%0 Journal Article %A Jiaojian Song %A George C.-Y Chan %A Xianglei Mao %A Jonathan D Woodward %A Robert W Smithwick %A T. Gregory Schaaff %A Ashley C Stowe %A Candace D Harris %A Ronger Zheng %A Vassilia Zorba %A Richard E Russo %B Spectrochimica Acta Part B: Atomic Spectroscopy %D 2018 %G eng %P 67 - 76 %R 10.1016/j.sab.2018.10.008 %T Multivariate nonlinear spectral fitting for uranium isotopic analysis with laser-induced breakdown spectroscopy %V 150 %8 12/2018 %! Spectrochimica Acta Part B: Atomic Spectroscopy %X

The analytical characteristics of U isotopic analysis with laser-induced breakdown spectroscopy (LIBS) were investigated by analyzing a series of U3O8–Li2B4O7 fused glassy disks. The 235U isotopic contents in the samples ranged from natural abundance at ~0.72 atom-% to a highly enriched level at ~94 atom-%. Isotopic information of the sample was extracted from the LIBS spectra through multivariate nonlinear spectral fitting, which is theoretical in basis and does not require calibration with isotopically enriched standards. A database consisting of twelve U-line pairs, which can be used for multiple-line-pair spectral fitting, was constructed for the spectral region from 424.300 to 424.500 nm. The use of twelve line pairs in the fitting was found to offer superior performance than the use of only a single line pair (424.412 nm for 235U and 424.437 nm for 238U). The analytical accuracies (biases) in absolute [235U / (235U + 238U)] ratios were about ±4% for the single-line-pair fitting, and improved to around ±1% for the multiple-line-pair fitting. In both cases, precisions, expressed as standard deviations from ten repetitive measurements, each with an accumulation of LIBS signals from 100 laser shots, were found to be similar and within 0.2% in absolute [235U / (235U + 238U)] ratios. Although hyperfine structure and Stark broadening can be incorporated into the algorithm as fitting parameters, the inclusion of these two effects caused overfitting, which degraded analytical accuracies. The best multiple-line-pair fitting algorithm was found to be the one with the least fitting parameters [i.e., one global linewidth (namely, neglecting Stark broadening) and without inclusion of hyperfine structure], which consequently mandates a detector delay time sufficiently long so that almost all the Stark effect subsides.