Fig. 8 . Observed δ18O values vs. previously published ones.
For the samples used the FO6 recipe, the oxygen yields of NBS28 were 98.2 – 99.3% and that of SWy-1 were 90.0 – 91.6%. NBS28 showed higher oxygen yields, which may be due to the longer CO peak with slight tailing rather than a higher reactivity than that of SWy-1. For the samples used the NiC_FO2 recipe, the oxygen yields of both NBS28 and SWy-1 are approximately 110%. These anomalous yields cannot be explained by the oxygen released from residues of standard samples. Thus, an additional source of oxygen was suspected (e.g., the oxidation of Ni metal in Ni/C during storage).
The samples with NaF at F/O = 2 in Run2 (Fig.3) showed a large deviation of δ18O values (16 – 22 ‰), while the FO6 and the NiC_FO2 recipes in Run4 resulted in the smaller deviation (0.29 and 0.23 ‰, respectively). SWy-1 tended to have a higher precision than NBS28 in both recipes. This is probably due to the smaller particle size of SWy-1 (< 2 μm) than NBS28. In addition, the NBS28_FO6 samples were analyzed in the first sequence among the other NaF-bearing samples in the series of measurement runs (Fig. 6), which could be related to the lower precision. By using a continuous-flow TC/EA-IRMS method and KF as the fluorine source, 0.33 ‰ of 1σ for NBS28 and 0.29 ‰ for NBS27 biotite were reported 11. The precision in this study was comparable to the above report, while the precisions of the conventional methods (dual-inlet system, BrF5 or CO2 laser fluorination) were at most 0.2‰25.
Although it was reported that δ18O values for NBS28 varied from 8.8 to 10.0‰ 25, 9.6‰ has been widely accepted. This study showed a discrepancy of +1.2‰ for the FO6 recipe and +0.4‰ for the NiC_FO2 recipe. The similar TC/EA-IRMS method reported 10.23‰ for NBS28 11, showing a similar trend to this study. For SWy-1, the reported values were very limited compared with those of NBS28. A laser fluorination method provided 18.6 ± 0.23‰4, while a conventional BrF5fluorination method showed 19.2‰ 26. Compared to these, the present study obtained lighter values; 17.8‰ for the FO6 recipe and 16.7‰ for the NiC_FO2 recipe.

3.5 Potentials and further improvements on the methodology

Although this study used at least 100 mg of each sample and additive in the process of homogenization in an agate mortar before encapsulation, one measurement is possible with about 200 µg of sample in a silver capsule. It is important to design pretreatment procedures for homogenizing tiny volumes of material. The measurement time was approximately 20–30 min for one silicate sample and 10 min for one non-silicate standard. Therefore, it has the potential to measure more than 50 unknown samples within 24 h. This is a major advantage of our proposed method. The conventional BrF5 fluorination method requires several milligrams of the sample and approximately 12 h for fluorination.
We also confirmed that hydrogen isotope ratios could be measured using the same instrument and configuration (data not shown). However, hydrogen isotope ratio measurement requires approximately 20 μg of hydrogen, and in the case of smectite or mica, it is equivalent to approximately 4–5 mg of the sample to be loaded. Therefore, the simultaneous measurement of hydrogen and oxygen isotope ratios for one capsule is difficult; however, by preparing samples for hydrogen and oxygen analyses, respectively, it is possible to measure both in a series of runs.
Despite the improvement in the oxygen yields from silicates in this study, problems persist. First, at the beginning of the series of samples containing fluorine, excessive amounts of oxygen yield occurred. This is probably due to the recapture of oxygen derived from standard samples’ residue that were pyrolyzed in the preceding sequence. Because the CO peak is recorded only once per sample in the continuous-flow IRMS system, the entire sample may be wasted without eliminating the effect of the preceding sample residue. Currently, a stable analysis can only be performed by checking the pyrolysis chromatograph and arbitrarily rejecting samples with abnormal behavior. Therefore, it is necessary to optimize the loading method to prevent the waste of samples. Next, the accuracy of the δ18O value needs to be examined with a wider variety of silicate minerals. However, there are few international standards for silicate minerals, which may hinder this verification. Inter-laboratory comparisons using multiple methods for silicate minerals other than NBS28, which are the most frequently reported, are needed. Because the main focus of this study was to find a promising recipe for separating oxygen from silicates, normalization based on international water standards such as VSMOW2 and SLAP2 were not achieved. To eliminate instrumental bias, VSMOW-SLAP scaling should be addressed in future studies. Third, optimization of a recipe with as little fluoride additive as possible may be required. In this study, the higher the F/O ratio, the better the reactivity with silicate. Although the IRMS detector did not show any fluoride contamination, the negative effects of the fluoride gas produced on the instrument cannot be completely ruled out. Regardless of whether it is in a continuous-flow system or a dual-inlet system, the optimization of the recipe will be performed in the future. The recipe produced in this study is anticipated to exhibit a comparable outcome for TC/EA-IRMS in the dual-inlet system.