Brounce_HSDP2_XANES_SupplementaryText_unformatted_v8

Supplementary material to accompany the manuscript, “

Redox variations in HSDP2

Mauna Kea lavas, the

oxygen fugacity of the Hawaiian plume

, and the role of volcanic

gases in Earth’s oxygenation

” by M. Brounce, E. Stolper, and J. Eiler.

METHODS

Fe XANES

We determined

ratios of pillow glass by micro

ray absorption near

edge structure (

XANES) spectroscopy at beamline 13

IDE, Advanced Photon Source,

Argonne National Laboratory. Spectra were collected in fluorescence mode from 7020

eV to 7280 eV

using a Si [111] monochromator and a defocused beam diameter of 50x50

Counts were recorded on a multi

element silicon drift detector x

ray spectrometer,

equipped with two Si drift diode detectors. Eight layers of aluminum foil were placed in

the path o

f the incident photon beam in order to decrease the intensity of the incident

photon beam prior to interaction with the surface sample, which could lead to auto

oxidation or auto

reduction of Fe species dissolved in the glass. The incident photon

beam inte

nsity resulted in on the order of 10

counts on Fe on a LW standard glass

(

)

Spectra were normalized and the pre

edge features were fit following techniques

outlined by

ref. 1

, with some exceptions.

Ref. 1

fit the pre

edge features using two

background functions and two Gaussian curves to fit the Fe

and Fe

peaks,

simultaneously solving for 11 curve parameters that are allowed to vary independently to

produce a fit spectra by minimizing error between the

fit and the data. In this study, we

have attempted to minimize uncertainty or bias that may be introduced by allowing curve

parameters to vary, which might be reasonably expected to be constant. For instance, the

width of the Gaussian curves should not be

variable in glasses with similar compositions,

but the intensities should vary as a function of the Fe

/Fe

ratio of the glass. To fix

these two curve parameters and reduce the number of free parameters in the fit solution to

9, we calculated the averag

e full

width half

maximum of both the high energy and low

energy Gaussian curves for all the samples analyzed in each beam session. We then re

fit

each spectrum, keeping the full

width half

maximum of both Gaussian curves fixed and

equal to the average val

ue of all samples analyzed in that session (supplementary figure

1, 2).

We use the ratio of the areas of each Gaussian curve in our calibration to

ratios, instead of the centroid positions that

ref. 1

used

Ref. 1

utilized beamline X26a at

the National Synchrotron Lightsource I, which has been decommissioned. Although

X26a and 13

IDE both have

XANES capabilities, the hardware and configuration of

IDE are sufficiently different from X26a that data collection ro

utines and spectra

handling require a modified approach. For example, there is not significant drift in

monochromator energy at 13

IDE, which results in greater energy reproducibility in the

absorption spectra collected over several hours in a beam session

, as well as greater

session

session reproducibility. This, in combination with

other factors,

results in

significantly improved energy resolution in the pre

edge region for Fe spectra, even when

using the Si [111] monochromator (supplementary figure 1)

. The consequence of this is

that the Gaussian

fit pre

edge features that correspond the 1s to 3d electronic transitions

in Fe

and Fe

are well defined compared to those collected earlier and render the use of

the centroid positions in Fe redox calibrat

ions less necessary than in previous works (e.g.,

ref. 2

). Here, we use the ratio of the area under the higher energy Gaussian curve to the

area under the lower energy Gaussian curve, because this ratio is closely related to the

physical phenomenon of

NES spectroscopy, and generate a calibration curve using

the Fe

/Fe

ratio determined by Mossbauer spectroscopy, reported by

ref. 1

We also

utilize the centroid calibration described in

ref. 1

, and note that the choice of centroid or

area ratio calibrat

ions does not impact the calculated

ratios of unknown samples

meaningfully.

We performed an error analysis of the spectral fit parameters of spectra collected

using the Si [111] monochromator using a bootstrap monte

carlo method. For a given

spectrum, we assigned an arbitrary uncertainty on the deadtime corrected, normalized,

det

ector measured counts on Fe per second for each data point in the pre

edge of 0.002.

