Determination of the triple oxygen and carbon isotopic composition of CO2 from atomic ion fragments formed in the ion source of the 253 Ultra high-resolution isotope ratio mass spectrometer.

RATIONALE: Determination of δ17 O values directly from CO2 with traditional gas source isotope ratio mass spectrometry is not possible due to isobaric interference of 13 C16 O16 O on 12 C17 O16 O. The methods developed so far use either chemical conversion or isotope equilibration to determine the δ17 O value of CO2 . In addition, δ13 C measurements require correction for the interference from 12 C17 O16 O on 13 C16 O16 O since it is not possible to resolve the two isotopologues.
METHODS: We present a technique to determine the δ17 O, δ18 O and δ13 C values of CO2 from the fragment ions that are formed upon electron ionization in the ion source of the Thermo Scientific 253 Ultra high-resolution isotope ratio mass spectrometer (hereafter 253 Ultra). The new technique is compared with the CO2 -O2 exchange method and the 17 O-correction algorithm for δ17 O and δ13 C values, respectively.
RESULTS: The scale contractions for δ13 C and δ18 O values are slightly larger for fragment ion measurements than for molecular ion measurements. The δ17 O and Δ17 O values of CO2 can be measured on the 17 O+ fragment with an internal error that is a factor 1-2 above the counting statistics limit. The ultimate precision depends on the signal intensity and on the total time that the 17 O+ beam is monitored; a precision of 14 ppm (parts per million) (standard error of the mean) was achieved in 20 hours at the University of Göttingen. The Δ17 O measurements with the O-fragment method agree with the CO2 -O2 exchange method over a range of Δ17 O values of -0.3 to +0.7‰.
CONCLUSIONS: Isotope measurements on atom fragment ions of CO2 can be used as an alternative method to determine the carbon and oxygen isotopic composition of CO2 without chemical processing or corrections for mass interferences.

INTRODUCTION

Oxygen has three stable isotopes,16O,17O and 18O, with average terrestrial abundances of 99.76%, 0.04% and 0.21%, respectively. These abundances can be changed by kinetic and equilibrium fractionation processes and other physicochemical effects. Variations in isotopic abundance are reported as deviations of a heavy‐to‐light isotope ratio in a sample relative to a reference material. In the case of oxygen isotopes, the two isotope ratios are 18R = [18O]/[16O] and 17R = [17O]/[16O] and the international standard is Vienna Standard Mean Ocean Water (VSMOW).

Since isotope variations are small, they are usually reported in per mill (‰). Most isotope fractionation processes depend on mass. For oxygen isotopes, this results in fractionation patterns where the fractionation in 17O is approximately half of the fractionation in 18O (Equation 3).

The factor λ ranges from 0.5 to 0.53 for such mass‐dependent fractionation processes.1, 2, 3 Ozone photochemistry is a well‐known exception to this rule, and O3 and related gases have a large oxygen isotope anomaly, expressed as Δ17O and referred to as mass‐independent fractionation. We use the logarithmic definition to calculate Δ17O of CO2 (Equation 4).2, 4, 5 Note that the choice of λ is arbitrary since a variety of sources contribute to the isotopic composition of tropospheric CO2 with different fractionations and different three‐isotope slopes. In this study we used a λ value of 0.528 to calculate the Δ17O of CO2 following Barkan and co‐workers6, 7 and the 17O‐correction algorithm by Brand et al.8

Since the discovery of mass‐independent fractionation,9 the Δ17O value has been used to study sources/sinks of atmospheric trace gases and chemical reaction pathways. Several studies have shown that CO2 acquires Δ17O from O3 via photochemical isotope exchange in the stratosphere.10, 11, 12, 13, 14, 15, 16, 17 When this CO2 re‐enters the troposphere18, 19, 20 the Δ17O is successively reduced by oxygen isotope exchange with leaf, soil and ocean water. Isotopic exchange of CO2 with leaf water is more efficient than with ocean water due to the presence of carbonic anhydrase in the leaves, and as a result the main sink for the Δ17O of CO2 is exchange with leaf water. Precise measurements of the Δ17O of CO2 may therefore help to better constrain the exchange of CO2 between the atmosphere and the biosphere/hydrosphere. For several processes it has been shown that Δ17O is a more suitable tracer than the δ18O value alone.21, 22, 23, 24

Determination of Δ17O in CO2 with traditional isotope ratio mass spectrometry techniques remains challenging due to the isobaric interference of 13C16O16O (exact mass 44.9932) and 12C17O16O (exact mass 44.9940). Resolving these two isotopologues requires a mass resolving power (m/Δm) of ~56,000, far beyond the resolving power of most traditional mass spectrometer systems. Different alternative techniques have been developed to measure the δ17O value of CO2: (1) CO2 fluorination and isotopic measurement of the released O2 25; (2) conversion of CO2 into H2O and CH4 followed by H2O fluorination and isotopic measurement of the released O2 26; (3) isotope exchange between CO2 and CeO2 27, 28, 29 or CuO30 with known oxygen isotopic composition and measurement of the δ45CO2 value before and after exchange to calculate the δ17O value of CO2; (4) isotope exchange between CO2 and CeO2 followed by isotope analysis of the equilibrated CeO2 by laser fluorination31; (5) equilibrium exchange of CO2 with H2O followed by fluorination of H2O and measurement of the isotopic composition of released O2 6, 32; (6) isotope exchange between CO2 and O2 over hot platinum and measurement of the isotopic composition of oxygen before and after exchange to calculate the δ17O value of CO2.7, 33 All these methods require either chemical conversion or isotope exchange, which can introduce procedural errors. In recent years, laser‐based absorption spectroscopy techniques to determine δ17O values and other isotope signatures of CO2 from air samples have been developed.34, 35, 36

Very small variations in the δ13C value are used to quantify fluxes between atmosphere and hydrosphere and/or ocean37, 38, 39, 40, 41. Due to the mass interference of 12C17O16O and 13C16O16O,8, 40, 42, 43, 44, 45, 46 the measurements of δ13C values require an appropriate correction for 17O‐interference. Different “17O correction” algorithms are in use to correct for the interference of 12C17O16O on the value of δ13C, causing discrepancies between different correction algorithms used. The discrepancies in the δ13C value introduced by different 17O correction algorithms (i.e. different λ, 17R, 13R) are explored by Assonov and Brenninkmeijer42 in detail. They reported a discrepancy of 0.058‰ for tropospheric CO2 with δ45(CO2) and δ46(CO2) values of −9.2‰ and +2.180‰ vs NBS19‐CO2 between the algorithm by Allison et al47 and that by Santrock et al45 due to differences in the values of 17R and λ. The discrepancies introduced by 17O correction algorithms depend on the δ46(CO2) values44 resulting in a different 17O correction for CO2 having the same δ45(CO2) value but a different δ46(CO2) value. By design, most of the 17O correction algorithms do not consider the Δ17O of the CO2 and the ones that do include Δ17O require precise measurement of the δ17O value of CO2. For instance, the algorithm of Allison et al47 introduces an error ranging from −0.78 to −0.13‰ for stratospheric CO2. Nevertheless, the error introduced to the δ13C value because of the use of different values of λ is different for CO2 with different Δ17O even if the same algorithm is used. It is desirable to use an alternative technique that enables the determination of the δ13C value without a bias introduced due to the 17O correction algorithm for better use of the δ13C values as a tracer to quantify fluxes between atmosphere and hydrosphere and ocean.

Recently developed high‐resolution isotope ratio mass spectrometers48, 49 are designed to overcome limitations of traditional isotope ratio mass spectrometer systems in terms of mass resolution and sensitivity. In this study, we present a technique to determine the isotope composition of CO2 from the C+ and O+ fragment ions, which are produced from CO2 in the ion source of two 253 Ultra (Thermo Fisher Scientific, Bremen, Germany) instruments installed at Utrecht University and the University of Göttingen.

