Oxygen Fugacity of Global Ocean Island Basalts

Mantle plumes contain heterogenous chemical components and sample variable depths of the mantle, enabling glimpses into the compositional structure of Earth's interior. In this study, we evaluated ocean island basalts (OIB) from nine plume locations to provide a global and systematic assessment of the relationship between fO2 and He‐Sr‐Nd‐Pb‐W‐Os isotopic compositions. Ocean island basalts from the Pacific (Austral Islands, Hawaii, Mangaia, Samoa, Pitcairn), Atlantic (Azores, Canary Islands, St. Helena), and Indian Oceans (La Réunion) reveal that fO2 in OIB is heterogeneous both within and among hotspots. Taken together with previous studies, global OIB have elevated and heterogenous fO2 (average = +0.5 ∆FMQ; 2SD = 1.5) relative to prior estimates of global mid‐ocean ridge basalts (MORB; average = −0.1 ∆FMQ; 2SD = 0.6), though many individual OIB overlap MORB. Specific mantle components, such as HIMU and enriched mantle 2 (EM2), defined by radiogenic Pb and Sr isotopic compositions compared to other OIB, respectively, have distinctly high fO2 based on statistical analysis. Elevated fO2 in OIB samples of these components is associated with higher whole‐rock CaO/Al2O3 and olivine CaO content, which may be linked to recycled carbonated oceanic crust. EM1‐type and geochemically depleted OIB are generally not as oxidized, possibly due to limited oxidizing potential of the recycled material in the enriched mantle 1 (EM1) component (e.g., sediment) or lack of recycled materials in geochemically depleted OIB. Despite systematic offset of the fO2 among EM1‐, EM2‐, and HIMU‐type OIB, geochemical indices of lithospheric recycling, such as Sr‐Nd‐Pb‐Os isotopic systems, generally do not correlate with fO2.


Introduction
In planetary systems, oxygen fugacity ( fO 2 ) is a chemical parameter that affects the speciation, geochemical behavior, and physical distribution of multivalent trace elements among key chemical reservoirs, such as the Earth's metallic core, rocky mantle and crust, liquid water ocean, and gaseous atmosphere.For example, mantle fO 2 influenced the speciation of volcanic gases that made up the Earth's early atmosphere (French, 1966;Hirschmann, 2012;Kump et al., 2001).Mantle fO 2 and dynamics may be linked to the oxygenation of the atmosphere (Andrault et al., 2018;Kadoya et al., 2020;Kasting et al., 1993;Ortenzi et al., 2020;O'Neill & Aulbach, 2022).Plate tectonics have been invoked to explain why Earth is more oxidized and has more fO 2 variability compared to Mars (Righter & Drake, 1996).Rocks from Earth's surface are mixed back into the mantle, potentially modifying and regulating the fO 2 of the interior (e.g., Evans, 2012;Kasting et al., 1993;Lécuyer & Ricard, 1999).If recycling of lithosphere is responsible for regulating and/or modifying mantle fO 2 , there may be a link between plate tectonics, mantle fO 2 , and planetary habitability (Cockell et al., 2016).
To probe the roles of distinct Earth materials in the evolution of fO 2 in the planet's interior, this study systematically explores the fO 2 of a diverse global suite of plume-derived lavas from the Canary, Samoa, La Réunion, Hawaii, Azores, Pitcairn, St. Helena, and Macdonald hotspots (Figure 1).Mangaia and the Austral Islands of Rapa Iti and Raivavae, which are all products of the Macdonald hotspot, are considered individually because the lavas from Mangaia have distinctly high fO 2 (and loss on ignition, indicating high degrees of alteration and warranting cautious interpretation).Despite forming part of a continental flood basalt province, lavas from Baffin Island are also included in this survey because they sample the proto-Iceland plume and offer useful insights relative to OIB due to their high-3 He/ 4 He-a signature of ancient mantle preservation (Starkey et al., 2009;Willhite et al., 2019).
This study leverages the olivine-melt partition coefficient of redox-sensitive V D ol/melt V ) to characterize the fO 2 of global mantle components, including depleted mantle, EM1, EM2, and HIMU.Vanadium primarily exists as a +3 or +4 cation in terrestrial magma systems (Borisov et al., 1987;Gaetani & Grove, 1997).Under oxidizing conditions, a higher proportion of V exists at the higher valence state, resulting in a net change in the size-tocharge ratio that renders V largely incompatible in olivine during melt crystallization (Canil, 1997).By contrast, under reducing conditions V 3+ can substitute more readily for Mg 2+ and Fe 2+ in the olivine lattice.Experimental studies have shown that V partitioning between olivine and melt is relatively insensitive to bulk  composition, pressure, and temperature during basalt petrogenesis (Canil, 1999;Canil & Fedortchouk, 2001;Righter, Leeman, & Hervig, 2006, Righter, Leeman, & Hervig, 2006;Suzuki & Akaogi, 1995;Wang et al., 2019).Olivine is an early crystallizing phase in magmatic systems and V is relatively immobile during metamorphism (Condie, 1976).Due to these characteristics, fO 2 signatures recorded by the most primitive (i.e., most Mg-rich and earliest crystallizing) olivine are representative of the original magma and resistant to post-magmatic processes, such as low-to-moderate degrees of oxide accumulation (<5 modal %), sulfur saturation, degassing, or metasomatism (Locmelis et al., 2019).

OIB in This Study
The 56 rocks studied here compose a global sample suite of OIB from nine mantle-plume-associated hotspots: Azores, Canary, Hawaii, Iceland (proto-Iceland plume lavas from Baffin Island), Macdonald (including Mangaia and the Austral Islands of Raivavae and Rapa Iti), Pitcairn, La Réunion, Samoa, and St. Helena.Samples were selected based on the availability of previously published major, minor, and trace element data as well as isotopic compositions (see Tables S1 and S2, respectively).Primitive samples with limited evidence of pyroxene fractionation were targeted.Forty-two samples have whole-rock MgO contents greater than 9.0 wt.%; seven samples from Baffin Island are vitrophyres with MgO wt.% between 7.9 and 9.4, and three samples (KOO-01; PIT-3, CE-13) are slightly more evolved (5.2-7.7 wt.% MgO).Four lavas do not have published whole-rock MgO wt.%.Both subaerial and submarine lavas are included (Table 1).Photomicrographs and scans of epoxy-mounted rock fragments are provided for a subset of the lavas (Figures S1 and S2 in Supporting Information S1).

