Efficient NMR measurement and data analysis supported by the Bayesian inference : The case of the heavy fermion compound YbCo2Zn20

Abstract

We propose a data-driven technique to infer microscopic physical quantities from nuclear magnetic resonance(NMR) spectra, in which the data size and quality required for the Bayesian inference are investigated. The $^{59}$Co-NMR measurement of YbCo$_2$Zn$_{20}$ single crystal generates complex spectra with 28 peaks. By exploiting the site symmetry in the crystal structure, the isotropic Knight shift $K_{iso}$ and nuclear quadrupole resonance(NQR) frequency $\nu_Q$ were estimated to be $K_{iso} = 0.7822 \pm 0.0090 \%$, $\nu_Q = 2.008 \pm 0.016$ MHz ( T = 20 K, H $\simeq$ 10.2 T) by analyzing only 30 data points from one spectrum. The estimation of $\nu_Q$ is consistent with the precise value obtained in the NQR experiment. Our method can significantly reduce the measurement time and the computational cost of data analysis in NMR experiments.