Assessment of the accuracy of the 17 O correction algorithm used in δ 13 C determinations by CO 2 mass spectrometry

High-accuracy δ C values are required for observations of greenhouse gases CO and methane, and, consequently, for international reference materials (RMs). Recently, another application, clumped isotope geothermometry of natural carbonates, has demonstrated the requirement for high-accuracy δ C values. δ C determinations by mass spectrometry use an O isobaric correction on m/z 45, where O abundance is calculated from the measured O with the O- O relationship assumed with λ = 0.528. This relationship is the key assumption of the algorithm proposed in 2003 and accepted by IUPAC in 2010. However, to date, this relationship and potential δ C biases have not been verified using O measurements. To verify the O correction and to estimate potential δ C biases, we compile measured O data for carbonate RMs, for a range of natural carbonates that are typically analyzed in clumped isotope geothermometry and for CO in isotope equilibrium with natural waters including plants and biota. δ C biases are calculated based on O deviation from the O- O relationship assumed in the O correction. To estimate δ C biases accurately, the VPDB-CO framework for expressing O excess is defined and linked to the δ C scale definition. δ C biases estimated for carbonate RMs are found within ±0.004‰; the biases estimated for natural carbonates and CO in equilibrium with natural waters are mostly within ±0.010‰ (bidirectional distribution around zero). In all cases, the estimated biases are found within the best instrumental uncertainty of modern isotope ratio mass spectrometry (IRMS) (around ±0.014‰, k = 2). For the first time, high accuracy of δ C data obtained by CO mass spectrometry using the O correction with fixed λ = 0.528 has been demonstrated using measured O data. δ C biases estimated are within the best IRMS precision (±0.014‰, k = 2) and can be neglected in most practical applications. To obtain high-quality δ C data, it is strictly necessary that all data are treated on the VPDB-CO scale.