A rapid high-precision analytical method for triple oxygen isotope analysis of CO2 gas using tunable infrared laser direct absorption spectroscopy.

RATIONALE: The simultaneous analysis of the three stable isotopes of oxygen-triple oxygen isotope analysis-has become an important analytical technique in natural sciences. Determination of the abundance of the rare 17 O isotope in CO2 gas using magnetic sector isotope ratio mass spectrometry is complicated by the isobaric interference of 17 O by 13 C (13 C16 O16 O and 12 C16 O17 O, both have mass 45 amu). A number of analytical techniques have been used to measure the 17 O/16 O ratio of CO2 gas. They either are time consuming and technically challenging or have limited precision. A rapid and precise alternative to the available analytical methods is desirable.
METHODS: We present the results of triple oxygen isotope analyses using an Aerodyne tunable infrared laser direct absorption spectroscopy (TILDAS) CO2 analyzer configured for 16 O, 17 O, and 18 O combined with a custom gas inlet system. We evaluate the sensitivity of our results to a number of parameters. CO2 samples with a wide range of δ18 O values (from -9.28‰ to 39.56‰) were measured and compared to results using the well-established fluorination-gas source mass spectrometry method.
RESULTS: The TILDAS system has a precision (standard error, 2σ) of better than ±0.03‰ for δ18 O and ±10 per meg for Δ'17 O values, equivalent to the precision of previous analytical methods. Samples as small as 3 μmol CO2 (equivalent to 300 μg CaCO3 ) can be analyzed with a total analysis time of ~30 min.
CONCLUSIONS: We have successfully developed an analytical technique for the simultaneous determination of the δ17 O and δ18 O values of CO2 gas. The precision is equal to or better than that of existing techniques, with no additional chemical treatments required. Analysis time is rapid, and the system is easily automated so that large numbers of samples can be analyzed with minimal effort.

INTRODUCTION

After the discovery of the temperature‐dependent fractionation between calcite and water over 70 years, numerous studies have used this relationship as a paleoclimate indicator to reconstruct ocean water temperatures. Despite the prevalence of measuring carbonates to reconstruct temperatures, researchers are still trying to develop new techniques to minimize the assumptions associated with carbonate palaeothermometry. These assumptions include the following: the original oxygen isotope composition of the ocean, the carbonate formed in oxygen isotope equilibrium with ocean water, and the carbonate is free of diagenetic alteration.

New carbonate isotope techniques have been developed, including for clumped isotopes and triple oxygen isotopes. The relatively new field of triple oxygen isotopes for carbonates uses the rare stable isotope of oxygen, 17O, in addition to the traditional measurements of 18O and 16O, to evaluate the equilibrium conditions under which the carbonates precipitated. For terrestrial processes, oxygen isotopes undergo mass‐dependent fractionations such that δ17O ≈ 0.5(δ18O), where δxO = R sample/R standard – 1)×1000 in per mil (‰) notation and R is 17O/16O or 18O/16O. Because the δ17O and δ18O values of terrestrial materials are generally plotted with a slope of ½, it was assumed that the δ17O value provided no additional information to the δ18O value alone. It is now known that the δ17O and δ18O values of terrestrial materials deviate slightly from this near‐linear relationship and that these deviations have geological or biological significance. Triple oxygen isotope data are presented in terms of Δ′17O (also called 17O excess), which are defined as follows: in per mil (‰) notation and in per meg notation, where λ is a reference slope (0.528 in this work) and δ’17O and δ’18O are linearized versions of δ17O and δ18O (δ’xO = 1000ln(δxO/1000 + 1)). The small but measurable Δ′17O values provide meaningful information on carbonates and other minerals. , , ,

The benefit of measuring both δ17O and δ18O of carbonates becomes apparent when plotted in triple oxygen isotope space with the carbonate–water isotope equilibrium curve. The triple oxygen isotope composition of marine carbonates can be used to assess the degree of postdepositional diagenesis and to estimate the temperature of deposition. Therefore, the triple oxygen isotope composition of carbonates provides us with an independent estimate of temperature and the ability to assess the preservation of a sample.

The development and wide usage of high‐precision triple oxygen isotope measurements of carbonates (Δ′17O ± 10 per meg) is hindered by the difficulty using current available analytical techniques. Traditional measurements of carbonates for δ13C and δ18O are made from CO2 produced by phosphoric acid digestion. However, due to the isobaric interference at mass 45 between 13C16O2 and 12C17O16O, the 17O contribution to CO2 at mass 45 cannot be determined using conventional magnetic sector mass spectrometry. A number of techniques have been developed to overcome this problem, including the following: (a) high‐temperature fluorination of CO2 to O2, which is then analyzed in a conventional gas source mass spectrometer ; (b) conversion of CO2 to H2O using Fischer–Tropsch process followed by fluorination of the resultant H2O to O2 , ; (c) high‐temperature Pt‐catalyzed exchange between CO2 and subordinate O2 gas, in which the O2 gas essentially “acquires” the isotopic composition of the CO2 gas , ; and (d) analysis of the O+ fragment of CO2 gas in an electron impact source ultrahigh‐resolution mass spectrometer. None of these methods is ideal for measuring relatively large numbers of samples due to cost, time of analysis, or limited precision. See Passey and Levin for a full review.

