The statistical physics analysis of learning parametric models was revisited by combining the replica method with a grand canonical ensemble and variational approach, enabling prediction error estimation for learning systems with real data.

This study developed a first-principles tight-binding framework that systematically evaluates higher-order spin interactions to enable the quantitative analysis and design of complex magnetic properties in materials.

In high-field ESR measurements, the quantum phase transition in the S = 1/2 one-dimensional Ising-like antiferromagnet, driven by transverse magnetic fields, is strongly related to the softening of the spinon excitation.

Topochemical exchange converts Li₂CoTi₃O₈ into metastable Co₂Ti₃O₈ with a diamond network, showing strong J₁–J₂ frustration, T_N near 4.4 K, and four magnetization steps up to 50 T.