Olivine and Basaltic Matrix Major Element Analyses
Basaltic samples were cut, mounted in one-inch diameter epoxy mounts, and abraded with alumina powder (down to 1 μm particle size) so that visible olivine crystals were exposed and polished.The polished samples were cleaned via Milli-Q water (18.2MΩ•cm) in an ultrasonic bath, and carbon coated for electron probe microanalysis (EPMA) using the JEOL 8900R electron microprobe in the Advanced Imaging and Microscopy Laboratory at the University of Maryland, USA.All analyses were performed using a 15 kV potential and 20 nA current measured at the faraday cup.Matrix and olivine analyses were acquired using a 10 and 2 micron diameter beams, respectively.
To determine the major element composition of basaltic matrices, the crystalline groundmass of each sample was analyzed using four lines comprising 10 equally spaced analytical spots.Analytical uncertainty was typically less than 2%.Care was taken to avoid phenocrystic olivine during the measurement of the matrix composition in order to determine the V partitions between olivine and the melt.Primary standards for the matrix analyses included Makapuhi Lava Lake basalt glass (USNM: VG-99), Indian Ocean basalt glass (USNM: 113716) and Broken Hill rhodonite (USGS PXBX).
Depending on the size of the phenocryst, each olivine grain was analyzed in one to three sites located away from features such as microfractures, reaction rims, and/or inclusions when observed; each site was characterized by four independent spot measurements.When only a single site was analyzed, the olivine core was sampled.For grains that were large enough, the core and rim were targeted to investigate chemical gradients in zoned olivine, though few samples exhibited appreciable differences in V concentrations between the core and rim.Primary standards for the olivine analyses included San Carlos (USNM 111312/444) and Rockport (USNM 85276) olivines, and Kakanui hornblende (USNM 122142).

Olivine and Basaltic Matrix Trace Element Analyses
First-row transition elements (FRTE; Sc, Ti, V, Cr, Mn, Fe, Co, Ni, Cu, Zn) and Ga and Ge were measured in olivine grains (n = 302 grains) and sample matrix (i.e., crystalline groundmass) using laser ablation (LA-) and medium-resolution inductively coupled plasma mass spectrometry (ICPMS) using either the New Wave UP213 laser system coupled to the Thermo Fisher Scientific Element 2 mass spectrometer housed in the Plasma Lab at the University of Maryland, or the Photon Machines Analyte G2 laser system coupled to the Nu AttoM mass spectrometer housed in the Planetary Environments Lab at NASA Goddard Space Flight Center, Maryland, USA.Each sample was characterized following the analytical protocol of Arevalo et al. (2011), whereby multiple reference materials (i.e., USGS basaltic reference glasses BHVO-2G, BIR-1G and BCR-2G) were used to build a sensitivity calibration curve, rather than relying on only a single bracketing standard to quantify elemental abundances.The measured isotopes for each element were 45 Sc, 47,49 Ti, 51 V, 52,53 Cr, 55 Mn, 56,57 Fe, 59 Co, 60,62 Ni, 63,65 Cu, 66,67,68 Zn, 69,71 Ga, 72,73,74 Ge, 75 As, 77 Se, and 43 Ca as the internal standard.Laser parameters used were 2-3 J/cm 2 fluence, 10 Hz repetition rate, and a spot size between 150 and 250 microns in diameter to maximize count rates.The plasma source of the mass spectrometer was tuned to maximize ionization, as monitored by 43 Ca and 232 Th count rates, while maintaining limited oxide production ( 232 Th 16 O/ 232 Th ≤ 0.20%).New olivine and matrix data were provided in Table S3 and S4 in Supporting Information S1, respectively.

Petrologic Modeling to Determine Olivine V Partition Coefficients
The measured V, MgO, and FeO abundances in each matrix and olivine pair were used as starting points to model the parental melt of each respective olivine.The aim of the model is to determine the V concentration and Mg# of each olivine's parental melt and determine D ol/melt V .To model the melt composition that is in equilibrium with each individual olivine, olivine is iteratively added or subtracted from the measured matrix composition.Each iteration adds 0.1% of the equilibrium olivine composition to the matrix.The FeO and MgO composition of the equilibrium olivine is calculated based on an Fe-Mg partitioning coefficient of 0.31 (Roeder & Emslie, 1970).Here, the total FeO of the matrix is used to calculate the Fe-Mg partition coefficient because the Fe 2+ /Fe 3+ of the matrix is unknown.This can result in a higher calculated fO 2 ; thus, the final calculated fO 2 may reflect a maximum estimate for the parental melt.Paired Fe 2+ /Fe 3+ and D ol melt V may be an improvement for future studies.The SiO 2 of equilibrium olivine in each step was calculated by subtracting the FeO and MgO wt.% from 100.The V concentration of the equilibrium olivine at each melt stage is based on empirical relationships observed between the olivine forsterite content and V concentration for each locality (Figure 2).As the equilibrium olivine composition changes throughout the model, the empirical regressions are used to determine the corresponding olivine V concentration.Regression parameters used to calculate V concentration as a function of olivine Fo# [Fo# = 100* Mg/(Mg + Fe) in mol%] for each locality are recorded in Table S5.The model is further described in Note.The fO 2 shown here is calculated using the partition coefficient of V between modeled parental melt and V measured in the most primitive (highest Fo#) olivine from each sample, which includes all grains whose Fo# is within analytical uncertainty of the highest Fo# olivine.Data for all olivines and other trace elements analyzed in this study can be found in the supplementary material.a Samples that do not have a mantle component listed are considered geochemically depleted (see Section 2.7).Fe and Mg were measured via electron probe microanalysis; olivine V concentrations were measured via laser-ablation medium-resolution inductively coupled plasma mass spectrometry (see methods).b Uncertainties represent either the external reproducibility of fO 2 calculated among primitive olivine from a single sample or the average prediction uncertainty of each calculated fO 2 (see Section 2.6.), whichever is greater.Samples with a single olivine analysis have only the prediction uncertainty from the calculated fO 2 .
the schematic in Figure S3 in Supporting Information S1 and the resulting parental melt V concentration and final olivine V concentration are reported in Table S6.
Other major elements (e.g., TiO 2 , Al 2 O 3 , MnO, CaO, Na 2 O, K 2 O, and P 2 O 5 ), do not affect the modeled olivine composition or the calculated V partition coefficient; therefore, the matrix was not characterized by EPMA in this study.In order to provide a parental melt composition for all major elements, the published whole-rock data are used.Throughout the petrologic model, olivine dilutes the TiO 2 , Al 2 O 3 , MnO, CaO, Na 2 O, K 2 O, and P 2 O 5 as the olivine FeO, MgO, and SiO 2 are added to the matrix.In each iteration of the model, the TiO 2 , Al 2 O 3 , MnO, CaO, Na 2 O, K 2 O, and P 2 O 5 content are reduced by 0.1% as 0.1% olivine is added.The full parental melt compositions are reported in Table S6.