New developments using laser‐based spectrometry show significant progress toward high‐precision triple oxygen isotope measurements directly from CO2. Previous studies using tunable infrared laser direct absorption spectroscopy (TILDAS) have successfully measured 17O/16O and 18O/16O of CO2 simultaneously but have not yet achieved the precision required for application to natural materials (~10 per meg for Δ′17O). ,

Here we report the simultaneous analyses of 17O/16O and 18O/16O of CO2 through the development of a new TILDAS instrument for high precision and rapid analysis of triple oxygen isotope composition. To obtain high‐precision results, it is necessary to rapidly (on the order of 1–2 min) switch between the sample and reference gas (to calibrate drift), and it is critical to balance the mixing ratios and pressures of the sample and reference gases. A sample preparation system was constructed to meet these objectives. This new instrument and the sampling system significantly reduce the analysis time and sample size while achieving the necessary precision of 10 per meg for Δ′17O. The cost of the triple oxygen CO2 analyzer and the sampling system is less than that of a traditional mass spectrometer required to acquire such precision. The rapidity and ease of analysis, combined with the potential for full automation, allow for large numbers of samples to be analyzed with reasonable effort, opening up the technique to numerous applications in the physical sciences.

EXPERIMENTAL

Tunable infrared laser direct absorption spectroscopy

Spectroscopic analyses were performed using a TILDAS CO2 isotope monitor for Δ17O–CO2 designed and manufactured by Aerodyne Research, Inc. (ARI, Billerica, MA, USA), configured for triple oxygen isotope analyses of CO2. The TILDAS instrument uses a mid‐infrared distributed‐feedback interband‐cascade laser (Nanoplus Nanosystems and Technologies GmbH) tuned to sweep a 0.4 cm−1 wide spectral region near 2349 cm−1 to measure the absorption of CO2 isotopologues 628 (C12O18O16), 627 (C12O17O16), and 626 (C12O16O16) (Figure 1). The notations 628, 627, and 626 used here follow those of the atmospheric database for high‐resolution transmission molecular absorption.

FIGURE 1

Sample CO2 triple oxygen isotope spectrum. Measured (green) and fitted‐model (black) transmission spectrum from the TILDAS (tunable infrared laser direct absorption spectroscopy) triple oxygen isotope analyzer for CO2 [Color figure can be viewed at wileyonlinelibrary.com]

Analyses of the three stable isotopes of oxygen directly from CO2 are possible using laser spectroscopy due to the well‐determined spectral lines for each isotopologue, which result from their highly characteristic rotational‐vibrational bands. Laser light is absorbed only when the frequency of the laser closely matches the absorption frequency of a CO2 isotopologue, and the amount of absorption is proportional to the concentration of that isotopologue, per Beer–Lambert law : where A is the absorbance, L is the path length of the laser within the absorption cell 36 m, N is the concentration of the isotopologue, and α is the absorption coefficient, which differs for each isotopologue and the spectral region. The narrow spectral region of the laser was selected by ARI to include one well‐determined spectral line for each isotopologue, all with similar absorbances at ambient concentrations (Figure 1).

The TILDAS instrument uses a scan rate of 2.4 kHz and averages the 2400 spectra acquired each second. The laser frequency is swept across the spectral region by ramping the electrical current supplied to the laser, and the temperature of the laser is maintained within 0.005 K by thermoelectric coolers and a recirculating liquid chiller (Oasis T‐Three Thermoelectric Chiller, Martinsried, Germany). The light signal is detected using a thermoelectrically cooled photovoltaic (HgCdTe) detector whose zero level is calibrated by briefly turning the laser off at the end of each sweep.

CO2 absorbs the laser beam in an absorption cell where the laser beam bounces 182 times between two astigmatic mirrors to lengthen the absorption path and thereby increase sensitivity. Most ARI TILDAS instruments have a flow‐through absorption cell that provides continuous monitoring of air. In the present configuration, the absorption cell was designed for discrete sampling: the cell was evacuated, filled with the sample or reference gas, and sealed for 30 s while the spectrum was measured. In this configuration, gas entered and exited the absorption cell through a single port to minimize the required sample volume. The optics outside the absorption cell (i.e., mirrors) were thermally stabilized and purged with nitrogen at a constant flow rate of 1 L/min to prevent any absorption of the laser light by CO2 outside of the absorption cell.

Spectrometer control and data acquisition were performed using TDL Wintel, a dedicated piece of software produced by ARI to run all its TILDAS instruments. A separate graphical user interface called IRIS, created by ARI for discrete sampling applications, coordinated with TDL Wintel to automate the inlet system and perform real‐time Δ′17O analysis (Figure 2). IRIS runs within the Igor Pro computing environment (Wavemetrics, Inc., Portland, OR, USA).

FIGURE 2

TILDAS (tunable infrared laser direct absorption spectroscopy) user interface. Sample image of the TILDAS user interface run in IgorPro. The user interface communicates with TDL Wintel, providing sampling system automation and real‐time data analysis [Color figure can be viewed at wileyonlinelibrary.com]

More details about the spectrometer design can be found in McManus et al and Wang et al, which describe nearly identical instruments used for other isotopologues.

