includes instrumental repeatability, calibration errors and uncertainty of reference materials. Consistency of the data obtained in different laboratories by different analytical techniques is considered a measure of data quality. The preferred reference values are in Table 8, which shows statistical uncertainty: 2stwo standard deviation and 2SEtwo standard deviation of the mean [sometimes incorrectly called 'standard error' (GUM 2008, Potts 2012, which corresponds to the 95% confidence level]. The latter is also expressed in relative % (Table 8).

Major elements (Si, Mg, Fe)
Measurement results for major elements (Si, Mg, Fe) are summarised in Table 5. For Fe and Mg, the isotope dilution analysis using ICP-MS is considered as the primary method with the highest metrological properties (Jochum et al. 2016). EPMA data obtained by both laboratories and XRF data are in good agreement with ID-ICP-MS and thus have been included in determination of reference values. The calculated reference values are in Table 8.

Minor and trace elements
To produce reference values, the data were treated with a filtering procedure proposed by the European Commission IRMM (Application Note 1 2010). This procedure considers the uncertainty of individual measurement results.
For each element, a 'global' average (C glob ) and standard deviation of the mean (U gl ) were first calculated for the values obtained by different techniques in different laboratories. We then compared this 'global' average with Table 6 (continued).
Minor and trace element measurement results  (Figures 3-5). The absolute difference between the mean value and 'global average' was calculated as: The uncertainty of Δ m (U Δ ) was calculated from the uncertainty of the 'global average' and uncertainty of the measurement result U (Table 7): A value is accepted if Δ m ≤ UΔ, if Δ m > UΔ the value is considered an outlier and is discarded (Figures 3-5). The preferred reference values were calculated as the mean of consistent values obtained by different methods in different laboratories (Table 8). Na, Al, P, Ca, Ti, Cr, Mn, Co, Ni and Zn For this group of ten elements with mass fraction levels greater than 10 lg g -1 , nine to eleven values were used to calculate the 'global' averages. Depending on the method used, a value represents either the mean of 120 individual analyses (for the in situ analytical methods) or the mean of 3-8 analyses (for the bulk analytical methods).
Ti, Mn, Cr, Co, Ni, Zn and Na: These seven elements show uniform distribution and good consistency between the mean values obtained by different microanalytical techniques and solution ICP-MS ( Figure 3). The accepted data for Mn, Cr, Co, Ni and Zn agree within 1-3% (2 RSDrelative standard deviation), whereas the accepted data for Ti and Na agree within 6-8% (2 RSE; Table 8). Only one such inclusion was observed at the surface of a polished olivine from more that 200 olivine fragments studied by SEM ( Figure S1). This small spinel is colourless and is therefore almost impossible to detect optically inside olivine. Our data show that the presence of rare micro-inclusions of spinel containing 57-58% w/w of Al 2 O 3 may be responsible for the increased aluminium content in the powders prepared for solution ICP-MS measurement. However, such

6 6
an inclusion is easily visible in a polished olivine fragment in reflected light or under electron beam and thus can be easily avoided in microanalysis. The calculated value for Al is 245 ± 13 lg g -1 .
-Calcium: Contents of Ca show large discrepancies within LA-ICP-MS and solution ICP-MS data obtained in different laboratories (Figure 4). In contrast, EPMA data for Ca (Lab 1, Lab 6) obtained on different instruments using different matrix correction methods and different sets of primary reference materials are in excellent agreement and are also consistent with the XRF results ( Figure 4). Given that ICP-MS-based methods have to use minor Ca isotopes that are subject to O -, OHand Nbased  (2) Na (µg g -1 ) Ti (µg g -1 ) Mn (µg g -1 ) Cr (µg g -1 ) Co (µg g -1 ) Ni (µg g -1 ) Zn (µg g -1 ) Al (µg g -1 ) blue horizontal field corresponds to two standard deviations of the mean (standard errors). The red thick horizontal lines represent the reference values derived in Table 8, and red field corresponds of two standard deviations of mean (standard error) of the reference value. The error bars for each value correspond to overall uncertainty of analytical method (2U) provided by each analytical laboratory.

6 8
interferences from Mg and Si, we used EPMA and XRF data only to derive the reference value for Ca (Table 8), which fit almost all measurements within their errors (Figure 4). To be conservative we assign 5% (2 RSE) to this value.
-Phosphorus: This element is heterogeneously distributed between olivine fragments and shows the highest uncertainties of the mean values (12-29% 2 RSD of individual analysis; Table 6). Solution ICP-MS data were provided only by one laboratory and show values that are higher than reported by other techniques (Figure 3). The mean has 2 RSD of 11%, and we consider this as an information value.

Sc, V, Cu, Ga and Li
For this group of elements with mass fractions between 0.1 and 10 lg g -1 , LA-ICP-MS, SIMS (except Ga) and solution ICP-MS data are available (Table 6, Figures 4 and 5). In general, data for Sc, V, Li and Ga are consistent between different laboratories and methods; however, in each case one or two outliers were observed (Figures 4 and 5).

6 9
Yttrium Four data sets of both LA-ICP-MS and solution ICP-MS analyses are available ( Figure 5) and, with the exception of two outliers, show an overall mean of 0.079 (± 0.005) lg g -1 (Table 8).

Strontium and zirconium
Solution ICP-MS data shows strong variation between laboratories and values are up to several orders of magnitude greater than LA-ICP-MS. This suggests that some sort of contamination occurred with the solution analyses. Some of this could be due to acid blanks or 'carryover' effects; however, we believe that the most likely explanation is the possible presence of secondary melt/fluid micro-inclusions in olivine, which contain these incompatible elements. The mean of four LA-ICP-MS data sets for Zr is 0.044 ( ± 0.005) lg g -1 and for Sr is 0.007 (± 0.0004) lg g -1 .
Dy, Ho, Er, Tm, Yb, Lu The mass fractions of these high atomic number REE are given in Table 7. For dysprosium, the solution ICP-MS data were excluded from consideration for the same reason as for Sr and Zr (see discussion above). For other HREE, the data sets of solution ICP-MS and LA-ICP-MS are in good agreement with each other.

Conclusions
The fragments of natural olivine separated from the inner part of a xenolith of mantle spinel lherzolite (MongOL Sh11-2) are sufficiently homogeneous to be used as a matrix-matched reference material for in situ microanalysis of olivine by EPMA, LA-ICP-MS and SIMS. Some 120 olivine fragments were studied, involving > 1020 in situ analyses by EPMA, LA-ICP-MS and SIMS. In addition, eight aliquots were analysed by ID-ICP-MS, nine aliquots were analysed by solution ICP-MS, and two aliquots by XRF. Analyses were performed in six different analytical laboratories. Wellcharacterised reference values were obtained for major elements (Si, Mg, Fe), minor elements (Ni, Mn) and trace elements (Li, Na, Al, Ca, Sc, Ti, V, Cr, Co, Cu, Zn, Y, Er, Tm, Yb and Lu).

7 0
Significant heterogeneity was detected for mass fraction of phosphorus (heterogeneity index 12.4). Minor heterogeneity (heterogeneity index of 1-2) was also detected for contents of Li, Na, Al, Sc, Ti and Cr.
The data obtained for Ga, Sr, Zr, Dy and Ho are considered information values because these elements show significant inconsistency between mass fractions obtained by LA-ICP-MS and solution ICP-MS. We interpret this as an indication of the possible presence of melt/fluid microinclusions in olivine, which affect solution ICP-MS data.