Oxygen Fugacity Calculation
The relationship between oxygen fugacity and D ol/melt V has been empirically derived (Canil, 1997;Canil & Fedortchouk, 2001;Mallmann & O'Neill, 2009, 2013;Wang et al., 2019).Experimental data are limited in the higher fO 2 range that is observed in OIB leading to higher uncertainties in the higher fO 2 range.This affects the Figure 2. Linear regressions for each locality used to carry out petrologic models (see methods for details).The shaded fields represent the prediction uncertainty (RMSE) obtained using the Jack-knifing technique described in Section 2.6.The prediction uncertainty is the error associated with using a Fo# to predict a V concentration to model the parental melt.This uncertainty is propagated through the melt model and calculation of fO 2 .Regression parameters for each locality are given in Table S1.
calculations reported here (details provided below).Experimental calibrations at higher fO 2 would increase the fidelity and reduce the uncertainty of quantitative fO 2 derivations in future work.This study uses the regression equation (Equation 1) from Nicklas et al. (2019) that includes data from previous experimental studies compiled in Nicklas et al. (2018).In this equation, fO 2 is calculated relative to the nickel-nickel oxide buffer (∆NNO).
The final reported fO 2 for each sample is the average determined by the most primitive (highest Fo#) olivine(s) from that rock (Table 1).Using only the most primitive olivines as representative of the earliest solids derived from the parental melt limits the potential effects of clinopyroxene and oxide fractionation on the olivine V concentration.Out of caution, olivine with anomalous Ti, Cu, or Cr contents (defined outside 3× the interquartile range) are not used to calculate oxygen fugacity to avoid the possibility that oxide or sulfide inclusion was sampled by the laser during LA-ICPMS.A total of 200 primitive olivines were used to calculate the fO 2 of the 56 lavas in this study (Table S7).In arc lavas, V partitioning is dependent on temperature (T ) and melt polymerization (quantified as the ratio of non-bridging oxygen to total tetrahedrally coordinated cations, NBO/Tot) in addition to oxygen fugacity (Wang et al., 2019), per the expression: Equation 2 is used to calculate fO 2 relative to the fayalite-magnetite-quartz buffer.For the subset of lavas in this study that have complete major element data sets (n = 50), we applied both Equations 1 and 2 to check for consistency (Table S8).All but one lava (Samoan sample AVON3-68-11) have overlapping fO 2 within uncertainty using the two empirical relationships (Figure S4 in Supporting Information S1).Sample AVON3-68-11 has the highest NBO/Tot and melting temperature calculated using Petrolog3 software (Danyushevsky & Plechov, 2011).Given that both equations give consistent fO 2 for almost all lavas, this study uses Equation 1 to include the lavas that do not have complete major element data.

Evaluation of Uncertainty
In order to investigate the statistical robustness of possible distinctions in fO 2 between different OIB localities (as well as individual samples from the same locality), a comprehensive analysis of uncertainties and error propagation is essential.To constrain the uncertainty associated with the linear regression for olivine V ppmw versus Fo# (Step 1 in the petrologic modeling; Figure S3 in Supporting Information S1), we employ cross validation (or jack-knifing) and Monte Carlo sampling of the analytical errors associated with each olivine and matrix measurement.To include the analytical errors associated with the measurement of olivine V concentration and Fo#, we employ Monte Carlo sampling of the V and Fo# uncertainties during each iteration of the Jack-knifing routine.
In each iteration of the Jack-knifing routine, each datum can fall anywhere within the 95% confidence interval of its analytical uncertainty.Jack-knifing iteratively leaves out one data point at a time and the number of iterations is equal to the total number of data.
The final regression parameters for each locality are the means of the sampled parameters, and the final error is the root mean square error (RMSE) of the regression line relative to the validation point (i.e., the datum that is left out in any given iteration).The RMSE is propagated through the petrologic model provided here to constrain the uncertainty on the modeled V partition coefficient between olivine and parental melt, which is then propagated through Equation 1.The parameter uncertainties for Equation 1 reported by Nicklas et al. (2018) are incorporated into the final calculation.The final uncertainty for each rock is either the external reproducibility (i.e., the standard deviation of the fO 2 from different olivines within the rock) or the prediction uncertainty from the fO 2 calculation (described here), whichever is greater.The fO 2 and uncertainty for each rock are reported in Table 1.
Olivine from St. Helena, Azores, Pitcairn, and Austral have regressions with high p values (Figure 2), which indicate that olivine Fo# and olivine V concentrations are not well correlated.This likely indicates that the individual samples defining each curve are not cogenetic.However, the data can still be used to model the parental melt because the petrologic model only needs to predict the V concentration of the equilibrium olivine in each step within the 95% confidence interval.Thus, the model still determines the V concentration of the equilibrium Geochemistry, Geophysics, Geosystems olivine for the parental melt within the confidence interval.High p values and scatter in the data will lead to higher uncertainties associated with the linear regression, which are propagated through the model.

Grouping Lavas Into Mantle Components
To simplify the investigation of the effect of recycled materials on fO 2 , we examine individual mantle components by grouping lavas from this study and our compilation of previous studies into HIMU, EM1, EM2, or geochemically depleted OIB to identify if these components have distinct fO 2 .Lavas are considered HIMU if their 206 Pb/ 204 Pb is > 20 (e.g., Jackson et al., 2018).EM1 lavas are identified by moderately to highly radiogenic 87 Sr/ 86 Sr and unradiogenic 143 Nd/ 144 Nd with a slope of ∼-0.28 in Sr-Nd isotope space; EM2 lavas are categorized by extreme 87 Sr/ 86 Sr and moderate 143 Nd/ 144 Nd with a slope of ∼ 0.06 in Sr-Nd isotope space (Zindler & Hart, 1986).Figure S5 in Supporting Information S1 illustrates the distinct trajectories that define the EM1 and EM2 lavas in this study.Lavas that have 87 Sr/ 86 Sr < 0.7044, 143 Nd/ 144 Nd > 0.5128, and 206 Pb/ 204 Pb is < 20 are grouped together as geochemically depleted lavas.The geochemically depleted group may include lavas that have isotopic compositions described as "FOZO" (focus zone), which is the isotopic composition where OIB Sr-Nd-Pb arrays appear to converge (Hart et al., 1992).Lavas that have "FOZO" compositions are grouped with geochemically depleted lavas here because they do not contain definitive evidence of recycled material in their sources based on canonical isotopic signatures.This is not meant to imply that all grouped lavas are derived from a uniform or shared physical reservoir.For the purpose of discerning fO 2 among types of recycled material, we assume individual lavas belong to only one mantle component (i.e., samples are not mixtures of multiple components).The component type assigned to each lava in this study is given in Table 1 and illustrated Figure S5 in Supporting Information S1.

Statistical Tests
To test the statistical independence of different localities and mantle components with respect to inferred source fO 2 quantified relative to the fayalite-magnetite-quartz buffer (∆FMQ), we used a one-way analysis of variance (ANOVA) test.The one-way ANOVA test compares mean values to test whether data in categorical groups (in this case, OIB components and localities) have the same mean (the null hypothesis) or if at least one group has a distinct mean.To test the significance of correlations between fO 2 and isotope systems (He-Sr-Nd-W-Os-Pb), we use the square of Pearson's correlation coefficients (r 2 ) and the associated p value.P-values quantify the probability that the regression can predict the dependent variable (i.e., an isotopic composition) by incorporating the independent variable (i.e., fO 2 ) better than by relying on a degenerate model (e.g., the average value of the dependent variable).In this case, p values greater than 0.05 indicate that the regression is not statistically significant at the 95% confidence interval, while p values less than 0.05 indicate that the correlation is significant at the 95% confidence interval.In this study, a significant p value means that fO 2 can predict a given isotopic, trace element, or major element composition better than relying on the average composition.Because Pearson's correlation coefficients assume a linear relationship, we also investigated non-parametric correlation tests, such as Spearman's correlation coefficient and Kendall's Tau; the findings remain the same.