Preparation of CO2 from carbonate samples

Carbonates were converted to CO2 using the well‐established phosphoric acid digestion system. The minimum sample required for the present system was ~3 μmol CO2, corresponding to 300 μg calcite. (For testing the precision of the TILDAS instrument, large samples [3 mg calcite] were reacted to minimize errors associated with the CO2 extraction step. The gas was then cut by a factor of 10 for analysis. Smaller samples were easily introduced into the TILDAS system by quantitatively freezing the CO2 gas into a cold finger in the mixing volume.) Calcite was loaded into a glass reaction vessel with a central glass divider; 1 mL of 100% phosphoric acid was loaded in the glass reaction vessel on the opposite side of the divider from the calcite. The vessel was evacuated and placed in a water bath held at a constant temperature of 25°C, and then phosphoric acid was introduced into the calcite section of the reaction vessel. The reaction vessel remained in the water bath at a constant temperature of 25°C for 16 h. The resulting gas was purified using a cryogenic trap to remove water and any noncondensable gases before the CO2 was cryogenically transferred to a separate gas sampling tube for transfer to the TILDAS analyzer.

Fluorination

To determine the actual δ17O value of our reference gases, CO2 was fluorinated to produce O2 to be measured for triple oxygen isotopes using the conventional nickel bomb fluorination method by Sharma and Clayton, modified by Wostbrock et al ; 1 mL of CO2 (~45 μmol CO2) was injected into the fluorination line through a septum and cryogenically transferred to a high‐purity nickel reaction vessel; 30 times stoichiometric excess of bromine pentafluoride (BrF5) was cryogenically transferred into the nickel reaction vessel with the CO2. The nickel reaction vessel was sealed with a valve and heated to 700°C for 5 days to ensure quantitative conversion of CO2 to O2 by the reaction CO2 + 4/5BrF5 = O2 + CF4 + 2/5Br2.

After the fluorination reaction, the reaction vessels were cooled with liquid nitrogen. The reaction by‐products CF4, Br2, and unreacted BrF5 remained frozen in the vessel, and the O2 released from fluorination was purified by passing through two liquid nitrogen traps and then a 100°C NaCl trap to remove any F2 by converting it to NaF and Cl2 gas, which were then trapped in a third liquid nitrogen–cooled trap. The O2 gas produced through fluorination was absorbed on a 5 Å molecular zeolite sieve cooled with liquid nitrogen. The molecular sieve containing the O2 was then isolated from the fluorination line, and the O2 was released into a stream of ultra‐high‐purity helium carrier gas and passed through a 5 Å molecular sieve gas chromatograph at a flow rate of 5 mL/min to remove any trace contamination of NF3 and N2. The O2 in the helium carrier gas was then collected on a second liquid nitrogen–cooled molecular sieve attached to the inlet port of a MAT 253+ isotope ratio mass spectrometer (Thermo Fisher, Bremen, Germany) configured for triple oxygen isotope analyses. Helium was pumped away through the inlet pump on the mass spectrometer, and the molecular sieve was heated to release the absorbed O2 into the sample bellows. The resultant O2 was analyzed in dual‐inlet mode relative to a reference gas calibrated to VSMOW (Vienna Standard Mean Ocean Water), providing a precision of better than ±0.005‰ (5 per meg) for Δ′17O.

Five CO2 samples were analyzed in this study. They included three IAEA (International Atomic Energy Agency) calcite standards (NBS‐18, NBS‐19, and IAEA‐603; see https://nucleus.iaea.org/sites/ReferenceMaterials/Pages/Stable-Isotopes.aspx and https://www-s.nist.gov/srmors/view_detail.cfm?srm=8544 for details) and two internal CO2 standards (CSI‐8628 and CSI‐040711). CSI‐8628 is a commercial CO2 tank (δ18O = 29.97‰), and CSI‐040711 is a very light sample produced by equilibrating CO2 gas with water from an Antarctic ice core (δ18O = −9.28‰). The δ13C and δ18O values of CSI‐8628 and CSI‐040711 were measured on CO2 gas (before fluorination) using a Thermo‐Fisher 253+ dual‐inlet isotope ratio mass spectrometer with a reference gas calibrated to IAEA standards NBS‐18, NBS‐19, and IAEA‐603. All data are presented in Table 1.