Isotope measurement of fragment ions is not a new concept. The method has been deployed, for example, to study the intramolecular distribution of 15N+ in N2O,50, 51, 52, 53, 54 to determine the site‐specific carbon isotopic composition of propane55 and to measure sulfur isotope ratios in COS.56

Here we establish an analytical method to determine the δ17O, δ18O and δ13C values of CO2 directly on the C+ and O+ fragment ions of CO2 without any chemical manipulation of the CO2 molecule. Notably, this method provides an independent technique to measure Δ17O of CO2 and the results are validated by comparison with the existing CO2‐O2 exchange method and by measuring CO2 with known Δ17O.

EXPERIMENTAL

The 253 Ultra instrument

The 253 Ultra is the commercial version of a high mass resolution gas source multi‐collector mass spectrometer, which was pioneered with the MAT 253 Ultra prototype in 2012.48, 57 The high mass resolution of the 253 Ultra enables the investigation of the abundance of isotopologues that suffer from isobaric interferences. The mass resolving power of the instrument can be tuned to m/Δm >35,000 and the peak stability over time is <5 ppm in mass; m/Δm is the width of a peak flank between 5% and 95% of the maximum peak signal. The instrument is controlled by the Qtegra™ software package (Thermo Fisher Scientific).

The ion source of the 253 Ultra is connected to a sample introduction system of four variable volume reservoirs that can be filled with sample or reference gases. The control of the ion source chemistry (adduct formation, fragmentation, formation of metastable ions, linearity and exchange reactions of the sample gas with adsorbed species at the inner ion source surfaces) is critical for accurate isotope ratio measurements. The differentially pumped ion source can be baked to high temperature and is fitted with a variable ion source conductance (VISC) window to adjust the source pumping conductance and to control the residence time of the sample gas in the ionization volume, which is one critical parameter for ion source chemistry. The source slit can be switched to three different slit sizes for low‐, medium‐ and high‐resolution settings. For the instruments at Utrecht University and the University of Göttingen the slit widths are 250 μm, 16 μm and 5 μm. The intermediate aperture at the entrance of the magnetic sector allows an extra‐high‐resolution mode to be selected to achieve m/Δm >35,000 mass resolving power. It should be noted that higher resolution comes at the cost of lower ion beam intensities.

The basic setup of the instrument follows a double‐focusing Nier Johnson geometry with a 90o deflection angle of the electrostatic sector (r = 22.4 cm) and the magnetic sector (r = 23 cm) as shown in Figure 1. Double focusing means that there is stigmatic focusing of the ions passing the source slit regardless of the angular and energy distribution in the ion beam. Usually low‐resolution sector mass analyzers are of the single‐focusing type, i.e. just a magnetic sector. The mass resolving power of a single‐focusing system is limited by the chromatic aberration caused by the energy spread of the ions generated in the ion source. Double focusing can overcome this limitation. In a properly designed double‐focusing system the electrostatic sector optics match the chromatic aberrations of the magnetic sector optics such that the combined system eliminates both, the angular and the chromatic aberrations up to the second order.58

Figure 1

Ion optical layout of the Thermo Scientific 253 Ultra high‐resolution isotope ratio mass spectrometer. In the ion source, the ions are accelerated to 5 keV onto the source slit. After the electrostatic analyzer the ions are accelerated to 10 keV just before passing the crossover. The switchable intermediate aperture behind the magnetic sector is used for extra high mass resolution settings and the zoom lens allows for fine adjustments of peak overlap. The variable multicollector assembly is mounted on the focal detector plane of the mass spectrometer system. The RPQ filter lens discriminates for scattered ions and reduces abundance sensitivity. It is located behind the focal plane right in front of the ion counting detector [Color figure can be viewed at wileyonlinelibrary.com]

In the 253 Ultra the ions are generated at a potential of 10 kV. The ions are accelerated to the source slit of the double‐focusing mass analyzer at a kinetic energy of 5 keV. After passing through the electrostatic analyzer the ions are further accelerated to 10 keV kinetic energy before they pass through the magnetic sector where the ion trajectories are split up according to their mass. Finally, the ions are focused along the focal detector plane of the mass analyzer. The two‐stage acceleration of the ion beam allows a very compact design of the electrostatic sector geometry, which otherwise would have required the radius of the electrostatic sector to be about twice as large as that of the magnetic sector. Due to its compact geometry, the ion optical setup of the 253 Ultra fits onto just one monolithic base plate. The resonance frequency of this rigid mechanical construction is very high and precise, which makes the system robust against low‐frequency vibrations that usually occur in buildings. In order to achieve ultimate stability, the complete mass analyzer and the electronics are housed in a shielded temperature‐stabilized cabinet to be robust against temperature fluctuations in the lab (±2°C).

The variable detector array supports eight moveable detector platforms, which are equipped with Faraday detectors that can be read out with selectable resistors with resistances between 3 × 108 Ω and 1013 Ω. The three collector platforms at the high mass end are additionally equipped with compact discrete dynode ion counting detectors59 next to the Faraday detectors. The axial detector channel is fixed in position and supports a dual‐detector arrangement, where the ion beam can be switched between a Faraday cup and an ion‐counting channel. The axial ion‐counting detector is equipped with a retardation lens (RPQ‐lens) to reject scattered background ions originating from scattering events along the ion optical flight path (apertures, residual gas particles) which leads to an abundance sensitivity in the ppb range.48

Characterization of the 253 Ultra for CO2 measurement

We investigated the effect of equilibration time, emission current, source conductance and signal intensity on the ionization of CO2 as suggested by Verkouteren et al58, 60 and Meijer et al.61 We characterized the scale contraction effect of the ion source of the 253 Ultra at Utrecht University using two CO2 gases (G1 and SCOTT, see Table 1 for details). The characterization of the instrument is performed at low resolution (250 μm entrance slit width, m/Δm ~2000) with five Faraday collectors that are read out with resistors of 3 × 108 Ω, 1 × 109 Ω, 3 × 1010 Ω, 1 × 1011Ω and 1 × 1011 Ω for m/z 44, 45, 46, 47 and 48. The corresponding collectors used for this measurement are L2, L1, Center, H1 and H2 for m/z 44, 45, 46, 47 and 48, respectively. Here, only data corresponding to m/z 44 to 46 are presented. The ion signal of the high intensity ion beam (m/z 44) is adjusted before each acquisition to 3.2 × 1011 cps (counts per second) with an allowed difference of 1 × 1010 cps between the two bellows that are used for the measurement. Under these conditions the ion source pressure is 2.5 × 10−7 mbar. The reference measurement is performed with 9.9 kV accelerating voltage, filament emission current of 1.8 mA, equilibration time of 60 s, integration time of 67.1 s and with the VISC window closed.

Table 1

Overview of names, suppliers and isotopic compositions of the CO2 and O2 working standards used in this study. All the CO2 gases used have a purity of 99.995% and O2 gases have a purity of 99.9998%

CO2 working reference gases
Name	Supplier	δ13C vs VPDB [‰]	δ18O vs VSMOW [‰]
G1	Air Products, Germany	−39.47 ± 0.012	4.843 ± 0.013
G2	Linde Gas, The Netherlands	−31.733 ± 0.008	34.998 ± 0.023
G5	Air Products, Germany	−10.445 ± 0.010	30.404 ± 0.020
SCOTT	Air Products, Germany	−2.900 ± 0.011	25.803 ± 0.015

To study the effect of equilibration time and source conductance, we measure the two gases with equilibration times of 10, 20, 30, 40, 50, 60 and 90 s with the VISC window open and closed. The effect of the emission current is quantified by setting the emission current to 1 mA, 1.5 mA and 1.95 mA. To investigate the effect of signal intensity (cps for m/z 44), three experiments with 2.5 × 1011 cps, 1.5 × 1011 cps and 9 × 1010 cps for m/z 44 are performed. Note that measurements to characterize the effect of emission control current and signal intensity are performed with an equilibration time of 30 s, so they cannot be directly compared with the reference measurement with an equilibration time of 60 s. The effect of cross contamination is calculated according to Meijer et al61 using Equation 5. To calculate the change in scale contraction with changes in equilibration time, we compare the relative difference of the two gases (in δ13C and δ18O values) measured at different equilibration times with the value obtained at 90‐s equilibration time. Similarly, the scale contraction due to the emission current is calculated with respect to the results obtained at an emission current of 1 mA. The cross contamination (η) is calculated as: where y is 13 (for δ13C) or 18 (for δ18O), the index a indicates the respective δ value under reference conditions (90‐s equilibration time and 1 mA emission current), and index m indicates the δ value at a different equilibration time or different emission current.