The units of this error are arbitrary, but we note that it is a liberal estimate based on

observations over 20 beam sessions at two synchrotron facilities, including 10 se

ssions

using the same detector array used in this study, 3 of which

sessions

were at the beamline

used in this study. A normal distribution of data was generated for each data point in the

pre

edge, with a standard deviation equal to the assigned analytica

l uncertainty and

centroid equal to the actual measured value. A synthetic spectrum was then constructed

by randomly sampling this distribution of data of each measured data point. This

synthetic spectrum was fit using the same routine described above, and

ref. 1

, and this

was repeated 100 times for each collected spectrum. This exercise generates quantitative

uncertainties for each spectral parameter, which can be propagated through the

calibration to generate an uncertainty for

ratios, which i

s +/

0.015 (absolute).

This can be compared to the empirical precision of the area ratios on the standard glasses,

which is +/

0.01 (n = 2 spots), corresponding to an uncertainty in

ratios of

0.006 (absolute)

Finally, we considered differenc

es between spectra produced at X26a and 13

IDE

that may arise due to major differences between beamlines, such as the energy of their

respective electron storage rings (~2.8 GeV at NSLS

I X26a, ~7 GeV at APS 13

IDE),

the storage ring sampling methods (bend

ing magnet at X26a, insertion device at 13

IDE),

and the intensity of the incident photon beam (orders of magnitude more intense at 13

IDE). The analytical facilities are sufficiently complicated at the synchrotron level that it

is not practical for this s

tudy to seek to understand the nature of each of these differences,

so we present an empirical comparison of spectra collected at X26a before NSLS

decommission with spectra collected at 13

IDE (supplementary figure 3). The spectra are

collected on two su

bmarine glass chips from the Gulf of Aden

(3)

. Following the

methods presented here, the first glass chip has calculated

ratios

of 0.171

(NSLS) and 0.188 (APS). The second glass chip has calculated

ratios

of 0.163

(NSLS) and 0.174 (APS).

The first glass chip has

ratios that are 0.017

(absolute) more oxidized when measured at APS than at NSLS, and the second glass chip

has

ratios that are 0.011 (absolute) more oxidized when measured at APS than

at NSLS. The origin of this

offset is not clear at this time, but in order to produce a

dataset that can be compared with previously collected data at NSLS

I, we “correct”

ratios calculated from spectra collected at APS by subtracting 0.02

units

from

each sample average.

XANES

We determined

S ratios of pillow glass by micro

ray absorption near

edge

structure (

XANES) spectroscopy, also at beamline 13

IDE, Advanced Photon Source,

Argonne National Laboratory. Spectra were collected in fluorescence mode from 2447

eV t

o 2547 eV, with a dwell time of two seconds on each point, using a Si [111]

monochromator and a defocused beam diameter of 50x50

Counts were recorded on a

multi

element silicon drift detector x

ray spectrometer, equipped with two Si drift diode

detecto

rs. All analyses were done in a helium atmosphere, to avoid interaction between

the incident photon beam and atmosphere. As with iron, to avoid significant beam

damage during analysis, the beamline hardware was tuned so that the incident photon

beam intens

ity was on the order of 10

counts on S on the submarine MORB glass

TR101

15D

8g, which has 1790 ppm S (supplementary figure 4a). Each sample was

analyzed with a stationary beam in triplicate, moving the beam position for each of the

three analysis spots.

In the absence of an indep

endent method for accurately determining

S ratios,

we follow the approach of

ref. 4

to calculate

S ratios from our absorption spectra.

We take a MORB glass with all sulfur present as S

and an experimental glass with all

sulfur present as S

our two endmember absorption spectra (supplementary figure 4a).

Each unknown is fit using linear combinations of the MORB and experimental glasses.