TABLE 1

Triple oxygen isotope data for CO2 standards

Sample	Analysis	Mole fraction CO2 (ppm)	δ17O	δ18O	δ′17O	δ′18O	∆′17O	δ17Ocorr a	δ18Ocorr a	δ′17Ocorr a	δ′18Ocorr a
CSI‐8628 1	Fluorination	–	15.119	29.077	15.006	28.662	−128	15.584	29.970	15.464	29.530
CSI‐8628 2	Fluorination	–	15.367	29.528	15.250	29.100	−115	15.597	29.970	15.477	29.530
CSI‐8628 3	Fluorination	–	15.366	29.537	15.249	29.109	−121	15.591	29.970	15.471	29.530
	Average	–	15.284	29.381	15.168	28.957	−121	15.591	–	15.470	–
	±1𝜎	–	0.143	0.263	0.141	0.256	7	0.005	–	0.005	–
CSI‐8628 1	TILDAS	451.620	15.618	30.017	15.497	29.575	−119	15.593	29.970	15.473	29.530
CSI‐8628 2	TILDAS	455.650	15.621	29.996	15.500	29.555	−105	15.607	29.970	15.487	29.530
CSI‐8628 3	TILDAS	456.900	15.556	29.876	15.436	29.438	−107	15.605	29.970	15.485	29.530
CSI‐8628 4	TILDAS	458.760	15.594	29.963	15.474	29.523	−114	15.598	29.970	15.477	29.530
	Average	455.733	15.597	29.963	15.477	29.523	−111	15.601	–	15.480	–
	±1𝜎	3.025	0.030	0.062	0.029	0.060	6	0.006	–	0.006	–
CSI‐040711 1	Fluorination	–	−4.189	−7.582	−4.198	−7.611	−179	−5.089	−9.280	−5.102	−9.323
CSI‐040711 2	Fluorination	–	−4.244	−7.689	−4.253	−7.719	−178	−5.088	−9.280	−5.101	−9.323
CSI‐040711 3	Fluorination	–	−4.877	−8.901	−4.889	−8.941	−168	−5.078	−9.280	−5.091	−9.323
	Average	–	−4.437	−8.057	−4.447	−8.090	−175	−5.085	–	−5.098	–
	±1𝜎	–	0.382	0.733	0.384	0.739	6	0.005	–	0.005	–
CSI‐040711 1	TILDAS	462.420	−4.763	−8.783	−4.774	−8.822	−116	−5.026	−9.280	−5.039	−9.323
CSI‐040711 2	TILDAS	460.600	−4.786	−8.844	−4.798	−8.884	−107	−5.017	−9.280	−5.030	−9.323
CSI‐040711 3	TILDAS	457.650	−4.872	−9.013	−4.884	−9.054	−103	−5.013	−9.280	−5.026	−9.323
CSI‐040711 4	TILDAS	458.490	−4.940	−9.091	−4.952	−9.132	−130	−5.040	−9.280	−5.053	−9.323
CSI‐040711 5	TILDAS	452.470	−4.817	−8.907	−4.829	−8.947	−105	−5.015	−9.280	−5.028	−9.323
	Average	458.326	−4.836	−8.928	−4.847	−8.968	−112	−5.022	–	−5.035	–
	±1𝜎	3.765	0.071	0.125	0.071	0.126	11	0.010	–	0.010	–
IAEA‐603 b	Fluorination		20.262	39.012			−147	20.262	39.012
IAEA‐603 1	TILDAS	461.79	20.3869	39.2682	20.1819	38.5168	−155	20.254	39.012	20.052	38.270
IAEA‐603 2	TILDAS	456.22	20.4034	39.2684	20.1980	38.5170	−139	20.270	39.012	20.068	38.270
IAEA‐603 3	TILDAS	458.58	20.3903	39.2794	20.1852	38.5276	−157	20.252	39.012	20.049	38.270
IAEA‐603 4	TILDAS	454.66	20.3571	39.192	20.1527	38.4435	−145	20.264	39.012	20.061	38.270
	Average	457.813	20.384	39.252	20.179	38.501	−149	20.260	–	20.057	–
	±1𝜎	3.103	0.020	0.040	0.019	0.039	9	0.008	–	0.007	–
NBS‐18 b	Fluorination		9.114	17.524			−100	9.072	17.524
NBS‐18 1	TILDAS	467.130	9.276	17.806	9.233	17.649	−86	9.128	17.524	9.087	17.372
NBS‐18 2	TILDAS	461.160	9.262	17.776	9.220	17.619	−83	9.131	17.524	9.089	17.372
NBS‐18 3	TILDAS	464.940	9.250	17.735	9.208	17.580	−74	9.140	17.524	9.098	17.372
NBS‐18 4	TILDAS	460.580	9.312	17.863	9.269	17.705	−79	9.135	17.524	9.093	17.372
	Average	463.453	9.275	17.795	9.232	17.638	−81	9.133	–	9.092	–
	±1𝜎	3.122	0.027	0.054	0.026	0.053	5	0.004	–	0.004	–
NBS‐19‐1 b	Fluorination		20.348	39.194			−155	20.348	39.194
NBS‐19 1	TILDAS	435.890	20.573	39.623	20.364	38.858	−153	20.350	39.194	20.146	38.445
NBS‐19 2	TILDAS	440.310	20.524	39.561	20.316	38.799	−170	20.333	39.194	20.129	38.445
NBS‐19 3	TILDAS	451.920	20.532	39.529	20.324	38.768	−145	20.358	39.194	20.154	38.445
NBS‐19 4	TILDAS	451.220	20.520	39.542	20.312	38.780	−164	20.339	39.194	20.135	38.445
NBS‐19 5	TILDAS	453.390	20.537	39.569	20.329	38.806	−161	20.342	39.194	20.138	38.445
	Average	446.546	20.537	39.565	20.329	38.802	−159	20.345	–	20.141	–
	±1𝜎	7.906	0.021	0.036	0.021	0.035	10	0.009	–	0.009	–

Notes: Results are provided for TILDAS and fluorination for CO2 sample gas, and CO2 extracted from carbonate standards using phosphoric acid digestion at 25°C. δ‐values are reported in percentage relative to VSMOW, and Δ′17O is reported in per meg relative to VSMOW. TILDAS data are presented without any further correction using Equation (3). All values are relative to VSMOW.