New OIB Olivine Analyses
New olivine FRTE and Ga and Ge compositions are reported in

Oxygen Fugacity of Global OIB
The average fO 2 of global OIB in this study is +1.2 ± 0.5 (2SD) ∆FMQ, in agreement with the EM-type and HIMU-type OIB average of +1.5 ± 0.8 ∆FMQ determined by Nicklas, Hahn, Willhite, et al. (2022).Though fO 2 uncertainties and heterogeneity within and among OIB groups are large, making it difficult to draw definitive conclusions, statistical tests indicate that some OIB localities and mantle components have distinct fO 2 .A summary of which plume locations are statistically distinct is illustrated in Figure S6 in Supporting Information S1.When the new data presented here are combined with a compiled OIB data set (Table S9), the mean fO 2 of OIB is +0.5 ± 1.1 ∆FMQ.Lavas designated as HIMU have the highest average fO 2 of +1.5 ∆FMQ (±3.0 2SD) and are distinctly oxidized compared to all other mantle components using a one-way ANOVA test (Figure 5).EM2 lavas have the second highest fO 2 (+0.7 ± 0.7 ∆FMQ) and are oxidized relative to geochemically depleted lavas ( 0.3 ± 1.1 ∆FMQ) but are not distinct from EM1 (+0.2 ± 0.6 ∆FMQ).The fO 2 of OIB is elevated relative to the range for MORB mantle (+0.3 to +0.9 ∆FMQ) estimated using the same proxy and methodology (Nicklas et al., 2019) as well as the global MORB average ( 0.1 ± 0.3 ∆FMQ) using a variety of other fO 2 proxies (Figure 5; Table S10).The variance of fO 2 among all OIB samples is greater than that among MORB samples, possibly related to OIB source heterogeneity and/or complex petrogenesis.
Using an independent samples t-test, there is no apparent difference between the fO 2 inferred from submarine (n = 23) versus subaerial (n = 31) OIB in the data set presented here.Lavas from Mangaia have notably high LOI -i.e., 6.8 wt.% and 8.8 wt.%-reflecting subaerial alteration (Figure S7 in Supporting Information S1).Though V is robust to subaerial alteration, the calculated fO 2 of the Mangaia samples in this study could be affected if SiO 2 , FeO T , or MgO were decreased or increased by alteration, as this would affect the calculated parental melt  composition to determine the V partition coefficient.The fO 2 of the Mangaia samples is within error when calculated using Equation 1 and Equation 2. Given that Equation 2 requires the ratio of non-bridging oxygen to total tetrahedrally coordinated cations, the consistency of the calculated fO 2 using Equations 1 and 2 for Mangaia gives confidence that the major element composition, and therefore, fO 2, has not been severely affected by subaerial alteration.The fO 2 of Mangaia lavas should be verified in fresh samples, which are unfortunately rare on this ∼20 Ma island.

Oxygen Fugacity Does Not Correlate With Isotopic Composition
There are no correlations between fO 2 and He-Sr-Nd-W-Os-Pb isotopic compositions when OIB is taken together globally (Figure 6).When OIB is separated by mantle component (e.g., HIMU, EM1, EM2) or by plume location, few correlations exist.The fO 2 of geochemically depleted OIB correlates positively with radiogenic 206,207,208

Offset Between fO 2 Determined by Different Proxies
In the compiled global OIB database there are 492 previously published fO 2 data: 38 data are derived from D ol/melt V , 390 data are calculated from X-ray absorption near edge structure (XANES) spectroscopic measurements, and 64 data correspond to other oxybarometers (see Table S9).This study adds an additional 56 fO 2 measurements via the D ol/melt V proxy.The new average fO 2 observed in the data set presented here is +1.2 ∆FMQ, similar to the average value for all OIB fO 2 measurements derived from D ol/melt V systemics (i.e., +1.5 ∆; FMQ; Nicklas et al., 2019;Nicklas, Hahn, Willhite, et al., 2022;Taracsák et al., 2022).In contrast, the average fO 2 determined in all OIB via XANES and micro-XANES is +0.4 ∆FMQ (Brounce et al., 2017(Brounce et al., , 2022;;Hartley et al., 2017;Helz et al., 2017;Moussallam et al., 2014Moussallam et al., , 2016Moussallam et al., , 2019;;Shorttle et al., 2015).Other techniques in the compilation (n = 64)-including titrimetric determination of Fe 2+ and X-ray fluorescence for Fe T and MgO thermogeobarometry, etc.-yield even lower fO 2 with an average of +0.3 ∆FMQ.Some previous studies (e.g., Taracsák et al., 2022) have shown that the fO 2 values recorded by OIB via the D ol/melt V proxy are higher than fO 2 determined by other techniques, such as XANES measurements of Fe +3 /Fe T and S speciation (Figure 5c).However, recent work has demonstrated that Fe XANES and D ol/melt V produce concordant fO 2 results in MORB glasses with greater than 8.3 wt.% MgO (Nicklas, Puchtel, & Baxter, 2024).Complex fractional crystallization, which can be difficult to appropriately correct for, can cause erroneous fO 2 calculation via D ol/melt V oxybarometry (Nicklas, Puchtel, & Baxter, 2024).In OIB, the maximum estimates by XANES are generally in agreement with the fO 2 determined using D ol/melt V .For example, using the 2SD of the average, the observed of fO 2 of Canary (2.5 ∆FMQ ± 2.0), Hawaii (1.2 ∆FMQ ± 0.8), and Reunion (1.1 ∆FMQ ± 1.1) determined by D ol/melt V overlap the maximum fO 2 observed using XANES for these localities (1.4,2.0, and 0.08 ∆FMQ, respectively).Thus, differences between  1) and previously published lavas (Table S9) grouped by mantle component (a) and by plume location (b).All individual rock data from this study only are superimposed as diamonds on top of each box.Data are considered outliers if they fall outside the "whiskers" defined by 1.5× the interquartile range.Outliers in these plots are not excluded from discussion or plots as they are considered real, high (or low) fO 2 recordings.Note that the top axis is fO 2 given as ΔNNO and the bottom axis is ΔFMQ.Boxes are ordered by median fO 2 .(c) Kernel probability density function for global MORB (Table S10) compared to global ocean island basalts (OIB)including this study and the compiled data set.The purple distribution is the fO 2 of OIB determined via the V-in-olivine proxy to illustrate that this method generally produces higher fO 2 compared to other proxies.average OIB fO 2 determined via D ol/melt V systematics versus that derived from (micro-)XANES techniques may represent the cumulative effects of a sampling bias in which prior XANES work focused on samples with variable extents of degassing to understand the evolution of fO 2 during petrogenesis, and inaccuracies in D ol/melt V determinations due to unaccounted for fractional crystallization of spinel or other oxides.
The average fO 2 for lavas from Kilauea in this study (+1.5 ∆FMQ) is within the range of fO 2 observations in the least degassed Kilauea melt inclusions and glasses (+0.7 -+2.0 ∆FMQ), as determined by XANES (Helz et al., 2017;Moussallam et al., 2016).Sulfur degassing at Kilauea reduces the residual melt; therefore, the maximum observed fO 2 may be closer to the source composition (Helz et al., 2017;Humphreys et al., 2022;Moussallam et al., 2016Moussallam et al., , 2019)).The olivines from Kilauea in this study are more primitive (Fo# 86.3 to 90.5) compared to those that hosted the melt inclusions examined via XANES (Fo# 77.5 to 82.5; Moussallam et al., 2016).Earlier crystallizing olivines are less affected by S degassing.Further, diffusive equilibration of Fe 3+ /∑Fe in olivine-hosted melt inclusions will create offset between the fO 2 that is "locked in" by the D ol/melt V proxy (Humphreys et al., 2022).The precision of the XANES method allows for tracking the rapid evolution of fO 2 in the melt system.In samples without fresh glass, D ol/melt V can be a useful oxybarometer; paired XANES and D ol/melt V observations provide greater context to the fO 2 systematics of OIB.

