Application of glow discharge mass spectrometry to multielement ultra-trace determination in ultrahigh-purity copper and iron: a calibration approach achieving quantification and traceability

A new approach was developed for quantitative calibration in GD-MS which can afford reliable and metrologically traceable results for many trace elements and was exemplified for pure copper and pure iron. It can be assumed that the technique can be further improved and applied to the analysis of other pure metals. Pressed copper and iron powder samples were used to calibrate the glow discharge mass spectrometry applied to the analysis of pure copper and iron. The new type of glow discharge mass spectrometer-the Element GD (Thermo Electron Corporation)-was used with a Grimm-type discharge cell for flat samples. Two series of powder samples were prepared for each of the copper and iron matrixes. The powders were quantitatively doped with solutions of graduated and defined concentrations of 40 or 20 analytes, respectively. The mass fractions of the analytes in the dried and homogenized metal powder samples ranged from mu g/kg levels up to 10 mg/kg levels. A special technique was developed to press the samples and to form mechanically stable pellets with low risk of contamination. Ion beam ratios of analyte ions to matrix ions were used as measurands. The calibration curves were determined and the linear correlation coefficients were calculated for different intervals of the curves. The linear correlation coefficients are very satisfactory for most of the calibration curves, which include the higher segments of mass fractions; however, they are less satisfactory for the lower segments of the calibration curves. Nevertheless, in many cases rather acceptable and rather promising values were achieved even for these lower segments, representing mass fractions of analytes at ultra-trace level. The comparison of the certified values of different reference materials with the measured values based on calibrations with the pressed powder samples led to deviations less than 30% for most of the considered examples.