Since each of those glasses represents an endmember speciation of S, we assume that the

intensity of absorp

tion spectra features for both S

and S

respond linearly to the

concentration of S

and S

dissolved in the sample glass, and report

S ratios equal

to the mixing proportions of the endmember spectra necessary to fit each unknown

spectra (suppleme

ntary figure 4b). The glasses in this study have low sulfur contents

compared to recent experimental glasses

(

)

, which results in a decreased signal to

noise ratio in collected spectra. We estimate that the uncertainty on these sulfur XANES

measurements

is +/

0.01 (absolute), based on the reproducibility of spectra from

individual natural samples.

To test the extent of possible beam damage, we collected several spectra in one

location on a series of back

arc basin glasses that have spectral features

indicative of the

presence of both S

and S

, exposing the same pool of glass to the incident photon beam

for 110 consecutive minutes (supplementary figure 5). There are noticeable changes in

the sulfur absorption spectra over this time period. The S

bsorption feature decreases

in intensity. At the same time,

the

broad S

absorption feature increases in intensity, but

at slightly higher energy than the center position of this feature. This could be due to the

generation of S

as the result of beam da

mage

(

5, 6

)

. The overall differences in intensity

of the absorption features between spot 1 and spot 10 (i.e., after 110 minutes of beam

exposure) are small, and unlikely to impact the calculated

S ratios determined here.

Nonetheless, we collect a sin

gle spectrum in eleven minute

and then

move the position

of the incident photon beam before collecting another spectrum on the same sample,

limiting the exposure time of any glass pool.

VOLATILES IN HSDP2 SUBMARINE GLASS

Supplementary figure 6 shows th

e key relationships between S and FeO*, and S

and H

O described in the main text.

The S contents of all but two glasses are below the

sulfur content of MORB magmas that are sulfide

saturated, from

ref. 7

, modified after

ref.

The sulfur contents range from ~200 ppm to ~1400 ppm within a narrow range of

FeO*, suggesting that sulfur degassing is taking place in a sulfide

undersaturated melt.

This is supported by petrographic descriptions from

ref. 8

that note these glasses do no

appear to have sulfide blebs present. Both suites have H

O and S contents that are

positively correlated. The low SiO

glasses appear to have higher initial H

O contents,

but similar initial S contents, relative to the low SiO

glasses (i.e. low SiO

gla

sses have

higher H

O/S ratios than high SiO

glasses; supplementary figure 6b), but both suites are

consistent with concomitant sulfur and water degassing.

COMPARISON TO OTHER DEGASSING

REDOX STUDIES

Supplementary figure 7 shows the relationship between

ratios

and

calculated magmatic

and S and H

O contents in submarine pillow glasses from this

work, and in two other recent studies of olivine hosted melt inclusions

(

9, 10

)

The

Agrigan melt inclusions are more oxidized than the HSDP2 submarine glasses in this

study

(Supplementary Fig. 12

The Agrigan melt inclusions are also more water rich

than the HSDP2 submarine glasses

(Supplementary Fig. 12

. The relationship betw

een

ratios and S and H

O contents is less systematic than is observed for HSDP2

submarine glasses

(Supplementary Fig. 12

a, b)

. This could be because the HSDP2

submarine glasses are more restricted in their major elemen

t compositions, or because

melt inclusion process

es introduce

variability in

ratios +/

O contents, +/

contents

of the Agrigan dataset

or some combination of the two.

The Agrigan melt

inclusions also record more oxidized magmatic

s than the HSDP2 submarine glasses

(Sup

plementary Fig. 12

f).

The Erebus me

lt inclusions span a wide range in

ratios and S contents, and a wider range in H

O contents than the HSDP2 submarine

glasses, but are not as water rich as the Agrigan melt inclusions

(Supplementary Fig. 12

a, b, d, e)

The relationship between

ratios and S and H

O contents in Erebus

melt inclusions are less systematic than the submarine glasses from this study, similarly

to the Agrigan melt inclusions

(Supplementary Fig. 12

a, b)

. The general sense

decreasing

ratios with increasing extents of degassing is broadly similar

between the three datasets, despite differences in sample types (i.e., submarine glass vs.