Petrologic Influence on Calculated fO 2 in OIB
While this study primarily aims to investigate the link between lithospheric recycling and mantle fO 2 , it is critical to evaluate the effects of OIB petrogenesis and the influence of ancient mantle domains on fO 2 in global OIB.Oxygen fugacity is influenced by mantle potential temperature such that higher mantle potential temperatures produce lower fO 2 even with a fixed Fe 3+ /∑Fe of the peridotite source (Gaetani, 2016).A difference in mantle potential temperatures among plume localities, and between plume and mid-ocean ridges, is unlikely to explain the variability observed in global OIB.First, fO 2 does not vary systematically with mantle potential temperature among plume localities; a negative correlation would be expected if mantle potential temperature was the main control of the average fO 2 of OIB (Figure S11 in Supporting Information S1).Second, the mantle plume localities studied here have mantle potential temperatures that are, on average, approximately 130°C hotter than ridges (Bao et al., 2022).Hotter mantle potential temperatures at plumes would predict lower fO 2; however, typically the plumes studied here overlap or have higher fO 2 than MORB (Figure 5).The effect of mantle potential temperature, if any, is less than the uncertainties of the fO 2 observed in this study.
Whether partial melting influences the fO 2 determined by the D ol melt V proxy in global OIB requires consideration.Within any hotspot location, the calculated fO 2 does not vary systematically with MgO, TiO 2 , or Na 2 O content of the parental melt (Table S6), which are sensitive to melt degree.La/Sm and La/Yb, which are inversely correlated with the degree of partial melting, do not vary systematically with fO 2 within individual hotspot locations in this study.Picritic rocks from Iceland have shown a relationship between fO 2 and La/Yb, which was inferred to reflect a high fO 2 , relatively fusible lithology hosted within a more refractory, lower fO 2 component in the Icelandic plume (Nicklas, Baxter, et al., 2024).There is a weak relationship between La/Sm and fO 2 in the global data set (r 2 = 0.1, p value < 0.1).The absence of correlations between fO 2 and other incompatible major element composition (both whole rock and parental) and trace elements indicates degree of partial melting does not control the fO 2 determined by D ol melt V within hotspot lavas.Therefore, the weak trend between fO 2 and La/Sm globally may instead be primarily related to source enrichment.If source enrichment has an effect on fO 2 , the relationship between La/Sm and fO 2 may be attenuated by the modification of the source La/Sm due to fractionation of La and Sm during partial melting.The mechanisms and conditions of melting as well as heterogenous lithologies among OIB may be important in controlling the fO 2 of plume-derived melts; further work is needed to assess these effects in detail.Effects less than one log unit are difficult to ascertain in this study given the uncertainties when using the D ol melt V proxy in OIB.
The CaO/Al 2 O 3 of the melt system remains relatively constant during olivine crystallization but decreases during pyroxene crystallization.Clinopyroxene (and orthopyroxene) generally has a higher D mineral melt V compared to olivine at the same fO 2 and temperature (Wang et al., 2019).Pyroxene fractionation after primitive olivine crystallization would remove V from the melt resulting in a higher measured D ol melt V (and therefore, lower calculated fO 2 ).A negative correlation between fO 2 and CaO/Al 2 O 3 would provide evidence that the calculated fO 2 was affected by clinopyroxene fractionation.Within the data set presented here, there are no global or local (i.e., within an individual plume locality) negative correlations between whole rock CaO/Al 2 O 3 and fO 2 .A detailed investigation of a greater number of cogenetic samples from each locality may better illuminate the effects of clinopyroxene fractionation on fO 2 determination using D ol melt V .However, there is a positive correlation between CaO/Al 2 O 3 and fO 2 in the global data set as well as within the Hawaiian and Canary plumes (e.g., Figure S12 in Supporting Information S1).Within all Hawaiian lavas and individual Hawaiian volcanic centers, there is no relationship between CaO/Al 2 O 3 and MgO content.This suggests that there is no significant effect from pyroxene fractionation in the Hawaiian lavas.Three of the four Canary lavas are described as ankaramites that derive from a pyroxenite-rich mantle source (Day et al., 2009).Therefore, rather than pyroxene fractionation during petrogenesis, the positive relationship between CaO/Al 2 O 3 and fO 2 in OIB may be related to the plume source (e.g., carbonated recycled materials and/or pyroxenite), which is discussed below.

The Role of Lithospheric Recycling in Elevated OIB fO 2
Subduction, accumulation, and redistribution of lithosphere in the Earth's mantle has been invoked to account for the chemical divergence between MORB-which are byproducts of decompression melting of the upper mantle-and OIB, that may sample mantle domains as deep as the core-mantle boundary (Foley, 2011;Gast, 1968;Hofmann & White, 1982;Morgan, 1971).It has been suggested that the subduction of lithosphere leads to higher fO 2 in the mantle (Brounce et al., 2019;Evans, 2012;Kasting et al., 1993;Lécuyer & Ricard, 1999).It has been shown that the subducted lithosphere retains most of its oxidized material during subduction (Brounce et al., 2019) and that a large mass of recycled lithosphere remains oxidized relative to the ambient mantle at ≥300 km depth due to the survival of carbonates in carbonated eclogites (Foley, 2011;Yaxley & Green, 1994).Thus, recycled lithosphere in plume sources has been cited to explain observations of elevated fO 2 in OIB compared to global MORB (Brounce et al., 2017;Hartley et al., 2017;Helz et al., 2017;Moussallam et al., 2014Moussallam et al., , 2016Moussallam et al., , 2019;;Nicklas, Hahn, Willhite, et al., 2022;Shorttle et al., 2015;Taracsák et al., 2022).
The offset between fO 2 observed in MORB and OIB may be attributed to the oxidation of plume sources by lithospheric recycling.The influence of recycled material on the fO 2 of mantle-derived magmas is complicated; the addition of recycled material may increase or decrease the fO 2 of melt systems.For example, prior studies have shown that geochemical enrichment of MORB sources by the addition of crustal sediments may serve to locally lower the fO 2 of enriched MORB (E-MORB), as evidenced by systematically lower fO 2 observed in E-MORB using XANES (Cottrell & Kelley, 2013).Reduced carbon from ancient anoxic ocean sediments leads to reduction of ferric iron during decompression melting and petrogenesis.The opposite is observed at the Reykjanes Ridge, where plume-influenced MORB become systematically oxidized and geochemically enriched as the ridge approaches mainland Iceland (Novella et al., 2020;Shorttle et al., 2015).The proximity of Reykjanes Ridge basalts to the Icelandic mantle plume suggests that the plume could be the source of higher fO 2 material; it has been interpreted that the oxidized component is entrained oceanic crust in the plume (Novella et al., 2020;Shorttle et al., 2015).
The two examples described above demonstrate that different crustal compositions can effectively reduce or oxidize the mantle sources.Ancient pelagic sediments, which contain abundant reduced carbon, may reduce the mantle while oceanic crust, which contains a greater proportion of ferric iron than ambient mantle, may be more likely to oxidize the ambient mantle.Since global OIB overlap MORB fO 2 but extend to much higher fO 2 , lithospheric recycling does not reduce plume sources to a greater extent than MORB sources as no known OIB have lower fO 2 than MORB.Lithospheric recycling may oxidize plume sources to a greater extent than is observed in the MORB mantle, perhaps because the proportion and/or chemical compositions of lithospheric material added to plume sources after subduction is distinct from recycled material in the shallower MORB mantle.
Mantle components, which are thought to result in part from the recycling of different types of lithospheric material, provide an opportunity to test whether different types of recycled materials cause net oxidation or reduction in plume sources.For example, EM1 lavas have been speculated to contain recycled oceanic crust with pelagic sediment from the seafloor or delaminated lower continental crust (McKenzie & O'Nions, 1983;Weaver, 1991;Garapić et al., 2015).The EM2 lavas have isotopic and trace element signatures consistent with a contribution of recycled terrigenous sediment or metasomatized lithosphere (Jackson et al., 2007;Workman et al., 2004).HIMU lavas are often attributed to recycling of a chemically-modified oceanic crustal package such as carbonated eclogite (Dasgupta et al., 2007;Hofmann, 1997;Moreira & Kurz, 2001;Stracke et al., 2005).Ocean island basalts that are geochemically depleted generally lack observable signatures from lithospheric recycling and are likely the least chemically modified by recycling compared to other OIB.Lavas in this study are grouped into these mantle components based on their isotopic compositions (see Section 2.7.).We investigated the fO 2 systematics of each group.

