| 研究実績の概要 |
Patchy particles are the particles of nanometer scale, have anisotropic interactions with other particles, and they can form complex structures such as dodecagonal quasicrystal (DDQC). However, it is difficult to control the DDQC structures at steady states because the growth of the DDQC involves several kinetic pathways.
We propose reinforcement learning to learn how to control the dynamical self-assembly of DDQC from patchy particles. In detail, we estimate the best policy of temperature control trained by the Q-learning method and demonstrate that we can generate dodecagonal quasicrystal structures using the estimated policy. The temperature schedule obtained by reinforcement learning can reproduce the desired structure more efficiently than the conventional pre-fixed temperature schedule, such as annealing.
|
| 今後の研究の推進方策 |
The patchy particles are capable of organizing themselves into complex structures, which are important for the development of innovative materials. One of the complex structures is quasicrystal. The quasicrystal is interesting in both theoretical and application aspects. The quasicrystal can be applied in various application such as advanced coatings, reinforced composites, magnetism. We have applied reinforcement learning to identify the best parameter control for the growth of dodecagonal quasicrystals. However, aside from the structural configuration, the functions or properties of the obtained structure should be controlled as well because of the practical application. This kind of problem also occurs for other assembled structures.
In the future plan, we aim to apply reinforcement learning to control the properties of the assembled structure, thereby investigating the underlying relation between the structure and property. We also expect that by changing the external driving force, we can further control the functions or properties of these structures.
|