olivine

hosted melt inclusions), magma compositions, and magma storage conditi

ons

(Supplementary Fig. 12

a, b)

The extent of reduction per ppm S loss (or per wt% H

lost) is greatest for Erebus melt inclusions and HSDP2 submarine glasses (i.e., the slope

of the relationship between S and H

O and

ratios is steepest for these samples

;

Supplementary Fig. 12

MODELING THE IMPACT OF DEGASSING ON MELT

f

We modeled the change in magmatic

with progressive degassing of a C

S vapor species using the gas

melt equilibrium model of

ref. 11

. This thermodynamic

model computes C, H, O, and S concentrations and speciation in coexisting fluid (i.e.,

gas) and silicate melt as functions of pressure, temperature, and

, based on

experimental calibrations of melt solubility and homogeneous equilib

rium in the gas

phase for H

, H

O, CO, CO

, SO

, H

S, and S

species. The melt is assumed in this model

not to crystallize any solids or to precipitate a separate sulfide phase. We modified the

solubility models of

ref. 11

so that they fit the experimental

results of

ref. 12

for H

O and

solubility. In Figure 3a and b, the calculation begins at a total pressure of 120 bar, a

temperature of 1190°C, and an

for the initial (i.e., undegassed) melt (QFM+1.1) that

is slightly higher than the highest

constraint from this study; the volatile contents of

the initial melt were set at 1.05 wt% H

O, 1370 ppm S, and 0 ppm CO

, near the highest

volatile contents measured in the samples from this study (except for the assumed zero

concentration of CO

; if the

initial melt is allowed to have CO

at the 100

200 ppm level,

the model curves predict less total reduction in

). We assume the melt has the major

element composition of sample SR0914

10.50, a high

SiO

glass with high S content

(1370 ppm S). The model

ing assumes that the initial melt is vapor saturated at the starting

pressure and temperature of the calculation. We calculated redox conditions and

coexisting vapor and silicate liquid compositions by progressively decreasing total

pressure (at a constant

temperature of 1190°C) from the assumed starting pressure of 120

bars down to 1 bar both for fractional (red curve, Fig. 3a, b) and batch (blue curve, Fig.

3a, b) degassing.

Supplementary figure 8 shows the relationship between

gas phase and melt

redox d

efined by the degassing models used in the main text. We show degassing

trajectories for a melt containing H

O as the only volatile e

lement (gray long dashed

line, Supplementary Fig.

8), a melt containing S as the only volatile element (solid

gray line,

pplementary Fig. 8

), a melt containing CO

as the only volatile element

(dash

dot

gray

line,

Supplementary Fig. 8

), a melt containing both H

O and S (short

dashed

gray

line,

Supplementary Fig. 8

), and a melt containing H

O, S, and CO

(solid

black line,

pplementary Fig. 8

It is possible (and indeed likely for CO

; e.g.,

ref. 13

) that the least degassed of

our samples were themselves the degassing products of more volatile

rich liquids. In

principle, the original magmatic

before degassing from such melts could be estimated

by incrementally adding the composition of the saturating vapor to the least degassed

melt composition, calculating the

of this volatile

enriched melt and finding the

pressure at which it is vapor sa

turated, and then repeating this up to arbitrarily high

volatile contents and pressures. However, the most volatile

rich glasses in this study are

approximately saturated with a liquid sulfide phase based both on thin section study

and

their position on an

S plot for basaltic melts (Supplementary Fig. 6a;

refs. 7, 8

Degassing (or the addition of gas to) a melt containing sulfide blebs cannot be as simply

modeled as degassing from a sulfide under

saturated melt. For example, when on

decompression, S tra

nsfers from the silicate liquid to a vapor phase, the silicate liquid

would respond by dissolving a fraction of the sulfide phase in order to maintain the

appropriate sulfur content at sulfide saturation. This not only would tend to buffer the S

content of

the melt even while degassing proceeds, but it would likely increase the FeO*

content of the melt (which in turn would influence the sulfide solubility) and would

therefore have feedback on

. These processes have not, to our knowledge, been

quantitativ

ely modeled, and they are beyond the scope of this study. Consequently, we

have not explored any degassing history and

for liquids more volatile

rich than the

HSDP2 samples with the highest S and H

O contents.

