When numerically estimating observables on a large-scale system with a complex-valued action, one needs an exponentially long computational time for precise estimation. After reviewing various approaches to this “sign problem”, we explain “the tempered Lefschetz thimble method”, which was introduced by MF and N. Umeda in [arXiv:1703.00861]. We argue that this has a potential to be a universal solution to the sign problem, by explicitly showing that this algorithm gives correct estimates for problems that have been difficult by other algorithms.