To link our results to international isotope scales, we use a set of isotopically different pure O2 and CO2 reference gases. Multiple aliquots of each gas were sent to Eugeni Barkan from the Hebrew University of Jerusalem (Jerusalem, Israel) for analysis. This research group also provides high‐precision δ17O values and has established a direct link between the oxygen isotope scales of O2 and CO2. The reported results were assigned to our reference gas cylinders, which were also measured extensively on the Thermo Scientific Delta Plus XL™ instrument in our laboratory and on the 253 Ultra. The appropriate scale contraction factors (see Section 4) are used to convert the raw data into the scale of the Hebrew University of Jerusalem.6, 62, 63

Fragment method

The 17O+ fragment ion measurements at Utrecht University are performed at medium resolution (16 μm entrance slit width, m/Δm >7500) with the “reference” source settings mentioned above, i.e., emission current of 1.80 mA, accelerating voltage 9.9 kV, VISC window closed. The ion signals are registered in three Faraday collectors (L3, Center, H3) that are read out with resistors of 1 × 1011 Ω, 1 × 1013 Ω and 1 × 1013 Ω for m/z 16, 17 and 18, respectively. The ion signal intensity is adjusted before each acquisition to 9.2 × 108 cps on m/z 16, which corresponds to a source pressure of ~2.5 × 10−7 mbar, with a tolerance of 3 × 106 between the bellows. Reasonable source pressures for fragment ion measurement are determined to fall between 2.0 and 4.5 × 10−7 mbar (resulting in major ion beam signals of 0.75 to 1.25 × 109 cps at medium resolution), corresponding to the linear portion of the source pressure vs signal intensity relationship for m/z 16 (Figure S1, supporting information). The integration and equilibration times are 67.1 and 60 s, respectively, which implies that in a measurement cycle both sample and reference are measured for 67.1 s out of 254.2 s, i.e., 26% of the time. Figure 2 shows the mass spectra covering the range of m/z 16, 17 and 18. The main interference for the 17O+ ion (mass 16.9991 u) is OH+ (mass 17.0027 u). The mass difference between these two ions is only 0.0036 u. With the 253 Ultra, they are sufficiently separated using the medium‐resolution slit to enable measurement of 17O+ on a narrow plateau without interference from OH+. In this study the medium‐resolution slit is chosen since the plateau is sufficiently flat and gives a sufficient signal to allow stable positioning for 17O+ measurement, as shown in Figure 2. The width of the plateau can in principle be increased by going to high mass resolution, but this would result in a reduction of the ion current by a factor of 3 and a corresponding increase in the required measurement time to reach a certain precision. For 18O+ (mass 18.9984 u) the mass difference to its main interference H2O+ (19.0148 u) is 0.0164 u which results in a broad shoulder even at medium mass resolution. The potential effect of other interferences is discussed below.

Figure 2

Medium‐resolution mass spectra for measurement of 16O+, 17O+ and 18O+ fragment ions of CO2. The shaded area shows the region of the shoulder where 17O+ is measured interference‐free, a magnified view is shown in the right panels. The mass scale (x‐axis) applies to the middle panels (17O) for the top and bottom panels; the mass scale is shifted one mass down or up, respectively [Color figure can be viewed at wileyonlinelibrary.com]

Small shifts in the mass scale regularly lead to a deterioration of measurement precision, when the mass position shifts away from the small 17O+ shoulder. This can be largely circumvented by resetting the mass scale at regular time intervals during the measurement. The present version of the Qtegra software does not allow automatic positioning on a shoulder of multiple overlapping peaks. Therefore, the collector configuration is carefully arranged such that the center of the m/z 16 peak is precisely located at the shoulder of the m/z 18 and m/z 17 peaks where 17O+ and 18O+ can be measured interference‐free. A peak centering is then performed on m/z 16 before each acquisition which is precise enough to relocate the system on the narrow shoulder of the m/z 17 peak. Nevertheless, instabilities in the mass scale are still considered a main contributor to the remaining error above counting statistics, and an automatic positioning routine that scans the 17O+ shoulder directly to reposition the peak might improve the precision.

All 17O+ fragment ion measurements on the 253 Ultra at the University of Göttingen are performed at medium resolution (16 μm entrance slit width, m/∆m ~7500) with 9.85 kV accelerating voltage and 1.85 mA emission current, with the VISC window closed. The integration and equilibration times are 67.1 and 12 s, respectively, which implies that in a measurement cycle both sample and reference are measured for 67.1 out of 158.2 s, i.e., 42.4% of the time. Three Faraday collectors (L3, Center, H3), equipped with 1 × 1010 Ω, 1 × 1013 Ω and 1 × 1012 Ω resistors, are used to detect the ion signals for m/z 16, 17 and 18, respectively. The signal intensity is adjusted per acquisition on m/z 16, with a target intensity of 1.2 × 109 cps (tolerance 0.2%), corresponding to a source pressure of 4.12 × 10−7 mbar.

The doubly charged 16O18O++ ion is very close in mass to 17O+ (Table S5, supporting information) and interferes at the lower mass shoulder of the 17O+ peak. Figure 3 shows mass spectra recorded at medium resolution using the compact discrete dynode (CDD) collector of the H2 collector unit of the 253 Ultra (H2‐CDD). The interference of 16O18O++ can be detected 0.002 mass units before the larger 17O+ peak starts. The 16O18O++ ion is formed in the ion source, probably from the recombination of 16O and 18O atom fragments. Therefore, the contribution of 16O18O++ to 17O+ depends on the 18O content of the gas, and it has to be corrected to avoid a systematic bias in the δ17O determination when the δ18O values of the sample and the working reference gas are different. Figure 3C shows that the 16O18O++ signal increases relative to the 17O+ and 18O+ signals towards lower source pressures but it is quite stable at pressures above 10−7 mbar. At 2.5 ×10−7 mbar, where our measurements were carried out, the 16O18O++ signal is 0.055% of the 18O+ signal, which results in a 16O18O++ contribution of about 0.3% to the 17O+ ion beam. Based on this correction factor, Figure 3D shows the calculated effect of 16O18O++ on the measured δ17O values, as a function of the δ18O difference between sample and working reference gas and for different source pressures. The correction is probably instrument and tuning‐dependent and should be determined regularly. We applied a corresponding correction to the data where we compare the results from the O‐fragment method and CO2‐O2 exchange method.

Figure 3

Interference of 16O18O++ on the measurement of the 17O+ fragment ion. A, Mass spectra at different source pressure. B, Zoom to the background signal where the interference of 16O18O++ can be detected starting around mass 17.445, 0.002 mass units before the larger 17O+ peak. The CDD background signals determined in the grey shaded area were subtracted from the signals in the dark shaded area to quantify the contribution from 16O18O++. C, Abundance of the 16O18O++ signal relative to the measured signals 17O+ m and 18O+ m (in %). For source pressures above 10−7 mbar, where our measurements were carried out, the 16O18O++ signal is 0.06% of the 18O+ signal, which results in a contribution of 0.3% to the 17O+ ion beam. D, Bias in the δ17O value introduced by 16O18O++ as a function of the difference in the δ18O value between sample and working gas for different source pressures [Color figure can be viewed at wileyonlinelibrary.com]

The 13C+ fragment ion is measured at Utrecht University at medium resolution (16 μm entrance slit width) with the same emission current, acceleration voltage, integration time and equilibration time as used for the 17O+ fragment method, again with the VISC window closed. The ion signals are registered in two Faraday collectors (L4 and Center) that are read out with resistors of 1.0 × 1011 Ω and 1.0 × 1013 Ω for 12C+ and 13C+, respectively. The mass spectra covering the range for 12C+ and 13C+ are shown in Figure 4. The main interference for 13C+ (mass 13.0034 u) is 12CH+ (mass 13.0078 u), which requires a mass resolving power of 2900. This is well resolved with the medium‐resolution slit of the 253 Ultra (m/Δm >7500).

