Title:	Homological determination of phonon mixtures
DNr:	Berzelius-2026-165
Project Type:	LiU Berzelius
Principal Investigator:	Wojciech Chachólski <wojtek@kth.se>
Affiliation:	Kungliga Tekniska högskolan
Duration:	2026-05-30 – 2026-12-01
Classification:	10199
Homepage:	https://www.kth.se/math/act/tda
Keywords:

Abstract

This project aims to develop a machine learning model for determining phonon mixtures in scattering data based on homological features. To this end, we have obtained simulated data (density functional theory, DFT) of the possible phonon modes in BiTe crystals. The training data consists of combinations of these modes plus noise. We have seen some initial promising results in the noise-free regime, running on smaller inputs. Our goal now is to scale up the training data in order to achieve a model that generalizes well across the space of phonon mixture configurations. The DFT data are from Oscar Grånäs's group in Uppsala, and we have access to experimental data via Jonas Weissenreider's group at KTH. Our first goal is to obtain a model on the simulated data alone. This model will then be tested on experimental data. We are in the process of preparing a bigger funding proposal for a potential multi-year project together with the aforementioned research groups and our group in Mathematics. For this reason, we are asking for expedited access to the default allocation on Berzelius-Ampere as a starting point, with the hope that we can get access to a larger allocation as part of the regular process.