Abbreviations: TILDAS, tunable infrared laser direct absorption spectroscopy; VSMOW, Vienna Standard Mean Ocean Water.

Corrected using the calibrated δ18O value of the sample gas by phosphoric acid digestion and analysis of CO2 by conventional mass spectrometry and the Δ′17O value from fluorination (see Section 3.1).

Data from Wostbrock et al.

TILDAS sample inlet system

A schematic of the sample system design is shown in Figure 3. The inlet system is represented by two primary components, the mixing volume (outlines in blue), the TILDAS (outlined in red), and the accompanying valves that are controlled through TDL Wintel (outlined in orange). The entire system was pumped by an oil‐free scroll pump (Agilent IDP‐15 Dry Scroll Pump, Santa Clara, CA, USA) to prevent any backflow of oil vapors into the vacuum line. The purpose of the mixing volume was to produce a CO2 sample gas diluted in CO2‐free dry air to obtain a reproduceable mixing ratio that matches that of the reference gas. The reference gas used in this study was prepared by filling an evacuated high‐pressure gas tank with CO2 and then dry CO2‐free air to achieve a mixing ratio of 456 ppm CO2.

FIGURE 3

Schematic diagram of the sample gas inlet system for the TILDAS (tunable infrared laser direct absorption spectroscopy) spectrometer. The system consists of a mixing volume for sample preparation (blue box), a small volume between two pneumatic valves (Swagelok bellows valves) to acquire aliquots of sample gas from the mixing volume (green box), the automated valves actuated on a schedule by IRIS (orange box), and the absorption cell in the TILDAS instrument (red box) [Color figure can be viewed at wileyonlinelibrary.com]

For analysis in the TILDAS system, CO2 produced by phosphoric acid digestion from calcite standards NBS18, NBS19, and IAEA‐603 and collected in glass sample tubes was connected to the sample inlet port using a glass vacuum seal. For standards CSI‐8628 and CSI‐040711, 0.07 mL of CO2 was introduced into the line using a syringe and injected through a septum on the inlet port. The CO2 was transferred into the mixing volume by freezing on a cold finger cooled with liquid nitrogen. The liquid nitrogen was removed, and the CO2 was heated to room temperature and expanded into the mixing volume. Motor‐controlled bellows 1 (salvaged from an old VG Prism II mass spectrometer) adjusted the bellows volume until a preset pressure for CO2 was attained on pressure sensor 1 (Kurt J. Lesker 300 series convection vacuum gauge, Jefferson Hills, PA, USA). Zero air was then added to the mixing volume until the pressure on gauge 2 (Baratron 122AA‐01000AD) reached the set pressure (~1 bar) that produced a mixture of 456 ppm CO2. (The convection vacuum gauge has a higher precision in the milliTorr to Torr range, whereas the Baratron has a higher precision in the range of 500–1000 Torr.) CO2 and zero air were mixed for 20 min by alternately opening and closing bellows 1 and 2. Incomplete mixing was observed when mixing time was less than 10 min, as evidenced by a drift in the mixing ratio of the sample gas measured in the TILDAS between sample injections.

The TILDAS portion of the sampling system (Figure 3) includes the instrument and the automated valve process controlled through TDL Wintel and the IRIS interface developed by ARI. The operation procedure is outlined here. The optical cell and the sample aliquot volume were evacuated and flushed with nitrogen. The absorption cell was filled to 30 Torr with dry nitrogen, and a 15 s background measurement was recorded for subtraction from future measured spectra. The absorption cell was then evacuated, and the working reference was introduced through a 0.004 inch critical orifice (O'Keefe IC‐PC4‐4‐SS precision metal orifice, Monroe, CT, USA) using an electronic valve automated to hit a target pressure of 38 Torr. Once the reference gas (or sample gas) was introduced into the absorption cell, the system was allowed to stabilize for 18 s, and then data were collected for 12 s. While the reference gas was being measured, an aliquot of the sample gas was prepared. The mixing volume pressure was adjusted by closing bellows 2 to hit a target pressure on pressure gauge 2 that would correspond to the same working pressure as the reference gas when expanded into the optical cell (38 Torr). Once the reference gas measurement was completed, the absorption cell was evacuated and then flushed with nitrogen and evacuated a second time. The sample aliquot was then expanded into the optical cell. After the sample measurement, the absorption cell was evacuated and flushed with nitrogen, and the process was repeated for 12 cycles. After the introduction of each sample gas, its pressure in the manifold was adjusted to the correct pressure by compressing the bellows. The isotope values of each sample measurement were determined relative to the reference gas by interpolating the prior and subsequent reference gas measurements in time. The total time required for 12 reference‐sample cycles was 30 min.