IMPACT OF PRESSURE AND TEMPERATURE ON SIL

ICATE MELT

f

RELATIVE

QFM

The mantle source for Hawaiian magmas undergoes melting at higher

pressure and temperatures than the mantle source for mid

ocean ridge magmas. We

calculated the effect that this will have on absolute

calculated from

∑

ratios, as well as

relative to the QFM oxygen buffer

, using the algorithm of

ref. 14

and the definition of the QFM oxygen buffer according to

ref. 15

Supplementary figure 9

shows the results of these calculations, and demonstrates that increasing

pressure from 1

to 3 GPa causes an increase in the absolute

of both silicate melt and the QFM oxygen

buffer, but that the absolute

of the silicate melt increases slightly more than that of

QFM, leading to a change in

relative to QFM of ~0.14 log units for the silicate melt

at 3 GPa.

The opposite is true for a silicate melt at higher temperature. The difference in

absolute

between a silicate melt and QFM decreases with increasing temperature by

<0.02 log unit from

1200°C to 1400°C.

We conclude that the higher pressures and

temperatures of equilibration for primary melts in Hawaii can account for ~0.13 log units

of the observed offset between the

of undegassed Hawaiian melts and undegassed

MORB.

THE EFFECT OF OL

IVINE REMOVAL ON Fe REDOX

Hawaiian primary magmas are thought to be in equilibrium with Fo90

Fo91

olivine, and as a result, likely crystallize significantly higher mass fraction of olivine

prior to eruption than MORB magmas at the same MgO content. We cal

culated the

∑

Fe ratios and

of primary mantle melts for MOR and Hawaiian settings by

taking an undegassed composition from each setting and adding the equilibrium

composition olivine back to the melt in 0.1% increments, treating Fe

and Fe

conservative elements

(Supplementary Fig. 10

)

. We use a K

Fe2+/Mg

between olivine and

melt of 0.34. We recalculate the equilibrium olivine composition at each increment, and

continue the calculation until the melt composition is in equilibrium with Fo8

9, Fo90,

and Fo91 olivine.

A MORB primary melt in equilibrium with Fo90 olivine has 16 wt%

MgO, 0.12 Fe

∑

Fe, and has

~QFM

0.3. A Hawaiian primary melt with the same

∑

Fe ratio and

is in equilibrium with Fo93.5 olivine and has 26 wt% MgO, both

of which are considered too high to be reasonable for Hawaiian mantle and melts. If

Hawaiian primary melts are in equilibrium with Fo91 olivine, they have 19 wt% MgO,

0.14 Fe

∑

Fe, and fO2 ~QFM+0.2. We conclude that the difference in primary melt

composi

tions between Hawaii and MOR settings can account for ~0.5 log units of the

observed offset between undegassed Hawaiian and MORB melts.

FIGURE CAPTIONS

Figure 1. A stacked plot of pre

edge features on the same

standard

glass, analyzed at

beamline X26a at

NSLS (light gray line) and beamline 13

IDE at APS (black line). The

spectrum at NSLS was collected using a Si [311] monochromator. The spectrum at

APS was collected using a Si [111] monochromator.

Figure 2. A plot of a pre

edge feature from standard glass LW

0 and the fit

parameters, obtained using the methods described here.

Figure 3. A comparison of

ratios calculated from spectra collected on the same

glasses from APS and NSLS, using area ratio calibrations described here. Error bars

represent the uncertainty on spectral fits using the monte carlo boot strap simulation

described here.

Figure 4. Sulf

ur absorption spectra of (a) endmember composition glasses, and (b)

unknown HSDP2 glass SR0915

0.4.The red dashed line in panel b represents the fit to

the unknown data, obtained using linear combinations of the spectra in panel a.

Figure 5. Two sulfur ab

sorption spectra demonstrating the effects of beam damage on the

intensity of the S

absorption feature.