Figure 4

Medium‐resolution mass spectra for measurement of 12C+ and 13C+ fragment ions of CO2. The shaded area shows the region where the isotope measurements were performed. Measurement of the C fragment is performed at medium resolution. The mass scale (x‐axis) applies to the middle and bottom panels (13C); for the top panel, the mass scale is shifted one mass down

To establish the scale contraction correction for fragment ion measurements, isotopically well‐characterized pure CO2 gases (see section 3.2) were analyzed both with the molecular ion method and with the fragment ion method. The CO2 and O2 working reference gases used in this study are summarized in Table 1. The two CO2 samples, G3 and G4, are prepared from G2 by adding isotopically anomalous CO2 generated by UV‐induced isotope exchange between CO2 and O3.

The reported internal precision of the fragment technique is compared with the expected error (precision) based on counting statistics (EECS), which is calculated as: where N is the average count rate (cps), t int is the integration time in seconds, n is the number of measurement cycles and the factor accounts for the fact that the reference and the sample both introduce the same error to the δ value. Throughout the manuscript the error of a single measurement series is reported as the standard error of the mean. When we quantify errors for more than one measurement (series), we report the standard error times the Student's t‐factor to cover the 95% confidence interval.

O2‐CO2 exchange method

A schematic diagram of the O2‐CO2 exchange experimental setup at Utrecht University is shown in Figure S2 (supporting information). The central part of the CO2‐O2 exchange system is the exchange reactor, which is made of quartz, while the other parts are made from borosilicate glass. The general design is similar to the one in Barkan et al,7 except for some modifications in the ways of introducing CO2 and O2 into the reactor.

Approximately 1.7 mL of pure CO2 with known (measured) δ18O value was expanded to the glass line and trapped cryogenically using liquid nitrogen (LN2) in the calibrated volume (CV, 2.319 mL). The amount of CO2 was precisely determined with a pressure sensor (PS9504, Geological and Nuclear Sciences Ltd, Lower Hutt, New Zealand). The CO2 sample was then transferred cryogenically to the quartz reactor. The trapping in the quartz reactor occurs at the horizontal tube that is continuously cooled using LN2 provided by a microdosing system (Norhof 900 series LN2 cooling system, Ede, The Netherlands). After introduction of the CO2 sample, an approximately equal amount of pure O2 (IMAU‐O2) with known δ17O and δ18O values is admitted to the small volume above the reactor and then expanded into the reactor. The CO2 is then released from the cold tube by stopping the LN2 microdosing system, and the gases are allowed to react for 30 min in the quartz reactor that contains 0.18 g of platinum sponge (99.9% purity, Sigma Aldrich, St Louis, MO, USA) at the bottom, which is heated to 750°C with a temperature‐controlled oven (CFH VC401A06A‐0000R, Kurval, Nieuw‐Vennep, The Netherlands). After 30 min, CO2 is extracted cryogenically in a double U trap, while O2 is collected behind this trap on 3 pellets of molecular sieve 13X (1.6 mm, Sigma Aldrich) at LN2 temperature. The isotopic composition of the exchanged O2 is measured using a dual‐inlet system on the DeltaPlusXL isotope ratio mass spectrometer (Thermo Fisher Scientific) using three Faraday collectors equipped with resistors of 3 × 108 Ω, 3 × 1010 Ω and 3 × 1011 Ω for m/z 32, 33 and 34, respectively. The value of δ17O (CO2) is then calculated from the change in the δ17O(O2) value before (index i = “initial”) and after (index f = “final”) isotope exchange with CO2 based on the following mass balance equation (Eequation 7), after Barkan et al7: where β is the molar ratio of CO2 to O2 and and are the 17O and 18O equilibrium fractionation factors between CO2 and O2 in the presence of the hot platinum catalyst.7 In our CO2‐O2 exchange setup the equilibrium fractionation factors are α17(CO2/O2) = 1.0006657 and α18(CO2/O2) = 1.000998, determined by measuring the isotopic composition of CO2 and O2 after isotope exchange was fully established.

Samples

Preparation of CO2 with known δ17O and δ18O values

At Utrecht University, CO2 with known isotopic composition is prepared by combusting a pure graphite rod (99.9995% purity, Alfa Aesar, Part No: 40765) (Thermo Fisher Scientific) in isotopically known pure IMAU‐O2 (Table 1). The graphite rod (3.05 mm × 32 mm) is wrapped in a sheet of platinum foil and platinum wire and placed inside a quartz reactor as shown in Figure S3 (supporting information). The experimental setup is similar to the one presented in Barkan and Luz,64 except for a modification in the way that CO2 is trapped. The graphite rod is conditioned by heating to 1000°C in vacuum for 2 days. The combustion experiment is performed at 750°C and the CO2 is trapped immediately at LN2 temperature using a collar trap (Figure S3, supporting information) to avoid fractionation due to possible exchange with the graphite. After the O2 has been fully combusted to CO2 (as indicated by the pressure), the reactor is cooled to below 200°C and the collar trap is heated to room temperature (25°C) to release the CO2. The CO2 is collected in a break seal tube at LN2 temperature. After each conversion experiment the graphite rod is re‐conditioned by heating at 900°C for 1 h to avoid contamination from remaining oxygen.

At the University of Göttingen, isotopically light CO2 was produced from combustion with isotopically depleted O2 using a slightly different setup. Instead of using platinum foil and wire as catalyst, the graphite rod was immersed in chloroplatinic acid and dried before being installed in the quartz reactor. Isotopically light oxygen for the reaction was provided by hydrolysis of Antarctic precipitation (Dronning Maud Land, δ2H = −341.1‰ vs SMOW and δ18O = −42.4‰ vs SMOW). After full combustion, the produced CO2 was transferred into a glass vial, which was kept at LN2 temperature.

Preparation of 17O‐enriched CO2

17O‐enriched CO2 is prepared by inducing oxygen isotope exchange between CO2 (G2) and O2 (IMAU‐O2) (via O3 and O(1D))65 using a Hg ultraviolet (UV) lamp (Oriel Instruments, Newport Corporation, Stratford, CT, USA). The borosilicate photolysis reactor is equipped with a UV‐transparent Suprasil™ finger in the center to place the lamp, as shown in Figure S4 (supporting information). 50 mbar of CO2 is expanded into the 2‐L reactor and O2 is then expanded into the reactor until the pressure reading reaches around 1 bar. The mixture is then allowed to photolyze for 18 h without regulating the temperature. Due to the heat produced by the UV light the temperature outside the reactor reaches 30°C during photolysis, and is much higher at the Suprasil finger, but this is only a preparative experiment where the exact conditions are not critical. After photolysis‐induced isotope exchange, CO2 is separated cryogenically in a glass spiral trap at LN2 temperature and O2 is pumped out. Finally, the CO2 is collected in a sample vial containing nickel foil (thickness 0.05 mm, 99.98% purity, Goodfellow Cambridge Ltd, Huntingdon, UK). O3 that is formed during photolysis is also condensed with CO2 and is decomposed to O2 by heating the sample vial with a heat gun at 500°C for 10 min. Ni foil catalyzes the decomposition of O3 to O2. The CO2 is then trapped again with LN2 and the O2 that has formed from O3 decomposition is pumped out. Finally, the CO2 is passed through a glass U‐trap at dry‐ice temperature (−78°C) to remove remaining traces of water. Heating the O3 and CO2 mixture above 200°C might cause isotope exchange between O3 and CO2,66 but it does not cause a problem for our purpose which is to prepare 17O‐enriched CO2.

