# Thermodynamic and optical properties of metallic nanostructures
Metallic nanostructures are of interest for many areas, including, e.g., catalysis as well as energy conversion and storage. Studies of the thermodynamic and optical properties of these structures are a core topic in our research group. The main computational tools in this context are density functional theory (DFT) and time-dependent density functional theory (TDDFT) calculations along with machine-learned potentials in connection with molecular dynamics and Monte Carlo simulations. We are have also integrated electrodynamic simulations using the finite-difference time-domain (FDTD) approach in our workflows.
In the coming years we will continue our research in this area targeting in particular multi-component ("nanoalloy") systems and strong light-matter coupling. Here, we will study, e.g., the plasmon induced creation of "hot carriers" in nanoalloys and their transfer to adsorbates (KAW project grant), the manipulation of chemical reactions through strong coupling ("cavity chemistry"; Marie Skłodowska Curie grant) or the shape evolution of nanoalloys as a function of the environment (EI Nano Excellence PhD grant).
# Transport properties
Understanding and manipulating the thermal and electrical transport properties of materials is of interest to a large number of applications, including electronic and optoelectronic devices, thermal management, thermoelectric energy generation and materials for quantum computing.
In this research are, we use both first-principles calculations based on DFT and machine-learned potentials as well as molecular dynamics simulations. To deal with the large number of degrees of freedom and the chemical complexity of these systems we are continuously developing our methods and software packages.
In the coming years we will continue our research on thermal transport on one-dimensionally disordered layered materials building on our recent work [Nature 597, 660 (2021)]. In addition we will expand our work toward hybrid perovskites, which combine inorganic and organic features. To this end, we will continue and deepen our activities in the development of machine learned models for the description of not only the potential energy surface but additional scalar and tensorial properties such as dipole moment or optical spectra. In addition to our custom CUDA implementations we will use packages such as pytorch.
## Hydrogen sensing
To dramatically reduce greenhouse gas emissions, large investments in H2 (energy) technologies are imminent. In this con- text, the availability of robust and fast H2 safety sensors is a key enabling technology. However, no currently existing H2 sensor technology meets all the necessary performance targets set. To overcome this challenge, we are developing machine learning models to radically improve the performance of plasmonic optical H2 sensors. Specifically, we are building neural network models based on, e.g., LSTM, GRU, and transformer architectures, which we implement via packages such as keras and tensorflow. Our preliminary results show that this approach does indeed enable a leap forward with respect to key features such as response time and stability toward environmental perturbations. In the coming years, we will extend and improve this approach in close collaboration with our experimental partners.