HIMU Lavas
In the global suite presented here, HIMU has the highest and most variable fO 2 .The average HIMU fO 2 (1.5 ± 3.0 ΔFMQ) is statistically higher than both enriched mantle types as well as the geochemically depleted lavas (Figure 5).HIMU lavas also host olivine with the greatest CaO wt.% and whole-rocks with the highest CaO/ Al 2 O 3 (Figure 4; Figure S12 in Supporting Information S1).The addition of volatile, oxidizing agents such as C 4+ Geochemistry, Geophysics, Geosystems 10.1029/2023GC011249 WILLHITE ET AL. and S 6+ hosted in subducted carbonated lithosphere to the mantle source has been invoked to account for the highly oxidized and volatile-rich HIMU lavas in El Hierro, Canary (Taracsák et al., 2022).The reduction of carbonate and sulfate will lead to the oxidation of silicates, as observed in arc settings (Rielli et al., 2017).Recycled carbonated oceanic crustal materials, such as pyroxenite or eclogite, in the HIMU source can explain the radiogenic Pb isotopic compositions and elevated CaO/Al 2 O 3 of HIMU-type lavas (Dasgupta et al., 2007;Jackson & Dasgupta, 2008).Relatively high olivine CaO wt.% has been previously argued to reflect carbonatitic metasomatism in the HIMU source (Weiss et al., 2016).Our global data set supports recycling of carbonated materials, such as oceanic crust in the form pyroxenite, to produce elevated fO 2 , CaO/Al 2 O 3 , and olivine CaO content in HIMU-type OIB.HIMU lavas show a positive correlation between fO 2 and 87 Sr/ 86 Sr and negative correlation with 143 Nd/ 144 Nd, which supports the existence of a geochemically-enriched, high fO 2 end-member in the plume source; however, there is no reason that a carbonated component would necessarily produce a correlation with 143 Nd/ 144 Nd.High La/Sm, associated with higher fO 2 (Section 4.1) in OIB, likely points to the link between material with long-term geochemical enrichment (like recycled oceanic or continental crust) and high fO 2 .Variations in Fe and C content and speciation in the subducting slab as well as subduction environment and timing of recycling likely play a role in decoupling fO 2 from tracers of recycled materials and creating a scatter in the observed data set.Furthermore, a correlation between fO 2 and radiogenic isotopic compositions is not necessarily expected, given that the isotopic compositions of recycled lithosphere reflect parent-daughter fractionation processes, such as partial melting (and time of radiogenic ingrowth), which have not been shown to significantly affect fO 2 .
HIMU lavas from St. Helena are distinctly less oxidized than HIMU lavas from the Macdonald (including Mangaia and Austral lavas) and Canary hotspots.Distinct fO 2 observed in St. Helena compared to other HIMU lavas may reflect a different composition of recycled material in the St. Helena plume source; however, high whole-rock CaO/Al 2 O 3 and olivine CaO content are observed in St. Helena lavas.Heavy Zn isotopic compositions, which are linked to recycled surficial carbonates, are observed in St. Helena and other HIMU-type lavas (Zhang et al., 2022).These observations support a carbonated recycled crust in the St. Helena plume source, so it is unclear why St. Helena lavas in this study are among the lowest fO 2 in the global data set (Figure 5).Complex subduction processes affect the redox potential of the subducting material and may also decouple fO 2 from timeintegrated radiogenic isotopic compositions.For example, partial melting during subduction could reduce the Fe 3+ /Fe tot and C content of the subducting slab before it is entrained in a mantle plume.

Enriched Mantle (EM1 and EM2) OIB
Previous studies of fO 2 in geochemically-enriched OIB have considered EM1 and EM2 lavas together (Nicklas, Hahn, Willhite, et al., 2022).Here, the fO 2 of EM1-type and EM2-type lavas are considered separately for the first time.The EM1 lavas (average = 0.2 ± 0.6 ΔFMQ) have statistically lower fO 2 than HIMU but are not distinct from EM2 or geochemically depleted OIB (Figure 5).The EM1 type of recycled material, likely recycled pelagic sediments based on the high Th/U and Lu/Hf required to explain the time-integrated Nd, Hf, and Pb isotopic compositions of EM1, may not have the potential to significantly oxidize the mantle source despite imparting radiogenic 87 Sr/ 86 Sr on the lavas.As discussed earlier, sediments can also lower mantle fO 2 depending on their depositional environment (Cottrell & Kelley, 2013).EM2 OIB (average = 0.7 ± 0.7 ΔFMQ) are oxidized relative to geochemically depleted OIB (average = 0.3 ± 1.1 ΔFMQ) but less so than HIMU.Though continental sediments have been invoked to account for the radiogenic 87 Sr/ 86 Sr signatures of EM2-type lavas, recycled sediments do not contribute as much to the net redox budget of recycle material compared with altered oceanic crust and serpentinized lithosphere (e.g., Evans, 2012).Continentally-derived sediments may be present in EM2 plumes but might not have the capacity to drive the oxidation of the mantle source.Recycled lithospheres that have been metasomatized by carbonatitic fluids can also account for the isotopic signatures of EM2 lavas in Samoa (Hauri et al., 1993;Workman et al., 2004).Like HIMU-type lavas, EM2-type OIB generally exhibits higher CaO/Al 2 O 3 and olivine CaO content than EM1 or geochemically depleted OIB (Figure 4).These chemical characteristics are similar to HIMU-type lavas and could result from carbonatite metasomatism in their mantle source or recycling of carbonated pyroxenite (Canil et al., 1994;Jackson & Dasgupta, 2008;Weiss et al., 2016).Despite few correlations between fO 2 and radiogenic isotopic compositions, there are statistical fO 2 distinctions among lavas linked to different mantle components.Decoupling of fO 2 and lithophile isotopic systems may occur during shallow plume dynamics, such as the separation and rapid ascent of carbonatite or volatile-rich fluids relative to the Geochemistry, Geophysics, Geosystems 10.1029/2023GC011249 WILLHITE ET AL. silicate plume that carries lithophile trace elements (Hammouda & Laporte, 2000;Valbracht et al., 1996).If carbonated recycled material increases the fO 2 of plume sources, then preferential melting of high fO 2 carbonatite or CO 2 -rich fluid in the shallow plume may cause physical separation from the lithophile trace elements in silicate that link to lithospheric recycling (Hammouda & Laporte, 2000;Hofmann et al., 2011;Valbracht et al., 1996).As discussed above, there is no apriori reason that isotopic compositions should correlate with fO 2 given that radiogenic isotopic systems track different processes than fO 2 .Nevertheless, the association, rather than correlation, of isotopic signatures of recycled lithologies with elevated fO 2 may link the process of recycling to higher fO 2 in plume sources.