The isotopic composition of the 17O‐enriched CO2 sample is measured with the 253 Ultra for both molecular ions (m/z of 44 to 46) to determine δ18O and δ13C values, and atom fragments to measure δ17O and δ18O values. By diluting the 17O‐enriched CO2 with pure non‐anomalous CO2 from the reference CO2 tank (G2), two gas mixtures are prepared with target Δ17O values of approximately 0.25‰ and 0.55‰. The two mixtures are finally measured both with the CO2‐O2 exchange method and with the fragment technique.

RESULTS

Instrument characterization and scale contraction

Scale contraction decreases with equilibration time and source pressure (signal intensity), when the variable conductance window is fully opened and when the emission current is decreased. A detailed investigation of these parameters is presented in the supporting information (Figures S5, S6, and S7, and Tables S1 and S2, supporting information). The effects of ion source pressure and emission control current are the major contributors to the scale contraction. Scale contraction can be minimized if the measurement is performed at high source pressure, low emission control current and with the VISC window open. The drawback of having a higher source pressure is potentially a reduction in the life time of the filament, while having lower emission control current reduces the ionization of the molecules which leads to a lower signal. We suggest following the recommendations of Verkouteren et al,60 to minimize cross contamination in dual‐inlet isotope ratio mass spectrometry measurements. In general, the different parameters affect the δ18O and δ13C values in the same way, but the effects are larger for the δ18O values than for the δ13C values. The origin of the qualitatively different behavior for δ18O and δ13C values could not be identified and requires further study.

By comparing the results of the molecular ion measurements on the 253 Ultra with the values assigned to our reference gases by the Hebrew University of Jerusalem, a scale contraction factor of 0.981 was established and applied for molecular ion measurements. The scale contraction factor is the ratio of the difference between the two CO2 gases (G1 and SCOTT) measured with the 253 Ultra at Utrecht University and the assigned relative difference by the Hebrew University of Jerusalem. Thus, the final values reported below are linked to the isotope scale of the Hebrew University of Jerusalem.6, 62, 63

The key parameter relevant for the validation of the fragment ion method is the scale contraction of a fragment ion measurement relative to a molecular ion measurement. This was determined by analyzing a set of three isotopically distinct pure CO2 gases both with the traditional CO2 + method and with the fragment method (both O+ and C+ fragments). For the traditional molecular ion measurements, the 17O‐correction procedure from Brand et al8 is used. Table 2 shows that the scale contraction for fragment ion measurements is slightly larger than the one for molecular ion measurements. The scale contraction seems to be also slightly larger for measurements on the C+ fragment than for those on the O+ fragment, but more measurements are required to quantify this more thoroughly. Note that each individual measurement series presented in Tables 3 and 4 (CO2 + molecule plus O+ fragment and C+ fragment) takes one full day. For the evaluation of the Δ17O measurements below we use the relative scale contraction of 0.997 determined for the value of δ18O between the traditional CO2 + method and the O‐fragment method (Table 2).

Table 2

δ13C and δ18O scale contraction factors for measurements with the fragment method relative to the traditional measurement technique on molecular ions, using the 17O correction algorithm from Brand et al.8 Both measurements were carried out on the 253 Ultra using three CO2 gases (G1, SCOTT and G2)

Measurement	Fragment (253 Ultra) vs molecule (253 Ultra)
	δ13C	δ18O
G1 vs G2	0.996	0.997
G1 vs SCOTT	0.993	0.997
SCOTT vs G2	0.996	0.997
Average ± SE*t	0.995 ± 0.0016	0.997

Table 3

Oxygen isotope composition of various CO2 reference gases measured with the 17O+ fragment method. δ17O and δ18O values are given relative to VSMOW; Δ17O is calculated according to Equation 4 using λ = 0.528. Individual errors are standard errors of the mean of the corresponding measurement series. The error for the mean is the standard error of the mean for the six experiments multiplied by Student's t‐factor for the 95% two‐sided confidence. Γ is the ratio between the measured precision and the precision expected from counting statistics for δ17O and n is the number of sample‐standard cycles. For δ18O, Γ ≈ 1 for individual measurement series, but the weighted mean error is similar to the one for δ17O, which indicates additional handling errors in sample introduction at the 0.01‰ level. The values in the parentheses are the isotopic compositions of oxygen used for combustion

Experiment	n	Γ	δ17O [‰]	δ18O [‰]	Δ17O [‰]
Reference CO2 [Figure 5A]
1	227	1.54	15.661 ± 0.037	30.406 ± 0.011	−0.276 ± 0.036
2	109	1.53	15.719 ± 0.048	30.419 ± 0.14	−0.225 ± 0.048
3	47	1.73	15.672 ± 0.082	30.444 ± 0.025	−0.284 ± 0.081
4	109	1.48	15.701 ± 0.047	30.397 ± 0.014	−0.231 ± 0.047
5	169	1.42	15.672 ± 0.038	30.380 ± 0.011	−0.251 ± 0.038
6	68	1.47	15.668 ± 0.057	30.379 ± 0.016	−0.255 ± 0.057
Mean ± SE*t			15.682 ± 0.019	30.404 ± 0.021	−0.254 ± 0.019
Reference O2 to CO2 [Figure 5B] (vs reference CO2)
1	64	1.1	−10.518 ± 0.028	−19.266 ± 0.017	−0.303 ± 0.026
2	64	0.8	−10.586 ± 0.021	−19.367 ± 0.009	−0.316 ± 0.020
3	64	1.2	−10.639 ± 0.035	−19.360 ± 0.010	−0.373 ± 0.036
4	64	1.1	−10.534 ± 0.027	−19.184 ± 0.009	−0.362 ± 0.028
5	64	1.0	−10.516 ± 0.026	−19.194 ± 0.011	−0.339 ± 0.026
6	64	1.2	−10.743 ± 0.030	−19.595 ± 0.010	−0.352 ± 0.030
7	64	1.2	−10.741 ± 0.030	−19.610 ± 0.007	−0.342 ± 0.030
8	64	1.3	−10.611 ± 0.34	−19.345 ± 0.009	−0.353 ± 0.034
			−10.611 ± 0.062	−19.365 ± 0.109	−0.342 ± 0.016
Reference O2 to CO2 [Figure 8A]
1	200	2.43	9.206 ± 0.071	18.510 ± 0.018	−0.520 ± 0.071
2	300	1.99	9.220 ± 0.048	18.539 ± 0.018	−0.522 ± 0.048
3	180	1.88	9.298 ± 0.042	18.495 ± 0.017	−0.423 ± 0.042
4	200	2.16	9.302 ± 0.048	18.465 ± 0.017	−0.403 ± 0.048
Mean ± SE*t			9.256 ± 0.059 (9.254 ± 0.007)	18.503 ± 0.035 (18.542 ± 0.008)	−0.467 ± 0.074 (−0.489 ± 0.008)
Light O2 to CO2 [Figure 8B]
1	216	2.13	−26.934 ± 0.097	−50.791 ± 0.024	0.219 ± 0.067
2	208	1.43	−26.611 ± 0.355	−50.075 ± 0.512	0.182 ± 0.059
3	256	1.34	−26.381 ± 0.231	−49.824 ± 0.318	0.311 ± 0.056
Mean ± SE*t			−26.666 ± 0.488 (−26.239 ± 0.002)	−50.329 ± 0.817 (−49.614 ± 0.002	0.237 ± 0.097 (0.279 ± 0.011)

When the appropriate scale correction parameters are applied, the δ13C and δ18O values obtained from the fragment and molecular ion measurements generally agree at the ~0.01–0.03‰ reproducibility level (except for one outlier in δ13C, G1 vs SCOTT = −36.665 ± 0.002‰ and −36.601 ± 0.020‰ for molecular and fragment ion measurements respectively (Figure S10, supporting information). Isotope ratio measurements on C and O fragment ions could be an independent method to validate/evaluate traditional isotope measurements and ion (17O) correction algorithms at a level of precision similar to the reported differences between different ion correction schemes.

Figures S8, S9 and S10 (supporting information) show that the fragment method returns the same value when two pure CO2 gases are measured directly, and via a third intermediate gas for δ13C, δ18O and δ17O values. Tables 3 and 4 show that isotope ratios based on the 13C+ and 18O+ fragment ions are both measured with a precision close to the counting statistics limit.