RESULTS AND DISCUSSION

The Δ′17O values of CSI‐8628 and CSI‐040711 determined by fluorination are presented in Table 1. Both CSI‐8628 and CSI‐040711 were fluorinated thrice with an average Δ′17O of −121 and −111 per meg with standard deviations of ±7 and ±6 per meg, respectively. The δ17O and δ18O values from fluorination are generally slightly less than the correct value (due to incomplete fluorination of the CO2 gas), but it has been shown that δ17O and δ18O covary with a λ value of 0.528, such that the Δ′17O value is constant and independent of the measured δ18O value (see Wostbrock et al for further details). Using the Δ′17O obtained using fluorination and the δ18O measured from the CO2 gas, we can back‐calculate to determine the δ17O value of our CO2 standard gases. The fluorinated triple oxygen isotope values for IAEA‐603, NBS‐18, and NBS‐19 of CO2 produced by phosphoric acid digestion at 25°C referenced in this study are from Wostbrock et al (Table 1), which was measured on the same extraction line.

TILDAS results

Precision

The five standard gases calibrated using the VSMOW‐SLAP (Standard Light Antarctic Precipitation) scale (CSI‐8628, CSI‐040711, IAEA‐603, NBS‐18, and NBS‐19) were analyzed using TILDAS and inlet system described earlier. Each sample was analyzed four or five times, with an analysis consisting of 12 sample‐reference cycles and a data filter threshold of 2𝜎. Comparison of the δ18O‐Δ′17O values obtained using TILDAS and conventional methods (phosphoric acid digestion for δ18O and fluorination for Δ′17O) is shown in Figure 4. The averaged δ18O values reported using TILDAS agree with values calibrated by measuring CO2 gas on the IRMS (Table 1) to within ±0.03‰ for all standards, except for CSI‐040711, where the averaged δ18O for all runs differs from the IRMS measurement by ±0.35‰. The standard deviations of the δ17O and δ18O values for CSI‐040711 were ±0.07‰ and ±0.12‰ (standard error, 2σ), respectively. All other references analyzed using TILDAS had a standard deviation of ≤0.03‰ for δ17O and ≤0.06‰ for δ18O. Despite the variation in the δ17O and δ18O values for CSI‐040711, δ17O and δ18O covary (Figure 5), and the standard deviation for Δ′17O is 11 per meg over five analyses. The overall standard deviation of all analyses performed is ±8 per meg for Δ′17O.

FIGURE 4

Results of TILDAS (tunable infrared laser direct absorption spectroscopy) δ18O and Δ′17O measurements of CO2 compared to those obtained using dual‐inlet mass spectrometry. For dual inlet, the Δ′17O value was measured using conventional fluorination on O2, and the δ18O value was measured using conventional phosphoric acid digestion and analysis of CO2 gas. The Δ′17O values of the TILDAS data were corrected following Equation (4) [Color figure can be viewed at wileyonlinelibrary.com]

FIGURE 5

Covariance in δ17O and δ18O of CSI‐040711. There is significant variation in the δ17O and δ18O values for subsequent analyses of CSI‐040711, with a standard deviation of 0.07‰ and 0.12‰, respectively. δ17O and δ18O covary, and the standard deviation for Δ′17O is 11 per meg for five analyses, giving us confidence in the Δ′17O values. Fluorination values are δ18O = −9.28‰ and δ17O = −5.035‰ [Color figure can be viewed at wileyonlinelibrary.com]

Testing analytical conditions for precision and accuracy

There are a number of variables that could affect the precision of the TILDAS system. These include the following: (a) fractionation during introduction of sample gas, (b) differences in the working pressure of the reference and sample gases, (c) differences in the CO2 mole fraction of the reference and sample gases, and (d) extreme differences in the δ18O values of reference and sample. We tested the sensitivity of the system to these parameters by intentionally measuring samples where the sample and reference values were different.

Potential fractionation during introduction of sample gas

Potential fractionation associated with the inlet system and gas introduction process was assessed by filling the mixing volume with the working reference gas to act as the “sample gas” and running it against itself (a so‐called zero enrichment measurement). The average δ17O and δ18O values of four zero enrichment analyses differ from the calibrated working reference value by ±0.005‰ for δ17O and ±0.007‰ for δ18O with standard deviations similar to those reported for the sample gases analyzed in this study (≤0.024‰ for δ17O and ≤0.036‰ for δ18O), excluding CSI‐040711 (Table 2). The Δ′17O values have a precision of 9 per meg (standard error, 2σ). With the results of the zero enrichment and the four other standards analyzed having reproducible results, it does not appear that there is fractionation associated with the sampling system and analysis procedure.

