This article introduces the recent changes of machine learning and the unchangeable principles of machine learning. First, after showing the definition of machine learning (ML), we describe the research area of ML. We overview these sub-areas from the viewpoints of validity, effectiveness, and efficiency. Then, as a changing part of ML, we briefly show the recent progress of ML. And, we describe three basic principles of ML. Finally, we show the difficulty of using ML techniques, and examples of application of the ML techniques to the natural science area.
We succeeded in observing the electron distribution of a Frontier orbital in a molecular crystal using synchrotron radiation single crystal X-ray diffraction data obtained at SPring-8. This method of extracting only the valence electron information by subtracting the core electron distribution from the total electron density was named the core differential Fourier synthesis (CDFS) method. In order to propose such a new method, it is indispensable to verify whether the electronic state of a well-known material investigated can be explained accurately. Therefore, we selected a typical quasi one-dimensional molecular conductor (TMTTF)2PF6 and examined the accuracy of this method. The first report of this molecular system was in 1978. The charge disproportionation in the charge-ordered phase suggested by the dielectric constant measurement could not observe direct structural evidence and was called “structure-less transition” as a mystery for 40 years. We clarified the electronic cloud of the HOMO of TMTTF molecule by CDFS method and elucidated the details of charge order. The CDFS method can be applied not only to molecular crystals but also to a variety of materials such as transition metal oxides. Here we introduce the principle, experimental conditions necessary for measurement and analysis methods.
U-based ferromagnetic superconductors of UCoGe and URhGe are reviewed. In these superconductors, there are various experimental evidences that superconductivity is induced by the longitudinal ferromagnetic fluctuations and that the spin triplet pairing is formed. We show these results and the theoretical calculations to explain the results based on the spin-triplet superconductivity mediated by ferromagnetic fluctuations.
This article explains recent explorations of M-theory, which is expected to be the theory of everything. We especially focus on the developments in the matrix model derived from the worldvolume theory of M2-branes. It is found that the parameter space characterizing the M-theory satisfies the symmetry of integrability and the exceptional group.
We present the time-dependent multiconfiguration self-consistent-field (TD-MCSCF) methods to simulate, from the first principles, multielectron dynamics in atoms and molecules subject to intense laser fields. Our TD-MCSCF methods introduce flexible orbital subspace decomposition, conforming to phenomena under investigation and desired accuracy. Infinite-range exterior complex scaling in addition to mask-function boundary is adopted as an efficient absorbing boundary. We show illustrative numerical examples.