Table 4

Comparison of δ13C and δ18O values obtained using the C‐fragment and O‐fragment techniques with results from the traditional molecular measurements for pure CO2 gases. For the measurements on the molecule, the 17O correction according to Brand et al8 is used. Γ is the ratio between measured precision and the precision estimated from the counting statistics and n is number of cycles for the fragment measurement

δ13C
Sample	Exp	n	Γ	δ13C [‰] (13C+ measurement)	δ13C [‰] 13CO2 + measurement
G1vs G2	1	45	1.0	−7.968 ± 0.015	−7.963 ± 0.001
	2	20	0.73	−7.967 ± 0.022	−7.984 ± 0.001
	3	38	0.74	−7.991 ± 0.016	−7.967 ± 0.001
	4				−7.981 ± 0.001
	5				−7.972 ± 0.001
	6				−7.978 ± 0.002
Average ± SE*t				−7.975 ± 0.023	−7.974 ± 0.007
G2 vs SCOTT	1	49	0.84	−28.933 ± 0.015	−28.881 ± 0.001
	2				−28.923 ± 0.001
	3				−28.916 ± 0.001
	4				−28.913 ± 0.001
	5				−28.915 ± 0.001
Average ± SE*t					−28.910 ± 0.016

Fragment measurement

δ17O, δ18O and Δ17O: reproducibility

Figure 5A shows Δ17O for a pure CO2 (G5) sample with six replicates measured using the O‐fragment method at Utrecht University. The δ17O and δ18O values of the CO2 are given in Table 3. The measurement times are between 3 and 12 h. The δ17O values are measured with an individual measurement error (standard error of the mean) ranging from 37 to 82 ppm, while the δ18O values have an individual measurement error of 11 to 25 ppm (standard error of the mean, SEM). The measurement precision for the δ17O values is worse than that expected from counting statistics by a factor of 1.42 to 1.73. As shown in Figure 5A and Table 3, from these six replicates the Δ17O reproducibility is 19 ppm (standard error times Student's t‐factor for 95% confidence). At the University of Göttingen the reproducibility experiment is performed using CO2 produced by combustion of a graphite rod with pure O2 (GU‐O2) (Figure 5B). The δ17O and δ18O values of the CO2 are given in Table 3 relative to the working reference. The δ17O values are measured with an individual measurement error (SEM) ranging from 21 to 35 ppm while the δ18O values have an individual measurement error of 7 to 17 ppm (SEM). As shown in Figure 5B and Table 3, from these eight replicates the Δ17O reproducibility is 16 ppm (standard error times Student's t factor for 95% confidence). The reproducibility for the δ17O and δ18O values is lower in this method due to incomplete combustion of the graphite rod.

Figure 5

A, Δ17O (CO2) measured with the O‐fragment method for a pure CO2 (G5, see Table 1), measured at Utrecht University. B, Δ17O (CO2) measured with the O‐fragment method for CO2 prepared by combusting graphite rod with pure O2 (GU‐O2) (δ17O = −10.611 ± 0.062‰ and δ18O = −19.365 ± 0.109‰, relative to the working standard) measured at the University of Göttingen. Error bars represent ±1 standard error of the mean (SEM). The red line shows the mean and the shaded area is the SEM times Student's t‐factor (95% confidence) [Color figure can be viewed at wileyonlinelibrary.com]

Due to the low ion counts very long measurement times are required to achieve a precision of the order of 10 ppm. A long‐term measurement of a zero enrichment cylinder reference gas at the University of Göttingen (Tyczka Industrie‐Gase GmbH, Mannheim, Germany) yielded a precision of 14 ppm for Δ17O and δ17O values (5 ppm for δ18O values) after a measurement time of 20 h (Figure 6). As mentioned above, a requirement is that the mass scale remains very stable over the entire measurement period. At Utrecht University we monitor the stability of the mass scale by recording a medium‐resolution mass spectrum at regular intervals during the measurement. Figures 7A and 7B show an example of a long‐term fragment measurement during which the mass scale was very stable. However, the mass scale is not always as stable, and mass instabilities are one limitation for measurements that require long measurement times. Instabilities in the mass scale are more likely to contribute to the larger errors than counting statistics, factor Γ in Table 3, in some measurements.

Figure 6

A long‐term zero enrichment experiment (Δ17O, δ17O and δ18O) at the University of Göttingen. After 20 h of measurement time a precision of 14 ppm for δ17O and Δ17O, and 5 ppm for δ18O is achieved [Color figure can be viewed at wileyonlinelibrary.com]

Figure 7

A, Medium‐resolution mass sweep for m/z 17 performed during the isotope measurement to monitor the stability of the mass scale. Each line represents a single mass spectrum that was recorded after each acquisition of 10 cycles of dual‐inlet isotope measurements. The separation between two mass sweeps is roughly 21 min. B, 2‐D projection of A, where the ion count rate is presented in color to show the stability of the plateau used for measurement of the 17O+ fragment (green section) [Color figure can be viewed at wileyonlinelibrary.com]

Δ17O accuracy

The accuracy of Δ17O and δ17O measurements using the O‐fragment method is evaluated by measuring CO2 with known δ17O and δ18O values, prepared from isotopically known O2 (see section 4.5.1) The results presented in Figure 8A and Table 3 show that Δ17O of the CO2 obtained by measuring the δ17O and δ18O values from the 17O+ and 18O+ fragment ions is indistinguishable within the experimental error from the isotopic composition of the O2 used for the preparation of the CO2. The assigned Δ17O value of the reference O2 used for combustion at Utrecht University is −0.489 ± 0.008‰ while the CO2 obtained by combustion has Δ17O = −0.467 ± 0.074‰ when measured with the fragment method (Figure 8A and Table 3). To enable easy comparison, the Δ17O of O2 and CO2 are both calculated with the same value of λ = 0.528. In addition, the individual δ17O and δ18O values agree with those of the source O2 within the errors. It should be noted that the discrepancy of Δ17O results within our measurement series is larger than the errors from the individual measurements, which indicates that sample handling errors have contributed to the rather large spread in the fragment measurements. The isotopically light O2 in Göttingen has assigned values of δ17O = −26.239 ± 0.002‰ and δ18O = −49.614 ± 0.002‰ relative to VSMOW, which yields Δ17O = 0.279 ± 0.006‰. The CO2 produced by combustion and measured with the O‐fragment method (Figure 8B, Table 3) shows a rather wide range of δ17O and δ18O values, indicating fractionation (and/or incomplete combustion) in the process of preparing the CO2. The effect on Δ17O is much smaller.

Figure 8

A, Δ17O of CO2 produced by combustion of a graphite rod (black points and red line showing the mean) and Δ17O of the pure O2 used for combusting the graphite (blue line), measured at Utrecht University. B, Similar results for CO2 that was prepared from isotopically depleted O2 at the University of Göttingen, plotted versus the m/z 16 signal intensity. Δ17O values obtained from the fragment method are indistinguishable from the Δ17O values of the combusted O2. The Δ17O is calculated using λ = 0.528 for both gases. Individual error bars represent ±1 standard error of the mean (SEM). The shaded area shows the SEM times Student's t‐factor (95% confidence) [Color figure can be viewed at wileyonlinelibrary.com]

The good agreement between the δ17O, δ18O and Δ17O values of oxygen and of the CO2 produced by combusting graphite shows that determination of the triple isotopic composition of CO2 using the O‐fragment method is not only reproducible but also accurate. Furthermore, the agreement in the triple isotopic composition of oxygen between O2 and CO2 (produced by combustion) suggests that our isotope scales for CO2 and O2 are very compatible.

