Searching for optimal detector materials with new machine learning techniques

Dark matter candidates that are lighter than a proton comprise a major challenge when it comes to their experimental detection. The problem is that their recoil energy is well below the sensitivity threshold of currently world leading experiments, such as XENONnT, which have amazing sensitivity to dark matter heavier than a GeV, but are essentially blind to lighter dark matter.

In this research project, we are aiming a identifying promising organic molecules that could serve as new detection materials for low mass dark matter. Such molecules are promising, as they naturally have energy excitations in the eV energy range, which is the expected recoil energy for sub-GeV dark matter candidates. As the variety of organic molecules is so vast, machine learning algorithms will be employed to create an analysis pipeline, which can be trained on large existing datasets of organic molecules, and used to identify the desired properties on new unknown molecules.

One strong feature, which will be employed is the generative power of so called variational autoencoders to create suggestions for yet unseen molecular candidates. This identified molecules will be analysed by deterministic methods, such as density functional theory and time dependent density functional theory. The most promising candidates will be tested in our partner laboratories at the University of Chicago.

The final goal is creating a prototype detector which will be based on an inexpensive newly identified organic material and thus can be easily scaled up in size in order to improve sensitivity to light dark matter candidates in the future.