Depleted Mantle OIB
Depleted mantle OIB extend to the most reducing conditions and overlap with MORB fO 2 .These lavas exhibited robust positive correlations between fO 2 and 206 Pb/ 204 Pb, 207 Pb/ 204 Pb, and 208 Pb/ 204 Pb, as well as a negative correlation between fO 2 and 187 Os/ 188 Os.However, these relationships are primarily controlled by Hawaiian lavas (Figure S10 in Supporting Information S1).Radiogenic Os and unradiogenic Pb in Ko'olau have been associated with recycled pelagic sediments that may have experienced U loss in oxidized marine environments (Lassiter & Hauri, 1998).The pelagic sediment signature trends toward lower fO 2 , which suggests that pelagic sediment reduces the plume source with a mechanism similar to that observed in E-MORB (Cottrell & Kelley, 2013).
Another recycling model has been suggested to explain the major element and lithophile isotopic compositions of Ko'olau lavas: the Ko'olau source may contain a greater proportion of recycled eclogite compared to Kīlauea and Hualālai (Hauri, 1996).Archean eclogites are reducing compared to the modern mantle (Aulbach et al., 2019) and if present as recycled material in the Ko'olau mantle source, could lower the fO 2 in Ko'olau lavas relative to Kīlauea and Hualālai.
The spatio-temporal trend toward higher fO 2 from Ko'olau to Hualālai and Kīlauea (Figure 7) may also reflect the introduction of an oxidized component, such as hydrothermally-altered lower oceanic crust or lithospheric mantle to the Hawaiian plume.Considering that Hawaiian lavas extend to higher fO 2 than MORB, a relatively oxidized component is required to explain fO 2 up to +1.7 ΔFMQ (Figure 5).Multiple processes and recycled materials may be at play in the Hawaiian plume to explain the observed range of fO 2 observed.
An inverse correlation between fO 2 and 207 Pb/ 204 Pb in Baffin Island is opposite to the positive trend observed in Hawaii (Figure S10 in Supporting Information S1).Geochemically depleted isotopic signatures (i.e., low 87 Sr/ 86 Sr and high 143 Nd/ 144 Nd) along with MORB-like 187 Os/ 188 Os in Baffin lavas can be attributed to a depleted mantle source that does not contain significant contributions from recycled materials.Further, since only 207 Pb/ 204 Pb correlates significantly with fO 2 , and not 206 Pb/ 204 Pb, 208 Pb/ 204 Pb, or any other radiogenic isotope system in this study, lithospheric recycling does not explain the variability in fO 2 of Baffin Island lavas.
Baffin Island is unique in the global suite because the plume lavas erupted through the Archean and paleo-Proterozoic continental crust (Saunders et al., 2013).Using Nb/Th and Ce/Pb as identifiers of crustal assimilation (e.g., Willhite et al., 2019), where low Nb/Th (<13) and low Ce/Pb (<20) are indicative of crustal assimilation, all of the Baffin Island lavas in this study have been at least moderately affected by crustal contamination (Figure S13 in Supporting Information S1).However, crustal contamination does not appear to have modified the oxygen fugacity of the Baffin lavas as there is no significant correlation between fO 2 and Nb/Th, Ce/Pb, 87 Sr/ 86 Sr, 206 Pb/ 204 Pb, 208 Pb/ 204 Pb, etc.The two lavas with the lowest Ce/Pb are offset to higher fO 2; however, those two samples have the highest Nb/Th, which is inconsistent with crustal assimilation.Thus, the Pb-fO 2 trend observed in Baffin does not likely reflect lithospheric recycling or crustal assimilation.

Influence From Ancient And/Or Core-Equilibrated Mantle Domains?
Helium and tungsten isotopes in plume-derived lavas have been used to identify contributions from one or more ancient, well-preserved and/or core equilibrated reservoir(s) in the lower mantle (Class & Goldstein, 2005;Hart et al., 1992;Mundl et al., 2017;Rizo et al., 2016).Lavas with elevated 3 He/ 4 He compared to typical MORB ( 3 He/ 4 He = 6-10 Ra; where Ra denotes the sample's measured 3 He/ 4 He relative to Earth's atmospheric 3 He/ 4 He) sample an ancient reservoir that has been partially or wholly preserved through geologic time (Kurz et al., 1982).
Discovery of μ 182 W anomalies in plume-derived lavas (Mundl et al., 2017) as well as a correlation between μ 182 W and 3 He/ 4 He in deep-rooted plumes (Mundl et al., 2017;Mundl-Petermeier et al., 2020) provide evidence for the preservation of Earth reservoirs created in the early Hadean while 182 Hf was extant (i.e., within ∼60 My of Solar System formation).OIB with the highest 3 He/ 4 He and greatest magnitude μ 182 W anomalies also appear to the least modified by the addition of recycled materials (Jackson et al., 2020).If the fO 2 of ancient/core-equilibrated plume reservoirs are distinct from the modern convecting mantle, then fO 2 may be coupled with high 3 He/ 4 He, anomalous μ 182 W, and a low D Sr Nd Pb (a parameter that generally increases with the amount of recycled material entrained in a plume; Jackson et al., 2020).
For all lavas with published He and/or W isotopes in this study, there is no statistically significant correlation with fO 2 (Figure 8); however, high 3 He/ 4 He lavas are limited in this data set, and only 10 lavas have 3 He/ 4 He above 10 Ra.In previous studies of noble gas systematics in OIB, all studied OIB had higher fO 2 than MORB, regardless of 3 He/ 4 He (Day et al., 2022).Typically, OIB with a higher proportion of recycled material (i.e., a greater D Sr Nd Pb ) fall along mixing lines between MORB and various recycled end members with higher Fe 3+ /∑Fe (Brounce et al., 2022).This indicates that OIB with less recycled material does not have inherently higher fO 2 than other OIBs and MORBs.Given the limited number of samples with paired fO 2 and elevated 3 He/ 4 He measurements or anomalous μ 182 W, a critical evaluation of the fO 2 of OIB with those signatures may still be warranted in future studies.It is not apparent that ancient and/or core-equilibrated mantle sampled by some OIB has distinct fO 2 from the ambient mantle or that the fO 2 of the deep mantle source would be preserved once entrained in a mantle plume.This demonstrates that the offset to higher fO 2 observed in OIB is not likely related to the mechanism(s) that produce and preserve high 3 He/ 4 He and anomalous μ 182 W. (b) Osmium isotopic compositions of the Hawaiian lavas in this study negatively correlate with fO 2 (r 2 = 0.37), whereas 206,207,208 Pb/ 204 Pb positively correlate with fO 2 (r 2 = 0.76, 0.51, 0.77, respectively).The colored contours reflect a linear interpolation among 208 Pb/ 204 Pb data.There is also a spatio-temporal trend with increasing fO 2 and 206,207,208 Pb/ 204 Pb and decreasing 187 Os/ 188 Os from Ko'olau (oldest) to Kīlauea (youngest).(c) CaO/Al2O3 and (d) Na/Ti of Hawaiian whole-rocks have robust correlations (i.e., Pearson's correlation coefficient has a p value < 0.1) with fO 2 .