As shown in Table S3 (supporting information), Δ17O is measured with an average standard error of 39 ppm (standard error of the mean) for four measurements (A3, B2, B3, C2) at an intensity for m/z 16 of 1.18 × 109 cps. When measurements are made at lower signal intensity than the linear range for source pressure vs signal intensity relation for m/z 16 (see above), measurement precision decreases. For instance, the precision drops from 39 to 83 ppm (average SEM for the four measurements shown in Table S3, supporting information) when the intensity on m/z 16 decreases from 1.18 × 109 to 4.70 × 108 cps. Measurement at higher signal intensity, outside the linear window, does not show a significant improvement in the precision of the Δ17O measurement relative to measurements with lower signal intensity in the linear window (Table S3, supporting information). This might be also due to statistics since we only have four measurements.

Comparison of the O‐fragment method with the CO2‐O2 exchange method

After confirming the accuracy and reproducibly of the O‐fragment method, we measured the δ17O, δ18O and Δ17O values of four CO2 gases both with the O‐fragment method and with the oxygen exchange method (see above). Two of the gases are commercial CO2 gases (G1 and G2, Table 1) and the other two (G3 and G4) were artificially enriched in 17O as described in section 3.5.2. As shown in Figure 9 and Table S4 (supporting information), the results obtained with the two totally independent techniques are indistinguishable within the error bars. The δ18O values are in the range of 4.8–35.0‰ vs VSMOW and values of Δ17O range from −0.3‰ to +0.7‰ (λ =0.528) which covers and extends the Δ17O range expected for tropospheric CO2 samples, including international carbonate standards.32 The Δ17O is determined by the O‐fragment method with a precision of 36–79 ppm (standard error times Student's t‐factor for 95% confidence). The excellent agreement between the two totally independent methods provides an independent validation of the fragment ion technique.

Figure 9

Comparison of Δ17O measured with the fragment method and the CO2‐O2 exchange method for four different CO2 gases. The δ18O values of the CO2 gases range from 4.48‰ to 35.00‰. The horizontal axis shows the number of experiments. Error bars for the fragment measurement represent ±1 standard error of the mean (SE). The red line shows the mean and the shaded area is the standard error of the mean times student t‐factor (95% confidence) [Color figure can be viewed at wileyonlinelibrary.com]

C‐fragment

The δ13C values of the two CO2 gases G1 and SCOTT were measured against G2 with the C‐fragment method and with the traditional measurement on the CO2 molecule (evaluated with the Brand et al8 procedure). As shown in Table 4, the δ13C values obtained from the C‐fragment method and molecular measurement are the same within the error (at the ≈ 0.01‰ reproducibility level). A possible challenge for measuring δ13C values with the fragment method is the interference from the 12CH+ adduct due to ion source chemistry (e.g. in the presence of water). The 12CH+ adduct is only 0.004 u separated from 13C+ as shown in the mass spectra (Figure 4). However, the figure also shows that this interference can be resolved at medium resolution.

DISCUSSION

Scale contraction

We observe a higher scale contraction when measuring on the fragment ions than with the measurements on the molecular ions (Table 2). The difference might be because fragment ions are more reactive than the molecular ions. High energy collisions between ions and the source material cause sputtering and implantation, which may be more effective for fragment ions. Therefore, fragment ions may remain effectively longer in the ion source causing the observed higher scale contraction. The difference in scale contraction between fragment measurement and molecular measurement requires further study.

Possible interferences

Oxygen isotope measurements on O fragment ions with low‐resolution mass spectrometers are mainly limited by the interference from water and its OH fragment ions. The background level of water in mass spectrometers is always significant, and it also generally varies when switching between bellows in dual‐inlet measurements. With the 253 Ultra, these interferences can be separated from the O+ fragments (Figure 2; Table S5, supporting information), even if the shoulder for interference‐free 17O+ measurements is narrow. H2 16O+ is the main interference for 18O+ and 16OH+ for 17O+. The two rare isotopologues of OH, 17OH and 16OD, could also interfere with 18O, but they are negligible in abundance compared with H2 16O and can be resolved at medium mass resolving power. Table S5 (supporting information) shows a list of other potential interferences with cardinal masses 17 and 18. The molecules made up of lighter atoms than O have masses that are always higher than the cardinal masses 17 and 18, because O is the lightest element where the exact isotope masses are lighter than the cardinal masses. Therefore, these interferences all fall on the high mass side of the O+ fragment ion, and they can also be resolved with the 253 Ultra at medium resolution (the mass resolving power required is lower than that for separating OH+ and H2O+). Therefore, only interferences from doubly ionized oxygen formed in the ion source (16O18O++) and other doubly ionized molecules with higher masses (e.g. 34S++ or 36Ar++, Table S5, supporting information) can potentially interfere at the low‐mass shoulders where we perform measurements. Formation of doubly ionized ions is usually suppressed by several orders of magnitude compared with the singly charged ions. Nevertheless, they interfere at the low‐mass shoulder of the O atom fragments. The interference of 16O18O++ on 17O+ depends on the δ18O value and source pressure as shown in Figure 3. At a source pressure of 2.5 × 10−7 mbar, the size of the correction in our instrument is about 0.5 ppm in the δ17O value (and thus Δ17O) per 1‰ difference in the δ18O value between sample and working reference gas. Thus, when the working reference gas is close in isotopic composition to the samples that are measured, the correction is negligible.

The other challenge to measuring the δ17O and δ18O values of CO2 using the fragment method is the possible interference of O fragment ions from other oxygen‐bearing impurities (OBI) such as H2O, O2 or N2O. The sample and the mass spectrometer background should be very clean to avoid any oxygen contribution from other molecules. The effect of an OBI on the values of δ17O, δ18O and Δ17O measurements of CO2 (δI imp) can be estimated using Equation 8. The magnitude of the interference depends on the isotopic composition, the fragmentation pattern (efficiency of producing O fragment ions relative to CO2), ionization efficiency and the abundance of the impurity relative to the CO2 (Equation 8). where I is 17 or 18, is the abundance ratio, Ω is the ratio of oxygen atoms in OBI to the oxygen atoms of CO2, ψ is the ratio in ionization efficiency of OBI to CO2 and φ is the ratio of O+ fragment formation of OBI versus CO2. As mentioned above, a water background is always present in mass spectrometers and therefore we estimate the effect of water on the δ17O, δ18O and Δ17O measurements of CO2 using Equation 8. For water Ω = 0.5 and φ = 0.1 because the O+ fragment production is only 1% for H2O, whereas it is 10% for CO2.67, 68 We assume a similar ionization efficiency between CO2 and H2O (i.e. ψ = 1) for the calculation. Table S6 (supporting information) shows the calculated effect of water impurity on the δ17O, δ18O and Δ17O values of CO2 measured with the O‐fragment method for different water levels and isotopic composition of the water. For instance, when the isotopic composition of the water impurity relative to the CO2 is δ17O = −20‰ and δ18O = −40‰, the effect on the δ17O and δ18O values of CO2 will be significant for ρ >0.3% and ρ >0.1%, respectively. Since the isotopic composition of the water is assumed (roughly) to be mass dependent, the effect on the Δ17O will be only significant when ρ >1%. When the isotopic composition is strongly mass independent (δ17O = δ18O = −40‰ relative to CO2), the effect on Δ17O will be significant for ρ >0.3% (Table S6, supporting information).

Future developments and applications

In the present state of development, the O‐fragment method can be used to quantify Δ17O of CO2 with a precision about of 37 ppm in about 12 h measurement time (67.1 s integration time and 60 s equilibration time). Higher precisions can be achieved by (i) increasing signal intensity; (ii) increasing observation/integration time of the 17O+ fragment ions (Figure 6); and (iii) achieving measurement precisions at the counting statistics limits. The signal intensity can be increased by increasing source pressure, but the present measurements are already at the upper end of the range where signal intensity increases linearly with source pressure (Figure S1, supporting information). Increasing the ion current will also shorten the filament lifetime. Observation time can be increased by simply extending the integration time, by reducing the time that is used for peak centering, pressure adjust, etc., and by reducing the equilibration time. Reducing the equilibration time introduces additional error due to cross contamination/mixing between sample and reference. Ideally, a LIDI (Long Integration Dual Inlet) technique where the sample‐reference switching is not performed at all would enable longer observation times of the sample.69 LIDI measurements were attempted with the 253 Ultra but not continued because of instability issues. An increase in stability may also enable measurements at the counting statistics limit, which would improve precision by a factor of 1.5.