Figure 6. Plots of S versus (a) FeO* concentrations, and (b) H

O concentrations in

HSDP2 submarine glasses. The dark black line in panel a is the sul

fur con

centrations

MORB that are sulfide saturated, from Mathez et al. (1974) and modified from Seaman et

al. (2004).

Figure 7.

(a) Fractional (black curve) and batch (gray curve) degassing calculations as in

(Fig. 3a, b main text),

where H

O is assume

d to be the only volatile component dissolved

in the melt in order to isolate the effects of degassing of H

O from melts from those of S

and CO

. White circles are submarine glass and melt inclusions from Mariana arc

volcanoes (7), gray circles are submari

ne glass from the Mariana Trough back

arc

spreading center (7), and black circles are submarine MORB glasses (44). (b) Fractional

(black curve) and batch (gray curve) degassing calculations as in (a), where S is assumed

to be the only volatile component in

the melt in order to isolate the effects of degassing of

S from melts from those of H

O and CO

Figure 8. Plot of magmatic

versus pressure for several

batch

degassing trajectories,

using the D

COMPRESS algorithm presented by Burgisser et al. (2015).

Figure 9

. Plots of (a) pressure and (b) temperature versus

for silicate melts (squares

and inverted triangles) and the QFM oxygen buffer (solid and dashed bl

ack lines).

The

is calculated according to Kress and Carmichael (1991) and QFM is defined

according to Frost (1991).

Figure 10

. Plot of

ratios versus MgO concentration for submarine glasses from

MORB (Cottrell and Kelley; 2011) and HSDP

2 (this study).

Calculations show olivine

addition trajectories for undegassed MORB and HSDP

2 samples, with endpoints at

melts in equilibrium with Fo89,

Fo90, and Fo91 olivine.

Figure 11. Plots with the results of fractional degassing models calculated using the D

COMPRESS software from Burgisser et al. (2015), showing pressure (bars) versus (a)

of the melt, and (b, c) the Holland

factor of the gas

phase. The solid black curves

represent modern degassing scenarios from OIB (stars, same degassing model from Fig.

3a, b, for HSDP2 glasses), MORB (circles; initial melt has

= QFM+0.3, 0.1 wt%

O, 200 ppm CO

, 1700 ppm S), and arc (triangles, initial

melt has

= QFM+1.5, 4.5

wt% H

O, 800 ppm CO

, 1982 ppm S) settings. All models assume no sulfide phase is

allowed to precipitate. We use the same solubility models as for the HSDP2 models

described in the text. The curve for MORB degassing extends to 1

bar pressure, however

the average pressure of MORB eruptions is 300 bars (marked with a horizontal gray line

on all panels to demonstrate the melt and gas phase chemistries of MORB magmas on

eruption to the seafloor). The arc degassing curve approximates

the trend in volatile

contents and

of melt inclusions from Agrigan volcano in the Mariana arc

(Kelley and

Cottrell, 2012)

. The gray curve is a fractional degassing scenarios for a reduced (initial

melt has

= QFM

2.3, 0.75 wt% H

O, 250 ppm CO

, 2500 ppm S) basaltic magma.

The

factor is calculated from the output gas chemistries from the D

COMPRESS

software of Burgisser et al. (2015), as in equation 1 from the text.

Figure 12.

Plots of

ratios versus (a) sulfur concentrations, (b) H

concentrations, (c) total Fe expressed as FeO, and calculated magmatic fO2 relative to

QFM versus (d) sulfur concentrations, (e) H

O concentrations, (f) total Fe expressed as

FeO in HSDP2 submarine glasses and olivine hosted melt inclusions from Moussalla

m et

al. (2014, black squares) and Cottrell and Kelley (2012, white squares). Calculated

for Agrigan and Erebus magmas are taken directly from the original publications. The

s for HSDP2 submarine glasses are calculated at 1200 C and 1 atm according

to the

algorithm of Kress and Carmichael (1991) relative to the position of the quartz

fayalite

magnetite oxygen buffer according to Frost (1991).