Conclusions
We provide fO 2 constraints for lavas derived from key mantle components (EM1, EM2, HIMU, geochemically depleted OIB) in a global framework to characterize global OIB using the same analytical techniques across our data set.We also provide a database of previously published MORB and OIB fO 2 from a variety of techniques and fO 2 proxies for cross-comparison.We find that D ol/melt V results in higher fO 2 than Fe 3+ /Fe T and other oxybarometers, but generally agrees with maximum fO 2 values determined by XANES Fe 3+ /Fe T measurements, perhaps due to degassing-related phenomena during progressive melt differentiation that lowers fO 2 observed by XANES in degassed samples.Though few robust correlations exist between radiogenic isotope compositions and fO 2 , lithospheric recycling remains a viable mechanism for the oxidation of plume source regions in the mantle.Despite overlap among HIMU, EM2, EM1, and depleted OIB, ANOVA tests reveal that HIMU-and EM2-type OIB are distinctly oxidized compared to depleted OIB and contain isotopic and petrologic evidence for recycling of carbonated oceanic crust.Even geochemically depleted OIB with limited evidence of lithospheric recycling show evidence for a geochemically-enriched oxidized component in their plume source.So far, there is limited evidence that primitive geochemical signals like elevated 3 He/ 4 He or negative μ 182 W are associated with distinct fO 2 .These findings support the link between lithospheric recycling and variable and elevated fO 2 in Earth's interior.(Jackson et al., 2020).Lavas with the lowest D Sr Nd Pb should represent mantle sources least affected by lithospheric recycling.

Figure 1 .
Figure 1.Map of global ocean island basalts localities with new fO 2 data presented in this study.The size of the circle at each site corresponds to the number of analyses from that hotspot.Earth image is from the NASA Earth Observatory Blue Marble series.

Figure 3 .
Figure 3. Olivine first-row transition element compositions normalized to the San Carlos olivine reference material.Olivines are grouped according to the mantle components of their host lava.

Figure 4 .
Figure 4. HIMU and enriched mantle 2 lavas have high olivine CaO and HIMU has high whole-rock CaO/Al 2 O 3 .(a) Analysis of variance (ANOVA) results showing the mean (circles) and 95% confidence interval (lines) for the olivine CaO wt.% of each mantle component.If the confidence intervals do not overlap, then the respective components have statistically distinct means.(b) ANOVA results for whole-rock CaO/Al 2 O 3 illustrating statistically higher CaO/Al 2 O 3 in HIMU lavas compared to geochemically depleted ocean island basalts and enriched mantle 1.

Figure 5 .
Figure5.Box and whisker plots illustrating the fO 2 distributions of lavas in this study (Table1) and previously published lavas (TableS9) grouped by mantle component (a) and by plume location (b).All individual rock data from this study only are superimposed as diamonds on top of each box.Data are considered outliers if they fall outside the "whiskers" defined by 1.5× the interquartile range.Outliers in these plots are not excluded from discussion or plots as they are considered real, high (or low) fO 2 recordings.Note that the top axis is fO 2 given as ΔNNO and the bottom axis is ΔFMQ.Boxes are ordered by median fO 2 .(c) Kernel probability density function for global MORB (TableS10) compared to global ocean island basalts (OIB)including this study and the compiled data set.The purple distribution is the fO 2 of OIB determined via the V-in-olivine proxy to illustrate that this method generally produces higher fO 2 compared to other proxies.

Figure 6 .
Figure 6.Oxygen fugacity plotted against isotopic compositions for each mantle component.See Figure S7 in Supporting Information S1 for isotope compilation and references.Error bars represent the 1SD. Figure S10 in Supporting Information S1 shows the same plots with data color coded by plume locality instead of mantle component.

Figure 8 .
Figure 8. Ancient geochemical signals, such as high 3 He/ 4 He (a) and anomalous μ 182 W (b), do not correlate with fO 2 .Vertical error bars represent 2SD (internal) for helium isotopes and 2SE (internal) for tungsten isotopes.(c) Increasing D Sr Nd Pb reflects a larger proportion of recycled material in the source(Jackson et al., 2020).Lavas with the lowest D Sr Nd Pb should represent mantle sources least affected by lithospheric recycling.

Table 1
New Oxygen Fugacity Data and Most Primitive Olivine Fo# Measured in Each OIB

Table 1 Continued
TableS3and visualized in Figure3.The forsterite number varies from 74 to 92.Olivines with the lowest and most variable Fo# are from HIMU-type lavas (81.4 ± 7.4 2SD).Geochemically depleted OIB have the highest average Fo# olivine (86.4 ± 7.2 2SD).Vanadium concentrations ranged from 2.7 to 13.7 ppmw in the global data set.Mean olivine V concentrations were statistically lower in HIMU and EM2 lavas according to the one-way ANOVA test.Olivines from EM1 and geochemically depleted lavas have similar mean olivine V concentrations.HIMU-and EM2-type lavas are also distinguished from EM1 and geochemically depleted OIB based on olivine CaO that is resolvedly higher in HIMU and EM2 OIB (Figure4).Olivine Ni concentrations are the highest and most variable in geochemically depleted OIB, with concentrations ranging from 1,100 to 4,800 ppmw Ni (0.14-0.61 wt.% NiO).The highest Ni contents in olivine in the global data set were from Hawaii and Baffin Island.Lavas that are HIMU-and EM1-type have olivine with the lowest average Ni concentrations.WILLHITE ET AL.
Pb/ 204Pb and negatively with radiogenic 187 Os/ 188 Os; however, this trend is primarily controlled by Hawaiian lavas.In the Canary Islands, fO 2 is positively correlated with 87 Sr/ 86 Sr; this correlation is also present when all HIMU-type lavas are grouped together.Global HIMU also has a negative correlation between fO 2 and 143 Nd/ 144 Nd.For Baffin Island lavas, only 207 Pb/ 204 Pb is correlated with fO 2 .Summaries of the Pearson's tests between fO 2 and each isotopic system to visualize statistically significant correlations in the current data set are given by location and by mantle component in FiguresS8 and S9in Supporting Information S1, respectively.