Sparse modeling approach for quasiclassical theory of superconductivity

Nagai, Yuki   ; Shinaoka, Hiroshi*

Sparse modeling technique, one of the machine learning techniques, is now one of the very important techniques in materials science and solid state physics. In this talk, I will report on the application of sparse modeling to the theory for superconductors and achieve a speed-up of nearly 100 times faster than the conventional method. In the conventional theory, it is necessary to cut off the infinite series sum required for self-consistent simulations to some extent, but by using sparse modeling, it is shown that the infinite series sum can be calculated with practically tens of pieces of information by taking advantage of the sparsity of the information. As a result, the computation time was dramatically reduced.