REFERENCES CITED

Cottrell, E, Kelley, K. A, Lanzirotti, A, Fischer, R. A (2009)

High

precision

determination

of iron oxidation state in silicate glasses using XANES.

Chem

Geol

268:

167

179, doi:10.1

016/j.chemgeo.2009.08.008

Berry

, A.J, O’Neill, H.S.C, Jayasuria, K.D, Campbell, S.J, Foran, G.J

(

2003

) XANES

calibrations for the oxidation state of iron in a silica

t glass.

Am Mineral

88(7):

967

977.

Kelley, K. A, Kingsley, R, Schilling, J.

G (2013)

Composition of plume

influenced mid

ocean ridge lavas and glasses from the Mid

Atlantic Ridge,

East Pacific Rise, Galápagos Spreading Center, and Gulf of Aden.

Geochem

Geophys Geosys

14:

223

242, doi:10.1002/ggge.20049

Jugo, P. J, Wilke, M, Botcharnikov, R. E (2010)

Sulfur K

edge XANES analysis

of natural and synthetic basaltic glasses: Implications for S speciation and S

content as function of oxygen fugacity.

Geochim

Cosmochi

Act

74:

5926

5938, doi:

10.1016/j.gca.2010.07.022

Klimm, K, Kohn, S. C, O'Dell, L. A

, Botcharnik

ov, R. E, Smith, M. E (2012)

The

dissolution mechanism of sulphur in hydrous silicate melts. I: Assessment of

analytical techniques in determining the

sulphur speciation in iron

free to

iron

poor glasses.

Chem Geol

322

323:

237

249,

doi:10.1016/j.chemge

o.2012.04.027

Klimm, K, Kohn, S. C, Botcharnikov, R. E (2012)

The dissolution mechanism of

sulphur in hydrous silicate melts. II: Solubility and speciat

ion of sulphur in

hydrous silicate melts as a function of

Chem Geol

322

323:

250

267,

doi:10.1

016/j.chemgeo.2012.04.028

Mathez, E.A (1976)

Sulfur solubility and magmatic sulfides in submarine

basalt glass.

J Geophys Res

81: 4269

4276

Seaman, C,

Sherman, S.B, Garcia, M.O, Baker, M.B, Balta, B, Stolper, E (2004)

Volatiles in glasses from the HSDP2 drill core.

Geochem Geophys Geosys

doi:10.1029/2003GC000596.

Moussallam, Y, Oppenheimer, C, Scaillet, B

, Gaillard, F

, Kyle, P, Peters, N,

Hartley, M,

Berlo, K

, &

Donovan, A (2014)

Tracking the changing oxidation

state of Erebus magmas, from mantle to surface, driven by magma ascent and

degassing.

Earth Planet Sci Lett

393: 200

209.

10.

Kelley, K,

Cottr

ell, E (2012)

The influence of magmatic differentiation

on the

oxidation state of Fe in a basaltic arc magma.

Earth Planet Sci Lett

329: 109

121.

11.

Burgisser, A, Alletti, M, Scaillet, B (2015)

Simulating the behavior of volatiles

belonging to the C

–

S system in silicate melts under magmatic

conditions with the

software D

Compress.

Comps

Geosc

14,

doi:10.1016/j.cageo.2015.03.002

12.

Dixon

, J.E, Stolper, E.M, Holloway, J.R

(

1995

) An experimental study of water

and carbon dioxide solubilities in mid

ocean ridge basaltic liquids. Part I:

Calibration and solubility models.

J Petrol

36(6): 1607

1631.

13.

Gerlach

, T.M,

Gra

ber

, E.J

(

1985

) The volatile budget of Kilauea volcano.

Nature

313

: 273

277.

14.

Kress, V. C, Carmichael, I. S. E (1991)

The compressibility of silicate liquids

containing Fe2O3 and the effect of composition, temperature, oxygen

fugacity and pressure on their redox states.

Contrib

Mineral

Petrol

108

: 82

92.

15.