TABLE 2

Sampling system zero enrichments

Sample	δ17O	δ18O	δ′17O	δ′ 18O	∆′17O
Zero 1	17.407	32.724	17.257	32.200	256
Zero 2	17.450	32.782	17.299	32.256	268
Zero 3	17.476	32.818	17.325	32.291	276
Zero 4	17.418	32.722	17.268	32.198	267
Zero 5	17.435	32.771	17.284	32.245	259
Zero 6	17.429	32.772	17.279	32.246	253
Average	17.436	32.765	17.285	32.239	263
±1𝜎	0.025	0.037	0.024	0.035	9
Assumed working reference Value	17.445	32.779	17.295	32.253	265

Notes: Results for zero enrichment analysis, where working reference gas was loaded in the mixing volume of the inlet system and analyzed as “sample” to test potential fractionation associated with sample introduction.

Total working pressure

The infrared‐absorption lines follow well‐established shapes whose mathematical fit is determined using a Voigt line shape profile. A limited range of working pressures for a given spectral region provides well‐defined peaks with sufficient absorption for all isotopologues within the spectral region. Very low working pressure results in insufficient signal and poorly defined absorption lines, whereas very high working pressure results in broadening the absorption peaks and interference between isotopologue spectral lines.

To validate a range of acceptable working pressures, we conducted zero enrichment analyses at working pressures of 30, 34, and 38 Torr for both the sample and reference gases. Working reference gas was analyzed against itself for the three different working pressures. The results for the zero enrichments at 30, 34, and 38 Torr are shown in Figure 6. The results for each working pressure are within the error of the Δ′17O value of the working reference and a standard deviation of <±2 per meg for the three analyses.

FIGURE 6

Effect of varying the working pressures in the sample cell. Results of zero enrichment analysis at working pressures of 30, 34, and 38 Torr. All analyses are within error of the accepted value of the working reference gas, Δ′17O = 265 per meg. Error bars are ±1𝜎 (8 per meg) [Color figure can be viewed at wileyonlinelibrary.com]

The consistency of results over a range of working pressures allows us to be flexible with sample size. All analyses were performed at 38 Torr, where an excess of the sample was reacted to allow for multiple analyses of the same gas. Reducing the working pressure from 38 to 30 Torr reduces the sample size required by approximately one‐third.

Working pressure mismatch

To determine the acceptable discrepancy between the sample and reference pressure, a series of zero enrichments were performed at mismatched pressures. The working pressure for the “sample gas” was adjusted for each subsequent aliquot over 12 measurement cycles, whereas the working reference pressure was held constant at 38 Torr. There is a strong linear relationship (R 2 = 0.935) between the difference in the measured Δ′17O from the accepted working reference value and the difference between the working pressure of the “sample” and the working reference (Figure 7).

FIGURE 7

Sample versus working reference pressure effect. Graphical results for zero enrichment analysis illustrating the importance of matching the working pressure of the sample and reference gases. Each dot represents one sample‐reference measurement (total 12 cycles). The working pressure of the sample was increased for each subsequent cycle, whereas the reference pressure remained constant. The y‐axis is the difference between the measured Δ′17O value and the accepted Δ′17O value of the working reference. To achieve a precision of 10 per meg for Δ′17O, the working pressure of the sample and reference pressure must match and be constant within ±20 mTorr, which is easily achieved with our instrument [Color figure can be viewed at wileyonlinelibrary.com]

The working pressure for the sample and reference gas during a measurement cycle must be within ±23 mTorr to achieve a precision of ±10 per meg (Figure 7). Using the bellows system in our sampling system, we can introduce the sample gas at a working pressure with high reproducibility (±10 mTorr). Generally, the range in working pressure for both sample and reference over 12 measurement cycles is less than 40 mTorr.

There are instances in which there are anomalous spikes or drops in the working reference pressure. These deviations in pressure result from fluctuations in the backing pressure of the working reference gas feeding through the critical orifice altering the flow rate, resulting in the valve being closed too early or late to hit the target pressure. Instances of spikes or drops in the working reference pressure are uncommon and, at most, occur for one reference measurement over a 12‐cycle sample measurement. The difference between the reference pressure and the target pressure introduces an isotopic shift such that the data points calculated using the anomalous reference measurement are excluded from data analysis.

Mole fraction

The TILDAS instrument must be operated with a near‐ambient dilution of CO2 in zero air (or dry N2). The precise dilution is not important as long as it is constant: the reference and sample gases must use the same dilution to avoid an artifactual dependence of the measured isotope ratios on the sample gas CO2 mole fraction. Such dependence arises not from the spectrometer per se but rather from the use of only a single reference gas for calibration. In general, TILDAS measurement error is partly multiplicative (i.e., gain error) and partly additive (i.e., zero‐offset error), and a single reference gas can be used only to correct for one kind of error or the other. The super ratio used here to express a sample gas isotope ratio relative to the reference gas effectively assumes that all measurement error is multiplicative. The goal is therefore to minimize the impact of additive error by subtracting spectral background measurements (as mentioned earlier) and by keeping the sample and reference mole fractions as close as possible. When the sample and reference gas mole fractions are the same, the additive errors cancel out. Conversely, as the sample and reference gas mole fractions diverge, the additive errors cause increasing bias.

