Description
Neural networks are a promising tool to approximate wave functions and solutions of partial differential equations in general. They were shown to be free from the curse of dimensionality in some cases. However, methods employing neural networks are far from being reliable because of the lack of convergence guarantees. Normalising flows are a promising tool where accurate numerical results as well as convergence guarantees can be obtained.
In this project, the student will learn about the general framework of approximating solutions to the Schrödinger equation via neural networks. The student will implement autoregressive normalising flows to compute molecular vibrational wave functions of small molecules (e.g., water or ammonia) and will analyse the developed numerical scheme.
Special Qualifications:
Background in machine learning, programming with Python and differential equations.
Field | A6: Theory and computing |
---|---|
DESY Place | Hamburg |
DESY Division | FS |
DESY Group | CFEL-CMI |