Compared with traditional δ13C measurements that require a 17O‐correction, the C‐fragment is not subject to the following uncertainties related to the 17O‐correction:

The use of different 17R, 13R and λ values in different algorithms introduces discrepancies that are larger than the precision of current isotope ratio mass spectrometry techniques42

Most of the correction algorithms used do not include the impact of Δ17O of CO2

The accepted values for 17R and 13R may require revision to meet the current measurement precision44

There is no single λ value that can be assigned to CO2 since different processes that contribute to the formation or removal of CO2 follow different three‐isotope slopes.

The fragment technique is simple and unlike other techniques does not require any additional chemical conversion or exchange steps to measure the δ17O value of CO2. Therefore, it can be used to independently assess discrepancies in δ17O values measured by different laboratories, such as the difference in δ17O of IAEA (International Atomic Energy Agency, Vienna, Austria) carbonate standard (NBS‐18) measured by Passey et al32 and Barkan et al.7 However, the signal intensities for rare isotopes of fragment ions are relatively small, especially when they have to be separated from near‐by mass interferences and require higher mass resolution, which reduces ion transmission in the 253 Ultra. Therefore, long measurement times are required to reach a precision of the order of 0.01‰. When this precision is reached, the fragment technique can also be useful to evaluate discrepancies introduced in δ13C measurements due to the use of different algorithms for 17O‐correction.

Isotope measurements of atomic ion fragments may have many applications for other molecules. A straightforward extension of the application presented here is the mass‐interference‐free measurement of 17O+ and 18O+ in other oxygen‐containing compounds, for example, CO or N2O. Current isotope techniques of these gases rely in many cases on an assumed relation of mass‐dependent fractionation between δ17O and δ18O values and (e.g. in the case of the CO) chemical conversion into CO2.70, 71, 72 Direct isotope ratio measurements on the O+ fragment can overcome these limitations and provide quantification of Δ17O. Similar to the case of CO2 presented here, the 13C+ content of CH4 and CO can be measured directly on the C+ fragment of these gases, without chemical conversion steps that are known to cause artifacts in traditional isotope techniques.70, 71, 72, 73 Furthermore, isotope measurements on atomic fragment ions may be combined with measurements of larger fragments of hydrocarbons to determine the position‐specific carbon isotope composition of hydrocarbons.55

The position‐specific 15N+ content of N2O is presently determined by measurement of the parent N2O molecule and the NO fragment, which allow the average δ15N value and the 15N content at the central nitrogen position to be quantified, and the δ15N value of the terminal N atom is derived by mass balance, which induces large errors.51, 52 In principle, the 15N+ content of the terminal N atom could be derived from the N+ fragment, which originates primarily from the terminal N atom in N2O. Similar to the case of O atoms shown here, this requires a very good vacuum system to avoid contamination from the main atmospheric gas N2.

In addition to these environmental applications, the analysis of atomic fragment ions of different compounds may be a useful tool to study fractionation processes in the ion source of an isotope ratio mass spectrometer. As discussed earlier, the scale contractions for isotopic measurements are different for the fragment ions and molecular ions of CO2. Examining these effects further may help to understand the chemistry and surface effects in the ion source of isotope ratio mass spectrometers by studying different fragments. In addition, analysis of fragment ions facilitates measuring the isotopic composition of two different chemical compounds versus each other (e.g. δ13C value in CH4 versus in CO2). This can on the one hand provide information on ion source effects associated with fragmentation, but on the other it may also help to directly compare isotope scales between different compounds.

Figure S1

Relationship between source pressure and signal intensity at m/z 16 (Faraday collector L3, equipped with a 1x1010Ω resistor). The linear range (shaded area) ends at a source pressure of 4.5x10−7 mbar (corresponding to a signal intensity of approx. 1.3x109 cps).

Figure S2 Schematic diagram of the O2‐CO2 exchange experimental setup. The quartz reactor has an outer diameter of 21 mm. PID stands for proportional‐integral‐derivative temperature controller.

Figure S3 Schematic diagram of the setup used for conversion of O2 to CO2 by combusting a graphite rod. The reactor is made out of quartz with 21 mm outer diameter.

Figure S4 Schematic diagram of the setup used for preparing CO2 with a positive Δ17O by photolysing a mixture of O2 and CO with UV light. BR: Borosilicate reactor (2 L), PRL: pen ray Hg vapor lamp, CT: CO2 trap (liquid nitrogen temperature), SV: sample vial, WT: water trap (operated at dry ice temperature), PS: pressure sensor, HVP: high vacuum pump, FVP: fore vacuum pump.

Figure S5 δ13C and δ18O values of SCOTT measured against G2. A and B) Effect of equilibration time with VISC window closed. C and D) Effect of equilibration time with VISC window fully opened; E and F) Effect of filament emission current on the relative difference between the two gases in δ13C and δ18O values. G and H) Effect of amount of gas in the ion source quantified by the signal intensity in cps for m/z 44. The emission current experiments are performed at 60‐seconds equilibration time and the sensitivity to the amount of gas was determined with 30‐seconds equilibration time. The VISC window was kept closed.

Figure S6 Effect of the equilibration time on the δ13C (A) and δ18O (B) differences between measurement with open VISC window (VISCo) and closed VISC window (VISCc).

Figure S7 Drop in signal intensity for at m/z 44 (corresponding to the source pressure) when the dual inlet valve was closed for different initial source pressures. A) decrease of the main ion signal of CO2 as a function of time. B) fraction of CO2 remaining as a function of time for the main ion signal of CO2. The emission current and accelerating voltage were 1.95 mA and 9.9 kV, respectively.

Figure S8 Comparison of δ17O differences between three different CO2 gases with the fragment technique using the 253 Ultra at Utrecht University The red arrows indicate that the respective measured δ values are combined for comparison with the directly measured third δ value.

Figure S9 Comparison of δ18O differences between three different CO2 gases using either the molecular CO2 ion technique (left) or the atom fragment technique (right) with the 253 Ultra at Utrecht University The red arrows indicate that the respective measured δ values are combined for comparison with the directly measured third δ value.

Table S1 Effect of equilibration time on the absolute isotopic difference of two gases measured on the 253 Ultra with VISC window open and closed. The errors given are standard errors of the mean.

Table S2 Effect of emission control current and source pressure on the absolute isotopic difference of two gases measured on the 253 Ultra (G1 and SCOTT). The errors given are standard errors of the mean. The integration time of individual measurements is 61.7 seconds The effects of source pressure (intensity of m/z 44) are determined at an equilibration time of 30 seconds and an emission current of 1.8 mA.

Table S3 Isotopic composition of CO2 produced by combustion with isotopically light O2 at the University of Göttingen. Each CO2 sample is analyzed four times at different intensity to investigate the effect of signal intensity on the precision of Δ17O measurement based on the relationship between source pressure and signal intensity at m/z 16.

Table S4: Comparison of the results obtained with the O2‐CO2 exchange method and O‐fragment technique. δ17O, δ18O and Δ17O values are in per mill (‰) with respect to VSMOW. The error of the mean is the standard error multiplied by the student t‐factor for the 95% two‐sided confidence interval while for the individual measurements it is the standard error. Γ is the ratio between measured precision and the precision calculated according to counting statistics, n is number of cycles, i stand for initial (before exchange) and f stands for final (after exchange). For δ18O values the measurement error is similar to the error calculated based on counting statistics.

Table S5 List of potentially interfering ions with masses close to 17O+ and 18O+, and the required resolution to avoid the interference. The interfering ions are ordered based on the resolving power requirement.

Table S6 Simulated effect of a H2O impurity (contamination level quantified by γ = [H2O]/[CO2] as source of oxygen during Δ17O measurement using the O‐fragment method for an intensity for m/z 16 of 4.05x109 cps. The signal of O atom fragments relative to molecular ions is 10% for CO2 and 1% for H2O5,6. For these conceptual calculations we assumed the same ionization efficiency for H2O and CO2.

Click here for additional data file.