A major challenge in designing the sample system was the ability to mix a sample gas of CO2 and dry air efficiently and reproducibly within several ppm CO2 of the reference gas. The target mole fraction is ~450 ppm CO2 in zero air (456 ppm in our system). To examine the effects of differences in the mole fraction between the sample and working reference gases, the mixing ratio of the sample gas was varied slightly above and below the target mole fraction. A mismatch of 20 ppm between the mixing ratio of reference and sample gases was found to have no apparent effect (Figure 8).

FIGURE 8

Sample versus working reference mole fraction. Dependance of Δ′17O based on the CO2 mole fraction of the sample and working reference matching. The y‐axis is the difference between the measured Δ′17O value of the analysis and the averaged Δ′17O for the standard from Table 1. The x‐axis is the difference between the mole fraction of CO2 in the sample gas (Table 1) and the mole fraction of CO2 in the reference gas (456 ppm). Our sampling system reproducibly produces a sample gas within ±2% of the target CO2 mole fraction, and differences in the CO2 mole fraction of ±2% do not measurably affect the Δ′17O value [Color figure can be viewed at wileyonlinelibrary.com]

Effect of different isotope values for sample and reference

To test the “linearity” of the system related to extreme differences in the δ18O values of the sample and reference on the measured Δ′17O value, the samples were specifically prepared to cover a wide δ18O range of 48‰. Figure 9 shows the offset in the Δ′17O values measured in the TILDAS unit versus those obtained by fluorination as a function of the δ18O value. There is a systematic difference between the Δ′17O values measured by both systems that vary linearly as a function of the difference between the δ18O value of the sample and reference. The larger the difference between the two δ18O values, the greater the Δ′17O value shifts from the true value. Over a range in δ18O values from −9‰ to +38‰, Δ′17O shifts by ~65 per meg. The linear relationship between the measured δ18O (per mil) and the offset in the Δ′17O (per meg) from the true value allows for using the equation (applicable to our instrument): The δ17O value of the sample gas can then be back‐calculated using the measured δ18O and the corrected Δ′17O values. Over a period of 6 months, we have not observed any significant changes to this correction formula.

FIGURE 9

Effect of Δ′17O on the δ18O value of the sample relative to working gas; offset in the Δ′17O value reported from the TILDAS (tunable infrared laser direct absorption spectroscopy) and the accepted value from fluorination. The y‐axis is the difference between the Δ′17O value from the TILDAS reference gas for each analysis and the accepted Δ′17O value from fluorination (Table 1). The lighter the δ18O value of a sample (the more different from the reference gas value of 32.77‰), the greater the Δ′17O shifts from the actual value. The linear relationship between the measured δ18O value and the shift in Δ′17O can be corrected using Equation (4) based on the linear best fit (red dashed curve) [Color figure can be viewed at wileyonlinelibrary.com]

The shift in Δ′17O over a large range of δ18O values is potentially a result of cross‐correlation in the spectral fit between the isotopologues in the spectrum. Theoretically, the dependance in the spectral fit between the isotopologues could be reduced or removed with improvements in the spectral fit and if the spectral lines for the isotopologues in the spectrum were completely independent from one another. But, due to subtle baseline curvature effects in the spectrum and spectral overlap between the isotopologues in the spectrum (Figure 1), the dependence in the spectral fit between the isotopologues cannot be determined, and calibration and correction of the Δ′17O values are necessary.

Additional considerations for the future involve the possibility of trace organic compounds in the CO2 sample gas due to phosphoric acid digestion of organic‐rich carbonates. The effect of this possible contaminant has not yet been evaluated.

CONCLUSION

We have demonstrated that the TILDAS (ARI) can measure the triple oxygen isotope composition of CO2 with a precision of ±10 per meg for Δ′17O and ±0.03‰ for δ18O values. The system is accurate as long as the sample and reference gases have similar mixing ratios and δ18O values. The effect of different δ18O values between reference and sample is linear and small (~1 per meg Δ′17O per 1 per mil δ18O) and easily corrected for.

With our sampling system configuration for the TILDAS instrument, sample size is reduced to a tenth of that needed to perform traditional fluorination reactions for triple oxygen isotope analyses. A 70 μL injection of CO2 needed for a 12‐measurement cycle analysis is equivalent to 300 μg of pure carbonate reacted with phosphoric acid. The system can easily be modified to reduce the requirement to 200 μg carbonate equivalent.

The TILDAS instrument's ability to measure triple oxygen isotopes directly from CO2 using laser spectroscopy removes the need to convert CO2 to O2 using methods such as fluorination to remove the isobaric interference at mass 45. Sample size (300 μg), analysis time (~30 min), and cost are significantly less using TILDAS, whereas precision for Δ′17O is ≤10 per meg, comparable to fluorination.

Operating parameters such as cell pressure and CO2‐N2 mixing ratio must be controlled to a high tolerance. We have developed an inlet system that allows for sample and reference gas cell pressures to be the same within 0.02 Torr and CO2 mixing ratios within 2% of each other. Differences of ±2% between the sample and working reference CO2 mixing ratios do not appreciably affect Δ′17O, but further work should be done to test this effect when mixing ratios are larger than this. The system can be fully automated so that a large number of analyses can be made with minimal effort.

PEER REVIEW

The peer review history for this article is available at https://publons.com/publon/10.1002/rcm.9391.