Frost, B. R (1991

)

Introduction to oxygen fugacity and its petrologic

importance.

Rev Mineral Geochem

: 1

7108

Energy (eV)

NSLS Si [311] mono

APS Si [111] mono

Standard glass LW-0

Supplementary Figure 1

Arbitrary absorption units

0.00

0.05

0.10

0.15

0.20

0.25

7106

7108

data, LW-0, this study

total fit

background fit curves

Supplementary Figure 2

Gaussian fit curves

Arbitrary absorption units

0.14

0.15

0.16

0.17

0.18

0.19

0.20

0.14

0.15

0.16

0.17

0.18

0.19

0.20

∑

Fe (APS 13-IDE)

∑

Fe (NSLS X26a)

1:1

Supplementary Figure 3

0.0

2.0

4.0

6.0

8.0

10.0

12.0

2445

2465

2485

2505

2525

2545

Energy (eV)

absorption (arbitrary units)

0.0

0.2

0.4

0.6

0.8

1.0

1.2

1.4

1.6

1.8

2.0

2445

2465

2485

2505

2525

2545

absorption (arbitrary units)

PC-35 glass

TR101-15D-8g

SR0915-0.4 data, 1310 ppm S

SR0915-0.4 fit

91.5% TR101-15D-8g

8.5% PC-35

Supplementary Figure 4

-0.5

0.5

1.5

2.5

2445

2455

2465

2475

2485

2495

2505

E (eV)

Mariana trough glass, 553 ppm S

spot 1, 11 minutes exposure to beam

spot 10, 110 minutes exposure to beam

Supplementary Figure 5

0.2

0.4

0.6

0.8

1.0

O (wt%)

200

400

600

800

1000

1200

1400

1600

1800

2000

9.0

10.0

1.0

12.0

13.0

14.0

S (ppm)

FeO* (wt%)

SCSS

Sulfur

degassing

high SiO

HSDP2 glasses

low SiO

HSDP2 glasses

O/S = 4

O/S = 7

Supplementary Figure 6

4.0

5.0

6.0

7.0

8.0

9.0

MgO (wt%)

200

400

600

800

1000

1200

1400

1600

1800

2000

S (ppm)

0.0

0.2

0.4

0.6

0.8

1.0

O (wt%)

4.0

5.0

6.0

7.0

8.0

9.0

MgO (wt%)

S (ppm)

SupplementaryFigure7

0.8

fractional

batch

Oonl

,4.5wt%

0.2

0.4

0.6

0.8

1.0

1.2

1.4

1.6

1.8

2.0

2000

4000

6000

∆QFM(1190ºC,P)

fractional

batch

Sonl

,6000ppm

S(ppm)

O(wt%)

∆QFM(1190ºC,P)

2000 bars

500

100

1000

2000 bars

1000

500

100

arc

back-arc

MORB

0.4

1.2

1.6

2.0

400

800

1200

1600

2000

2400

2800

-0.5

0.0

0.5

1.0

1.5

2.0

2.5

(bar)

∆

QFM, melt

Supplementary Figure 8

O only,

5.0 wt %

only,

1000 ppm

S only,

6000 ppm

O-S

5.0 wt % H

2000 ppm S

O-S-CO

1.5 wt % H

2500 ppm S

1700 ppm CO

0.00

1.00

2.00

3.00

4.00

5.00

6.00

-8

-7

-6

-5

-4

-3

-2

-1

Pressure (GPa)

log (

)

QFM, 1300

C (constant)

MORB, 1300

C (constant)

QFM, 1450

C (constant)

undegassed HSDP,

1300

C (constant)

undegassed HSDP,

1450

C (constant)

150

1200

1250

1300

1350

1400

-8

-7

-6

-5

-4

-3

-2

-1

QFM, 1 GPa (constant)

MORB, 1 GPa (constant)

QFM, 3 GPa (constant)

undegassed HSDP,

1 GPa (constant)

undegassed HSDP,

3 GPa (constant)

log (

)

Temperature (

Supplementary Figure 9